You are on page 1of 252

Frontiers in Mathematics

Advisory Editorial Board

Luigi Ambrosio (Scuola Normale Superiore, Pisa)


Leonid Bunimovich (Georgia Institute of Technology, Atlanta)
Benot Perthame (Ecole Normale Suprieure, Paris)
Gennady Samorodnitsky (Cornell University, Rhodes Hall)
Igor Shparlinski (Macquarie University, New South Wales)
Wolfgang Sprssig (TU Bergakademie Freiberg)
Leonid A. Kurdachenko
Javier Otal
Igor Ya. Subbotin

Artinian
Modules
over
Group
Rings

Birkhuser Verlag
Basel . Boston . Berlin
Authors:

Leonid A. Kurdachenko Igor Ya. Subbotin


Department of Algebra Department of Mathematics and Natural Sciences
School of Mathematics and Mechanics National University
National University of Dnepropetrovsk Los Angeles Campus
Vul. Naukova 13 3DFLF&RQFRXUVH'ULYH
Dnepropetrovsk 49050 /RV$QJHOHV&$
Ukraine USA
e-mail: lkurdachenko@hotmail.com
e-mail: isubboti@nu.edu
Javier Otal
Departamento de Matemticas
Universidad de Zaragoza
Pedro Cerbuna 12
50009 Zaragoza
Spain
e-mail: otal@unizar.es

0DWKHPDWLFDO6XEMHFW&ODVVLFDWLRQ33(()))
))))

/LEUDU\RI&RQJUHVV&RQWURO1XPEHU

Bibliographic information published by Die Deutsche Bibliothek


'LH'HXWVFKH%LEOLRWKHNOLVWVWKLVSXEOLFDWLRQLQWKH'HXWVFKH1DWLRQDOELEOLRJUDH
detailed bibliographic data is available in the Internet at <http://dnb.ddb.de>.

,6%1%LUNKlXVHU9HUODJ%DVHO%RVWRQ%HUOLQ

7KLVZRUNLVVXEMHFWWRFRS\ULJKW$OOULJKWVDUHUHVHUYHGZKHWKHUWKHZKROHRUSDUWRIWKHPDWHULDOLV
FRQFHUQHGVSHFLFDOO\WKHULJKWVRIWUDQVODWLRQUHSULQWLQJUHXVHRILOOXVWUDWLRQVUHFLWDWLRQEURDG-
FDVWLQJUHSURGXFWLRQRQPLFUROPVRULQRWKHUZD\VDQGVWRUDJHLQGDWDEDQNV)RUDQ\NLQGRIXVH
permission of the copyright owner must be obtained.

%LUNKlXVHU9HUODJ32%R[&+%DVHO6ZLW]HUODQG
Part of Springer Science+Business Media
&RYHUGHVLJQ%LUJLW%ORKPDQQ=ULFK6ZLW]HUODQG
3ULQWHGRQDFLGIUHHSDSHUSURGXFHGIURPFKORULQHIUHHSXOS7&)f
Printed in Germany
,6%1; H,6%1
,6%1H,6%1

  www.birkhauser.ch


To Tamara, Marisol, and Milla
Contents

Preface ix

1 Modules with chain conditions 1

2 Ranks of groups 13

3 Some generalized nilpotent groups 25

4 Artinian modules and the socle 37

5 Reduction to subgroups of nite index 45

6 Modules over Dedekind domains 53

7 The KovacsNewman theorem 63

8 Hartleys classes of modules 81

9 The injectivity of some simple modules 93

10 Direct decompositions in artinian modules 111

11 On the countability of artinian modules over F C-hypercentral groups 131

12 Artinian modules over periodic abelian groups 153

13 Nearly injective modules 161

14 Artinian modules over abelian groups of nite section rank 167

15 The injective envelopes of simple modules over group rings 179


viii Contents

16 Quasinite modules 189

17 Some applications: splitting over the locally nilpotent residual 211

Bibliography 227

Index 245
Preface
Let G be a group and suppose that G has an abelian normal subgroup A. If
H = G/A, then H acts on A by ah = ag , where h = gA H and a A, and this
action transforms A into a ZH-module (see all details below). If A is periodic, then
very often we may replace A by one of its primary pcomponents. This allows us
to assume that A is a p-subgroup, where p is a prime. This way we arrive at a
p-module over the ring ZH. In this case, the structure of the lower layer
P1 = 1 (A) = {a A | pa = 0}
of A has a signicant inuence on the structure of A. Since P1 is an elementary
abelian p-subgroup, we may think of P1 as a module over the ring Fp H, where Fp
is a prime eld of order p.
The above approach allows one to employ module and ring-theoretical meth-
ods for the characterization of the groups considered. This relatively old idea has
shown itself to be very eective in the theory of nite groups. Progress in the study
of nite groups naturally led to the implementation of this approach in innite
groups that are closely related to nite groups, specically, in innite groups with
some niteness conditions. It is well known that in the theory of rings many signif-
icant results are related to niteness conditions, especially the classical conditions
of minimality and maximality. Thus, both artinian and noetherian rings are the
main subjects of the largest and the richest branches of the theories of commuta-
tive and non-commutative rings. The minimality and the maximality conditions
were introduced in groups side by side and played a crucial role in the development
of the theory of innite groups. The study of groups with the maximal condition
on all subgroups (the Max condition) led to the fundamental theory of polycyclic
groups and applications to other areas (see, [255]). On the other hand, the explo-
ration of groups with the minimal condition on all subgroups (the Min condition)
was extremely fruitful and generated among many other the research associated
with the well-known problems stated by O.Yu. Schmidt and S.N. Chernikov (see
[43, 311]).
After very detailed investigations, many group theorists conjectured that
groups with the Max condition should be polycyclic-by-nite and the groups with
the Min condition should be Chernikov. Refuting this, A.Yu. Olshanskii surpris-
ingly developed a series of his famous monsters ([208, Chapter 28]), emphasizing
the inexhaustible wealth of Innite Group Theory.
These examples also solved other famous problems in Group Theory. Other
innite, nitely generated, residually nite, periodic groups having many unusual
properties have been constructed by R. I. Grigorchuk (see, for example the survey
[93]). Observe that for his rst example [92], R.I. Grigorchuk produced a clear
and relatively simple construction which compares favorably with examples that
existed before. These brilliant creations of A.Yu. Olshanskii and R.I. Grigorchuk
show that methods and approaches that are traditional for the theory of general-
ized soluble groups do not work eectively beyond this theory. They developed a
x Preface

clear understanding that the theory of generalized soluble groups is just a proper
specic part of the general group theory. Along with other branches such as nite
groups, abelian groups, and linear groups, the theory of generalized soluble groups
has its own specic subject of research, rich history and sophisticated methodology.
Together with the ordinary maximal and minimal conditions on all sub-
groups, the maximal and minimal conditions on normal subgroups (Max-n and
Min-n) began to be studied. It turns out that in locally nilpotent groups these
latter conditions coincided with the respective ordinary conditions. However, even
soluble groups with Max-n and Min-n required new approaches. The classical pa-
pers of P. Hall [95, 96, 98] played a major role in the implementation of both
module and ring-theoretical methods in the study of soluble groups. The investi-
gation of abelian-by-nilpotent groups that satisfy the maximal condition on normal
subgroups led P. Hall to the consideration of noetherian modules over a ring of the
form ZH, where H is a nitely generated nilpotent group. The basis connection
here between groups, rings and modules is established by the remarkable theorem
due to P. Hall: If R is a noetherian ring and G is a polycyclic-by-nite group, then
the group ring RG is likewise noetherian. This result stimulated further develop-
ment of the theory of group rings of polycyclic- by-nite groups as well as the
theory of modules over polycyclic-by-nite groups (see, for example, C.J. Brookes
[28], K.A. Brown [30], S. Donkin [57, 58], D.R. Farkas [68], K.W. Gruenberg [94],
A.V. Jategaonkar [118, 120] , I.N. Musson [195, 196], D. Passman [216, 217, 218],
J.E. Roseblade [248, 249, 250, 251, 252], R.L. Snider [267, 268], and many others).
Exploration of the dual condition that is, investigation of groups with the
minimal condition on normal subgroups began signicantly later. The rst in-
vestigations showed that there exist metabelian groups satisfying the minimal con-
dition on normal subgroups that are not Chernikov. Investigation of such groups
determined the necessity of the study of artinian modules over rings of type ZH,
where H is an abelian Chernikov group. The excellent paper by B. Hartley and D.
McDougall [110] contains the description of such modules and groups. This paper
was also a starting point for the investigation of artinian modules over group rings.
It is worth noting here that the situation with artinian modules is rather dier-
ent. Actually, the group ring of a group that is not polycyclic-by-nite loses the
valuable property of being noetherian. Therefore, we have no well-developed deep
theory of the respective group rings, and we need to build one. At rst glance, the
criteria of complementability of modules, which arise from the classical theorems
of Maschke and Fitting, seem as likely candidates for this. A result of L.G. Kovacs
and M.F. Newman [137] is one of the rst important generalizations of Maschkes
Theorem on innite groups. Since then, the following steps of the theory were cor-
related with the criteria of semisimplicity of artinian modules, which are related
to conditions of injectivity of simple modules. Investigation of the questions men-
tioned above developed approaches to the description of some artinian modules
over a ring of the form F G, where F is a eld. Note that the transition from a
scalar eld to a ring signicantly complicates the problem because there are not
Preface xi

too many situations in which one can get a good description of artinian modules
over group rings. Sometimes it is only possible to obtain the description of their
injective envelopes, and just in the cases in which they are also artinian modules.
This way we arrive at the following important problem: to nd conditions under
which an injective envelope of an artinian module is likewise artinian.
At present the theory of modules over group rings is a very well developed
algebraic theory that is rich in many important results and has its own goals and
themes of a dierent nature. Many famous algebraists have made their contribu-
tions to this theory. The main aim of this book is to highlight some important
results within the framework of the described circle of issues outlined above. Be-
cause of the voluntarily limited scope of this book, we were unable to include
all valuable accomplishments. We focused our study on artinian modules because
noetherian modules are presented well enough elsewhere. The last chapter is ded-
icated to some group theoretical results about the splitting of a group over its lo-
cally nilpotent residual. Such theorems about splitting of a group over its abelian
generalized nilpotent radical are very useful in many dierent investigations. In
particular, we found them to be very eective in the study of just non-X-groups
(see L. A. Kurdachenko, J. Otal and I.Ya. Subbotin [157] ). Among these results
we chose a general theorem proven by D.J.S. Robinson [246]. In [246] the authors
proof is based on homological methods. In the current book we develop a new
proof having applied only results pertaining groups and artinian modules. This
aptly illustrates the eectiveness of the artinian condition.
We want to note that originally many important results have been obtained
for integral group rings. However, quite often in the applications, one needs to
deal with other group rings RG. For example, the cases in which R = F x is the
group ring of an innite cyclic group over a (nite) eld F , or R = Zp is the ring
of integer p-adics, or R = F [[X]] is the ring of the power series over a (nite) eld
F frequently occur. Therefore, in the majority of situations, we consider artinian
modules over a group ring DG, where the ring D of scalars is a Dedekind domain.
This requires the insertion of some valuable results of the theory of modules over
Dedekind domains.
Of course, the choice of content has also been determined by the interests
and tastes of the authors. Our selections have been inuenced by the work of many
people, and the authors especially owe their gratitude to B. Hartley and D.I. Za-
itsev. The contribution of B. Hartley to the development of the theory of artinian
modules over group rings is dicult to overestimate. His well-known papers in this
and other areas, as well as the work of his numerous students and collaborators
around the world, have been incredibly inuential. Many of his important contri-
butions are mentioned in this book, but many others are not. For example, we
have omitted the detailed construction of the fundamental important counterex-
amples of uncountable artinian modules over certain nilpotent and soluble groups.
D.I. Zaitsevs interest in the implementation of ring and module-theoretical re-
sults to groups, and his outstanding achievements in this area, inspired a series of
xii Preface

works dedicated to the development of his productive methods (see the survey L.S.
Kazarin and L.A. Kurdachenko [132]). In many cases, his inuence determined the
content of this book. For the rst author, D.I. Zaitsev was a mentor and a good
friend. Unfortunately, the Chernobyl Nuclear Catastrophe undermined his health,
and D.I. Zaitsev passed away in 1990 at the age of forty eight. This catastrophe
was also an original cause of death of our friends and colleagues V.E. Goretsky,
S.S. Levishenko and V.V. Pylaev, who were at the top of their careers when pass-
ing away due to dierent medical complications brought on by this disaster. The
stress experienced during the Chernobyl Catastrophe was the main reason for the
heart attack that took away the life of one of the main founders of Innite Group
Theory and the head of the Kiev Group Theory School, S.N. Chernikov. This
great mathematician was a teacher for many Ukranian algebraists, including the
authors of this book. His inuence on the development of Innite Group Theory
could not be overestimated.
We, with genuine gratitude, remember another brilliant algebraist, whose
extremely various scientic interests and personality made a great impact on the
forming of the rst and the third authors, namely Z.I. Borevich. He was a profes-
sor of St Petersburg University, but he was born in Ukraine and was interested
in the development of Ukrainian algebra and greatly supported many Ukrainian
mathematicians, who have became leading researches nowadays.
We would like to express our great appreciation to the ocial research com-
mittees of Spain (CICYT) and Arag on (CONSI+D) for the nancing of this
project. We also are extremly grateful to the National University (California,
USA), and to the University of Zaragoza (Spain) for their thorough support of
the authors work.
Chapter 1

Modules with chain conditions

Finiteness conditions in Algebra play a key role in the evolution from nite to
innite objects. Historically, these conditions have been introduced in rings and
modules where they have shown their eectiveness. At the beginning, the niteness
conditions have been most fruitfully used as the minimal and maximal conditions.
They shaped the development of modern Algebra, and still have a signicant value
today. The maximal condition under the form of the ascending chain condition was
initially used by R. Dedekind [51], who introduced this subject in his study of ide-
als in algebraic numbers elds. On the other hand, the descending chain condition
was introduced by J.M. Wedderburn [279], and explicitly dened by W. Krull, E.
Noether and E. Artin. Their applications have been decisive in studying maxi-
mal and minimal conditions in groups, modules, Lie algebras, and other algebraic
structures. The investigation of dierent algebraic systems satisfying maximal and
minimal conditions still remains very active.
An ordered set M = (M, ) is said to satisfy the maximal condition if
every non-empty subset of M has a maximal element. It is rather easy to see that
M satises the maximal condition if and only if M satises the ascending chain
condition, that is, given an ascending chain

a1 a2 an

of elements of M, there is some k N such that ak = ak+n for any n N.


Dually, M satises the minimal condition if every non-empty subset of M
has a minimal element, which is equivalent to M satises the descending chain
condition, that is, given a descending chain

a1 a2 an

of elements of M, there is some k N such that ak = ak+n for any n N.


Throughout, unless specied otherwise, module means right module. We re-
call that any assertion about right modules has a similar assertion for left modules.
2 Chapter 1. Modules with chain conditions

For a ring R and an R-module A, we consider the lattice LR (A) consisting of all
R-submodules of A ordered by inclusion. We say that A is a noetherian R-module
or it satises the maximal condition on its submodules if LR (A) satises the maxi-
mal condition, and similarly A is an artinian R-module or A satises the minimal
condition on its submodules if LR (A) satises the minimal condition. As we men-
tioned above, this is equivalent to the members of LR (A) satisfying the ascending
chain condition (and the descending chain condition respectively).

A ring R is called right noetherian (in short, R satises Max-r) if the right
R-module R is noetherian, and R is called right artinian (R satises Min-r) if
the right R-module R is artinian (note, that a ring means a ring with identity).
Similarly, we dene left noetherian and left artinian modules and the conditions
Max-l or Min-l. Since in this book we consider modules rather than rings, it is
worth noting that we shall use the terms noetherian and artinian rings to denote
right noetherian and right artinian rings. Modules with chain conditions are stan-
dard topics of many books, such as F. Anderson and K. Fuller [2], M. Atiyah and
I.G. MacDonald [3], N. Bourbaki [15, 16, 19], A.W. Chatters and C.R. Hajarmavis
[41], C. Faith [66, 67], J.S. Golan and T. Head [85], B. Hartley and T.O. Hawkes
[109], T. Hungerford [114], A. Kertesz [136], A.G. Kurosh [177], T.Y. Lam [178],
S. Lang [179], J.C. MacConnel and J.C. Robson [190], D.G. Northcott [207], D.S.
Passman [219], and R. Sharp [263], for example. For the readers convenience, we
will review some elementary results on them, which we will freely use in what
follows.

Every eld and the ring Z of all integers are examples of noetherian rings.
By Hilberts Basis Theorem the ring of polynomials R[X] over a commutative
noetherian ring is likewise noetherian. As a consequence we obtain that a com-
mutative ring R such that R is nitely generated over its noetherian subring S
with the same identity is noetherian. Moreover, the ring of formal power series
R[[X]] over a commutative noetherian ring is likewise noetherian. There are dis-
tinct generalizations of Hilberts Basis Theorem on non-commutative noetherian
rings. For example: let R be a ring and S be a subring of R with the same identity.
If U (R) (the group of all invertible elements of R) includes a polycyclic-by-nite
subgroup G such that S = S G and G generates R over S, then R is noetherian
[217, Theorem 10.2.7]. The famous Halls theorem follows from this: the group ring
RG over a noetherian ring R is also noetherian [95]. By Hopkins theorem (see,
for example, [136, Corollary 59.3]) every artinian ring is noetherian. It follows that
every artinian ring has a nite composition series of (right) ideals. In particular, a
direct sum of nitely many division rings is artinian; a matrix ring over a division
ring is artinian.
Let R be a ring, and let A be an R-module. If B is a subset of A, by denition
the R-annihilator of B is the left ideal of R

AnnR (B) = {x R | ax = 0 for all a B}.


Chapter 1. Modules with chain conditions 3

In particular, the annihilator of A is the two-sided ideal



AnnR (A) = AnnR (a).
aA

Lemma 1.1. Let R be a ring, and A be an R-module.


(1) If A is noetherian and B A, then B and A/B are noetherian.
(2) If B is a submodule of A such that B and A/B are noetherian, then A is
noetherian.
(3) If A has a nite series of submodules

0 = A0 A1 An = A,

every factor of which is noetherian, then A is noetherian.


(4) If A = A1 + + An and A1 , . . . , An are noetherian, then A is noetherian.
(5) A is noetherian if and only if every submodule of A is nitely generated.
(6) If R is a noetherian ring, then any nitely generated R-module is noetherian.
Proof. (1) is obvious and (3) is an immediate consequence of (2).
(2) Let
A1 A2 An
be an ascending chain of submodules of A. Since B and A/B are noetherian, there
is some m N such that Am B = Am+n B and Am + B = Am+n + B for all
n N. Therefore

Am+n = Am+n (Am+n + B) = Am+n (Am + B)


= Am + (Am+n B) = Am + (Am B) = Am

for all n N, and then the result follows.


(4) We proceed by induction on n. Assume that B = A1 + + An1 is
noetherian. Then by (1) A/B = (An + B)/B = An /(An B) is also noetherian.
By (2), A is noetherian.
(5) Suppose that A is noetherian, and let B be a non-zero submodule of A.
Pick 0 = b1 B, and put B1 = b1 R. If B1 = B, then pick 0 = b2 B \ B1 , and
put B2 = b1 R + b2 R. Since b1  B1 we have that B2 = B1 . Proceeding in this way,
we construct an ascending chain of nitely generated submodules of A,

B1 B2 Bn ,

such that every Bn is nitely generated. By hypothesis, there is some k N such


that Bk = Bk+m for all m N. It follows that B = Bk , and thus B is nitely
generated.
4 Chapter 1. Modules with chain conditions

Conversely, let
A1 A2 An

be an ascending chain of submodules. If B = nN An , then there exist b1 , . . . , bm
A such that B = b1 R + + bm R. Choose r N such that b1 , . . . , bm Ar . We
have that B = Ar , and then Ar = An for all n r so that A is noetherian.
(6) Let A = a1 R + + at R. Given 1 i n, since ai R = R/AnnR (ai ),
each ai R is noetherian by (1). By (4), A is also noetherian. 
The corresponding result for artinian modules is similar.
Lemma 1.2. Let R be a ring and A an R-module.
(1) If A is artinian and B is a submodule of A, then B and A/B are artinian.
(2) If B is a submodule of A such that B and A/B are artinian, then so is A.
(3) If A has a nite series of submodules

0 = A0 A1 An = A,

every factor of which is artinian, then A is likewise artinian.


(4) If A = A1 + + An , where Ai is artinian for 1 i n, then A is artinian.
Now we focus our study on two topics. The rst one is the establishment of
the existence and the uniqueness of direct decompositions of certain modules. To
do so, we consider endomorphisms of modules with chain conditions.
Lemma 1.3. Let R be a ring and A an R-module. If is an endomorphism of A
and n 1, we put En = Im n and Rn = Ker n . Then:
(1) If E1 = E2 , then R1 + E1 = A.
(2) If R1 = R2 , then R1 E1 = 0.
(3) If A is artinian, then there is some n N such that A = Rn + En .
Proof. (1) If E1 = E2 , then for every a A there is some b A such that
a = b2 . It follows that a b Ker = R1 , and so a E1 + R1 . Consequently,
A = E1 + R1 .
(2) Let R1 = R2 . If a R1 E1 , then a = 0, and there is some b A such
that a = b. It follows that b Ker 2 = R2 = R1 , and hence a = b = 0.
(3) If A is artinian, then there is some m 1 such that Em = E2m , and
therefore A = En + Rn by (1). 
Corollary 1.4. Let R be a ring and A an R-module, and let be an endomorphism
of A.
(1) If A is artinian, then is an automorphism if and only if Ker = 0.
(2) If A is noetherian, then is an automorphism if and only if Im = A.
Chapter 1. Modules with chain conditions 5

Proof. Suppose that A is artinian and Ker = 0. Then m is a monomorphism


for every m N, that is, Ker m = 0 for m N. By Lemma 1.3, there is some
n N such that A = Im n . Then A = Im , i.e. is an automorphism.
Suppose that A is noetherian, and A = Im . Then A = Im m for every
m N. By the niteness of ascending chains, there is some n N such that
Ker n = Ker n+1 . Taking reciprocal images, we obtain that
(Ker )(n )1 = Ker n+1 = Ker n = {0}(n )1 .
Since n is an epimorphism,

Ker = ((Ker )(n )1 )n = (0 (n )1 )n = 0 ,

and so is an automorphism. 
Proposition 1.5 (Fitting lemma). Let A be an artinian and noetherian module
over a ring R. If is an endomorphism of A, then A = B C, and the following
conditions hold:
(1) B B and C C.
(2) The restriction of on B is an automorphism of B.
(3) The restriction of on C is nilpotent.
Proof. Since the chains

A A Ak

and
0 Ker Ker k
have to be nite, there exists some r 1 such that Ar = Am , and Ker r =
Ker m for all m r. Put B = Ar , and C = Ker r . Then

B = (Ar ) = Ar+1 = Ar = B,

C = (Ker r ) = (Ker r+1 ) Ker r = C.


Let 1 (respectively 2 ) be the restriction of on B (respectively on C). By
Corollary 1.4, 1 is an automorphism of B since it is clearly surjective. Moreover

Cr = (0 (r )1 )r = 0 ,

and then 2 is nilpotent.


If b B C, then br = 0. Since br = b(1 )r and 1 is an automorphism
of B, (1 )r is also an automorphism. It follows that b = 0; that is, B C = 0.
Finally, applying Lemma 1.3 we obtain that

A = Im r + Ker r = B + C.

This nishes the proof. 


6 Chapter 1. Modules with chain conditions

An R-module A is said to be indecomposable if it has no non-trivial direct


decompositions; that is, if A = B C, then either B = 0 or C = 0. Otherwise
A is said to be decomposable. As a consequence of Proposition 1.5, we have
Corollary 1.6. Let A be an indecomposable R-module over a ring R, and let
be an endomorphism of A. If A is noetherian and artinian, then either is an
automorphism of A or is nilpotent.
A non-zero R-module A is said to be simple if A has no non-zero proper sub-
modules. For an arbitrary R-module A, the intersection of all non-zero submodules
of A is called the R-monolith of A, and denoted by R (A). If R (A) = 0 , then
A is said to be monolithic. In this case, it is clear that R (A) is the unique simple
R-submodule of A.
Theorem 1.7. Let A be an artinian R-module, where R is a ring. If

A = A1 An = B1 Br

are two direct decompositions of A, and all the submodules Ai and Bj are R-
monolithic, then n = r, and there are a permutation Sn , and an automorphism
of A such that (Bj ) = Aj for 1 j n.
Proof. For every 1 i n and 1 j r, let i : A Ai and j : A Bj be
the canonical projections. It will suce to prove the following assertion:
There are an injective mapping
(t) : {1, . . . , t} {1, . . . , r},

and an automorphism t of A such that, if 1 j t, then (Bj )t = Aj(t) .


We proceed by induction on t. If t = 1, then it suces to take 1 = A , the
identity automorphism of A. Suppose that t > 1, and we are given an injective
mapping (t 1), and an automorphism t1 , which satisfy the above conditions.
Reordering if necessary, and replacing Bj by an isomorphic image, we may assume
that B1 = A1 , . . . , Bt1 = At1 . Given a A, we have that a = a1 + + an .
It follows that
at = (a1 )t + + (an n )t
and so
t = 1 t + + n t .
We note that the restriction of t on Bt is the identity automorphism of Bt . On the
other hand, if the kernel of an endomorphism of a monolithic module is non-zero,
this kernel must include the monolith. It follows that the sum of nitely many
endomorphisms with non-zero kernels has a non-zero kernel.
Since Bt is monolithic, there is some 1 m n, such that the kernel of the
restriction of m t on Bt is zero. By Corollary 1.4, this restriction is in fact an
Chapter 1. Modules with chain conditions 7

automorphism of Bt . If 1 j t 1, then j = j so that j t j t = 0. This


means that m t.
Dene 
j, if 1 j t 1,
j(t) =
m, if j = t,
and
t = eA t + t m .
If a Ker t , then

0 = (at )t = at (at )t + (at m )t


= at at + (at )m t = (at )m t .

Since Ker m t = 0, it follows that at = 0. Thus

a = at + at (at )m = at = 0

and therefore Ker t = 0. Then, by Corollary 1.4, t is actually an automor-


phism. Since 
A = (Bt )t ( Bj )t and (Bt )t Am ,
j=t

we conclude that 
Am = (Bt )t (Am ( Bj )t ).
j=t

But Am is monolithic, so that, in particular, Am is indecomposable. Hence Am =


(Bt )t . Taking t = n, we obtain the required result. 
Let A be a group with a set of (multiplicative) operators . This means that
with each there is associated a certain endomorphism of A so that to each
a A there corresponds a A such that (ab) = a b for all a, b A. We say
that A is a -operator group or A is a -group. Note that dierent operators of
can correspond to the same endomorphism, that is, it can be = and a = a
for every a A. This is the great advantage of considering groups with operators
instead of groups with a xed set of endomorphisms. A subgroup B of A is called
-invariant if for every b B and , we have b B. It follows that we
can consider -invariant subgroups of a group A as groups with the same set
of operators.
Suppose that A is a normal subgroup of a group G. If = G, we may consider
A as a -operator group dening an action of G on A by (a, g) g 1 ag = ag ,
a A, g G. In this case, the -invariant subgroups are called G-invariant.
If A is an abelian -group and = R is a ring, then a typical example
here is a module over the ring R. In this case -invariant subgroups are exactly
R-submodules.
8 Chapter 1. Modules with chain conditions

A -operator group A is said to satisfy the minimal condition on its -


invariant subgroups (in short Min-) if the ordered-by-inclusion set of its -
invariant subgroups satises the minimal condition. Dually, A is said to satisfy the
maximal condition on its -invariant subgroups (in short Max-) if the ordered-
by-inclusion set of its -invariant subgroups satises the maximal condition.
Suppose that A is a normal subgroup of a group G, = G and G acts on
A by conjugation. If A satises Min- (respectively Max-), then we say that A
satises the condition Min-G (respectively Max-G). In particular, if A = G, then
we obtain the minimal condition for normal subgroups (Min-n) and the maximal
condition for normal subgroups ( Max-n).
The case when A = G and = {G } is opposite. If A satises Min- (respec-
tively Max-), then in this case G satises the minimal condition for all subgroups
or the condition Min (respectively, the maximal condition for all subgroups or the
condition Max). We also say that a group G is artinian (respectively noetherian), if
G satises Min (respectively Max). We will denote by M the class of all artinian
groups and by M the class of all noetherian groups.
As we already noted above, the minimal and the maximal conditions have
been introduced by German algebraists. The study of groups satisfying these con-
ditions was initiated later than the corresponding investigations in rings and mod-
ules. It is not hard to see that a locally (soluble-by-nite) groups satisfying Max
is polycyclic-by-nite. Distinct characteristics of noetherian groups have been ob-
tained by R. Baer [6]. In connection with this R. Baer formulated the following
problem:
Is every noetherian group polycyclic-by-nite?
A prominent Soviet algebraist S.N. Chernikov was a pioneer in investigation
of artinian groups. He was a student of the famous founders of the Soviet Algebra
School A.G. Kurosh and O.Yu. Schmidt. It is worth noting, that after WW I,
fruitful and cross inuential connections between German and Soviet algebraists
were established. Some Soviet researchers visited Germany at that time and many
prominent German algebraists, including E. Noether, delivered lectures at Moscow
State University. Unfortunately, in a few years the Soviet government prohibited
these and any other contacts with foreign scientists.
Let p be a prime and
Cp =< an | ap1 = 1, apn+1 = an , n N > .
The group Cp is called a Pr ufer p-group. A group G is called a Chernikov
group if G includes a normal subgroup of nite index which is a direct product of
nitely many Pr ufer p-groups. It is not hard to see that every Chernikov group
satises Min. We will denote by C the class of all Chernikov groups and by P
the class of all polycyclic-by-nite groups. These groups are among the oldest
classical subjects in the theory of innite groups. The properties of Chernikov
Chapter 1. Modules with chain conditions 9

and polycyclic-by-nite groups have been considered in detail in many books (for
example, S.N. Chernikov [43], M.I. Kargapolov and Yu.I. Merzlyakov [129], O.H.
Kegel and B.A.F. Wehrfritz [134], A.G. Kurosh [176], D.J.S. Robinson [230, 234,
235, 242], and D. Segal [255]). The following description of locally soluble artinian
groups was one of the rst results of S.N. Chernikov:
A locally soluble group satises Min if and only if it is Chernikov
(see, for example [43, Theorem 1.1]). V.P. Shunkov and O.H. Kegel and B.A.F.
Wehrfritz independently obtained the following theorem which one can consider
as the latest positive result on artinian groups:
Every locally nite group satisfying Min is Chernikov
(see, for example [134, Theorem 5.8]). S.N. Chernikov formulated the following
problem:
Is every artinian group Chernikov?
Negative answers both to Baers and Chernikovs questions have been ob-
tained by A.Yu. Olshanskii who constructed exotic examples of torsion-free groups
whose proper subgroups are cyclic and periodic groups whose proper subgroups are
nite (such groups are called quasinite); in particular, an example of an innite
p-group whose proper subgroups have order p (the Tarsky Monster ). Furthermore,
there exist uncountable artinian groups (see A.Yu. Olshanskii [208, chapters 28, 35,
38]). In connection with this we observe that R.I. Grigorchuk [92, 93] constructed
very interesting examples of nitely generated p-groups with highly diverse prop-
erties, which are some kind of dual to the quasinite groups. These groups are just
innite; that is, they have only nite proper factor-groups.
Let A be an abelian group. The generalization of the concept of the layer of
an abelian p-group leads us to the following denition. If is an endomorphism
of A, for each n N, we dene

,n (A) = {a A | an = 0}.

Note that ,n (A) = Ker n is a subgroup of A, and ,n (A) ,n+1 (A) for
any n N. In terms of these subgroups, D.I. Zaitsev [306] proved the following
criterion, which is very useful in the study of artinian modules.

Proposition 1.8. Let A be an abelian -group, where is a semigroup of distributive


endomorphisms of A. Suppose that we are given an element such that
(i) = for every ; and

(ii) A = nN ,n (A).
Then the group A satises Min- if and only if ,1 (A) satises Min-.
10 Chapter 1. Modules with chain conditions

Proof. Since satises (i), each ,n (A) is a -invariant subgroup of A. Therefore,


if A satises Min-, then ,1 (A) satises Min-.
Conversely, suppose that B = ,1 (A) satises Min-. Let

A1 A2 An

be a descending chain of -invariant subgroups of A. If k, m N, we put

Bk,m = (Ak ,m+1 (A))m .

Obviously each Bk,m is -invariant, Bk,m Bk+1,m , and Bk,m Bk,m+1 for any
pair k, m N. It follows that Bk,m Bs,t whenever s k, and t m. Since B
satises Min-, there are some q, r N such that Bk,m = Bq,r whenever k q
and m r. The mapping : a at , a ,t+1 (A) is a -endomorphism, and
so Im , and Ker = ,t (A) are -invariant. Therefore

,t+1 (A)/,t (A)


= Im ,1 (A).

It follows that ,t+1 (A)/,t (A) satises Min-, and therefore so does
,n (A), for any n N. Therefore, there is some q(1) N such that

Ak ,r+1 (A) = Aq(1) ,r+1 (A)

for every k q(1). Let q(2) be the maximum of {q, q(1)}. If k q(2) and n r+1,
then we claim that the following expression holds:

(E) : Ak ,n (A) = Aq(2) ,n (A).

To see this, we proceed by induction on n. Suppose that (E) has been shown for
some n r + 1. Pick a Aq(2) ,n+1 (A). Then an Bq(2),n = Bk,n so that
an = cn for some c Ak ,n+1 (A). It follows that (a c)n = 0; that is,

a c Aq(2 ) ,n (A) = Ak ,n (A).

In particular, a Ak , that is a Ak ,n+1 (A). Consequently,

Ak ,n+1 (A) = Aq(2) ,n+1 (A),

and then our claim follows. Then we have


 
Ak = Ak A = Ak ( ,n (A)) = (Ak ,n (A))
nr+1 nr+1
 
= (Aq(2) ,n (A)) = Aq(2) ( ,n (A)) = Aq(2) ,
nr+1 nr+1

and so A satises Min-. 


Chapter 1. Modules with chain conditions 11

Let P be an abelian p-group, where p is a prime. If n N, by denition, the


nth -layer of the p-group P is the characteristic subgroup

n (P ) = {x P | |x| pn }.

We clearly have that

1 (P ) 2 (P ) n (P )

and P = nN n (P ). The rst layer 1 (P ) concentrates a lot of information
about P, and it is simply known as the lower layer of the p-group P . Let be the
endomorphism of P given by : a pa, a P . Applying Proposition 1.8, we
deduce the following elegant characterization due to B. Hartley and D. McDougall
[110].
Corollary 1.9. Let A be an abelian p-group, where p is a prime, and suppose that
A is a -group, where is a semigroup of distributive endomorphisms of A. Then
A satises Min- if and only if 1 (A) satises Min-.
Chapter 2

Ranks of groups

In this and successive chapters we consider some extensions of the class of nite
groups whose members are groups G related to certain numerical invariants, which
are generically known as ranks of the group G. The concept of rank has appeared
in the theory of modules over arbitrary rings as a natural generalization of the idea
of dimension of a vector space over a eld. Recall that an abelian group is exactly
a module over the ring Z of integers. Since the theory of abelian groups is the
initial part of the theory of groups, it is not surprising that the inherited notion of
rank of a module has been successfully employed in many areas of Group Theory.
In fact, these ideas have become the foundation for new important extensions of
nite groups, most of which were created by A.I. Maltsev [181, 182] and D.J.S.
Robinson [230]. Some of such extensions are our next goals. However, we are not
going to consider in full the properties of groups of nite rank, which has been
done already in a wide variety of dierent papers. Note that one can nd some
fundamental results about groups of dierent nite ranks in the books of M.R.
Dixon [52], M.I. Kargapolov and Yu.I. Merzlyakov [129], A.G. Kurosh [176], B.I.
Plotkin [221], D.J.S. Robinson [230, 234, 235, 242], and others. We are going to
focus on some important denitions and results that are suitable for our purposes.
A group G is said to have nite 0-rank r0 (G) = r if G has a nite subnormal
series with exactly r innite cyclic factors, the others being periodic. We note that
every renement of one of these series has only r factors that are innite cyclic.
Since any two nite subnormal series have isomorphic renements, we conclude
that 0-rank is independent of the chosen series. This numerical invariant is also
known as the torsion-free rank of G. In this general form the concept of the 0-
rank was introduced by D.I. Zaitsev [303]. Polycyclic-by-nite groups are rst
examples of groups having nite 0-rank. For these groups 0-rank is exactly its
Hirsch number .
Lemma 2.1. Let G be a group of nite 0-rank. Suppose that H is a subgroup of G
and L is a normal subgroup of G. Then
(1) H has nite 0-rank; concretely, r0 (H) r0 (G); and
(2) G/L has nite 0-rank; concretely, r0 (G) = r0 (L) + r0 (G/L).
14 Chapter 2. Ranks of groups

Proof. Suppose that r0 (G) = r. Then G has a nite subnormal series


1 = G0  G1   Gn = G
such that r n, r factors are innite cyclic and the remaining n r factors are
periodic.
(1) If H G, then
1 = G0 H   Gn H = H
is a subnormal series of H. Since the factors of this series are isomorphic to sub-
groups of factors of the given series, we deduce immediately that r0 (H) r0 (G),
as claimed.
(2) Given L  G, we have the subnormal series
1 = G0 L   Gn L = L = LG0  LG1   LGn = G.
Since {Gj L | 0 j n} is a subnormal series of L , and {Gj L/L | 0 j n}
is a subnormal series of G/L with exactly r innite cyclic factors, the others being
periodic, we nd out that r0 (G) = r0 (L) + r0 (G/L), as required. 
If A is an abelian group, then
r0 (A) = dimQ (A Z Q).
In fact, an abelian group A has nite 0-rank r if and only if A/t(A) is isomorphic
to a subgroup of the additive group
Q Q
 
r

(see L. Fuchs [78, Lemma 24.3]), where t(A) is the periodic part of A (that is,
the unique maximal periodic subgroup of A). Note also that r0 (A) is exactly the
Z-rank of the Z-module A/t(A). Related to abelian groups of nite 0-rank are the
soluble groups of nite 0-rank. This class of groups is exactly the class of soluble
A1 -groups, which was introduced by A.I. Maltsev [182]. The linear group UTn (Q)
(even UTn (F ) where F is a nite eld extension of Q) is a natural example of a
non-abelian group of nite 0-rank.
Let A be an abelian group. If p is a prime, then the p-rank rp (A) of A is
dened as follows. Let P be the Sylow p-subgroup of A , and consider 1 (P ) as a
vector space over the prime eld Fp = Z/pZ. By denition,
rp (A) = dimFp 1 (P ).
We note that rp (A) is nite if and only if the Sylow p-subgroup P of A is Chernikov
(see L. Fuchs [78, Theorem 25.1]).
Following D.J.S. Robinson [230, 6.1], we dene the classes A0 and G0 as
follows:
Chapter 2. Ranks of groups 15

An abelian group A belongs to the class A0 (or A is an A0 -group) if and only


if r0 (A) is nite and rp (A) is nite for all primes p.
A soluble group G belongs to the class S0 (or G is an S0 -group) if and only if
G has a nite subnormal series, every factor of which is an abelian A0 -group.
A direct product U DrpP Cp , where U is a direct product of nitely many
copies of Q and Cp is a direct product of nitely many copies of a Pr ufer p-groups
for all p P (here P is the set of all primes), is a natural example of an abelian
group from the class A0 . Moreover, every group from the class A0 is isomorphic to
some subgroup of a direct product of this kind. We can nd some natural examples
of groups from class S0 among linear groups. If F is a eld and p is a prime such
that p = char F , then every p-subgroup of GLn (F ) is Chernikov (see, for example
[283, 9.1]). It follows that every periodic soluble subgroup S of GLn (F ) such that
char F / (S) belongs to S0 .
These classes A0 and S0 admit the following almost obvious characteriza-
tions.

Proposition 2.2. (1) An abelian group A is an A0 -group if and only if every ele-
mentary abelian p-factor of A is nite, for each prime p.
(2) A soluble group G is an S0 -group if and only if every elementary abelian
p-section of G is nite, for each prime p.
The above result shows that S0 -groups can be called groups of nite (abelian)
section rank . In fact Proposition 2.2 provides us the possibility of extending the
concept of p-rank from abelian groups to arbitrary groups in the following way.
We say that a group G has nite p-rank rp (G) = r if every abelian p-section of G
is nite of order pr , and there is an abelian p-section U/V such that |U/V | = pr .
Let A be an abelian group. Following D.J.S. Robinson [230, 6.5], we dene
the reduced rank of A by
rred (A) = r0 (A) + max{rp (A) | p (A)}.

The abelian group A belongs to the class U (or A is a U-group)


if and only
if rred (A) is nite. In other words, there is some m N such that r0 (A) m and
rp (A) m, for each p (A). A soluble group G belongs to the class S (or G is

an S-group) if and only if G has a nite subnormal series, every factor of which

is an abelian A-group.
A group G is said to have nite special rank r (G) = r if every nitely gen-
erated subgroup of G can be generated by r elements and r is the least positive
integer with this property. This notion is due to A.I. Maltsev [181]. The special
rank of a group is called also the Pr
ufer Maltsev rank
To characterize these classes in a simpler way, we need the next almost ob-
vious result.
16 Chapter 2. Ranks of groups

Lemma 2.3. Let G be a group. If H is a subgroup of G and L is a normal subgroup


of G, then:
(1) If G has nite special rank, then H and G/L have nite special rank, and
r(H) r(G), r(G/L) r(G).
(2) If L and G/L have nite special ranks, then so do have G.
Corollary 2.4. (1) An abelian group A is an A -group if and only if A has nite
special rank.
(2) A soluble group G is an S -group if and only if G has nite special rank.
This result shows that the class S is exactly the class of soluble A2 -groups,
formerly introduced by A.I. Maltsev [182].
Let A be an abelian group. Following D.J.S. Robinson [230, 6.2], we dene
the total rank of A as

rtot (A) = r0 (A) + rp (A).


p(A)

An abelian group A belongs to the class A1 (or A is an A1 -group) if and


only if rtot (A) is nite. In particular, this means that (A) is nite. In other
words, A A1 if and only if A has nite 0-rank , and the periodic part of A
is a Chernikov group.
A soluble group G belongs to S1 (or G is an S1 -group) if and only if G has
a nite subnormal series, every factor of which is an abelian A1 -group.
The class S1 is exactly the class of soluble A3 -groups, which was introduced
by A.I. Maltsev [182]. Note that Maltsev also considered the subclass of soluble
A4 -groups, which consists of the groups having a nite subnormal series, every
factor of which is an abelian A1 -group with a nite periodic part.
A group G is called minimax if it has a nite subnormal series, all factors
of which belong to M or M . The class of soluble minimax groups is denoted
by S2 (D.J.S. Robinson [230]). In particular, S2 F is exactly the class of groups
having nite subnormal series, every factor of which is Chernikov or polycyclic-by-
nite. The following combinatorial characteristic of groups relates to the minimax
groups. Following D.I. Zaitsev [295] we say that a group G has nite minimax
rank rmmx (G) = m, if for each nite chain of subgroups

1 = G0 G1 Gn = G

such that all indexes |Gj+1 : Gj | are innite, we have n m and there exists a
chain of this kind for which n = m. If such a number does not exist, then the
minimax rank of the group is innite; if G is nite, then rmmx (G) = 0. Observe,
that in the paper [295], D.I. Zaitsev used another term the index of minimality.
Chapter 2. Ranks of groups 17

This term happened to be not suitable. Therefore Zaitsev stopped using it and
introduced instead the minimax rank. As it is proved by Zaitsev [291] a locally
(soluble-by-nite) group has a nite minimax rank if and only if it is minimax.
The theory of soluble-by-nite minimax groups is well developed now. These
groups have been studied by many authors from dierent points of view. Gen-
eral properties of soluble minimax groups have been considered in the paper
[229] (see, also [235, 10.3, 10.4]). Minimax groups appear in the study of dis-
tinct niteness conditions (see, for example, the following papers: R. Baer [7],
G. Cutolo and L.A. Kurdachenko [50], M. Karbe [126], M. Karbe and L.A. Kur-
dachenko [127], L.S. Kazarin and L.A. Kurdachenko [132], L.S. Kazarin, L.A.
Kurdachenko and I.Ya. Subbotin [133], P.H. Kropholler [138], L.A. Kurdachenko
[141, 142, 144, 145, 146, 147, 150], L.A. Kurdachenko and V.E. Goretsky [153],
L.A. Kurdachenko and H. Smith [165, 166, 167, 168], L.A. Kurdachenko, A.V.
Tushev and D.I. Zaitsev [175], M.L. Newell [202, 203], H. Smith [266], D.I. Zait-
sev [289, 290, 291, 292], D.I. Zaitsev, L.A. Kurdachenko and A.V. Tushev [312]).
D.J.S. Robinson showed that every nitely generated hyperabelian group of -
nite special rank (respectively nite section rank) is minimax [228, 233, 238, 243]
(see also D.I. Zaitsev [293]). Properties of subnormal and permutable subgroups
in minimax groups have been considered by D.C. Brewster and J.C. Lennox [24],
D.J. McCaughan [189], D. McDougall [191] and H.H.G. Seneviratne [262]. S. Fran-
ciosi S. and F. de Giovanni [72] considered maximal subgroups in soluble minimax
groups. In the papers by L.A. Kurdachenko [149] and L.A. Kurdachenko and J.
Otal [156] the groups with minimax conjugacy classes have been studied; J. S.
Wilson [286, 287] investigated a soluble product of minimax groups; in the arti-
cles by M. Karbe and L.A. Kurdachenko [127], D.I. Zaitsev, L.A. Kurdachenko
and A.V. Tushev [312], D. Segal [256, 257], D.I. Zaitsev, L.A. Kurdachenko and
A.V.Tushev [312] the authors explored modules over minimax groups.
Now we consider relations among dierent ranks of groups. As usual, if G
is a group, by (G) we denote the set of all primes that occur as divisors of the
orders of periodic elements of G.

Lemma 2.5. Let G be a group and let H be a periodic subgroup of G. If H is


ascendant in G, then H O (G), where = (H).
Proof. Let
H = H0  H1  H  H+1  H = G
be an ascending series. We will prove by transnite induction on that H
O (H ). If = 1, this inclusion is obvious. Suppose that we have already
 proved
that H O (H ) for all < . If is a limit ordinal, then H = < H and
hence H < O (H ) = O (H ). Suppose now that 1 exists. Since H1
is normal in H and O (H1 ) is a characteristic subgroup of H1 , O (H1 )
is normal in H and therefore O (H1 ) O (H ). Hence H O (H1 )
O (H ). 
18 Chapter 2. Ranks of groups

If G is a group, then by t(G) we will denote the largest normal periodic


subgroup of G.
Corollary 2.6. Let G be a group and suppose that G includes a non-identity as-
cending periodic subgroup. Then t(G) = 1.
Let A be an abelian torsion-free group of nite 0-rank and let T be a periodic
automorphism group of A. Then T is nite (see, for example, [280, Theorem 9.33]).
Moreover, if r0 (A) = r, then there is a function f1 : N N such that |T | f1 (r).
Lemma 2.7 ([55]). Let G be a group and suppose that t(G) = 1. If G has an
ascending series H = H0  H1   H  H+1   H = G such that
H is an abelian torsion-free subgroup of nite 0-rank and the rest of the factors
are locally nite, then G includes a characteristic abelian torsion-free subgroup A
such that r0 (A) = r0 (H), H A and |G/A| f1 (r0 (H)).
Proof. We proceed by transnite induction on . Let = 1 and put K = H1 ,
L = CK (H). By the above remark |K/L| f1 (r0 (H)). The subgroup L is a
central extension of H by a locally nite group. By the generalized Schur theorem
(see, for example, [234, Corollary to Theorem 4.12], the subgroup [L, L] is locally
nite. It follows that t(L) is locally nite, and L/t(L) is abelian and torsion-free.
Moreover, r0 (L/t(L)) = r0 (H). Since t(L) is a characteristic subgroup of L, t(L)
is normal in K. If we suppose that t(L) = 1, then Corollary 2.6 yields that
t(G) = 1. This show that L is an abelian and torsion-free subgroup of 0-rank
r0 (H). In this case let A be the locally nilpotent radical of K. Then clearly A is
an abelian torsion-free subgroup of 0-rank r0 (H), H A and |K/A| f1 (r0 (H)).
Assume that we have already proved that H includes a normal subgroup
L such that |H /L | f1 (r0 (H)), and L is a torsion-free abelian subgroup
of 0-rank r0 (H) for all < . Assume rst that is not a limit ordinal. Put
H = U, H1 = V, L1 = W . Since |V /W | = s f1 (r0 (H)), V s W . In
other words, U includes a normal torsion-free abelian subgroup V s such that
r0 (V s ) = r0 (H) and U/V s is locally nite. By the above argument the locally
nilpotent radical D of U is a torsion-free abelian subgroup of 0-rank r0 (H) such
that |U/D| f1 (r0 (H)). Since W is a subnormal abelian subgroup of U , W D.
Suppose now that is a limit ordinal. Theabove proof shows that we can
assume that L L+1 for all < . Put A = < L . Clearly, A is a normal
abelian torsion-free subgroup of 0-rank r0 (H) such that H /L is locally nite.
But this case has been considered above. So the result is proved. 
Corollary 2.8 ([55]). Let G be a group and suppose that t(G) = 1. If G has an
ascending series
H = H 0  H 1  H  H +1  H = G
such that r factors of this series are innite cyclic and the rest of the factors are
locally nite, then G has a nite series
1 = K 0 L1 K 1 L2 K 2 Ln K n = G
Chapter 2. Ranks of groups 19

of normal subgroups such that Lj+1 /Kj are abelian torsion-free subgroups of nite
0-rank, 0 j n 1, n r, and the factors Kj /Lj are nite of order at most
f1 (r).
Proof. We proceed by induction on r. Since t(G) = 1, Lemma 2.5 yields that
H1 is an innite cyclic subgroup. Since G includes an ascendant abelian subgroup,
its locally nilpotent radical L is non-identity. Moreover, H1 L. Note, that the
set of all elements of nite order of a locally nilpotent group is a characteristic
subgroup. Together with the equation t(G) = 1 it implies that L is torsion-free.
By Maltsevs theorem [182] a subgroup L is nilpotent, in particular, Z = (L) =
1.
Put K1 /Z = t(G/Z); then K1 /Z is locally nite. By Lemma 2.7, K1 includes
a characteristic abelian torsion-free subgroup L1 of nite 0-rank such that K1 /L1
is nite and |K1 /L1 | f1 (r). By the choice of K1 we have t(G/K1 ) = 1, and
G/K1 has an ascending series, in which the r 1 factors are innite cyclic and
the rest factors are locally nite. Hence we may apply to G/K1 the inductive
hypothesis. 

Let Mn be the set of all non-isomorphic nite groups of order at most n and
let An = {Aut(G) | G Mn }. Then the set An is nite and each member of An is
nite. Hence there is D An such that |D| is maximal. Put a(n) = |D|. Clearly,
a(n) n!.
Corollary 2.9 ([55]). Let G be a group and suppose that t(G) = 1. If G has an
ascending series

H = H 0  H 1  H  H +1  H = G

such that r factors of this series are innite cyclic and the rest of the factors
are locally nite, then G includes a normal soluble subgroup S of nite index.
Moreover, |G/S| a(f1 (r))r .
Proof. By Corollary 2.8, G has a nite series

1 = K 0 L1 K 1 L2 K 2 Ln K n = G

of normal subgroups such that Lj+1 /Kj are abelian torsion-free subgroups of nite
0-rank, 0 j n 1, and the factors Kj /Lj are nite of order at most f1 (r),
1 j n, n r. Put Cj = CG (Kj /Lj ), 1 j n. Then G/Cj is nite, moreover,
Gj /Cj a(f1 (r)).

Put S = 1jn Cj . By Remaks theorem G/S
Dr1jn G/Cj , in partic-
ular, G/S is nite and |G/S| |G/S1 | |G/Sn | a(f1 (r))n a(f1 (r))r . By its
choice, S is soluble. 

If G is a soluble group, then denote by s(G) the class of solubility of G. Let


G be a nite soluble group and |G| = n = pk11 pkmm . Put d(n) = k1 + + km .
20 Chapter 2. Ranks of groups

Then clearly s(G) d(|G|). We have

d(n) = k1 + + km
= logp1 (pk11 ) + + logpm (pkmm )
log2 (pk11 ) + + log2 (pkmm ) = log2 (pk11 pkmm ) = log2 n.

Thus d(n) log2 n.


Let A be an abelian torsion-free group and G be an automorphism group of
A. We say that A is rationally G -irreducible if for every non-identity G-invariant
subgroup B of A the factor-group A/B is periodic.
Lemma 2.10. Let G be a group and suppose that G includes a nite normal sub-
group T such that G/T is an abelian torsion-free group of nite 0-rank r. Then G
includes a characteristic abelian torsion-free subgroup A such that G/A is nite.
Moreover, if |T | = t, then |G/A| a(t)t2r+1 .
Proof. Put C = CG (A); then G/C is nite and |G/C| d = a(t). If g C ,
then clearly g t (C), so that |C/(C)| tr . A subgroup (C) is abelian and
has a nite periodic part, so that (C) = K Z for some torsion-free subgroup
Z where K = T (C) ( see, for example, [78, Theorem 27.5]). It follows that
A = ((C))t Z, in particular, A is torsion-free. Clearly, A is a characteristic
subgroup of G. Finally, |G/A| a(t)tr ttr = a(t)t2r+1 . 
Lemma 2.11 ([55]). Let G be a group and suppose that t(G) = 1. Assume also
that G has an ascending series

H = H 0  H 1  H  H +1  H = G

such that r factors of this series are innite cyclic and the rest of the factors
are locally nite. Then G has normal subgroups L K S G such that L is
nilpotent and torsion-free, K/L is abelian torsion-free and nitely generated, G/K
is nite and S/K is soluble. Moreover, there are functions f2 , f3 : N N such
that |G/K| f2 (r) and s(S) f3 (r).
Proof. By Corollary 2.9, G includes a maximal normal soluble subgroup S of nite
index such that |G/S| a(f1 (r))r . Let L be the locally nilpotent radical of G.
Since t(G) = 1 and the set of all elements having nite order of a locally nilpotent
group is a characteristic subgroup, L is torsion-free. By Maltsevs theorem [182]
the subgroup L is nilpotent. It is not hard to prove that L has a nite series

1 = A0 A1 An = L

of G-invariant subgroups which are L-central and all the factors Aj+1 /Aj are
abelian torsion-free and rationally G-irreducible, 0 j n 1.
Put Cj = CG (Aj+1 /Aj ), 0 j n 1. Then we can consider Gj = G/Cj as
an irreducible soluble-by-nite subgroup of GLk (Q) where k = r0 (Aj+1 /Aj ). Let
Chapter 2. Ranks of groups 21

Sj be a normal soluble subgroup of Gj having nite index. Then we can consider


Sj as a completely irreducible soluble subgroup of GLk (Q). By a Maltsev theorem
[182] Sj is abelian-by-nite. By a result due to V.S. Charin [36], Sj includes a
normal free abelian subgroup of nite index. Hence Gj includes a normal free
abelian subgroup Uj such that Gj /Uj is nite. Put Dj = CGj (Uj ), then Gj /Dj
is nite, moreover |Gj /Dj | f1 (r). A subgroup Dj is central-by-nite, therefore
by a Schur theorem (see, for example, [234, Theorem 4.12]) the set Tj of all its
elements having nite order is a nite characteristic subgroup. By the choice of Gj
we have |Tj | f1 (r). By Lemma 1.7, Dj includes a characteristic abelian torsion-
free subgroup Kj such that Dj /Kj is nite. Moreover, |Dj /Kj | a(t)t2r+1 , where
t = f1 (r). Hence Gj /Kj is nite and |Gj /Kj | a(t)t2r+2 .
Put C = 0jn1 Cj . By Remaks theorem G/L Dr0jn1 G/Cj , in
particular, G/C includes a nitely generated torsion-free abelian subgroup K/C
such that G/K is nite. Moreover,

|G/K| |G/Kj |n (a(t)t2r+2 )n (a(t)t2r+2 )r


= (a(f1 (r))r (f1 (r))(2(r+1)r = f2 (r).

It is not hard to prove that C L. Finally, clearly L is nilpotent of class nilpotency


at most r 1. We have now

s(S) s(K) + s(S/K) = r 1 + 1 + s(S/K) = r + s(S/K).

The factor-group S/K is nite and soluble, hence s(S/K) d(|S/K|). From
the choice of S it follows that s(S/K) d(a(f1 (r))(f1 (r))2(r+1) ). Now we may put
f3 (r) = r + d(a(f1 (r))(f1 (r))2(r+1) ) r + log2 (a(f1 (r))(f1 (r))2(r+1) ). 

Another way of calculating the function f3 (r) follows from the proof above.
By a Maltsev theorem [182] an irreducible soluble subgroup of GLr (Q) includes
a normal abelian subgroup of nite index at most (r) for some function :
N N . It follows that s(S) r + d((r)). Note that (r) r!(r2 a(r2 ))r (see,
for example, [280, pp. 45]), so that we may put f4 (r) = r + d(r!(r2 a(r2 ))r )
r + log2 ((r!(r2 a(r2 ))r ). Thus s(S) f4 (r).
Theorem 2.12 ([55]). Let G be a group and suppose that G has an ascending series

H = H 0  H 1  H  H +1  H = G

such that r factors of this series are innite cyclic and the other factors are locally
nite. Then G has normal subgroups T L K S G such that T is
locally nite, L/T is nilpotent and torsion-free, K/L is abelian torsion-free and
nitely generated, G/K is nite and S/K is soluble. Moreover, there are functions
f2 , f3 : N N such that |G/K| f2 (r) and s(S/T ) f3 (r).
Proof. Put T = t(G) and apply Lemma 2.11 to the factor-group G/T . 
22 Chapter 2. Ranks of groups

A group G is said to be generalized radical , if G has an ascending series whose


factors are locally nilpotent or locally nite. It is not hard to see that a generalized
radical group G has an ascending series of normal subgroups, whose factors are
locally nilpotent or locally nite. Note that a periodic generalized radical group is
locally nite. Therefore a periodic locally generalized radical group is locally nite
too. Thus we have
Corollary 2.13 ([55]). Let G be a locally generalized radical group. If G has nite
0-rank r, then G has normal subgroups T L K S G such that T is
locally nite, L/T is nilpotent and torsion-free, K/L is abelian torsion-free and
nitely generated, G/K is nite and S/K is soluble. Moreover, |G/K| f2 (r) and
s(S/T ) f3 (r).
Corollary 2.14 ([55]). Let G be a locally generalized radical group and T be the
maximal normal periodic subgroup of G. If G has a nite 0-rank r, then G/T
has nite special rank. Moreover, there is a function f5 : N N such that
r(G) f5 (r).
Indeed we can put f5 (r) = r + f2 (r).
Corollary 2.15 ([55]). Let G be a locally (soluble-by-nite) group. If G has a nite
0rank r, then G has normal subgroups T L K S G such that T is
locally nite, L/T is nilpotent and torsion-free, K/L is abelian torsion-free and
nitely generated, G/K is nite and S/K is soluble. Moreover, |G/K| f2 (r) and
s(S/T ) f3 (r).
This assertion is an essential specication [73, Lemma 2.12].
Corollary 2.16 (D.J.S. Robinson [245]). Let G be a locally (soluble-by-nite) group
of nite 0-rank. Then r0 (G) r(G).
Proof. Put r0 (G) = r. By Corollary 2.15, G has normal subgroups T L K
S G such that T is locally nite, L/T is nilpotent and torsion-free, K/L is
abelian torsion-free and nitely generated, G/K is nite and S/K is soluble. By
[293, Theorem 1], r0 (K/T ) = r(K/T ). It follows that r0 (G/T ) r(G/T ), and
because r0 (G/T ) = r0 (G) we have r0 (G) r(G). 

We consider now the relationship between the 0-rank, and the p-rank of a
group. If A is an abelian group of nite p-rank, then, obviously, A has nite 0-
rank, and moreover r0 (A) rp (A). Now we show that the situation is the same
for soluble groups.
Lemma 2.17. Let G be a nilpotent torsion-free group of nite p-rank for some
prime p. Then G has nite 0-rank. Moreover r0 (G) 12 (rp (G) + 1)rp (G).
Proof. The group G has a central series

1 = Z0 Z1 Zn = G
Chapter 2. Ranks of groups 23

every factor of which is torsion-free (see D.J.S. Robinson [234, Theorem 2.25]).
Since every free abelian group of 0-rank s has an elementary abelian factor-group
of order ps , every factor of this series has nite 0-rank. Therefore, G has nite
0-rank. Put rp (G) = k. Let A be a maximal abelian normal subgroup of G. Then
CG (A) = A, and A is a pure subgroup. Since G is nilpotent, G/CG (A) can be
embedded in the unitriangular group U Tm (Q), where m = r0 (A). It follows that
1
r0 (G/CG (A)) (m 1)m
2
and so
1 1 1
r0 (G) m + (m 1)m = (m + 1)m (k + 1)k,
2 2 2
as required. 
Theorem 2.18. Let G be a soluble group of nite section p-rank for some prime p.
Then G has nite 0-rank. Moreover r0 (G) 12 (rp (G) + 3)rp (G).
Proof. If U/V is a torsion-free abelian section of G, then clearly r0 (U/V ) rp (G).
It follows that G has nite 0-rank. Then G has a series of normal subgroups
T L G satisfying the following conditions: T is periodic, L/T has a nite
subnormal series, every factor of which is torsion-free abelian locally cyclic, and
G/L is nite. Without loss of generality, we may assume that T = 1. By V.S.
Charin [36, Theorem 5], G has a series of normal subgroups L S G such that
L is a torsion-free nilpotent subgroup, S/L is a free abelian group of nite 0-rank,
and G/S is nite. By Lemma 2.9,
1
r0 (L) (rp (G) + 1)rp (G),
2
and hence
1
r0 (G) r0 (L) + r0 (S/L) (rp (G) + 1)rp (G) + rp (G)
2
1
(rp (G) + 3)rp (G).
2 
Chapter 3

Some generalized nilpotent groups

In this book we will consider modules over group rings when the groups have cer-
tain features, for example, are near to nilpotency. In particular, we will consider
generalized nilpotent and generalized hypercentral groups, whose specic proper-
ties are the goals of this chapter. Nilpotent and hypercentral groups have arisen
from abelian groups with the help of the upper central series. We use a similar
construction for the groups that seem to be natural extensions of abelian groups.
We begin with the following generalization of an abelian group. Let G be a
group, and let S be a G-invariant subset of G. This means that for every a S
and g G, we have ag S. It follows that CG (S) is a normal subgroup of G.
The corresponding factor-group CocG (S) = G/CG (S) is called the cocentralizer
of the G-invariant set S in the group G (L.A. Kurdachenko [149]). Note that the
group G/CocG (S) is isomorphic to some subgroup of Aut(S). The inuence of the
cocentralizers of many objects related to a group on the structure of the group itself
is a subject of study in many branches of Group Theory. For instance, in the theory
of nite groups, many examples have been developed with the consideration of the
cocentralizers of chief factors, which play there a signicant role. For example, we
can mention Formation Theory since local formations are dened by cocentralizers
of the chief factors of the groups involved. In the theory of innite groups, many
types of groups appear in studying cocentralizers of their conjugacy classes. This
is our point of origin.
Let X be a class of groups. A group G is said to have X-conjugacy classes (or
G is anXC-group) if CocG (gG ) X for each g G. Thus, if X = I is the class of
all identity groups, then the class of IC-groups is exactly the class A of all abelian
groups. Therefore, the class of XC-groups is a generalization of the class of abelian
groups associated to the given class X. If X = F is the class of all nite groups,
then the FC-groups are the groups with nite conjugacy classes or F C-groups. This
class is a suitable extension both of F and A, so that it inherits many properties
of these classes. Currently, the theory of F C-groups is one of the best-developed
branches of the theory of innite groups, and many authors have made signicant
contributions to it (see the books Yu.M. Gorchakov [90], S.N. Chernikov [43],
26 Chapter 3. Some generalized nilpotent groups

D.J.S. Robinson [234, 242], M.J. Tomkinson [272], and the updated survey paper
M.J. Tomkinson [274]). Also the inuence of F C-groups on the structure of some
rings was a widely popular topic (see G. H. Cli and S.K. Sehgal [44], I.N. Herstein
[113], D.S. Passman [216, 217, 218], C. Polcino-Milies [222, 223], C. Polcino-Milies
and S.K. Sehgal [224], S.K. Sehgal [258, 259], S.K. Sehgal and H.J. Zassenhaus
[260, 261]). If X = C is the class of all Chernikov groups, then CC-groups are
the groups with Chernikov conjugacy classes or CC-groups. Ya.D. Polovicky [225]
introduced this class and obtained some initial results. Although CC-groups have
not been investigated as far as F C-groups, they are subjects of many recent papers
(see J. Alc
azar and J. Otal [1], S. Franciosi, F. de Giovanni and M.J. Tomkinson
[77], M. Gonz alez and J. Otal [86, 87, 88], M. Gonz alez, J. Otal and J.M. Pena
[89], J. Otal and J.M. Pena [209, 210, 211, 212, 213], J. Otal, J.M. Pe
na and M.J.
Tomkinson [214]). The study of XC-groups for other important classes X has only
recently begun (see S. Franciosi, F. de Giovanni and L.A. Kurdachenko [73], S.
Franciosi, F. de Giovanni and M.J. Tomkinson [76], L.A. Kurdachenko [149, 151],
L.A. Kurdachenko, V.N. Polyakov and I.Ya. Subbotin [161], L.A. Kurdachenko
and J. Otal [156], and L.A. Kurdachenko and I.Ya. Subbotin [169]).
The nature of F C-groups allows one to construct distinct types of such groups
using nite and abelian groups. Note that a direct product of F C-groups is also
an F C-group. In particular, any direct product of nite groups and abelian groups
is an F C-group. Moreover, let G = Cr G be the Cartesian product. Put

Cdr G = {(g ) | g (G ) for all \g where g is nite}.

The group Cdr G is called the central direct product of the groups G , .
Clearly, a central direct product of nite and abelian groups is an F C-group.
Central-by-nite groups and groups with nite derived subgroups (nite-by-abel-
ian groups) deliver other examples of F C-groups. However, not every abelian-by-
nite group is an F C-group (but it is F C-nilpotent). Other classes of XC-groups
are more diverse. For example, a central direct product of Chernikov groups is a
CC-group, and a central-by-Chernikov group is a CC-group. However, not every
group with a Chernikov derived subgroup is a CC-group. This is true only for
periodic groups. This fact follows from the assertion that a periodic automorphism
group of a Chernikov group is Chernikov (see, for example [43, Corollary 1.18]).
Let G be a group. If X is a class of groups, then we put

XC(G) = {x G | CocG (xG ) X}.

In general XC(G) is not a subgroup of G, although, depending on X, the set XC(G)


enjoys richer properties. We select the most common classes of groups to study.
A class of groups X is said to be a formation of groups if the following conditions
are satised:
(F1) If G X , and H is a normal subgroup of G, then G/H X; and
Chapter 3. Some generalized nilpotent groups 27

(F2) If H, L are normal subgroups of a group G such that G/H, G/L X, then
G/(H L) X.
Actually, we have
Lemma 3.1. Let X be a formation of groups. If G is a group, then XC(G) is a
characteristic subgroup of G.
Proof. Since
CG (x1 G ) = CG (xG )
and
CG (xyG ) CG (xG ) CG (yG ),
for every x, y G, and X is a formation of groups, it readily follows that XC(G)
is a subgroup of G. If Aut G, then

CG (xG ) = CG (xG )

and thus
G/CG (xG )
= G/CG (xG )
for each x G. Thus XC(G) is a characteristic subgroup of G. 

If X is a formation of groups, then the subgroup XC(G) is said to be the


XC-center of the group G. The reason for this term appears to be very clear. If
X = I, F and C, we obtain the ordinary center (G), the F C-center F C(G) , and
the CC-center CC(G) of the group G, respectively. Of course, it is worth noting
that the group G is an XC-group if and only if G = XC(G).
Starting from the XC-center of a group G, we construct the upper XC-central
series of the group G,

1 = C0 C1 C C+1 C

where C1 = XC(G), C+1 /C = XC(G/C ) if < , and XC(G/C ) = 1.


The term C of the series is said to be the th -XC-hypercenter of G, whereas the
last term C is called the upper XC-hypercenter of G, and denoted by XC (G).
If XC (G) = G, then G is said to be XC-hypercentral . If moreover, is -
nite, then G is called XC-nilpotent . Once again, if X = J, X = F, and X = C,
then XC (G) = (G), XC C = F C (G), and XC (G) = CC (G), which
are called the upper hypercenter , the upper F C-hypercenter , and the upper CC-
hypercenter of the group G, respectively.
Lemma 3.2. Let X be a formation of groups closed under taking subgroups (that
G G
is, X = SX). If G is a group, g1 , . . . , gs XC(G), and H = g1  gs  , then
H is a central-by-X-group.
28 Chapter 3. Some generalized nilpotent groups

Proof. Clearly,

C = CG (g1 G ) CG (gs G ) H (H).

Moreover
G G
H/C
H/(H CG (g1  )) H/(H CG (gs  )).

Since

H/(H CG (gj G )) G G G
= HCG (gj  ))/CG (gj  ) G/CG (gj  )),

1 j s, we obtain that H/(H) X, as required. 


Corollary 3.3. Let X be a formation of locally nite groups closed under taking
subgroups, and let G be an XC-group. Then
(1) The commutator subgroup [G, G] is locally nite.
(2) If G is torsion-free, then G is abelian.
(3) If H is a torsion-free normal subgroup, then H (G).
G
(4) If g G and L = g , then either L is locally nite or L includes a G-
invariant locally nite subgroup T such that H/T is innite cyclic.
G G
Proof. (1) Let g1 , . . . , gs G , and put H = g1  gs  . By Lemma 3.2, H
is a central-by-X-group. As a consequence of a result due to Schur (see D. J.
S. Robinson [234, Corollary of Theorem 4.12]), [H, H] is locally nite. Therefore
[G, G] is locally nite.
(2) and (3) are obvious.
(4) By (1), [L, L] is locally nite. It is immediate that the elements of nite
order of the abelian group L/[L, L] form a characteristic subgroup T /[L, L]. It
follows that T is a G-invariant subgroup of L , and either L/T is a non-identity
torsion-free abelian or T = L. In the rst case, we note that L/T (G/T ) by
G/T
(3). Note that L/T = gT  so that L/T = gT . 
G G
Corollary 3.4. Let G be a group, x1 , . . . , xs F C(G), and H = g1  gs  .
Then:
(1) H is a nitely generated subgroup.
(2) H is central-by-nite.
(3) Either H is nite or H includes a G-invariant nite subgroup T such that
H/T is a nitely generated free abelian group.
Proof. Since each xj F C(G), G/CG ((xj )G ) is nite, in particular, |G : CG (xj )|
is nite. It follows that each xj has nitely many conjugates in G, and then H is
nitely generated, which gives (1).
Chapter 3. Some generalized nilpotent groups 29

(2) follows from Lemma 3.2.


(3) By a Theorem of Schur (see D. J .S. Robinson [234, Corollary of Theorem
4.12]), the commutator subgroup [H, H] is nite. The group H/[H, H] is nitely
generated and abelian, hence its elements of nite order form a characteristic
nite subgroup T /[H, H]. Then H/T is a torsion-free nitely generated abelian
group. 
Our next goal is to obtain some elementary properties of the class of XC-
groups, when X is a formation of groups, which is a slight extension of the class F
of nite groups.
Lemma 3.5. Let G be a group, and suppose that G/C is an abelian Chernikov
group, where C = (G). Assume that H is a subgroup of G such that H C,
H/C = h1 C hn C and |hj C| = psj , where p is a prime and sj N for
1 j n. Then [G, H] is a nite p-subgroup of C, and rp ([G, H]) nrp (G/C).
Proof. If 1 j n, we consider the mapping j : G C given by gj =
[g, hj ], g G. We have

(g1 g2 )j = [g1 g2 , hj ] = (g2 )1 [g1 , hj ]g2 [g2 , hj ]


= [g1 , hj ][g2 , hj ] = g1 j g2 j

because [g1 , hj ] (G). Hence j is a homomorphism so that Im j = [G, hj ] and


Ker j = CG (hj ) are normal subgroups of G. Furthermore,

[g, h2j ] = [g, hj ](hj )1 [g, hj ]hj = [g, hj ]2


sj
and it follows that [g, htj ] = [g, hj ]t for every t N. Since hpj C, we have
sj
psj
1= [g, hpj ] = [g, hj ] . By the isomorphism

[G, hj ] = Im j
= G/ Ker j = G/CG (hj )

and the inclusion C CG (hj ), we obtain that [G, hj ] is a nite p-subgroup, and
rp ([G, hj ]) rp (G/C).
It is easy to see that

[G, H] = [G, h1 ] [G, hn ].

It follows that [G, H] is a nite p-subgroup, and rp ([G, H]) nrp (G/C). 
Corollary 3.6. Let G be a group, and suppose that G/C is an abelian Chernikov
group, where C = (G). If P/C is the Sylow p-subgroup of G/C, where p is a
prime, then [G, P ] is a Chernikov p-subgroup of C, and rp ([G, P ]) rp (G/C)2 .
Proof. We have
P/C = U1 /C Un /C
30 Chapter 3. Some generalized nilpotent groups

where U1 /C, . . . , Un /C either are Pr


ufer p-subgroups or cyclic p-subgroups. There-
fore, P has an ascending chain of subgroups

G = P0 P1 Ps

such that Ps /C is a direct product of n cyclic subgroups , and P = sN Ps . It
follows that [G, P ] has an ascending chain of subgroups

[G, P1 ] [G, P2 ] [G, Ps ] ,



such that [G, P ] = sN [G, Ps ] . By Lemma 3.5, [G, Ps ] is a nite p-subgroup of
C , and rp ([G, Ps ]) n2 , for each s N. Thus rp ([G, P ]) n2 . This implies that
[G,P] is Chernikov (see L. Fuchs [78, Theorem 25.1]). 
Corollary 3.7. Let G be a group, and suppose that G/C is an abelian Chernikov
group, where C = (G). Then the commutator subgroup [G, G] is Chernikov.
Proof. Suppose that
G/C = Drp(G/C) Sp /C
where Sp /C is the Sylow p-subgroup of G/C. By Corollary 3.6, each [G, Sp ] is a
Chernikov p-subgroup. It is easy to see that

[G, G] = Drp(G/C) [G, Sp ].

Since (G/C) is nite, [G, G] is Chernikov. 


Lemma 3.8. Let G be an abelian-by-nite group, and put C = (G). If G/C is
Chernikov, then [G, G] is Chernikov.
Proof. Let A be a normal abelian subgroup of nite index, and suppose that
{x1 , . . . , xs } is a transversal to H in G. For every 1 j s, we consider the
mapping j : A A given by aj = [gj , a], a A. Since

(a1 a2 )j = [gj , a1 a2 ] = [gj , a2 ]a1


2 [gj , a1 ]a2
= [gj , a1 ][gj , a2 ] = a1 j a2 j ,

j is an endomorphism of A. Therefore

[gj , A] = Im j
= A/ Ker j = A/CA (gj ).

Since CA CA (gj ), A/CA (gj ) is Chernikov , and it follows that for any 1 j s
[gj , A] is Chernikov too. Put

B = [g1 , A] [gs , A]

so that B is likewise Chernikov. If g G, then there is some 1 j s such that


g gj A , and so g = gj u, for some u A. Then

[g, a] = [gj u, a] = u1 [gj , a]u[u, a] = [gj , a] [gj , A].


Chapter 3. Some generalized nilpotent groups 31

This means that [G, A] = B. Thus A/B (G/B), and, in particular, G/B is
central-by-nite. By a classical result due to Schur (see D.J.S. Robinson [234,
Theorem 4.12]), the commutator subgroup K/B = [G/B, G/B] is nite. Hence K
is Chernikov, and G/K is abelian. 
Theorem 3.9. Let G be a group. If G/(G) is a Chernikov group, then the com-
mutator subgroup [G, G] is likewise Chernikov.
Proof. Put C = (G), and let D/C be the divisible part of G/C. By Corollary
3.7, [D, D] is Chernikov. Since G/[D, D] is abelian-by-nite, it suces to apply
Lemma 3.8. 
The above result is a generalization of the already mentioned theorem of
Schur. Now we are making use of it to describe the normal closures of elements of
the CC-groups.
Lemma 3.10 (L.A. Kurdachenko [144]). Let G be a hypercentral group. Suppose that
A is an abelian normal p-subgroup of G, where p is a prime, such that G/CG (A)
has no subgroups of index p. Then A (G).
Proof. Suppose the contrary, that is (G) contains no A. Then A(G)/(G) is a
non-identity normal subgroup of the hypercentral group G/(G), and hence there
exists some a (A 2 (G)) \ (G). Since A is a p-subgroup, we may assume that
ap (G). Then the mapping : g [g, a], g G is an endomorphism of G such
that [G, a] = Im = 1 since a  (G). Actually, since ap (G), [G, a]p = 1,
and so Im is an elementary abelian p-subgroup. Since
[G, a] = Im
= G/ Ker = G/CG (a),
we deduce that G/CG (a) is an elementary abelian p-group as well. Since CG (A)
CG (a), and G/CG (A) has no subgroups of index p, this is a contradiction. 
Lemma 3.11. Let G be a CC-group, and suppose that A is a normal abelian p-
subgroup of G, where p is a prime. Then G/CG (A) has no abelian divisible sub-
groups.
Proof. Suppose the contrary, and let D/CG (A) be a non-identity abelian divisible
subgroup. Pick g D \ CG (A). Then the mapping g : a [g, a], a A is an
endomorphism of A. For each n N, we put An = p,n (A). Since G is a CC-
group, G/CG (gG ) is a Chernikov group, and, in particular, An /(An CG (gG ))
is Chernikov. This means that An /(An CG (g)) is likewise Chernikov. Since the
orders of the elements of An /(An CG (g)) are bounded, this factor-group has
to be nite. As D/CG (A) is abelian, g is in fact a ZD-endomorphism. Now
D/CD (An /(An CG (g)) is nite, and CD (An /(An CG (g)) CG (A), so, we
deduce D = CD (An /(An CG (g)). In other words, [D, An ] An CG (g). As
the above argument holds for any n N, we conclude that [D, A] CA (g). This
means that 
[D, A] CA (g) = CA (D),
gD
32 Chapter 3. Some generalized nilpotent groups

so that [D, [D, A]] = 1. By Lemma 3.8, A (D), a contradiction. 


Corollary 3.12. Let G be a CC-group, and suppose that A is a periodic abelian
normal subgroup of G. Then G/CG (A) has no abelian divisible subgroups. In par-
ticular, if G/CG (A) is Chernikov, then it is nite.
Proof. Suppose the contrary, and let D/CG (A) be a non-identity divisible abelian
subgroup. If p (A), we decompose A = Ap Ep , where Ap is the Sylow p-
subgroup of A, and Ep is the Sylow p -subgroup of A. By Lemma 3.11, [D, A] Ep ,
and then 
[D, A] Ep = 1 .
p(A)

Therefore D CG (A), a contradiction. 


Theorem 3.13 (Ya.D. Polovicky [225]). Let G be a CC-group. If g G and H =
< g >G , then either H is a Chernikov subgroup or H has a G-invariant Chernikov
subgroup T such that H/T is innite cyclic.
Proof. By Lemma 3.2, H is central-by-Chernikov, and, by Theorem 3.9, [H, H] is
Chernikov. Thus, we may assume that H is abelian. Let T be the periodic part
of H (denition below). By Corollary 3.5, H/T is an innite cyclic group. If H
is periodic, then G/CG (H) is nite by Corollary 3.12. It follows that g F C(G).
Thus H is nite by Corollary 3.4.
Suppose now that H = T . Then H = g, T , and the order of g has to
be innite. Let D/CG (H) be the divisible part of G/CG (H). By Corollary 3.12,
D CG (T ). Also H/T (G/T ) by Corollary 3.3. Consider the mapping g :
d [g, d], d D. If d1 , d2 D, then

(d1 d2 )g = [g, d1 d2 ] = [g, d2 ]d1


2 [g, d1 ]d2
= [g, d1 ][g, d2 ] = d1 g d2 g

and therefore g is an endomorphism of D. Thus, Im g = [g, D], and Ker g =


G
CD (g) are D-invariant. Since G is a CC-group, G/CG (g ) is a Chernikov group,
so that D/(D CG (g)) is Chernikov too. As

[g, D] = Im g
= D/ Ker g = D/CD (g),
[g, D] is Chernikov. Let x G. Since H/T (G/T ), there exists some v T
such that g x = gv. If d D, then

[g, d]x = [g x , dx ] = [gv, dx ] = [g, dx ]v [v, dx ]


= [g, dx ] [g, D],

since D is normal. This means that [g, D] is a normal subgroup of G. Since


D CG (T ) , and H = g, T , D/[g, D] CG/[g,D] (H/[g, D]). In other words,
(G/[g, D])/CG/[g,D] (H/[g, D]) is nite. By Corollary 3.4, T /[g, D] is nite, and
then T is Chernikov. 
Chapter 3. Some generalized nilpotent groups 33

Lemma 3.14. Let G be a CC-group. If (G) = 1 , then G is periodic.


Proof. Since (G) = 1,
 G
CG (g ) = 1 .
gG

Moreover, every factor-group G/CG (gG ) is Chernikov since G is a CC-group.


Then, if S G a nite subset of non-identity elements of G, there is a normal
subgroup U of G such that S U = , and G/U is a Chernikov group. Let
1 = x G, and put X = xG . By Theorem 3.13, either X is Chernikov or X has
a G-invariant Chernikov subgroup Y such that X/Y = xY . Let C =SocG (X).
Since C Y , C is nite. Therefore, there exists a normal subgroup H such that
H C = 1, and G/H is Chernikov. Put B = H X, and K = [G, B]. By
Corollary 3.3, the commutator subgroup [G, G] is locally nite , and then K Y .
If K = 1, then K C = 1. But B H, and H is a normal subgroup of G,
hence K H , and it follows that K C = 1. This means that [G, B] = 1,
that is B (G) = 1 , and H X = 1. Since

X
= X/(X H)
= XH/H G/H,

X is Chernikov. In particular, the order of x has to be nite. 


By Theorem 3.13, every CC-group is F C-hypercentral. To nish this chapter,
we deduce some elementary properties of F C-hypercentral groups.
Lemma 3.15. Let X be a formation of groups. Given a group G, we put P =
XC(G), and Q = XC (G). If H is a non-identity G-invariant subgroup of Q,
then H P = 1.
Proof. Let
1 = C0 C1 C C+1 C = Q
be the upper XC-central series of G. There exists an ordinal such that H C =
1. Let be the least ordinal with this property. Clearly cannot be a limit,
G
so that H C1 = 1. Pick 1 = x H C , and put X = x . Since
xC1 C /C1 , G/CG (XC1 /C1 ) X. Let g CG (XC1 /C1 ). Then
[g, X] C1 . On the other hand [g, X] H since H is a normal subgroup of G.
Thus [g, X] H C1 = 1, which means that CG (XC1 /C1 ) CG (X).
In particular, G/CG (X) X since X is a formation. Hence X P , and this gives
the required result. 
Corollary 3.16. Let X be a formation of groups. If A is a non-identity normal
subgroup of an XC-hypercentral group G, then A XC(G) = 1.
Lemma 3.17. Let X be a formation of nite groups closed under taking subgroups,
and let G be a nitely generated XC-nilpotent group. Then G is a nilpotent-by-X-
group. In particular, G is polycyclic-by-nite.
34 Chapter 3. Some generalized nilpotent groups

Proof. Let
1 = F0 F1 Fn = G
be the upper XC-central series of G. It suces to show that G/CG (Fj+1 /Fj ) X,
for every 0 j n 1. We carry this out by induction on n.
If n = 1, then G is an XC-group , and it suces to apply Lemma 3.2. Let
now n > 1, and assume that we have already proved that G/CG (Fj+1 /Fj ) X
for every 1 j < n. Put

H = CG (F2 /F1 ) CG (Fn /Fn1 ).

Then G/H X, and H/F1 is nilpotent (see O. H. Kegel and B.A.F. Wehrfritz [134,
Theorem 1.C.1]). In particular, G/F1 is nitely presented (see D.J.S. Robinson
[234, Corollary 1.43]), and it follows that
G G
F1 = g1  gs 

where g1 , . . . , gs XC(G) (D.J.S. Robinson [234, Corollary 1.43]). Put


G G
U = CG (g1  ) CG (gs  ).

Then G/U X, and U = CG (F1 ). If C = H U , then G/C X, and C is nilpotent


(see O.H. Kegel and B.A.F. Wehrfritz [134, Theorem 1.C.1]). 
Corollary 3.18. A nitely generated F C-nilpotent group G is nilpotent-by-nite.
In particular, G is polycyclic-by-nite.
Proposition 3.19. Let X be a formation of nite groups closed under taking sub-
groups, and let G be an XC-hypercentral group. If Y is a nitely generated subgroup
of G, then Y is a nilpotent-by-X-group.
Proof. Let
1 = G0 G1 G G = G
be the upper XC-central series of G. We consider the following set of ordinals,

= { | Y /(Y G ) is a nilpotent-by-X-group}.

Since 
Y = (Y G ),

= . Let be the least element of . If = 0, then Y is a nilpotent-by-X-group,


as claimed. Suppose that > 0. If is not a limit, then

(Y G )/(Y G1 ) = (Y G )/(G Y G1 )

= (Y G )G1 /G1 ,
Chapter 3. Some generalized nilpotent groups 35

and, in particular,

(Y G )/(Y G1 ) XC(Y /(Y G1 ).

Since Y /(Y G ) is XC-nilpotent, by Lemma 3.17, Y /(Y G1 ) is XC-nilpotent


too. This fact contradicts the choice of .
Hence has to be a limit ordinal whenever > 0. Since Y /(Y G ) is nitely
presented,
Y G = y1 , . . . , ys G ,
for some y1 , . . . , ys Y G (see D.J.S. Robinson [234, Corollary 1.43] again).
Then there exists an ordinal < such that y1 , . . . , ys G , so that Y G =
Y G , and we get a new contradiction, which shows that the case > 0 cannot
occur. 
Corollary 3.20. Let G be an F C-hypercentral group. If X is a nitely generated
subgroup of G, then X is nilpotent-by-nite.
Chapter 4

Artinian modules and the socle

In studying further properties of artinian modules, the concept of the socle of the
module has appeared to be very functional. We develop here a slight extension of
this concept, which in its original form is due to R. Remak [226].
Let R be a ring, and let A be an R-module. The submodule SocR (A) gener-
ated by all minimal R-submodules of A is said to be the R-socle of A. If A has no
such minimal submodules, we dene SocR (A) = 0.
Lemma 4.1. Let R be a ring, and let A be an R-module. If M is a family of minimal
R-submodules of A and B is the submodule generated by all members of M, then
B is a direct sum of some elements of M.
Proof. If = S M, then S is said to be independent if the submodule S
generated by all members of S is the direct sum of them, that is, S = XS X.
Clearly, every atomic subfamily is independent. Let L be the set of all independent
subfamilies of M. If S is a linearly ordered (by inclusion) subset of L, we put

C= H
HS

and denote by C the submodule generated by all members of C. Given a nite


subset Z = {X1 , . . . , Xr } of C, pick some c Xj Xi | l = j. Since S is linearly
ordered, there is some D S such that Z D. As the submodule generated
by all members of D is the direct sum XD X, it follows that c = 0, so that
C = XC X, and therefore C is also independent. By Zorns Lemma, M contains
a maximal independent
subfamily
R.
Put E = XR X = XR X. If we assume that E = B, then there is a
minimal R-submodule U M such that E contains no U . Then U E = 0 , and
so E + U = E U . This means that R {U } is an independent family, which
contradicts the choice of R. This contradiction shows that E = B, as required. 
Corollary 4.2. Let A be a module over a ring R. Suppose that A =SocR (A). If B
is an R-submodule of A, then:
38 Chapter 4. Artinian modules and the socle

(1) A/B =SocR (A/B).


(2) A = B M whenever M is a maximal R-submodule of A under M B = 1.
(3) If M denotes the family of all minimal R-submodules of B, then B is gener-
ated by all members of M.
Proof. (1) is obvious.
(2) Suppose that A = B M . Then A/M = BM/M . Since A/M is gener-
ated by minimal R-submodules, there is a minimal R-submodule U/M such that
BM/M contains no U/M . Hence U/M BM/M = 0 , and then U BM = M .
Thus M = U BM = M (U B); that is, U B M . It follows that

U B = M (U B) = (M U ) B = M B = 0 ,

which contradicts the choice of M . Therefore A = B M .


(3) is an immediate consequence of (1) and (2). 
Corollary 4.3. Let R be a ring, and let A be an R-module. Then the following
statements are equivalent.
(1) A is a sum of simple R-submodules.
(2) A is a direct sum of simple R-submodules.
(3) For every R-submodule B, there is an R-submodule C such that A = B C.
Proof. (1) (2) has been proved in Lemma 4.1.
(2) (3). Let B be a non-zero R-submodule of A. By Corollary 4.2, B =
M , where M is a simple R-submodule for each . This means that the
family {M | } is independent. Then there is a maximal independent
family
M, such that {M | } M. We have already seen that A = MM M . If
L = M \ {M | } , and C = ML M , then A = B C, as required.
(3) (1). Let 0 = a A, and put E = aR. Obviously a does not lie
in any proper R-submodule of E. Therefore the union of any ascending chain of
proper submodules of E cannot reach E itself. By Zorns Lemma, E has a maximal
proper R-submodule V . In particular, E/V is a simple R-module. By (3), there
is a submodule U such that A = V U . Since V E, the modular law gives
that E = V (E U ). Since E/V = E U , E U is a simple R-submodule. In
particular, S =SocR (A) = 0. If S = A, then A = S W for some R-submodule
W . Again we choose an element 0 = w W , and consider the submodule wR.
We have already proved that wR has a simple R-submodule M . It follows that
M S. But M wR W and W S = 0. This contradiction shows that
A =SocR (A). 
A non-zero module A over a ring R is said to be a semisimple R-module if A =
SocR (A). By Lemma 4.1, A is a semisimple R-module if and only if A is a (direct)
sum of simple R-submodules so that semisimplicity is an equivalent condition to
the corresponding condition of Corollary 4.3. Moreover, it is immediate to obtain
Chapter 4. Artinian modules and the socle 39

Corollary 4.4. Let A be a semisimple R-module. Then:


(1) If B is a an R-submodule of A, then B and A/B are likewise semisimple
R-modules.

(2) If A = A , where every A is a simple R-submodule, and C is a simple
R-submodule of A, then there is an index such that C
= A .
Proposition
4.5. Let R be a commutative ring and A an R-module. Suppose that
A = A , where the A are simple R-submodules isomorphic each to other.
If 0 = a A, then aR is a simple R-submodule.
Proof. We have
a = a1 + + ak ,
where ai A(j) , and (j) , 1 j k. Suppose that a1 = 0. Put B = aR ,
and let be the restriction to B of the canonical projection from A to A(1) . If
ax Ker , then
0 = (ax) = (a)x = a1 x.
Let : A(1) A(1) be the mapping given by b = bx if b A(1) . Then
Ker = 0 because a1 x = 0. Since R is commutative, Ker is an R-submodule
of A(1) . It follows that Ker = A(1) ; that is, A(1) x = 0. Since A(1)
=R A
for every , A x = 0, and so

ax = a1 x + + ak x = 0.

Thus Ker = 0, and hence is an isomorphism from B to A(1) . 

With the aid of the socle we obtain the next characterization of artinian
modules. The group-theoretical analogue of this result was obtained by R. Baer
[5]. Recall that an R-module A has nite composition length if there is a nite
series of submodules
0 = A0 A1 An = A
whose factors Ai+1 /Ai are all simple (a nite composition series). It is known that
a module has nite composition length if and only if it is noetherian and artinian.
However we have
Theorem 4.6. An R-module A is artinian if and only if the socle of every non-zero
factor-module of A is non-zero, and has nite composition length.
Proof. If A is artinian and B is a submodule of A, by Lemma 1.2, A/B is artinian.
It follows that SocR (A/B) = 0 and, by Corollary 4.3, the socle clearly has nite
composition length.
Conversely, let

S0 = 0 , S1 = SocR (A), S+1 /S = SocR (A/S ),


40 Chapter 4. Artinian modules and the socle


and S = < S for every limit ordinal . Let B be a non-zero submodule of
A. Then there is a least ordinal such that C = B S = 0. Clearly, cannot
be a limit ordinal. In particular, S1 B = 0 , and then

C
= C/(C S1 )
= (C + S1 )/S1 S /S1 .

By Corollary 4.4, C is a semisimple submodule. In particular, C contains a non-


zero simple submodule, and then B S1 = 0. Hence every non-zero submodule
of A has non-zero intersection with the R-socle.
Suppose that A has an innite strictly descending chain of submodules

A1 > A2 > > An > .



Replacing A by A/( nN An ), we may assume that nN An = 0. Since S1 has
nite composition length, and An S1 = 0 for each n N, there is some m N
such that Aj S1 = Am S1 for every j m. However, in this case

An Am S1 .
nN

This contradiction proves that A is an artinian module. 


Let A be an R-module. Starting from the socle we dene the upper socular
series or the ascending Loewy series of A as

0 = S0 S1 S S ,

whereS1 =SocR (A), and S+1 /S =SocR (A/S ) for a given ordinal . Note that
S = < S for any limit ordinal . The least ordinal such that S = S+1 is
called the socular height of A.
Lemma 4.7. Let R be a ring and A an R-module. Suppose that

0 = S0 S1 S S

is the upper socular series of the R-module A. If B is a submodule of A having an


ascending series of submodules

0 = B0 B1 B B = B

all of whose factors are semisimple, then B S .


Proof. We proceed by transnite induction. Clearly, B1 SocR (A) = S1 . Let
> 1 , and suppose that B S for all ordinals < . If is a limit ordinal,
then
 
B = < B < S = S .
Chapter 4. Artinian modules and the socle 41

Suppose now that 1 exists. We have


(B + S1 )/S1
= B /(B S1 )
and
B S1 B1 .
By Corollary 4.4, (B + S1 )/S1 is semisimple, and hence
(B + S1 )/S1 SocR (A/S1 ) = S /S1 .
Therefore B S in this case. Applying induction, B = B S . 
Corollary 4.8. Let B be a submodule of an R-module A. If
0 = B0 B1 B
and
0 = S0 S1 S
are the upper socular series of B, and A respectively, then B = S B for any
.
Proof. Indeed B S by Lemma 4.7. Since S B has an ascending series of
length with semisimple factors, Lemma 4.7 yields that S B B , and
then we obtain the equality S B = B . 
Concerning the height of the socular series of A and related questions, we
will show some interesting results due to B. Hartley [105].
Lemma 4.9. The socular height of an R-module A is the supremum of the socular
heights of the cyclic submodules of A.
Proof. Let be the supremum of the socle heights of all cyclic submodules of A.
Let
0 = S0 S1 S S
be the upper socular series of A. Given an ordinal < , then there exists a cyclic
submodule B having an upper socular series
0 = B0 B1 B
such that > . In particular, B = B+1 . By Corollary 4.8,
B = B S < B+1 = B S+1 , so that S = S+1 .
On the other hand, assume that S = S+1 . Then S+1 has a cyclic submodule
D such that D
 S . Let
0 = D0 D1 D
be the upper socular series of D. By Corollary 4.8, D = S D = S+1 D =
D+1 , which contradicts the choice of . Consequently, , and the result has
just been shown. 
42 Chapter 4. Artinian modules and the socle

Corollary 4.10 ([105]). Let R be a ring with cardinality m, and let be the least
ordinal of cardinal greater than m. Then the socular height of any R-module is at
most .
Proof. Let A be an R-module, and pick a cyclic submodule C of A. Then C =
R/AnnR (C). It follows that |C| m, and then the socular height of C is an ordinal
of cardinal at most m. Since is an upper bound for all these ordinals, the result
follows from Lemma 4.9. 

Corollary 4.11 ([105]). Let R be an innite ring with cardinality m, and let m+ be
the least cardinal greater than m. Then every artinian R-module has cardinal at
most m+ .
Proof. Let A be an artinian R-module, and let

0 = S0 S1 S

be the upper socular series of A. Since A is an artinian module, S = A and


S+1 /S has nite composition length for every < . By Corollary 4.10, ,
where is the least ordinal of cardinal m+ . Every factor S+1 /S is a direct sum
of simple R-submodules, and nitely many of them can appear. Since a simple
R-module is a homomorphic image of the R-module R, its cardinal is at most m.
Therefore |S+1 /S | m for each < . Since the number of these factors is at
most m+ , we conclude |A| mm+ = m+ as required. 

We are now in a position to deduce a result due to P.H. Neumann (see B.


Hartley [105]).
Theorem 4.12. An artinian module A over a commutative ring R has socular
height at most , the rst innite ordinal. In particular, if R is countable so is A.
Proof. Let a A. Then aR = R/AnnR (a). Since R is commutative, J = AnnR (a)
is a two-sided ideal of R, and then R/J becomes an artinian ring. It follows
that R/J is also noetherian (see R. Sharp [263, Theorem 8.44]). Hence aR is
simultaneously artinian and noetherian, and therefore it has a nite composition
series. In particular, the upper socular series of aR is nite. It follows from Lemma
4.9 that the socle height of A is at most .
Since A is artinian, A itself is the last term of the upper socle series. Hence
A has an ascending series of submodules

0 = A0 A1 Am = A

whose factors are simple. In particular, An+1 /An is R-isomorphic to some factor
module of the R-module R.
Suppose that R is countable. Then every factor An+1 /An is countable. Hence
A is likewise countable. 
Chapter 4. Artinian modules and the socle 43

Even in the case of artinian modules over countable group rings, we cannot
guarantee that an artinian module is countable, as the following result shows.
Recall that a module A is called uniserial if the set of all submodules of A is well
ordered by inclusion. The ordinal corresponding to the order type of the set of
the proper submodules of a uniserial module is called the length of this module.
Clearly every uniserial module is artinian.
Proposition 4.13 ([105]). Let F be an arbitrary eld, and let G be the free group
freely generated by a countably innite subset X. Then there exists a uniserial mod-
ule A over the group ring F G of length , where denotes the rst uncountable
ordinal.
Proof. Suppose that X = {xn | n N}, and let A be a vector space over F with
basis {a | < }. For each 0 < < , the set of ordinals < is countable, and
so we may choose a bijective mapping f from N onto that set. For each n N,
we dene a linear transformation n of A by
a n = a + af (n) if 0 < < , and a0 n = a0 .
Since f (n) < , every is annihilated by some power of n 1, and so the
power series for (1 + (n 1))1 represents a well-dened linear transformation of
A that will be the inverse of n . Therefore n is an F -automorphism of A, and we
can make A into an F G-module via the mapping xn n , n N.
For each < , let A = < F a . Clearly every A is an F G-submodule
of A, and the proof of this proposition will be completed by showing that the set
of all submodules of A is exactly {A | < }. To do this, let a = < a be
an arbitrary non-zero element of A, where all but nitely many of the coecients
are zero, F , < . We put
Supp a = { | = 0}.
We dene the height of a to be the largest ordinal such that = 0, and prove by
transnite induction that any element of height generates A+1 as F G-module.
This clearly holds with = 0. Let 0 < < , and suppose that the claim holds
for all < . Let

b= a
<

be an element such that Supp b = {1 , . . . , k , }. We may assume that


1 < < k < .
Since f is surjective, f (t) = k for some t N. Then b(t 1) = c + a , where
= k and c is a linear combination of certain a with < k . The F G-submodule
B generated by b contains an element b(t 1), which has height k , and hence
this submodule contains A+1 by induction hypothesis. But then B also contains

a = b a ,

44 Chapter 4. Artinian modules and the socle

where = {1 , . . . , k }, and hence a itself. Finally B contains a (n 1) = a


where = f (n) for all n N. Since f is surjective, A+1 B. Equality is clear.
It now follows that if T is an arbitrary submodule of A and is the largest
ordinal such that A T , then every non-zero element of T has height less than
; that is, T A . Hence T = A , and the proof is complete. 
B. Hartley [105] has also constructed examples of uncountable artinian mod-
ules over group rings. For the readers convenience, we mention here some of his
important results without proof.
Let p = q be primes, and let Q be a Pr ufer q-group. There exists a simple
Fp Q-module A such that CQ (A) = 0 (see L.A. Kurdachenko, J. Otal and I.Ya.
Subbotin [157, Corollary 2.4]). Consider the natural semidirect product G = AQ.
This group is called the Charin group (it was constructed by V.S. Charin [33]).
Theorem 4.14 ([105]). Let G be a Charin group, and let r be a prime such that
r  (G). Then there exists a uniserial Fr G-module of length .
Let D = d  c be the dihedral group of order 8, such that |d| = 4, |c| = 2,
and dc = d1 . For every n Ndenote by Dn = dn   cn  be an isomorphic copy
of D. Put R =DrnN Dn , E = d2n d2n+1 | n N , and U = R/E.

Theorem 4.15 ([105]). Let F be a eld of characteristic other than 2. Then there
exists a uniserial F U -module of length .
Chapter 5

Reduction to subgroups of nite index

Let R be a ring, G be a group, and A be an RG-module. If H is a subgroup of G,


then A is trivially an RH-module, and the following questions naturally arise:
What properties of the RG-module A does the RH-module A inherit?
What can be said of A as an RH-module?
For example, if A is noetherian or artinian as an RG-module, it is rather easy
to see that these properties are no longer true when A is viewed as an RH-module.
Despite this, a famous result due to J.S. Wilson [284] shows that the properties of
being artinian and being noetherian are inherited when H has nite index in G.
This result is the key to signicant applications and further results in the study
of simplicity, the existence of important decompositions, and many other topics.
Also we consider here another main result, the classical Maschkes theorem, which
we will discuss in the most general form.
We will give here a module version of Wilsons theorem.
Let R be a ring, G a group, and A an RG-module. If B is an R-submodule
of A and T is a subset of G, then we dene BT as the RT -submodule generated
by B; that is,

BT = Bt.
tT

Lemma 5.1. Let R be a ring, G a group and A an RG-module. Suppose that H is a


subgroup of G and T is a subset of G such that G = HT . If B is an RH-submodule

of A, then {Bx | x G} = {Bt | t T } and the subsets BT and tT Bt are
RG-submodules.
The proof is immediate.
Theorem 5.2 (J.S. Wilson [284]). Let R be a ring, G a group and A an RG-module.
Suppose that H is a normal subgroup of G having nite index. If A is an artinian
RG-module, then A is an artinian RH-module.
46 Chapter 5. Reduction to subgroups of nite index

Proof. Suppose the contrary: that is, A does not satisfy Min-RH. Then the family
M of all RG-submodules B of A that does not satisfy Min-RH is not empty. Since
A is an artinian RG-module, M has a minimal element M . Let S be the set of all
non-empty nite subsets X of G with the following property: If {Cn | n N } is
a strictly innite descending chain of RH-submodules of M , then M = Cn X for
all n N.
Let T be a transversal to H in G. Then G = HT . By Lemma 5.1, Cn T is an
RG-submodule of M . We claim that Cn T = M . Otherwise, by the choice of M ,
Cn T is an artinian RH-module. It follows that there exists some m n such that
Cm+j = Cm for every j N, contradicting the choice of the chain {Cn | n N}.
Thus our claim has just been proven and so Cn T = M for all n N. Therefore
T S and then S = .
We now choose a minimal element X of S. If x X, then Xx1 S. In
particular, Xx1 is a minimal element of S containing 1. Replacing X by Xx1 , we
may assume that 1 X. Clearly X = 1, so that Y = X \1 is a non-empty nite
subset of G. By the minimality of X, Y  S. Given a strictly innite descending
chain of RH-submodules {Cn | n N} of M , we put En = Cn Cn Y for every
n N. Clearly En is an RH-submodule and En En+1 for every n N. Suppose
that there exists some k N such that Ek+1 = Ek . Since X S, Ck+1 X = M
and

Ck = Ck M = Ck Ck+1 X = Ck (Ck+1 + Ck+1 Y )


= Ck+1 + (Ck Ck+1 Y ) Ck+1 + (Ck Ck Y )
= Ck+1 + Ek = Ck+1 + Ek+1 = Ck+1 .

Thus Ck = Ck+1 , contradicting the choice of {Cn | n N}. This contradiction


shows that En = En+1 for every n N. In other words, {En | n N} is a strictly
innite descending chain of RH-submodules of M . Consequently, En X = M for
all n N and

Cn = Cn M = Cn En X = Cn (En + En Y )
= En + (Cn En Y ) En + (Cn Cn Y ) = En .

We have now

Cn X = Cn + Cn Y = En + Cn Y = (Cn Cn Y ) + Cn Y = Cn Y

for every n N. This means that Y S, which contradicts the choice of X. Hence,
A is an artinian RH-module, as required. 
Theorem 5.3 ([284]). Let R be a ring, G a group, and A an RG-module. Suppose
that H is a normal subgroup of G having nite index. If A is a noetherian RG-
module, then A is a noetherian RH-module.
Proof. Suppose the contrary: that is, A does not satisfy Max-RH. Then the family
M of all RG-submodules B of A such that A/B does not satisfy Max-RH is not
Chapter 5. Reduction to subgroups of nite index 47

empty. Since A is a noetherian RG-module, M has a maximal element M . Let S


be the set of all non-empty nite subsets X of G with the following property: If
{Cn | n N} is a strictly innite ascending chain of RH-submodules containing
M , then

M= Cn x = CoreX Cn
xX

f or all n N.
Let T be a transversal to H in G. Then G = HT . Let n N. By Lemma
5.1, CoreT Cn is an RG-submodule containing M . We claim that CoreT Cn = M .
Otherwise, by the choice of M , A/CoreT Cn is a noetherian RH-module. It follows
that there exists some m n such that Cm+j = Cm for every j N, contradicting
the choice of the chain {Cn | n N}. Thus our claim has just been proven and so
CoreT Cn = M for all n N. Therefore T S and then S = .
We now choose a minimal element X of S. As in the proof of Theorem 5.2, we
may assume that 1 X. Clearly X = 1, so that Y = X \1 is a non-empty nite
subset of G. By the minimality of X, Y  S. Given a strictly innite ascending
chain of RH-submodules {Cn | n N} including M , we put En = Cn +CoreY Cn
for every n N. Clearly En is an RH-submodule and En En+1 for every
n N. Suppose that there exists some k N such that Ek+1 = Ek . Since X S,
CoreX Ck+1 = M and

Ck = Ck + CoreX Ck+1 = Ck + (Ck+1 CoreY Ck+1 )


= Ck+1 (Ck + CoreY Ck+1 ) Ck+1 Ek = Ck+1 Ek+1
= Ck+1 (Ck+1 + CoreY Ck+1 )
= Ck+1 + (Ck+1 CoreY Ck+1 ) = Ck+1 .

Thus Ck = Ck+1 , which contradicts the choice of {Cn | n N}. This contradiction
shows that En = En+1 for every n N. In other words, {En | n N} is a
strictly innite ascending chain of RH-submodules including M . Consequently,
CoreX Ek+1 = M for all n N. Hence,

Cn = Cn + CoreX En = Cn + (En CoreY En )


= En (Cn + CoreY En ) En

and so Cn = En . Thus

M = CoreX Cn = Cn CoreY Cn = En CoreY Cn


= (Cn + CoreY Cn ) CoreY Cn = CoreY Cn

for every n N. This means that Y S. This contradicts the choice of X. Hence,
A is a noetherian RH-module, as required. 

We will study the behaviour of the semisimplicity.


48 Chapter 5. Reduction to subgroups of nite index

Lemma 5.4. Let R be a ring, G be a group, and A be an RG-module. Suppose that


H is a normal subgroup of G and let B be an RH-submodule of A.
(1) If g G, then Bg is likewise an RH-submodule. Moreover:
(a) If B is noetherian, then so is Bg.
(b) If B is artinian, then so is Bg.
(c) If B is simple, then so is Bg.
(2) If
A is a simple RG-module and B is a simple RH-submodule, then A =
gS Bg, for some S G. In particular, A is a semisimple RH-module.

Proof. (1) Since H is normal in G, gRH = RHg, for each g G. Therefore

(Bg)RH = B(gRH) = B(RHg) = (BRH)g = Bg,

and, then Bg is stable under RH.


(a) Let {Cn | n N} be an ascending chain of RH-submodules of Bg.
Then {Cn g 1 | n N} is an ascending chain of RH-submodules of B. Since B is
noetherian, there is some m N such that Cm g 1 = Cm+n g 1 , for all n N. It
follows that Cm = (Cm+n g 1 )g = Cm+n , for all n N, which shows the required
result.
(b) This proof is similar.
(c) Let C be a non-zero RH-submodule of Bg. If Cg 1 = 0, then C = 0,
1 1
Cg = B and so C = (Cg )g = Bg. Hence, Bg is simple.
a contradiction. Thus
(2) Put C = gG Bg so that C is an RG-submodule of A. Since C = 0,
we have that C = A. Then it suces to apply Corollary 4.3. 
The following result appears in the papers of B. Hartley [102] and D.I. Zaitsev
[296].
Theorem 5.5. Let R be a ring, G be a group, and A be a simple RG-module. If
H is a normal subgroup of G and |G : H| = n is nite, then A has a simple
RH-submodule B and A = gS Bg, for some nite subset S such that |S| n.
Proof. Obviously A is an artinian RG-module. By Theorem 5.2, A is an artinian
RH-module. It follows that the family of all non-zero RH-submodules of A has
a minimal element B. Then B is clearly simple. Let X be a transversal to H in
G so that |X| = n. By Lemma 5.4, A = gX Bg. It suces to apply Corollary
4.3. 
Corollary 5.6. Let R be a ring, G a group, and H a normal subgroup of G of nite
index. If A is a semisimple RG-module, then A is a semisimple RH-module.
This extends the celebrated theorem of Cliord (H. Cliord [45]) to innite
groups.
The paper of D.I. Zaitsev [297] contains the following extension of Theorem
5.5.
Chapter 5. Reduction to subgroups of nite index 49

Theorem 5.7. Let F be a eld, G an abelian group, and H a periodic subgroup of


G. If A is a simple F G-module, then A is a semisimple F H-module.

Proof. Let L be a local system of nite subgroups of H and choose K L. If


0 = b A, then dim
F bF K is nite and so bF K has a simple F K-submodule B.
By Lemma 5.4, A = xM Bx, for some subset M G. Since F G is commutative,
B and Bx are isomorphic for every x M . By Proposition 4.5, aF K
= B,for each
0 = a A. In
 particular, aF K is a simple F K-submodule. Since H = KL K
and aF H = KL aF K, we deduce that aF H is further a simple F H-submodule.
Then, it suces to apply Lemma 5.4. 

If A is a simple ZG-module, then the underlying additive group of A either is


a divisible torsion-free abelian group or is a p-elementary abelian group, for some
prime p. In the rst case we may think of A as a QG-module whereas in the second
one we think of A as an Fp G-module. Thus, in any case, we deduce from Theorem
5.7 the following result.

Corollary 5.8. Let G be an abelian group and H a periodic subgroup of G. If A is


a simple ZG-module, then A is a semisimple ZG-module.

The next results describe other criteria of semisimplicity obtained with the
aid of the well-known theorem of Maschke. We are considering here one of the most
general versions of this result, which was obtained in the paper of S. Franciosi,
F. de Giovanni and L.A. Kurdachenko [74], and from it we will deduce several
consequences.

Theorem 5.9. Let R be a ring, G a group and H a normal subgroup of G such


that G/H has nite order n. If A is an RG-module and B is an RG-submodule
admitting an R-complement, then there exists an RG-submodule E such that nA
B + E and n(B E) = 0.

Proof. By hypothesis, there is an R-submodule C such that A = B C. Let


: A C be the canonical projection. If x, y G satisfy Hx = Hy, then there
is some h H such that y = hx. If a A, then we express a = b + c, where b B
and c C. Then we have

a(h1 h) = (ah1 )h = (bh1 + ch1 )h = (ch1 )h = c = a

and, hence,

a(y 1 y) = a((x1 h1 )hx) = ((ax1 )(h1 h))x


= ((ax1 ))x = a(x1 x).

Given f = Hg F = G/H, we dene an R-endomorphism (f ) : A A by


a(f ) = a(g 1 g), if a A. Put now = f F (f ). If x G, then Hx = t F
50 Chapter 5. Reduction to subgroups of nite index

and, if a A, we obtain


(a )x = ( a(f ))x = ( a(g 1 g)x = (ag 1 g)x


f F HgF HgF

1 1
= (ax)(x g gx) = (ax)(f t) = (ax),
HgF f F

because the mapping f f t, f F is a permutation of F . Therefore is an


RG-endomorphism of A and so A is an RG-submodule of A. If a A, we have

na a = (a a(f )) = (a a(g 1 g))


f F HgF

= (ag 1 a(g 1 ))g.


HgF

On the other hand, ag 1 a(g1) B and then (ag 1 a(g 1 ))g B. Thus
na a B.
Let b B. Then bg 1 B and (bg 1 ) = 0. Hence,

b = b(f ) = ((b(g 1 ))g = 0.


f F HgF

If a nA, then a = nu, for some u A. Then a u = nu u B so that


a B + A ; that is, nA B + A .
If b B A, then b = v , for some v A. Again, nv v B; that is,
nv B. Therefore 0 = (nv) = n(v ) = nb. This means that n(B A ) = 0.
Hence, it suces to dene E = A to obtain the required result. 
Corollary 5.10. Let G be a nite group and suppose that A is a ZG-module whose
additive group has nite special rank. If B is a ZG-submodule admitting a Z-
complement, then there is a ZG-submodule E such that A/(B + E) and B E are
nite.
Proof. It follows immediately from Theorem 5.9 since a bounded abelian group of
nite special rank is nite. 
The following simple consequence is due to D.I. Zaitsev [294] and only makes
use of standard properties of divisible abelian groups.
Corollary 5.11. Let G be a nite group and suppose that A is a ZG-module, and B
is a ZG-submodule of A. If the underlying additive groups of A and B are divisible,
then there exists a ZG-submodule E such that A = B + E and n(B E) = 0,
where n = |G|. In particular, if B is Z-torsion-free, then A = B E.
Corollary 5.12. Let G be a nite group. Suppose that A is a ZG-module and B is
a ZG-submodule of A. If the underlying additive group of A is nitely generated,
then there exists a ZG-submodule E such that B E and A/(B E) are nite.
Chapter 5. Reduction to subgroups of nite index 51

Proof. Since the factor-group A/B is nitely generated,

A/B = T /B C/B,

where T /B is nite and C/B is torsion-free. If |T /B| = t, then t(A/B) = U/B is


torsion-free. Moreover, the ZG-submodule U has nite index. Since U/B is a free
abelian group, U = B E for some subgroup E. It suces to apply Corollary 5.10
to the ZG-submodule U . 
Corollary 5.13. Let G be a nite group. Suppose that A is a ZG-module whose
additive group is periodic such that (A)(G) = . If B is a ZG-submodule of A
admitting a Z-complement, then there is a ZG-submodule C such that A = B C.
Corollary 5.14. Let G be a nite group, F a eld, A an F G-module, and B an F G-
submodule of A. If char F = 0 or charF  (G), then there is an F G-submodule
C such that A = B C.
Corollary 5.15. Let A be an F G-module, where G is a nite group and F is a eld.
If char F = 0 or char F  (G), then A is a semisimple F G-module.
Given a group G and a ring R, the R-homomorphism : RG R given
by

( xg g) = xg ,
gG gG

where all but nitely many xg are zero (i. e. both sums are nite), is called the unit
augmentation of RG or simply the augmentation of RG. We denote the kernel of
by RG. It is a two-sided ideal called the augmentation ideal of RG. This ideal
is generated by the elements {g 1 | 1 = g G}.
Corollary 5.16. Let A be an F G-module, where G is a nite group and F is a eld.
If char F = 0 or char F  (G), then A = CA (G) A(F G).
Proof. Indeed, by Corollary 5.14, there exists an F G-submodule E such that

A = CA (G) E.

If a A, then a = c + u, where c CA (G) and u E. Then

a(g 1) = c(g 1) + u(g 1) = u(g 1) E

for each element g G; that is, A(F G) E. By Corollary 5.14 again, we have
the decomposition
E = A(F G) V,
where V is an F G-submodule. If v V , then v(g 1) A(F G) V = 0. It
follows that E = A(F G). 
Lemma 5.17. Let G be a nite group and suppose that A is a ZG-module of com-
position length 2. If (A) (G) = , then A = CA (G) A(ZG).
52 Chapter 5. Reduction to subgroups of nite index

Proof. Let B be a simple ZG-submodule of A such that A/B is also simple. If


B = B(ZG) and A/B = (A/B)(ZG), then A = A(ZG). Therefore we can
assume that either A/B = (A/B)(ZG) or B = B(ZG). We consider these cases
separately.
Suppose rst that B = B(ZG). Since B simple, B(ZG) = 0; that is,
B CA (G). If B = CA (G), then A = CA (G) and all is done. Hence, we may
assume that B = CA (G). It is not hard to see that B and A/B have to be nite
elementary abelian groups. If (B) (A/B) = , then all is done. Suppose that
the additive group of A is a p-group, where p is a prime. If pA = 0, we may
think of A as an Fp G-module and apply Corollary 5.16. Suppose that pA = 0.
In this case B = 1 (A), the lower layer of A. The mapping : a pa, a A is
a ZG-endomorphism, so that Im and Ker are ZG-submodules of A such that
Im =ZG A/ Ker . It follows that Im = B and Ker = B; that is, B
=ZG A/B.
In particular, (A/B)(ZG) = A/B. Let a A and g G. Then ag = a + b for
some b B. We have ag m = a + mb for every m N. In particular, ag p = a. Since
(|g|, p) = 1, g CA (G). This contradiction shows that pA = 0.
Consider now the case in which B = B(ZG) and A(ZG) = B. Considering
again the mapping : a pa, a A, we obtain another time that pA = 0. In
this case, it suces to apply Corollary 5.16. 
This implies immediately the following consequence.
Corollary 5.18. Let G be a nite group and suppose that A is a ZG-module of
nite composition length. If (A) (G) = , then A = CA (G) A(ZG).
Proposition 5.19. Let G be a nite group and suppose that A is a ZG-module whose
additive group is periodic. If (A) (G) = , then A = CA (G) A(ZG).
Proof. Let L be a local system of nite subgroups of A and let

Q = {LG | L L}.

Since G is nite, every member B Q is likewise nite. By Corollary 5.18, B =


CB (G)B(ZG). If C Q satises B C, then CB (G) CC (G) and B(ZG)
C(ZG). These inclusions imply
 
CA (G) = CB (G) and A(ZG) = B(ZG).
BQ BQ

It follows that A = CA (G) A(ZG). 


Chapter 6

Modules over Dedekind domains

Many chapters of this book are dedicated to modules over a group ring of the
form DG, where D is a Dedekind domain. Thus we shall need some results about
Dedekind domains and modules over Dedekind domains. Certain complications
arise here there is no single book that contains all the necessary material suitable
for our specic purposes. Many of these results can be found only in journal
articles, where they have been obtained for modules over wider classes of rings; but
most of the time we only need particular cases of those general results. Therefore,
it would be useful to present a modicum of such important information in this
chapter. In subsequent chapters we will provide results as needed along with the
appropriate references, and supply proofs for results that are not stated in the
needed form (for example, the primary decompositions of periodic modules over
Dedekind domains).

There are many books and papers dedicated to Dedekind domains and mod-
ules over Dedekind domains. Coming up with an exhaustive list would be very
dicult. Below we mention some of them, which include proofs of the most fre-
quently used and eective classic results: N. Bourbaki [20], Z.I. Borevich and A.R.
Shafarevich [21], S.U. Chase [32], C. Curtis and I. Reiner [47, 48], I. Fleichsner
[70], L. Fuchs and S.B. Lee [80], L. Fuchs and L. Salce [81], R. Gilmer [84], I.
Kaplansky [121, 122, 123, 124, 125], G. Karpilovsky [131], M.D. Larsen and P.J.
McCarthy [180], E. Matlis [183, 184, 185, 186, 187, 188], W. Narkiewicz [201],
D.G. Northcott [206], D.S. Passman [219], O.F.G. Schilling [254], R. Sharp [263],
and D.W. Sharpe and P. Vamos [265].

The idea of the Dedekind domain naturally arose within the process of the
development of Algebraic Number Theory. In his famous Supplement XI to Dirich-
lets book, R. Dedekind [51] proved that every ideal of the ring of integers of a eld
of algebraic numbers can be decomposed into prime factors and this factorization
is unique. He created the fractional ideal and also proved that non-zero fractional
ideals formed an abelian group. There are many denitions of Dedekind domains.
We will consider some of them.
54 Chapter 6. Modules over Dedekind domains

If R is a commutative ring, we recall that the prime spectrum Spec(R) of R


(or simply the spectrum of R) is the set of all prime ideals of R.
Let R be an integral domain, and let K be the eld of fractions of R. A non-
zero R-submodule A of K is called a fractional ideal of R if there is some a R such
that aA R. This means that all elements of A can be expressed with fractions
having a as denominator, that is, all of them have a common denominator. Note
that the fractional ideals of R contained in R itself are ordinary ideals of R and
conversely. If A and B are fractional ideals, then the set of all nite sums a1 b1 +
+ an bn , ai A, bj B forms a fractional ideal, known as the product AB of A
and B. For, if aA R and bB R, then abAB R. Obviously A(BC) = (AB)C,
for fractional ideals A, B, C of R. This means that the set F I(R) of all fractional
ideals of R is a commutative semigroup, which admits R itself as its identity
element.
If A is a fractional ideal of R, then

A = {x K | xA R}

is a fractional ideal of R and AA R. We say that A is invertible, if AA = R.


In this case, A is the unique fractional ideal of R satisfying AA = R and we
write A1 instead of A . An integral domain D is called a Dedekind domain, if
every fractional ideal of D is invertible, which is equivalent to the fact that the
semigroup F I(D) is a group. Every principal ideal domain is a Dedekind domain
and, in particular, the ring Z of all integers is a Dedekind domain.
Theorem 6.1. Let D be a Dedekind domain. Then:
(1) D is noetherian.
(2) Every non-zero prime ideal of D is maximal.
Let R be an integral domain and let L be a eld including R, so that L
includes the eld of fractions K of R. We recall that an element c L is said to be
integral over R (or shortly R-integral ) if there are a0 , a1 , , an1 R such that

a0 + a1 c + + an1 cn1 + cn = 0.

In other words, c is a root of a monic polynomial with coecients belong to R. It


is well known that the set C of all elements of L which are integral over R is a
subring. Let L be a eld and R be a subring of L such that 1 R. The subring of
L consisting of all elements which are integrals over R is called the integral closure
of R in L. The integral domain R is said to be integrally closed if R is its integral
closure in the eld of fractions.
The following result proves the full characterization of a Dedekind domain
by means of these concepts and the conditions of Theorem 6.1.
Theorem 6.2. Let D be an integral domain. Then D is a Dedekind domain if and
only if the following conditions are satised.
Chapter 6. Modules over Dedekind domains 55

(1) D is noetherian.
(2) Every non-zero prime ideal of D is maximal.
(3) D is integrally closed.
As we can see every principal ideal domain is a Dedekind domain. Among
principal ideal domains the following are the most important for applications in
Group Theory: the ring Z of all integers, the ring F x of an innite cyclic group
x over a (nite) eld F ; the ring Zp of integer p-adics; the ring F [[x]] of power
series over a (nite) eld F. Observe also, that if R is an integral noetherian domain
whose non-zero prime ideals are maximal, F is the eld of fractions of R, and K
is a nite eld extension of F , then the integral closure of R in K is a Dedekind
domain (see, for example, [20, Chapter 7.2, Corollary 2 of Proposition 5].
Beyond all doubt, the next results describe the most important features of
Dedekind domains.
Theorem 6.3. Every proper non-zero ideal of a Dedekind domain can be represented
uniquely as a product of non-zero prime ideals (up to ordering of these)
Corollary 6.4. Let D be a Dedekind domain. Then the group F I(D) is a free
abelian group generated by the non-zero prime ideals of D.
Given two ideals A and B of a Dedekind domain D, we say that A is divisible
by B (or that B divides A), if there exists an ideal C of D such that A = BC.
Corollary 6.5. Let D be a Dedekind domain. Then every non-zero ideal can be
divisible only by nitely many distinct ideals.
It follows that every fractional ideal A of a Dedekind domain D can be
uniquely written in the form
a(P1 )
A = P1 Pra(Pr ) ,

where P1 , . . . , Pr Spec(D) and a(P1 ), . . . , a(Pr ) Z. Sometimes, it is more con-


venient to write this in the general form

A = P Spec(D)\
0 P a(P ) ,

where a(P ) Z and a(P ) = 0 for all but nitely many values of a(P ). Note that
A is an ordinary ideal of D if and only if a(P ) 0 for all P .
Let A and B be (ordinary) ideals of the Dedekind domain D. By denition
the greatest common divisor of A and B is the (unique) ideal GCD(A, B) of D
that satises the following:
(i) GCD(A, B) divides A and B.
(ii) If C is an ideal of D and C divides A and B, then C divides GCD(A, B).
56 Chapter 6. Modules over Dedekind domains

In fact, if
0 = A = P Spec(D)\
0 P a(P )
and
0 = B = P Spec(D)\
0 P b(P ) ,
then A divides B if and only if a(P ) b(P ), for every P Spec(D)\ 0. It follows
that
GCD(A, B) = P Spec(D)\
0 P c(P ) ,
where c(P ) = max{a(P ), b(P )}.
The least common multiple of A and B is the unique ideal LCM(A, B) that
satises the following properties:
(i) A and B divide LCM(A, B).
(ii) If C is an ideal of D and A and B divide C, then LCM(A, B) divides C.
It is easy to see that
LCM(A, B) = P Spec(D)\
0 P d(P ) ,
where d(P ) = max{a(P ), b(P )}.
Let R be a commutative ring. We say that the proper ideals A and B of R
are relatively prime if A + B = R. Now we are in a position to obtain the other
accustomed properties of a Dedekind domain.
Proposition 6.6. Let D be a Dedekind domain. Then the following assertions hold.
(1) If A and B are fractional ideals, then B divides A if and only if A B.
(2) If A is a fractional ideal, then there exists a principal fractional ideal aR such
that (aR)A1 R.
(3) If A and B are ideals of R, then GCD(A, B) = A + B and LCM(A, B) =
A B.
(4) If A and B are ideals of R, then AB = (A + B)(A B).
(5) If A and B are relatively prime ideals of R, then AB = A B.
Lemma 6.7. Let R be a commutative ring and let A and B be ideals of R. If A
and B are relatively prime, then so are Ak and B m , for every k, m N
Lemma 6.8. Let R be a commutative ring. The ideals A1 , . . . , Ar of R are pairwise
relatively prime if and only if for every 1 i r we have that
Ai + A1 Ai1 Ai+1 Ar = R.
Corollary 6.9. Let D be a Dedekind domain. If the ideals A1 , . . . , Ar of D are
pairwise relatively prime, then
A1 Ar = A1 Ar .
Chapter 6. Modules over Dedekind domains 57

Theorem 6.10 (Chinese Remainder Theorem). Let D be a Dedekind domain. If


A1 , . . . , Ak are pairwise, relatively prime, non-zero ideals of D and a1 , . . . , ak D,
then there exists some a D such that ai + Ai = a + Ai , for each 1 i k.
Moreover, if b satises b + Ai = ai + Ai , for each 1 i k, then
a + A1 Ak = b + A1 Ak .
Corollary 6.11. Let D be a Dedekind domain. If A1 , . . . , Ak are pairwise, relatively
prime, non-zero ideals of D, then
D/(A1 Ak )
= D/A1 D/Ak .
Corollary 6.12. Let D be a Dedekind domain. If I is a non-zero ideal of D and
I = P1m1 Pkmk , where P1 , . . . , Pk are dierent prime ideals of D, then
D/I
= D/P1m1 D/Pkmk .
As we have already mentioned, every principal ideal domain is a Dedekind
domain. The following results show that Dedekind domains are very close to prin-
cipal ideal domains.
Proposition 6.13. Let D be a Dedekind domain and let Q be a non-zero prime ideal
of D. Suppose that B is an ideal such that B Qm , for some m N. Then there is
some k m such that B = Qk . In particular, if y Q \ Q2 , then Qt = y t D + Qn ,
for every n N and each t n.
Proof. By Theorem 6.3,
B = P (Spec(D))\
0 P a(P ) ,

and then there exists some k N such that B = Qk A, where A is an ideal of D. In


particular, B Qk by Proposition 6.6. If k m, then B Qk = Qkm Qm Qm ;
that is, B = Qm and we are done. Therefore k m. Suppose that A = D
so that A and Q are relatively prime by Lemma 6.7 and Lemma 6.8. We have
BA1 Qk D, whence A1 B 1 and then B A. It follows that Qm A.
By Lemma 6.7, A + Qm = D, which implies that A = D, a contradiction. Hence
A = D, and then B = Qk .
Since y Q, yD Q. This and Qn Q for each n N give that yD + Qn
Q. By the result shown in the above paragraph, yD + Qn = Qs , for some s n.
We claim that s = 1. For, if s > 1, then
yD + Qn Qs = Q2 Qs2 Q2
and so yD Q2 , which contradicts the choice of y. Therefore s = 1, and so
yD + Qn = Q. Pick x1 , x2 Q. Then xj = yzj + uj , where zj D and uj Qn
(j = 1, 2). Thus
x1 x2 = (yz1 + u1 )(yz2 + u2 )
= y 2 z1 z2 + yz1 u2 + yz2 u1 + u1 u2 y 2 D + Qn .
58 Chapter 6. Modules over Dedekind domains

It follows that Q2 y 2 D + Qn , and hence Q2 = y 2 D + Qn . By induction, Qt =


y t D + Qn , for every t n. 

Corollary 6.14. Let A be a non-zero ideal of a Dedekind ring D. Then every ideal
of D/A is principal.

Proof. Let B be an ideal of D such that B A. By Theorem 6.3,

A = P1k1 Ptkt ,

where P1 , . . . , Pt Spec(D). By Corollary 6.12,

D/A
= D/P1k1 D/Ptkt .

By Proposition 6.13, there exist some a1 , . . . , at D such that


k k
B + Pj j = aj D + Pj j ,

for every 1 j t. Using Theorem 6.10, we nd an element a such that


k k
a + Pj j = aj + Pj j ,

for every 1 j t. Now it is not hard to see that B = aD + A and then B/A is
principal. 

Corollary 6.15. Let A be a non-zero ideal of a Dedekind domain D. There exist


some elements a1 and a2 such that A = a1 D + a2 D.

Proof. Let 0 = a1 A and put B = a1 D. By Corollary 6.14,

A/B = (a2 + B)(D/B) = (a2 D + B)/B,

for some a2 A. Then A = a2 D + a1 D, as required. 

Corollary 6.16. Let A and B be non-zero ideals of a Dedekind domain D. Then


there exists a non-zero ideal C such that B + C = D and AC is principal.

Proof. Since D contains no zero-divisors, AB = 0. By Corollary 6.14, A/AB =


(aD + AB)/AB, for some a A. Since aD A, by Proposition 6.6, aD = AC,
for some ideal C = 0. Thus AC + AB = A and, multiplying by A1 , we obtain
that C + B = D. 

Corollary 6.17. Let A and B be non-zero ideals of a Dedekind domain D. Then


the D-modules D/A and B/AB are isomorphic. In particular, if k N , then D/A
and Ak /Ak+1 are isomorphic D-modules.
Chapter 6. Modules over Dedekind domains 59

Proof. By Corollary 6.16, there exists a non-zero ideal C such that C + A = D


and BC = aD, for some a D. Dene the mapping : D B/AB by x =
ax + AB, x D. We have
(x + y) = a(x + y) + AB = ax + ay + AB
= (ax + AB) + (ay + AB) = x + y
and
(dx) = a(dx) + AB = (ax)d + AB
= (ax + AB)d = (x)d (for each d D),
and so is a D-homomorphism. Furthermore,
aD + AB = CB + AB = (C + A)B = DB = B.
Thus Im = B/AB. If x Ker , then ax AB and so axC ABC = aA. It
follows that xC A. Since A + C = D, there are some elements a1 A and c C
such that 1 = a1 + c. Thus x = x 1 = xa1 + xc A. Conversely, if x A, then
ax AB and so x = ax + AB = AB. Therefore Ker = A. 
Let R be a ring and let A be an R-module. We dene
tR (A) = {a A | AnnR (a) = 0}.
This subset has a nice structure provided R has no zero-divisors, as the
following result shows.
Lemma 6.18. Let R be an integral domain. If A is an R-module, then tR (A) is an
R-submodule of A.
Proof. Let a1 , a2 tR (A). Put I1 =AnnR (a1 ) and I2 =AnnR (a2 ). Then I1
I2 AnnR (a1 a2 ). Since R contains no zero-divisors, I1 I2 = 0. Since I1 I2
I1 I2 , AnnR (a1 a2 ) = 0.
Given a tR (A) and 0 = x R, if u AnnR (a), then
(ax)u = a(xu) = a(ux) = (au)x = 0x = 0.
It follows that AnnR (a) AnnR (ax), that is ax tR (A). 
Let R be an integral domain. Then the submodule tR (A) of A is called the
R-periodic part of A. A is called R-periodic if A = tR (A) and R-torsion-free if
tR (A) = 0. Note that, for A arbitrary, then tR (A) is R-periodic and A/tR (A) is
R-torsion-free.
The following concept is dual to the concept of R-annihilators of subsets of
A. If L R is a subset of R, then the A-annihilator of L is
AnnA (L) = {a A | aL = 0}.
If R is commutative, then it is customary that AnnA (L) is an R-submodule of A.
60 Chapter 6. Modules over Dedekind domains

Lemma 6.19. Let A be an R-module, where R is a commutative ring and let W


be the set of all ideals H of R such that AnnA (H) = 0. If W has a maximal
element P = R, then P is a prime ideal of R.
Proof. All we have to show is that P is a prime ideal. Let B =AnnA (P ). Since R
is a commutative ring, B is an R-submodule. Let x, y be elements of R such that
xy P and let 0 = b B. Then b(xy) = 0. If bx = 0, then xR W. In this case,
xR + P AnnR (b) and, in particular, AnnA (xR + P ) = 0; that is, xR + P W.
By the choice of P , it follows that xR + P = P ; that is, x P . If bx = 0, then
AnnA (yR) = 0. Repeating the previous arguments, we obtain y P . This means
that P is a prime ideal. 
Corollary 6.20. Let R be a noetherian commutative ring and suppose that A is an
R-module such that tR (A) = 0. Then R has a non-zero prime ideal P such that
AnnA (P ) = .
Let A be an R-module, where R is a commutative ring. We dene the R-
assasinator of A as the set AssR (A) consisting of all prime ideals P of R such that
AnnA (P ) = 0.
Now we nish this chapter showing the existence of the primary decomposi-
tion of certain modules over a Dedekind domain. Again, inspired by the ordinary
layer of a p-group, we consider the following construction. Let I be an ideal of a
commutative ring R. If k N, we dene
I,k (A) = {a A | aI k = 0}.
It is easy to see that I,k (A) is an R-submodule and
I,1 (A) I,2 (A) I,k (A) .

The R-submodule AI = I,k (A) is said to be the I-component of A. If A = AI ,
then A is said to be an I-module. If I = P is a prime ideal, a P -module is
generically called a primary module.
Lemma 6.21. Let A be a D-module, where D is a Dedekind domain. If U and V
are relatively prime ideals of D, then AU AV = 0.
Proof. Pick u U and v V such that 1 = u + v. Suppose that AU AV = 0.
Then U,1 (A) V,1 (A) = 0. If
0 = a U,1 (A) V,1 (A),
then a = a 1 = a(u + v) = au + av = 0. This contradiction shows that AU AV =
0. 
Corollary 6.22. Let A be a D-module, where D is a Dedekind domain. Suppose
that U and V are two relatively prime ideals of D such that A = AU AV . If B
is a submodule of A, then
(A/B)U = (AU + B)/B, (A/B)V = (AV + B)/B,
Chapter 6. Modules over Dedekind domains 61

and A/B = (A/B)U (A/B)V .


Proof. Since A = AU AV ,

A/B = (AU + B)/B + (AV + B)/B.

It easy to see that

(AU + B)/B (A/B)U and (AV + B)/B (A/B)V .

By Lemma 6.21,
(AU + B)/B (AV + B)/B = 0 .
It follows that
A/B = (AU + B)/B (AV + B)/B
and then
(A/B)U = (AU + B)/B and (A/B)V = (AV + B)/B. 
Corollary 6.23. Let D be a Dedekind domain. Suppose we are given a family U of
proper, non-zero, pairwise, relatively prime, ideals of D. If A is a D-module, then


AU = AU .
UU UU

Proof. It suces to show the result for every nite subset V U. Suppose that
|V| = k and proceed by induction on k. If k = 2, then the result follows from
Lemma 6.21. Let k > 2. If V V, by induction


AU = AU .
UV\{V } UV\{V }

Suppose that 
AV ( AU ) = 0 .
UV\{V }

Thus, for every U V, there are some elements aU AU such that


aV = aU
UV\{V }

and aV = 0. Then there exists some V = W V such that aW = 0. Pick rW , rV


N such that aV V rV = 0 and aW U rW = 0. By Lemma 6.7, W rW + V rV = D,
and so there are xW W rW and xV V rV such that 1 = xW + xV . Then

aV = aV 1 = aV (xW + xV ) = aV xW + aV xV

= aV xW = aU xW .
UV\{V }
62 Chapter 6. Modules over Dedekind domains

Since AU is a D-submodule of A, aU xW AU . But aW xW = 0, so that


aU xW
UV\{V }

contains in fact k 2 < k summands. By induction, aV = 0. 


Corollary 6.24. Let D be a Dedekind domain. Suppose we are given a family U
of proper, non-zero, pairwise, relatively prime, ideals of D. Let A be a D-module
such that 
A= AU .
UU

If B is a D-submodule of A, then

A/B = (AU + B)/B.
UU

If A is a D-module over a Dedekind domain, then the set AssD (A) = by


Corollary 6.20, because D is noetherian by Theorem 6.1. Also, if A is D-periodic,
then AssD (A) is a family of proper, non-zero, pairwise, relatively prime, ideals of
D. Therefore, applying the above results, we obtain the primary decomposition of
A.
Corollary 6.25. Let D be a Dedekind domain. If A is a D-periodic D-module, then

A= AP .
P AssD (A)

If B is a D-submodule of A, then (A/B)P = (AP + B)/B and



A/B = (AP + B)/B.
P AssD (A)

Proof. Indeed, if a A, then L =Ann


D (a) = . Hence aD = D/L and it suces
to apply Corollary 6.12 to obtain a = P AP , where =AssD (A). 
Chapter 7

The KovacsNewman theorem

A major part of this book is dedicated to artinian modules over group rings.
Artinian modules are one of the oldest subjects of investigation in Algebra. They
are named after E. Artin, who initiated the study of rings and modules with the
minimal condition. The minimal condition (as well as the maximal condition)
has shown to be very useful and truly eective in many branches of Algebra,
particularly in Group Theory. A natural way of connecting groups and modules
can be described as follows. Let G be a group, and suppose that G has an abelian
normal subgroup A. If H = G/A, then H acts on A in the following way: if
h = gA H and a A, then we dene a h = ag . Since A is abelian, this
denition only depends on h. If n Z, we put

anh = (an )h = (ah )n .

Every element x ZH can be expressed as a (nite) sum as


x= nh h,
hH

where nh Z for every h H and nh = 0 for all but nitely many h H. Then,
the law 
ax= anh h
hH

(which will be briey written by juxtaposition) transforms A into a ZH-module.


Since additive notations are widely employed in Module Theory, from now on we
will use the additive notations and write ax instead ax when we think of A as a
ZH-module and keep the multiplicative notations when we think of A as a normal
subgroup of G.
Very often, the subgroup A satises some niteness condition, which turns
out to be linked with the minimal condition for G-invariant subgroups. In other
words, the ZG-module A is artinian.
64 Chapter 7. The KovacsNewman theorem

In the theory of innite groups, artinian modules rst appeared in the study
of metabelian groups with the minimal condition on normal subgroups (Min-n).
The rst interesting result in this direction can be found in the paper [33], in
which V.S. Charin constructed an example of a metabelian group satisfying Min-
n but not Min. Subsequent studies showed that this example enjoyed very typical
properties. Almost simultaneously P. Hall began his study of soluble groups with
the maximal condition on normal subgroups (the condition Max-n). Unfortunately,
after this rst step, V.S. Charin stopped his study of metabelian groups with Min-
n, a strange event especially considering that a very intensive research of groups
with niteness conditions had been developed in the former USSR by that time.
V.S. Charin was a student of S.N. Chernikov, to whom we owe one of the more
important accomplishments in this area, namely, the investigation of groups with
Min. Since then, in his next research, V.S. Charin [35] explored the question: for
what classes of groups does Min-n imply Min? ; but he did not come back to groups
with Min-n anymore. The description of metabelian groups satisfying Min-n was
obtained much later by B. Hartley and D. McDougall [110]. More precisely, B.
Hartley and D. McDougall described artinian modules over Chernikov groups.
That paper has stimulated investigations of artinian modules over group rings
as well as many other research projects in this area. At this point, it is worth
mentioning a fundamental distinction between the above topic and its dual theory
of noetherian modules over group rings. P. Hall [95] showed that the group ring of
a polycyclic-by-nite group over a noetherian ring is noetherian itself. This result
is fundamental in the theory of noetherian modules: it led to the developing of
the rich theory of group rings, which brought up many signicant achievements in
the study of noetherian modules over such rings. Among many other publications
we can mention, for example, C.J.B. Brookes [26, 27, 28], C.J.B. Brookes and
J.R. J. Groves [29], K.A. Brown [30, 31], J. Cornick and P.H. Kropholler [46],
S. Donkin [57, 58], D.R. Farkas [68], K.W. Gruenberg [94], P. Hall [95, 96, 98],
A.V. Jategaonkar [118, 120], P.H. Kropholler, P.A. Linnel and J.A. Moody [139],
I. Musson [195, 196, 197], D.S. Passman [216, 217, 218], J.E. Roseblade [248, 249,
250, 251, 252], R.L. Snider [268].

Apart from this, in the case of artinian modules, the properties of a group
ring are not very useful. A group ring of a non-polycyclic group with niteness
conditions (even one that is very close to being polycyclic) dees deep descrip-
tion, and there are not too many papers devoted to it (see S.D. Berman [10], S.D.
Berman and A.G. Podgorny [11], S.D. Berman and N.I. Vishnyakova [12, 13],
C.J.B. Brookes [25], B. Hartley [100], H. Heineken and I.J. Mohamed [112], D. Se-
gal [256], A.V. Tushev [276, 277]). Therefore the study of artinian modules requires
dierent approaches. We collect here some of the most important tools, which will
be considered in detail throughout the book. One of these tools is the study of the
semisimplicity conditions of artinian modules, whose main source is the celebrated
theorem of Maschke. One can nd many important results in this direction in the
following papers: D.R. Farkas and R.L. Snider [69], B. Hartley [101, 102, 103, 107],
Chapter 7. The KovacsNewman theorem 65

B. Hartley and D. McDougal [110], L.G. Kovacs and M.F. Newman [137], D.I.
Zaitsev [294, 301].
Another direction, specically available for artinian modules over group rings,
is based on the Fitting lemma and detects direct decompositions of modules that
are determined by a group action. This runs parallel to the study of direct de-
compositions of groups relative to formations (see R. Baer [8]), which allows us
to unify the points of view on many results. In this direction, we can mention the
following papers: Z.Y. Duan [59, 60, 61, 62, 63], Z.Y. Duan and M.J. Tomkinson
[64], B. Hartley and M.J. Tomkinson [111], L.A. Kurdachenko, J. Otal and I.Ya.
Subbotin [158], L.A. Kurdachenko, B.V. Petrenko and I.Ya. Subbotin [159, 160],
D.I. Zaitsev [301, 302, 304, 307, 309, 310], D.I. Zaitsev and V.A. Maznichenko
[313].
Finally, the description of artinian modules over group rings, perhaps, is the
rst problem in this list. Since the situation here is far from being satisfactory, it is
better to reformulate this problem as follows: for what groups G is the description
of artinian modules over the group ring ZG possible? As we have already noted,
artinian modules over Chernikov groups have been described by B. Hartley and
D. McDougall [110]. On the other hand, S.A. Kruglyak [140] pointed out that
the description of nite modules (and hence artinian modules) over a free abelian
group of 0-rank at most 2 is wild from the representation theory point of view.
The following analogy could be useful at this point. The description of torsion-
free abelian groups of nite 0-rank at most 2 is also wild even though the Z-
injective envelope of these groups is a direct sum of nitely many copies of the
full rational group (Q, +). This leads to the question on the description of the
injective envelope of artinian modules over group rings, a partial case of which
is the description of simple modules over group rings. These topics have been
reected in many papers. We would like to mention some of them: D.R. Farkas
and R.L. Snider [69], B. Hartley [101, 102, 103], B. Hartley and D. McDougall [110],
A.V. Jategaonkar [119], L.A. Kurdachenko [148, 150], L.A. Kurdachenko and J.
Otal [155], L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157], L.A. Kurdachenko
and I.Ya. Subbotin [170], I. Musson [194, 195, 196, 197, 198, 199, 200], R.L. Snider
[267, 269, 270], B.A.F. Wehrfritz [281, 282, 283].

We begin with the problem of existence of complements for submodules of


modules over group rings. Let A be a module over a ring R. A submodule B of A
is said to have a complement in A (or B is a complemented submodule) if there is
an R-submodule C such that A = B C. In the theory of modules over a group
ring RG, the following question arises very often. Let B be an RG-submodule of
A such that A = B C, for some R-submodule C. Under which conditions does B
have an RG-complement? It is well known that such an RG-complement does not
always exist. In some cases, only a supplement exists: an RG-submodule E such
that both A/(B +E) and B E are good in some sense, they are both nite. One of
the most celebrated positive results for modules over nite groups is the theorem
of Maschke. In Chapter 5, we considered general forms of this result. Hence, the
66 Chapter 7. The KovacsNewman theorem

question on the existence of analogous theorems in some classes of innite groups


appears to be very plausible. Now we consider other useful results, which imply, in
particular, semisimplicity criteria. The next one is the KovacsNewman theorem
[137]. We formulate this result in a slightly generalized form, namely, we extend
the scalar ring on Z to an arbitrary Dedekind domain. For this reason, some results
on modules over Dedekind domains will be needed.
Let R be a ring and let A be an R-module. If 0 = x R, we say that
A is x-divisible if given an element a A there exists some c A such that
cx = a, that is, A = Ax. A module A is said to be R-divisible if A is x-divisible
for every 0 = x R. The concept of a divisible module is tightly connected
with the concept of the injective module, which is dual to the concept of the
projective module. A module E over a ring R is called injective if for every R-
monomorphism : A B and every R-homomorphism : A E, there
exists an R-homomorphism : B E such that = . Now we formulate
some useful properties of injective and divisible modules. By the same reasons
mentioned in Chapter 6, and for the readers convenience, we will supply some of
them with suitable proofs. The proof of all omitted results can be directly found
in the books cited in Chapter 6.
Theorem 7.1 (Baers criterion). Let R be a ring and let E be an R-module. Then
E is an injective module if and only if for every right ideal J of R and for every
R-homomorphism : J E there exists an R-homomorphism : R E
such that x = x for each x J.
Proposition 7.2. Let R be a ring.

(1) Let {E | } be a family of R-modules. Then E is injective if
and only if E is injective for every .

(2) Let {E | } be a family of R-modules. If E is injective, then E
is injective for every .
(3) Let {E | } be a family of
R-modules such that set is nite. If E is
injective for each , then E is injective.
(4) An R-module A is injective if and only if A has an R-complement in every
R-module E A, that is there exists an R-module L such that E = A L.
The next theorem elucidates the signicance of injective modules.
Theorem 7.3. Let R be a ring. Then every R-module can be embedded in some
injective R-module.
Let A be a module over a ring R. Then an R-module E is said to be an
injective envelope of A if E satises the following conditions:
(IE1) A E;
(IE2) E is injective; and
Chapter 7. The KovacsNewman theorem 67

(IE3) if B is an injective R-module such that A B E, then B = E.


An R-module C is called an essential extension of an R-submodule A if
S A = 0 for every non-zero R-submodule S of C. An essential extension C
of a module A is called maximal , if there is no essential extension E of A and
monomorphism : C E satisfying a = a for every a A and C = E.
Theorem 7.4. Let A be a module over a ring R. Then there exists an R-module E
satisfying the following equivalent conditions:
(1) E is injective and E is an essential extension of A.
(2) E is a maximal essential extension of A.
(3) E is a minimal injective extension of A.
Moreover, if U and V are minimal injective extensions of A, then there exists an
isomorphism : U V .
In other words, for every R-module A there exists an injective envelope and
this injective envelope is unique (up to isomorphism).
The R-injective envelope of A is denoted by IER (A).
Proposition 7.5. Let R be a ring and suppose that y R is not a right zero-divisor.
If E is an injective R-module, then E is y-divisible.
Proof. Given 0 = e E, we dene a mapping : Ry E by (xy) = ex, x
R. By the choice of y, is a well-dened R-homomorphism and Ker = 0.
Therefore there exists an R-homomorphism : R E such that (xy) =
(xy) = ex. We have that

e = e = y = (1 y) = (1)y

so that e Ey. Therefore E = Ey, as required. 


We also will consider other properties of divisible modules.
Lemma 7.6. Let R be an integral domain.
(1) If E is an R-divisible R-module and C is an R-submodule of E, then E/C
is R-divisible too.

(2) If E is an R-divisible R-module for every , then E and E
are also R-divisible.
Proof. (1) Let e E and x R. There is some a E, such that ax = e. Then
e + C = ax + C = (a + C)x, that is,
 (E/C)x = E/C.
(2) Suppose that (e ) E , where 0 = x R. For every ,
there is some a E , such that a x = e . Therefore ((a ) )x = (e ) that
is,  
( E )x = E .

68 Chapter 7. The KovacsNewman theorem

Suppose now that S =Supp((e ) ) is nite. If , we dene



a , if S,
b =
0, if  S.

E
Then Supp((b ) ) = S. In particular, (b ) . By construction,
we have that ((b ) )x = (e ) , and hence ( E )x = E . 
Proposition 7.7. Let R be an integral domain and let E be an injective R-module.
Then E is an R-divisible module.
Proof. Let 0 = x R and e E. We dene the mapping : Rx E by
(ux) = eu, u R. Clearly, is an R-homomorphism. Since E is R-injective, by
Theorem 7.1, there is an R-homomorphism : R E extending . We have

e = x = x = (1 x) = (1 )x.

Thus Ex = E, i. e., E is R-divisible. 


We now consider an abelian group as a Z-module. An abelian group A is said
to be divisible, if A is a Z-divisible module; that is, A = kA for each 0 = k Z.
Proposition 7.7 has the following converse.
Proposition 7.8. Let R be an integral domain and let E be an R-module. If E is
R-divisible and R-torsion-free, then E is an injective module.
Proof. Let L be an ideal of R and let : L E be an R-homomorphism.
We assume that L = 0 , and then we pick an element 0 = x L. Since E is
R-divisible, there is an element e E such that ex = x. If y L, then

(y)x = (yx) = (xy) = (x)y = (ex)y


= e(xy) = e(yx) = (ey)x.

It follows that (y ey)x = 0, which means that y = ey since E is R-torsion-


free. Dene now the mapping : R E by ug = eu for every u R. Then
y = ey = y, for each y L, so is an extension of . By Theorem 7.1, E is
injective. 
We need to have a careful look at the case in which R is a Dedekind domain.
This case is extremely important for our purposes. In this case, as we will show,
the concepts of injectivity and divisibility are coincide.
Theorem 7.9. Let D be a Dedekind domain and let E be a D-module. If E is
divisible, then E is injective.
Proof. Let L be an ideal of D. Again we assume that L = 0. Let : L E
be a D-homomorphism. Since D is a Dedekind domain, there exists L1 , and so
there are some x1 , . . . , xk L and y1 , . . . , yk L1 such that

1 = x1 y1 + + xk yk .
Chapter 7. The KovacsNewman theorem 69

There is no loss if we assume that x1 , . . . , xk = 0. Since E is D-divisible, there are


some e1 , . . . , ek E such that xj = ej xj for every 1 j k. Pick x L. Since
xyj LL1 = D,

x = (x 1) = (xx1 y1 + + xxk yk ) = (x1 )xy1 + + (xk )xyk


= (e1 x1 )xy1 + + (ek xk )xyk = (e1 x1 y1 + + ek xk yk )x.

Since
x1 y1 + + xk yk LL1 = D,
e = e1 x1 y1 + + ek xk yk E.
Dene the mapping : D E by y = ey whenever y D. It is easy to see
that is a D-homomorphism and x = ex = x, and so is an extension of
. Then it suces to apply Theorem 7.1. 
Corollary 7.10. Let R be a principal ideal domain and let E be an R-module. Then
E is injective if and only if E is R-divisible.
Corollary 7.11 (R. Baer). An abelian group A is divisible if and only if A is an
injective Z-module.
Let D be a Dedekind domain and let A be a D-module. We say that a D-
divisible module E is a minimal divisible module for A if E contains A and every
proper D-submodule B of E containing A is not D-divisible.
Corollary 7.12. Let D be a Dedekind domain.
(1) Each D-module A can be embedded in a divisible D-module.
(2) Let A be a D-module and let E be a divisible D-module such that A E.
Then E includes a minimal divisible submodule for A.
(3) Let A be a D-module. Then there exists a minimal divisible module for A,
and two minimal divisible modules for A are isomorphic.
A minimal divisible module for A (unique up to an isomorphism) will be
called the divisible envelope of A.
Lemma 7.13. Let D be a Dedekind domain and let A be a D-module.
Suppose that
{E } is a family of D-divisible submodules of A. Then E = E is also
D-divisible.
Proof. Indeed, a sum of submodules is a homomorphic image of the direct sum of
them, and then it suces to apply Lemma 7.6. 
Corollary 7.14. Let D be a Dedekind domain and let A be a D-module. If A has
D-divisible submodules, then A has a unique largest D-divisible submodule, say div
(A).
Proof. In fact, div (A) is the sum of all D-divisible submodules of A. 
70 Chapter 7. The KovacsNewman theorem

If A has no D-divisible submodules, then we put div(A) = 0. The submod-


ule div (A) dened above is called the D-divisible part of A.
Lemma 7.15. Let R be a ring and let A and B be R-modules.
(1) If there are R-homomorphisms : A B and : B A such that =
A , then is a monomorphism, is an epimorphism, and B = Im Ker .
In particular, A is isomorphic to a direct summand of B.
(2) If B = A1 A2 and A1 = A, then there exists an R-monomorphism :
A B and an R-epimorphism : B A such that = A .
Proof. (1) If b = a Im Ker , then
a = aA = a = b = 0
and so b = 0. Hence Im Ker = {0}. On the other hand, if b B, then
b b Ker since
(b b) = b bA = 0.
Since (b) Im , b Im + Ker , that is
B = Im + Ker
and hence
B = Im Ker ,
as required.
(2) This is obvious. 
Theorem 7.16. Let D be a Dedekind domain. Suppose that A is a D-module in-
cluding two submodules E and B, which satisfy the following conditions:
(1) E is D-divisible.
(2) B E = 0.
Then there exists a D-submodule C such that B C and A = E C.
Proof. Let 1 : E E B and 2 : E B A be the canonical embeddings.
Then the mapping : E B E given by (e + b) = e, where e E and
b B, is a D-homomorphism. By Proposition 7.7, E is injective. Hence there is a
D-homomorphism : A E such that 2 = . Then
1 2 = 1 = E .
By Lemma 7.15,
A = Im 1 2 Ker .
Obviously, Im 1 2 = E. If C = Ker , then A = E C. Note that, if b B,
then
b = (b2 ) = b = 0
and so B Ker = C. 
Chapter 7. The KovacsNewman theorem 71

Corollary 7.17. Let D be a Dedekind domain and let A be a D-module. Then A =


div (A) B, where B is a D-submodule including no D-divisible submodules.
Let F be the eld of fractions of the Dedekind domain D. If a F and
0 = x D, we have a = (ax1 )x. This means that the D-module F is divisible.
Also, note that if A is a D-module, by Corollary 7.12, there exists a divisible
module E such that E A.
Lemma 7.18. Let D be a Dedekind domain. Suppose that A is a D-module and put
T = tD (A). If A is divisible, then T is likewise divisible.
Proof. If a T , then AnnD (a) = J = 0. Pick 0 = x D. Since A is divisible,
there is some b A such that bx = a. If 0 = y J, since xy = 0 and 0 = ay =
(bx)y = b(xy), we have 0 = xy AnnD (b); that is, b T . Thus T is divisible, as
required. 
Corollary 7.19. Let D be a Dedekind domain. Suppose that A is a D-module and
E is a minimal divisible module for A.
(1) If A is D-periodic, then E is likewise D-periodic.
(2) If A is D-torsion-free, then E is likewise D-torsion-free.
Proof. Put T = tD (E). By Lemma 7.18, T is divisible. If A is D-periodic, then
A T and the minimality of E assures that E = T . If A is D-torsion-free, then

A T = 0 .

By Theorem 7.16, there exists a D-submodule E1 A such that



E = E1 T.

Since E1
= E/T , E1 is divisible by Lemma 7.6. By the choice of E again, E = E1 .
Thus the result is proved. 
Lemma 7.20. Let D be a Dedekind domain. Suppose that C = cD is a cyclic D-
module and that E is a minimal divisible module for C. If C is D-torsion-free,
then E
=D F , where F is the eld of fractions of D.
Proof. Since C is D-torsion-free, the mapping : D C, dened by the rule
x = cx, x D, is a D-homomorphism. Corollary 7.19 implies that E is D-torsion-
free. Let c1 C, 0 = y D, e1 , e2 be elements of E such that e1 y = c1 = e2 y.
Then
0 = e1 y e2 y = (e1 e2 )y
and, since E is D-torsion-free,

e1 e2 = 0,

that is, e1 = e2 . This assures that if c1 C and 0 = y D, there is a unique


e E such that ey = c1 . If c1 = cx for some x D, then put e = c xy .
72 Chapter 7. The KovacsNewman theorem

Extend the mapping to a mapping 1 : F E by the rule x


y 1 = c xy .
Let x1 C, 0 = y1 D and e1 = c xy11 . Then
ey = cx, e1 y1 = cx1

and we obtain
eyy1 = cxy1 , e1 y1 y = cx1 y
and
(e + e1 )yy1 = cx1 y + cxy1 = c(x1 y + xy1 ).
It follows that
x1 y + xy1
e + e1 = c .
yy1
Hence
x x1 x1 y + xy1 x1 y + xy1
( + )1 = ( )1 = c
y y1 yy1 yy1
x x1 x x1
= e + e1 = c + c = ( )1 + ( )1 .
y y1 y y1
Let u D. Put e2 = c ux
y . Then

e2 y = cux = (cx)u = (ey)u = (eu)y.


It follows that (e2 eu)y = 0 and, then e2 = eu. Thus
x x x
e2 = (u )1 = eu = c u = ( )1 u.
y y y
This shows that 1 is a D-homomorphism. Let x
y Ker 1 , that is c xy = 0. Then
x
cx = c y = 0,
y
i.e., x Ann D (C) = 0. This means that xy = 0 and therefore Ker 1 = 0.
As Im 1 is a divisible submodule of E and C Im 1 , by the choice of E, we
conclude that E = Im 1 . Hence E
=D F . 
Let D be a Dedekind domain. If A is a simple D-module, then there exists a
maximal ideal P of D such that A = D/P . By Corollary 6.17, D/P k and P/P k+1
are isomorphic as D-modules for every k N. In particular, for every k N the
D-module D/P k can be embedded in the D-module D/P k+1 . Therefore we may
consider the injective limit of the family of D-modules {D/P k | k N}. We put

lim {D/P k | k N}.


CP =
The D-module CP is called a Pr
ufer P -module. By construction, CP is a P -

module such that P,k (CP ) =D D/P k and

P,k+1 (CP )/P,k (CP )


= (D/P k+1 )/(P/P k+1 )
= D/P
Chapter 7. The KovacsNewman theorem 73

for every k N. Hence, if C is a proper D-submodule of Cp , then there exists


some k N such that C = P,k (CP ). In fact, if b
/ P,k (CP ), then C = bD.
We also note that the Pr ufer P -module CP is monolithic and its monolith is
P,1 (CP ).
Lemma 7.21. The Pr
ufer P -module is divisible.
Proof. This has been established in the book by L.A. Kurdachenko, J. Otal and
I.Ya. Subbotin [157, Lemma 5.1]. 
Corollary 7.22. Let D be a Dedekind domain. Suppose that A is a simple D-
module and let E be a minimal divisible module for A. Then E
=D CP , where
P = AnnD (A).
Proof. By Corollary 7.19, E is a D-periodic module. By Corollary 6.25, E =

Q EQ , where EQ is the Q-component of E and =Ass D (A). In particular,
E = EP C, where C is the P -component of E, that is C = Q=P EQ . Then
Lemma 7.6 and the isomorphism EP = E/C at once give that EP is divisible. By
the choice of E, E = EP since A EP . Hence E is a P -module. By Proposition
6.13, if y P \ P 2 , then P = yD + P 2 . Let 0 = a1 A. Since E is divisible, there
are elements a2 , a3 , . . . , an , . . . A such that

an+1 x = an ,

for every n N. Obviously, Ann D (a2 ) =Ann D (a2 D) = P and so Ann D (a2 D) =
P k1 , where k1 > 1. Thus a2 D
= D/P k1 and, in particular, A2 = a2 D contains
copies of submodules isomorphic to

D/P
= P k1 1 /P k1
= A, . . . , D/P k1 1
= P/P k1 .

Put A3 = a3 D. As above, Ann D (a3 ) = Ann D (A3 ) = P k2 and k2 > k1 . Proceeding


 if Ann D (an ) = P
in this way, kn1
, we have that k1 < k2 < < kn < .
Therefore nN An has a submodule B which is isomorphic to

lim {D/P k | k N}.



Hence B = E. 
Lemma 7.23. Let R be a ring. Given a family of R-modules {A | }, suppose
that, for every , E is an essential extension of A . Then E is an
essential extension of A .
Proof. Put  
A= A and E = E .


Given 0 = e E, e = e , where =Supp ((e ) ) is nite. Let
such that 0 = e E . Since E is an essential extension of A , there is
some x1 R such that 0 = e x1 A . If \ {} satises e x1 = 0, then
74 Chapter 7. The KovacsNewman theorem

0 = ex1 = e x1 A, that is eR A = 0. Therefore we may assume that there


exists some \ {} such that e x1 = 0. Since E is an essential extension of
A , there is some x2 R such that

0 = e x1 x2 A .

Note that e x1 x2 A . If \ {, } satises e x1 x2 = 0, then

0 = ex1 x2 = e x1 x2 + e x1 x2 A

and so eR A = 0. Proceeding in the same way, we obtain after nitely many
steps, that eR A = 0. It means that E is an essential extension of A, as
required. 
Corollary 7.24. Let D be a Dedekind domain. If {A | } is a family of D-
modules and, for every ,
E is a D-divisible envelope of A , then E
is a D-divisible envelope of A .

Proof. By Lemma 7.6, U = E is divisible. By Corollary 7.12, U contains
the divisible envelope E of A = A . By Theorem 7.16, there exists a D-
submodule B such that U = E B. In particular, B A = 0 and then, by
Lemma 7.23, U is an essential extension of A. This means that B = 0 and so
E = U. 
Let D be a Dedekind domain, and let A be a D-module. If P is a maxi-
mal ideal of D, let AP be the P -component of A. Since Ann D (P,1 (AP )) = P ,
P,1 (AP ) may be considered as a vector space over the eld D/P . By denition,
the corresponding dimension rP (A) =dimD/P (P,1 (AP )) is called the P -rank of
A. Note that rP (A) is an invariant of the module A.
Let R be an integral domain, and let A be an R-module. If B is a maximal
free subset of A, then |B| is called the torsion-free R-rank of A or the 0-rank of the
R-module A and it is denoted by r0 (A). It is worth mentioning that this invariant
for R-modules is analogous to the dimension of a vector space on a eld.
Theorem 7.25. Let D be a Dedekind domain and let F be the eld of fractions of
D. If E is a divisible D-module and = AssD (E), then E = ( P EP ) U and
(1) EP is the P -component of E;

(2) EP = (P ) C , where C is the Pr ufer P -module for every (P );

(3) U = U , where U =D F , for every .
Moreover, |(P )| = rp (E) and || = r0 (E).
Proof. Let T = tD (E). By Lemma 7.18, T is divisible, and, by Theorem 7.16, there
exists
a D-submodule U such that E = T U . Further, by Corollary 6.25, T =
P EP , where EP is the P -component of E and =Ass D (E). Since
EP is di-
visible by Lemma 7.6, there exists a set (P ) such that P,1 (EP ) = (P ) M ,
Chapter 7. The KovacsNewman theorem 75

where each M is a simple D-module; that is, M = D/P for every (P ).


By Corollary 7.12, EP contains a divisible envelope C(P ) of P,1 (EP ) and, by
Corollary 7.24, 
C(P ) = C ,
(P )

where, for every (P ), each C is a divisible envelope of M , and C = CP


(by Corollary 7.22). Moreover, by Theorem 7.16, there exists a submodule A(P )
such that
EP = C(P ) A(P ).
Since P,1 (EP ) C(P ) and EP is an essential extension of P,1 (EP ), A(P ) = 0;
that is, EP = C(P ). By the denition of (P ), |(P )| = rP (A).
On the other hand, since U = E/T , r0 (E) = r0 (U ). Let S be a maximal D-
free subset of U and dene V = aS aD. If u U , then uD V = 0 , and so U
is an essential extension of V . Since r0 (U ) = |S|, r0 (E) = |S|. Since C
= E/T , by
Lemma 7.6, U is divisible. By Corollary 7.12, U contains a D-divisible envelope
W of V , and, by Corollary 7.24, W = aS Va , where each summand Va is a
D-divisible envelope of aD. By Lemma 7.20, Va =D F . Finally, by Theorem 7.16,
there exists a submodule M such that U = W M and, since V W and U
is an essential extension of V , M = 0; that is, U = W . Thus the theorem is
proved. 
Corollary 7.26. Let D be a Dedekind domain and let F be the eld of fractions of
D. Suppose that A is a D-module and let E be a divisible envelope of A.

(1) E = ( P EP ) U , where if = AssD (E), EP = (P ) C , and C is
ufer P -module for every (P ).
the Pr

(2) U = U , where U =D F , for every .
(3) |(P )| = rP (A) and || = r0 (A).
The above result nishes our preliminary work, and we are now ready to
prove the Kovacs-Newman theorem. The next lemma is the basis for that proof.
We remark that this lemma is originally shown for the ring D = Z, although we
can slightly modify its proof in such a way that it remains true for an arbitrary
Dedekind domain.
Lemma 7.27 (L.G. Kovacs and M.F. Newman [137]). Let D be a Dedekind domain,
G a periodic central-by-nite group, A a DG-module. Suppose that A satises the
following conditions:
(i) A is a monolithic DG-module with monolith M .
(ii) AnnD (A) = P = 0.
(iii) Either char(D/P ) = 0 or char(D/P )  (G).
76 Chapter 7. The KovacsNewman theorem

Then:
n
(1) A = A , where either A = D/P or A is a Pr
ufer P -module for
every . In particular, A is D-periodic and AssD (A) = P .
(2) If E is a proper DG-submodule of A, then there exists some k N such that
E = P,k (A).
Proof. (a) Put
C = (G), S = DC, L = P,1 (A) and A0 = tD (A).
Since the P -component of A contains the monolith, by Corollary 6.25, A0 is exactly
the P -component of A. Given some x Ann S (M ), we choose a nite subgroup
Cx C such that x DCx . By Theorem 5.9, there exists a DCx -submodule M1
such that L = M M1 . Then
Lx = M x M1 x M1 x M1 .
If Lx = 0, then M Lx M1 , a contradiction. Hence Lx = 0; that is,
AnnS (M ) =AnnS (L).
(b) Put T =AnnS (M ) so that T is an ideal of S. Since M P = 0, SP T .
By Theorem 5.5,
M = M0 M0 g1 M0 gm ,
where M0 is a simple S-module and g1 , . . . , gm G. It is rather easy to prove that
T =AnnS (M0 ). Since M0 is a simple S-module, there exists some a A such that
M0 = aS and Ann S (a) = T so that M0 = S/T . Moreover, T is a maximal ideal
of S, i.e. the factor-ring S/T is a eld. If b L and bS = 0, then T Ann S (b)
since AnnS (L) = Ann S (M ) = T . Since T is a maximal ideal, T = Ann S (b). This
means that bS is a simple S-module.
(c) Let B be a maximal S-submodule of L under B M = 0. If b L \ B,
then B does not include bS. Since bS is a simple S-submodule, bS B = 0. By
the choice of B, (bS + B) M = 0 , and so there are some x S and b1 B such
that 0 = b2 = b1 + bx M . Then bx = b2 b1 M + B. If bx = 0, then b1 = b2 .
Since M B = 0, b1 = b2 = 0, a contradiction. Hence bS (M + B) = 0. It
follows that bS B + M, since bS is simple. In other words, L = M B.
Let F be a nite subgroup of G such that G = CF . By Theorem 5.9, there
exists a DG-submodule B1 such that L = M B1 . Since M is the DG-monolith
of A, L = M .
(d) Pick y P \ P 2 and a M . Suppose that there is some n N such that
xy = a has solutions but xy n+1 = a has not. If 0 = b M , then there is some
n

u DG such that bu = a. If there is some b1 A such that b1 y n+1 = b, then


(b1 u)y n+1 = (b1 y n+1 )u = bu = a,
a contradiction. By Proposition 6.13, P = Dy + P n+1 , and therefore either A0 =
P,n (A) or A0 is y-divisible. In the latter, A0 is D-divisible, cf. L.A. Kurdachenko,
J. Otal and I.Ya. Subbotin [157, Proposition 7.2].
Chapter 7. The KovacsNewman theorem 77

For every j N, the mapping j : a ay j , a P,j+1 (A), is a DG-


endomorphism. Hence Im j and Ker j are DG-submodules; moreover Im j
P,1 (A) = M . If A0 = P,j+1 (A), then Im j = 0 , and so Im j = M . In
this case, Ker j = P,j (A), and so P,j+1 (A)/P,j (A)
= M . In particular, this
factor is a simple DG-module. Therefore, if E is a DG-submodule of A0 , then
either E = P,j (A) for some j N or E = A0 .
(e) If Ann D (A0 ) = P t , then

A= A ,

where A = D/P t , for every (see D. W. Sharpe and P. Vamos [265, Theorem
6.14]). If A is D-divisible, then, by Theorem 7.16, A = B , where B is a
Prufer P-module for every . In the rst case, there exists some D-submodule
U such that A = A0 U (see I. Kaplansky [123]). In the second one, by Theorem
7.25, there exists a D-submodule U0 such that A = A0 U0 . Therefore, in any
case, A0 is a direct summand of A; that is, there exists a D-submodule W such
that A = A0 W .
(f) Suppose that Ann D (A0 ) = P t . If A = A0 , then the complement W of
A0 in A is a non-zero torsion-free D-submodule. Moreover,

AP t = A0 P t W P t = W P t W,

and, in particular, AP t is a non-zero torsion-free DG-submodule. However, M


AP t , a contradiction that shows that A = A0 .
(g) Now suppose that A0 is D-divisible. Assume also that M is the S-monolith
of A. If 0 = a A, then aS is a non-zero S-submodule, and so M aS. Thus,
there is some x S such that 0 = ax M . Pick a nite subgroup Cx such
that x DCx . Since A0 is D-complemented in A, by Theorem 5.9, there is a
DCx -submodule V such that

kx A A0 + V and kx (A0 V ) = 0 ,

where kx = |Cx |.
Suppose that char D = p > 0 so that char(D/P ) = p. It follows that (kx , p) =
1 and, then kx A = A and A0 V = 0. In other words, A = A0 V .
Suppose now that char D = 0 but char(D/P ) = p > 0. In this case, the
additive group of A0 is a p-group. Since (kx , p) = 1, A0 V = 0; that is,
A0 + V = A0 V . The mapping : a kx a, a A, is a DCx -endomorphism of
A so that Ker and Im are DCx -submodules. Since the Z-periodic part of A is
A0 ,

Ker = 0 , Im = kx A A0 V,
and
A0 = kx A0 Im .
78 Chapter 7. The KovacsNewman theorem

Therefore
Im = A0 (V Im ),
and we notice that V Im is a DCx -submodule. Since A = Im , we obtain
A = A0 V1 , where A0 = A0 1 and V1 = (V Im ) 1 .
Finally, suppose that char(D/P ) = 0. Consider P,1 (A) as a vector space over
the eld D/P . Since char (D/P ) = 0, P,1 (A) is Z-divisible. By the isomorphism

P,1 (A)
= P,j+1 (A)/P,j (A)

for every j N, A0 is likewise Z-divisible. Therefore A0 = kx A0 . Since the factor-


module A/A0 is D-torsion-free and char D = 0, A/A0 is Z-torsion-free. Hence A
is Z-torsion-free. It follows that the mapping

: a kx a, a A

is then a DCx -monomorphism. Moreover, A0 V = 0. We nd

A
= Im = A0 (V Im )

again, which proves that A0 is also DCx -complemented in A.


Therefore, in any case, A = A0 V2 , for some DCx -submodule V2 . We claim
that A0 x = 0. Otherwise ax Ax = A0 x V2 x = V2 x V2 . Since ax M and
M A0 , this leads to a contradiction, which shows our claim.
Put B =AnnA (x). Since x DC, B is a DG-submodule of A. Since A0 x =
0, B A0 is a proper DG-submodule of A0 . We have already proved that B
A0 = P,j (A), for some j N. It follows that Ann D (B A0 ) = 0 , and then
B = (B A0 ) B1 , where B1 is a D-torsion-free D-submodule provided B1 = 0
(see I. Kaplansky [123]). Then BP j = B1 P j B1 and, in particular, BP j is D-
torsion-free; this conditions assures that BP j = 0. For, otherwise M BP j and
so M B1 , a contradiction. Then B A0 . Since x S, A0 x is a DG-submodule
and, since A0 x = 0, M A0 x. Then ax A0 x; that is, ax = bx for some b A0 .
Thus (ab)x = 0, and so ab B A0 . It follows that a A0 . Therefore A = A0
also holds in this case.
(h) Now suppose that A0 is D-divisible but M is not S-monolith. By Theorem
5.5, there are a simple S-module M0 and elements g1 , . . . , gm G such that

M = M0 M0 g1 M0 gm .

Dene
M 1 = M 0 g 1 , . . . , Mm = M 0 g m .
For every 0 j m, choose Rj to be a maximal S-submodule of A under
s=j Ms Rj and Rj Mj = 0. Then every A/Rj is an S-monolithic module
with S-monolith (Mj + Rj )/Rj . Since Ann D (Mj ) =Ann D (M0 ) = P , by applying
Chapter 7. The KovacsNewman theorem 79

the result shown in the last paragraphs, we deduce that A/Rj is D-periodic and
Ass D (A/Rj ) = {P }. Put

X = R0 R1 Rm .

If X = 0, then X M = 0 and, as we showed above in (c), M = P,1 (A). In


particular, X is D-torsion-free. By Remaks theorem

A/X
A/R0 A/R1 A/Rm ,

and so A/X is D-periodic. It follows that A/Xgj is likewise D-periodic, for every
0 j m. If
X1 = X Xg1 Xgm,
then, by Remaks theorem, again,

A/X1
A/X A/Xg1 A/Xgm ,

which shows that A/X1 is likewise D-periodic. Since X1 is a DG-submodule such


that X1 X, X1 is also D-torsion-free. Consequently, X1 = 0 and hence A is
D-periodic in this case.
Thus, the proof is now completed. 
Theorem 7.28 (L.G. Kovacs and M.F. Newman [137]). Let D be a Dedekind do-
main, G a periodic central-by-nite group, A a DG-module. Suppose that B is a
DG-submodule of A satisfying the following conditions:
(1) B is DG-monolithic with monolith M .
(2) AnnD (M ) = P = 0.
(3) Either char(D/P ) = 0 or char(D/P )  (G).
If B is D-complemented in A, then B is DG-complemented in A.
Proof. By hypothesis, there exists a D-submodule C of A such that A = B C.
Let E be a maximal DG-submodule of A under E M = 0. If there exists
some j N such that BP j = 0, then AP j = CP j . In particular, AP j B = 0.
Clearly, we may choose E such that AP j E. By Lemma 7.27, A/E is D-periodic,
and Ass D (A/E) = {P }. Since B also satises the conditions of Lemma 7.27, there
exists a decomposition 
B= B ,

where either B = D/P t for every , or B is a Pr


ufer P -module for every
. In the rst case, by the choice of E, we have (A/E)P j = 0. By Lemma
7.27, there exists a decomposition

A/E = C ,

80 Chapter 7. The KovacsNewman theorem

where C
= D/P t for every . Since

P,1 (A/E) = (M + E)/E = P,1 ((B + E)/E),

A/E = (B + E)/E, and so A = B E.


Suppose now that B is D-divisible. By Lemma 7.27, A/E is likewise D-
divisible. Since
P,1 (A/E) = P,1 ((B + E)/E),
as above,
A/E = (B + E)/E and A = B E,
as required 
Corollary 7.29. Let D be a Dedekind domain, G a periodic central-by-nite group,
A a DG-module. Suppose that B is a DG-submodule of A satisfying the following
conditions:
(1) B = B1 Bn , where every Bj is DG-monolithic with monolith Mj .
(2) For every 1 j n, AnnD (Mj ) = Pj = 0.
(3) Either char(D/Pj ) = 0 or char(D/Pj )  (G), for every 1 j n.
If B has a D-complement, then B has a DG-complement.
Corollary 7.30. Let F be a eld, G a periodic central-by-nite group, A an artinian
F G-module. If char F = 0 or char F  (G), then A is a semisimple F G-module.
Chapter 8

Hartleys classes of modules

This chapter deals with some specic criteria of semisimplicity that were obtained
by B. Hartley in [101, 103]. In these papers, Hartley considered a specic class
of modules that is wider than artinian modules over Chernikov groups. We will
consider here some results of [101] only.
Let us begin with some technical lemmas.
Lemma 8.1 ([101]). Suppose that r, m, n, k N satisfy m2 | rn 1. Then

1 + rn + + rn(k1) k(mod m2 ).

Proof. We have that

1 + rn + + rn(k1) k = (rn1 1) + + (rn(k1) ) 1

and, for each 1 j k 1,

rnj 1 = (rn 1)(rn(j1) + + 1), j 1

which is divisible by m2 . 
Let G be a group, H a subgroup of G, R a ring and A an RH-module. We
dene
AG = A RH RG.
We consider now the eect of inducing simple modules from a subgroup of an
abelian group to the whole group.
Lemma 8.2 ([101]). Let G be a periodic abelian group and suppose that K is a
subgroup of G such that G/K is a Chernikov locally cyclic group. Let F be a nite
eld such that char F  (G). Then there exists a subgroup H K satisfying the
following properties:
(1) H/G0 is nite.
82 Chapter 8. Hartleys classes of modules

(2) If A is a simple F H-module such that CH (A) = K, then AG is a simple


F G-module.
Proof. It suces to consider the case K = 1. Having dealt with this case, we
may be assured that there is a nite subgroup H/K G/K such that if A is a
simple F (H/K)-module and CH/K (A) = 1 , then AG/K is a simple F (G/K)-
module. In this case, let V be a simple F Hmodule such that CH (V ) = K. We
may naturally view V as a faithful simple F (H/K)-module, and, then the induced
F (G/K)-module V G/K remains simple. We may think of this module as an F G-
module so that, as consequence of the denitions, it is isomorphic to V G . Thus
V G is simple.
Therefore, we suppose that K = 1. Then G is a Chernikov locally cyclic
group, and we may assume that G is innite. Suppose (G) = {p1 , . . . , pk }. Then

G = G1 Gk ,

ufer pj -subgroup, 1 j k. Let H be


where Gj is a cyclic pj -subgroup or a Pr
any nite subgroup containing all elements of G of order dividing p21 p2k . The
group G has an ascending chain

H = H0 H 1 H j

of nite cyclic subgroups such that



G= Hj ,
jN

and all the indexes |Hj+1 : Hj | = qj are prime. Since qj {p1 , . . . , pk }, qj (H),
and hence qj2 divides |Hj |. Therefore Hj has elements of order qj2 , and then

(qj )2 divides |H0 | for every j N. (8.1)

Now we proceed to choose H in a particular way. Let

L = Drp(G) p,2 (G).

Suppose that |F | = r = pt and let n be the smallest natural number such that |L|
divides rn 1; the existence of n is guaranteed since (p, |L|) = 1. Let H be the
unique largest subgroup of G of order dividing rn 1. Then L H, and so n is
the smallest natural number such that |H| divides rn 1. Put q0 = n. We show
by induction on j N that if nj = q0 q1 qj , then Hj is the largest subgroup of
G of order dividing rnj 1 and nj is the smallest natural number j such that |Hj |
divides rj 1.
Suppose that we have already shown this for some j 0. Let X be any nite
subgroup including H such that |X| divides rnj+1 1. Then

rnj+1 1 = rnj qj+1 1 = (rnj 1)(rnj (qj+1 1) + + 1),


Chapter 8. Hartleys classes of modules 83

and so the index |X : Hj | divides

rnj (qj+1 1) + + 1.

Let s be a prime divisor of |X : Hj |. By [1], s2 divides |H0 |, and so s2 divides


rnj 1. By Lemma 8.1,

rnj (qj+1 1) + + 1 qj+1 (mod s2 ). (8.2)

Since s divides the left-hand side of (8.2), we nd that s = qj+1 . It also follows, that
(qj+1 )2 does not divide that left-hand side of (8.2), and hence either |X : Hj | = 1
or |X : Hj | = qj+1 . Since G has at most only one subgroup of any given nite
order, it follows that X Hj+1 . By Lemma 8.1, (8.1) assures that (8.2) holds
with s = qj+1 . Therefore |Hj+1 | divides rnj+1 1, and so Hj+1 itself is the largest
subgroup of G with this property. Furthermore, if |Hj+1 | divides rd 1, then, by
induction, d has the form d = nj u, for some u N. Then qj+1 divides

rnj (u1) + + 1 u(mod qj+1


2
),

and hence qj+1 divides u. This completes the inductive proof.


Let now Fj be the nite eld of order rnj . Since |Hj | divides (rnj 1), there
is a monomorphism : Hj U (Fj ). The additive group of Fj is made into
an F Hj -module by the action ah = a(h), where a Fj and h Hj . Moreover,
CHj (Fj ) = 1. Since |Hj | | rd 1 if and only if nj | d, it follows that Hj generates
the eld Fj ; and so we obtain in this way a simple F Hj -module. Since every simple
module with trivial centralizer arises in this way, we have just seen

dimF A = nj , if A is any simple F Hj -module


such that A has an identity centralizer in Hj . (8.3)

If A is such a module, then the induced module AHj+1 is an F Hj+1 -module of


dimension nj qj+1 = nj+1 over F . Since the F Hj -module AHj+1 is a direct sum
of copies of A, every element of prime order acts xed-point freely on AHj+1 .
Consequently, every simple submodule of AHj+1 is faithful for Hj+1 ; that is, its
corresponding centralizer is trivial. By (8.3), AHj+1 is a simple F Hj+1 -module.
Iterating the above argument, we nd that AHd is simple for every d > j. Since
AG is the union of the F Hd -modules AHd for every d > j, AG is simple. It suces
to take j = 0. 

Further we will use some ring-theoretical results.

Lemma 8.3. Let L be a nil-ideal of a ring R, and let : R R/L the canonical
epimorphism. If a is an element of R such that a is an idempotent, then there is
some b R such that aba = e is an idempotent and a = e.
84 Chapter 8. Hartleys classes of modules

Proof. We have

0 = a (a)2 = a (a2 ) = (a a2 ),

and so aa2 L. Then there exists some n N such that (aa2 )n = 0. Obviously,
we may assume that n 2. Moreover, (1 a)n = 1 ac, for some c R such that
ac = ca. In particular,

1 (ac) = (1 ac) = ((1 a)n ) = (1 a)n = 1 a,

and then (ac) = a. Thus

0 = (a a2 )n = an (1 a)n = an (1 ac);

that is, an = an ac. It follows that an = an (ac)j for every j 1. Making j = n, we


obtain an = an (ac)n = a2n cn , since ac = ca. Put e = (ac)n . Then

e2 = ((ac)n )2 = (ac)2n = a2n cn cn = an cn = (ac)n = e

and so e is an idempotent. Further,

e = ((ac)n ) = ((ac))n = (a)n = a.

Since n 2, e = a(an2 cn )a, as required. 


Let R be a ring, e1 , e2 idempotents of R. We say that e1 , e2 are orthogonal if
e1 e2 = e2 e1 = 0. Let e be a non-zero idempotent of the ring R. We say that e is a
primitive idempotent if e cannot be expressed in a sum of two non-zero orthogonal
idempotents.
If R is a ring, then J(R) denotes the Jacobson radical of R.
Lemma 8.4. Let R be an artinian ring.
(1) If e is a non-zero idempotent of R, then the R-module eR is directly inde-
composable if and only if e a primitive idempotent.
(2) If e is a primitive idempotent, then eR includes an unique maximal R-
submodule eJ(R); in particular, eR/eJ(R) is a simple R-module.
Proof. (1) If e = e1 + e2 and e1 , e2 are non-zero orthogonal idempotents, then
eR = e1 R + e2 R. Let a e1 R e2 R. Then there are x1 , x2 R such that
a = e1 x1 = e2 x2 . Thus e1 x1 = e1 e1 x1 = e1 e2 x1 = 0, and so eR = e1 R e2 R.
Conversely, suppose that R1 , R2 are R-submodules of R such that eR = R1 R2 .
Since R1 , R2 are R-submodules, e1 e2 , e2 e1 R1 R2 = 0. Clearly, e1 , e2 are
idempotents, so that e1 , e2 are orthogonal idempotents.
(2) Consider R/J(R) as a right R-module, and let M be the set of maximal
right ideals of R. We claim that M is nite. Otherwise, M has a countably innite
subset {Mn | n N} such that

M1 Mn = M1 Mn Mn+1 ,
Chapter 8. Hartleys classes of modules 85

for every n N. Hence we may construct an innite strictly descending chain of


right ideals
M1 > M1 M2 > > M1 Mn >
that contradicts the fact that R is artinian. Thus the set M is nite, and it follows
that 
R/J(R)
R/M.
MM

Since each R/M is a simple R-module, R/J(R) is a semisimple R-module of nite


composition length; Corollary 4.3 also yields that
(eR + J(R))/J(R)
=R (eR)/(eR J(R))
is a semisimple R-module. In particular, eR J(R) J(eR). The mapping :
x ex, x R is an R-endomorphism; therefore eR/eJ(R) is likewise a semisimple
R-module, so that eJ(R) J(eR). On the other hand, J(R) J(R), and thus
eJ(R) = J(eR). Since J(R) is a two-sided ideal, eJ(R) J(R). Hence eJ(R) =
J(R) eR. Since eR contains an idempotent e, eR = eJ(R). Let B be an R-
submodule of eR such that eJ(R) does not include B. Then J(R) does not include
B, and (B + J(R))/J(R) is a non-zero R-submodule of the semisimple module
R/J(R). By Corollary 4.3, (B + J(R))/J(R) is a direct summand of R/J(R).
Hence (B + J(R))/J(R) contains a non-zero idempotent, say f + J(R). Since
R is artinian, J(R) is a nilpotent ideal (see D. S. Passman [219, Lemma 5.11,
Lemma 5.1]). By Lemma 8.3, there exists an idempotent 0 = g B such that
f +J(R) = g+J(R). In other words, we may assume that f is an idempotent. Since
f + B eR, there exists some y R such that f = ey. Then ef = e2 y = ey = f .
Put z = f e B. Since ez = ef e = f e = f ee = ze, z 2 = f (ef )e = f (f e) = f e = z.
It follows that
e = z + (e z), and
z(e z) = ze z = (f e)e z 2 = f e z = z z = 0;
2

(e z)z = ez z 2 = ze z = z z = 0;

(e z)2 = (e2 ze) (ez z 2 ) = e2 f ee = e f e = e z.


Thus e = z + (e z) is a sum of two orthogonal idempotents. Moreover, since
0 = f = f 2 = (ef )(ef ) = e(f e)f = e(zf ),
z = 0. Since e is a primitive idempotent, e z = 0; that is, e = z. But in this case
e B, thus B = eR. It follows that eJ(R) contains every proper submodule of
eR. This means that eR/eJ(R) is a simple R-module, as required. 
Lemma 8.5 ([101]). Let G be an abelian group and suppose that G has a nite
subgroup K such that G/K is a Chernikov locally cyclic group. Let F be a nite
eld such that char F  (G). Then there exists a nite subgroup H K satisfying
the following properties:
86 Chapter 8. Hartleys classes of modules

(1) If e is a primitive idempotent in B = F H such that CH (eB) = K, then e


is a primitive idempotent in R = F G, and eR is a minimal ideal of R such
that CG (eR) = K.
(2) If A is a simple R-module such that CG (A) = K, then A
=R eR, for some
primitive idempotent e B such that CH (eB) = K.
Proof. We choose a subgroup H satisfying the properties (1) and (2) of Lemma
8.2. Thus H is nite because K is nite.
(1) Since char F  (H), B is a semisimple F H-module by Corollary 5.15.
In particular J(F H) = 0. By Lemma 8.4, eB is a simple F H-submodule. Now
it is immediate from the denition that eR is isomorphic as right R-module to a
module obtained by including the F H-module eB up to G. Since CH (eB) = K,
by the choice of H and Lemma 8.2, eR is a simple R-module; and so is a minimal
ideal of R. By Lemma 8.4, e is a primitive idempotent of F G. Since eR
= (eB)G ,
it is straightforward to obtain that CG (eR) = K.
(2) Since A is a simple R-module, there exists some maximal ideal M of R
such that A =R R/M . By Corollary 5.15, the F H-module A is semisimple. Let C
be a simple F H-submodule of A. By Lemma 5.4, there is a subset S of G such
that 
A= Cx.
xS

Since G is abelian, Cx
=F H C. It follows that A and every F H-submodule of A is
a direct sum of copies of C, and hence CH (C) = K. Since B is nite dimensional
and semisimple, by Corollary 4.3,

B = (B M ) D1 Dk ,

where D1 , . . . , Dk are simple F H-submodules of B. Given 1 j k, by Lemma


8.4, there exists a primitive idempotent ej such that Dj = ej B. Since ej B are
pairwise non-isomorphic as B-modules, it follows that k = 1. Now B/(B M ) is
naturally isomorphic to an F H-submodule of the F H-module A. Hence, if e = e1 ,
then eB = C and CH (eB) = K. By (1), eR is a minimal ideal of R. Since M does
not include eR, R = M eR, and hence R/M =R eR. 

Lemma 8.6 ([101]). Let G be a Chernikov abelian group having a locally cyclic sub-
group of nite index. Suppose that F is a nite eld such that char F  (G) and
let A be a simple F G-module such that CG (A) is nite. Then there is an uniquely
determined primitive idempotent e F G such that A =F G e(F G). This corre-
spondence determines a bijection between the isomorphism classes of the simple
F G-modules A with CG (A) nite and the primitive idempotents in F G.
Proof. The factor-group G/CG (A) is locally cyclic (see L.A. Kurdachenko, J. Otal
and I.Ya. Subbotin [157, Theorem 2.3]). Thus, the existence of e follows from
Lemma 8.5. Since G is abelian, eF G and (1 e)F G are mutual annihilators in
Chapter 8. Hartleys classes of modules 87

F G, whence, if e1 is an idempotent in F G. Therefore, if e1 F G is an idempotent


such that e1 F G
=F G eF G, we obtain

(1 e)F G = (1 e1 )F G and eF G = e1 F G.

Since both e and e1 are the identity of the subring eF G, we have e = e1 .


Finally, if e is any primitive idempotent of F G, then e F H, for some nite
subgroup H of G and since eF G = (eF H)G , we have thar CG (eF G) H. 

Lemma 8.7 ([101]). Let G be a periodic abelian group such that (G) is nite, and
suppose that K is a nite subgroup of G. If F is a nite eld, then the number of
isomorphism types of simple F G-modules A such that CG (A) = K is nite.
Proof. We clearly may assume that K = 1. Then G is a locally cyclic p -group,
where p =char F (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Theorem
2.3]). Since (G) is nite, G is Chernikov. By Lemma 8.5, A
= eF G, where e is a
primitive idempotent in F H and H is a nite subgroup satisfying the conditions
of Lemma 8.2. Since H is nite, the set of such primitive idempotents is nite.
This shows the result. 

Let R be a ring, G a group, A an RG-module. The set S of RG-submodules


(called the terms of S) of A is called a series of A if it satises the following
conditions:
(i) S is linearly ordered by inclusion.
(ii) If 0 = a A, then there are terms of S that do not contain a, and the union
of all such terms is a term V (a) of S.
(iii) If 0 = a A, then there are terms of S that contain a, and the intersection
of all such terms is a term (a) of S.
(iv) Each term of S has the form V (a) or (a), for some 0 = a A.
(v) 0 , A S.
Thus a belongs to \ V (a). The corresponding factor-module (a)/V (a) is
called a factor of S. If S and S1 are series of A, S1 is said to be a renement of S
if every term of S is a term of S1 . A series that has no renement other than itself
is called a composition series (more precisely, an RG-composition series). Clearly,
in such series every factor is a simple RG-module. As in groups, every series of
submodules can be rened to a composition series.
Lemma 8.8 ([101]). Let G be a periodic abelian group, and let F be a eld with
char F = p, where, if p > 0, then p  (G). Suppose that A is an F G-module and
put L = AnnF G (A). If A has a composition series whose factors fall into nitely
many isomorphism types, then F G/L is a direct sum of nitely many elds, and
A is a semisimple F G-module.
88 Chapter 8. Hartleys classes of modules

Proof. Let S be a composition series of submodules of A. Suppose that each factor


of S is isomorphic to one of the nitely many pairwise non-isomorphic F G-modules
X1 , . . . , Xn . For each 1 j n, we put Lj =Ann F G (Xj ) and
U = L1 Ln .
Let
: F G/U (F G/L1 ) (F G/Ln ) = B
be the homomorphism given by (x + L) = (x + L1 , . . . , x + Ln ), x F G.
Clearly, is both a ring homomorphism and an F G-homomorphism. Since
the F G-modules A/L1 , . . . , A/Ln are simple and pairwise non-isomorphic, every
F G-submodule of B is a direct sum of some of them and, since the image of
F G/U in B projects onto each summand F G/Lj , it must be the whole of B.
Hence F G/U = B as rings. Since every Xj is a simple F G-module, F G/Lj is a
eld and so F G/U is a direct sum of nitely many elds.
Obviously, U L, and we claim that equality holds. This will show that
F G/L is a direct sum of nitely many elds. Hence, it is a semisimple artinian
ring. Therefore A, which is naturally an F G/L-module, is semisimple.
Let u U . Then there is a nite subgroup H of G such that u F H. By
Corollary 5.15, the F H-module A is semisimple. Let C be a simple F H-submodule
of A, and choose 0 = c C arbitrary. Then c (c) \ V (c), where (c), V (c) S.
Since C is a simple F H-submodule, (c) C and V (c) C = 0, so that C is
isomorphic to an F H-submodule of (c)/V (c). Since u U F H, it follows that
Cu = 0. Since A is a direct sum of copies of the F H-submodule C, it follows
that Au = 0. Therefore u L, and U = L as claimed. 
Lemma 8.9 ([101]). Let G be a Chernikov abelian group and suppose we have an
innite set {K | } of distinct subgroups. Then there exists a subgroup B G
such that
(1) B is contained in only nitely many subgroups K ; and

(2) B = nN Bn , where
B1 B2 Bn
and each Bn is contained in innitely many subgroups K .
Proof. We proceed by induction on the sum of the ranks of the Sylow p-subgroups
of G. Since G contains nitely many elements of prime order, it follows that some
of such elements lie in innitely many subgroups K . Then there exist a non-
identity subgroup X1 and an innite subset 1 such that X1 K , for every
1 . This means that innitely many subgroups K /X1 are distinct. Applying
the same argument, there exist a non-identity subgroup X2 /X1 and an innite
subset 2 1 such that X2 /X1 K /X1 , for every 2 . Iterating, we obtain
a tower of subgroups of G
1 = X0 < X1 < < Xn < ,
Chapter 8. Hartleys classes of modules 89

each
 of which is contained in innitely many of the subgroups K . Put X =
nN Xn . If X lies in nitely many subgroups K , then it suces to dene B = X
and Bn = Xn , for every n N. Otherwise, the result follows by applying induction
to G/X. 
Let G be a locally nite group. Suppose that F is a eld of characteristic
p 0, and let A be an F G-module. Then A is said to be an Mc -module over F G if
for each p -subgroup H of G the set of centralizers in A of subgroups of H satises
the minimal condition (B. Hartley [101]) where,as usual, by 0 we mean the set
of all primes. It is straightforward that A is an Mc -module over F G if and only
if each p -subgroup H of G has a nite subgroup K such that CA (H) = CA (K).
Note that every artinian F G-module is trivially an Mc -module.
Theorem 8.10 ([101]). Let G be a Chernikov group. Suppose that F is a nite eld,
and let A be an Mc -module over F G. If char F  (G), then A is semisimple F G-
module.
Proof. First we suppose that G is abelian. Let S be any composition series of A. If
0 = a A, then we put Ka = CG ((a)/V (a)). Suppose, if possible, that innitely
many distinct subgroups of G lie in K = {Ka | 0 = a A}. By Lemma 8.9, there
exists an ascending chain of subgroups

B1 B2 Bn

such that if B = jN Bj , then B is contained in nitely many subgroups Ka
whereas each Bj is contained in innitely many of them. Since A is an Mc -module,
there is a nite subgroup E of B such that CA (B) = CA (E). Then there exists
some j N such that E Bj , and so there is some a A such that E Ka , but
Ka does not include B. If C and D are F G-submodules of A such that C D and
we put C1 /D = CC/D (E), by Corollary 5.15, there exists some F E-submodule
C2 such that C1 = C2 D. By the choice of C1 , C2 CA (E), and therefore
CC/D (E) = (CC (E) + D)/D. Since E Ka , the above argument gives that

(a)/V (a) = C(a)/V (a) (E) = (C(a) (E) + V (a))/V (a)


= (C(a) (B) + V (a))/V (a) C(a)/V (a) (B).

Hence, B Ka , a contradiction. Consequently, only nitely many subgroups of G


occur among Ka . This and Lemma 8.7 give that the composition factors (a)/V (a)
fall into nitely many isomorphism types as F G-modules. By Lemma 8.8, A is a
semisimple F G-module.
In the general case, let G0 be a normal abelian subgroup of G having nite
index. Then A is an Mc -module over F G0 , and so A is a semisimple F G0 -module.
Let C be a non-zero F G-submodule of A. Since C is an F G0 -submodule, by
Corollary 4.3, there exists an F G0 -submodule D such that A = CD. By Theorem
5.9, there exists an F G-submodule U such that A = C U . Applying Corollary
4.3 again, we obtain that A is a semisimple F G-module, as required. 
90 Chapter 8. Hartleys classes of modules

Lemma 8.11 (A.D. Gardiner, B. Hartley and M.J. Tomkinson [82, Lemma 3.2]).
Let G be a locally nite group, and suppose that F is a eld with char F = p > 0.
If A is a simple F G-module such that CG (A) = 1, then Op (G) = 1.

Proof. Assume the contrary, and choose 1 = x Op (G). Put X = x , and let a
X
be an element of A such that ax = a. Dene A0 = a . Since X is nite, A0 is a
nite subgroup of the additive group of A. If 0 = u A0 , then uF G = A. Hence,
there exists a nite subgroup Gu G such that u(F Gu ) A0 . Put

H = x, Gu | 0 = u A0  .

Then H is nite and C = A0 F H is nite dimensional over F . Since u(F Gu ) A0 ,


every non-zero element of A0 generates C as F H-submodule. Thus if we choose an
F H-submodule C1 of C maximal under C1 A0 = 0, then C1 is a maximal F H-
submodule of C and C/C1 is a simple F H-module. Now x Op (G) H Op (H).
Since C/C1 is a simple F H-module, cf. K. Doerk and T.O. Hawkes [56, Theorem
10.5], Op (H) CH (C/C1 ). However A0 = (A0 + C1 )/C1 as X-groups, and so x
acts as the identity on A0 . This contradiction proves the result. 

If A is a module over a ring R, then the intersection of all maximal R-


submodules of A is called the Frattini submodule of A and denoted by F rattF G (A).

Theorem 8.12 ([101]). Let G be a Chernikov group. Suppose that F is a nite eld,
and let A be an Mc -module over F G. Then A is a semisimple F G-module if and
only if FrattF G (A) = 0.

Proof. If A is a semisimple F G-module, obviously F rattF G (A) = 0. Conversely,


suppose that F rattF G (A) = 0. We clearly may assume that CG (A) = 1. If char
F = p and L = F Op (G) is the augmentation ideal of the group-ring F Op (G),
by Lemma 8.11, AL M , for every maximal F G-submodule M. Hence, AL
F rattF G (A) = 0. It follows that Op (G) = 1.
Let H be a normal abelian subgroup of G of nite index in G. Since Op (H) =
1, p  (H). Clearly, A is an Mc -module over F H, and so, by Theorem 8.10,
A is a semisimple F H-module. Let B be a simple F H-submodule of A. Choose T
to be a transversal to H in G, and put

C= Bx.
xT

By Lemma 5.4, Bx is a simple F H-module for every x T . Since T is nite,


by Lemma 1.1 and Lemma 1.2, C satises Max-F H and Min-F H. But C is an
F G-submodule, so it actually satises Max-F G and Min-F G. If M is a maximal
F G-submodule of A, then either M C, or A = M + C. In the second case, we
note that
C/(C M )
=F G A/M
Chapter 8. Hartleys classes of modules 91

is a simple F G-module, and so C M is a maximal F G-submodule of C. It follows


that F rattF G (C) = 0. Since C satises Min-F G, there are nitely many maximal
submodules M1 , . . . , Mn of C such that

M1 Mn = 0 .

By Remaks theorem,
C
C/M1 C/Mn .
By Corollary 4.4, C is a semisimple F G-submodule. It follows that C SocF G (A).
From the choice of C we obtain that A =SocF G (A), that is A is a semisimple F G-
module. 
It is worth mentioning that Theorem 8.10 and Theorem 8.12 can fail if (G)
is innite. Specic examples of this were constructed by B. Hartley in [101]. For
the readers convenience, we quote them here but we have omitted their proofs.
Theorem 8.13 ([101]). Let p be a prime. Then there exists a group G satisfying
the following conditions:
(1) G =Drq gq , where |gq | = q and is an innite set of primes.
(2) there are 20 pairwise non-isomorphic simple F G-modules which are faithful
for G, where F = Fp is the nite eld of p elements.
(3) the ring F G contains no primitive idempotents and no minimal ideals.
(4) There is an Mc -module A over F G such that F rattF G (C) = 0 but A is
not semisimple.
B. Hartley [103] continued the study of Mc -modules over group-rings. In this
case, the underlying group has a periodic abelian subgroup of nite index. We also
quote the main result of the above paper, omitting the proof as well. We will need
the following denition.
Suppose that G is a locally nite group, F is a eld and A is an F G-module.
Then A is called a strong locally semisimple module if G has a nite subgroup
K such that A is a semisimple F H-module whenever H is a nite subgroup such
that H K (B. Hartley [103]).
Theorem 8.14. Let G be a periodic abelian-by-nite group. Suppose that F is a
eld, and let A be an Mc -module over F G. Then the following three conditions
are equivalent:
(1) F rattF G (A) = 0.
(2) A is a strong locally semisimple module.
(3) every composition factor of A has an F G-complement.
Chapter 9

The injectivity of some simple modules

In this chapter we consider a more general topic than the semisimplicity of mod-
ules over group rings. In a semisimple module all submodules are complemented
(Corollary 4.3), and so, in particular, every simple submodule has a complement.
On the other hand, if a simple module A has a complement in every module E in-
cluding A, then A is injective (see Proposition 7.2). The commutative rings R such
that every simple R-module is injective have been already described. By a result
due to I. Kaplansky, such rings are necessarily von Neumann regular rings. We are
not going to study the non-commutative case in depth here. We will focus on mod-
ules over group rings and our objectives in this chapter are some important results
that were obtained by D.R. Farkas and R.L. Snider [69] and B. Hartley [107]. If
A is a simple RG-module where R is a commutative ring, then L = AnnR (A) is
a maximal ideal of R, and we may think of A as an F G-module where F = R/L
is a eld. Therefore we will focus on modules over a group ring F G, where F is a
eld.
The rst problem that arises here is the characterization of the injectivity
of the trivial F G-module. A eld F can be thought as a right F G-module under
the trivial action. Specically, if a F and x F G, then ax = a(x), where
denotes the augmentation map. If H is a subgroup of the group G, then, by
considering maps from ideals of F H to F , we can easily show the following result.
Lemma 9.1. Let F be a eld, and let G be a group. If H is a subgroup of G and
F is F G-injective, then F is likewise F H-injective.
Lemma 9.2 (D.R. Farkas and R.L. Snider [69]). Let G be a nite group. Then a
eld F is F G-injective if and only if charF  (G).
Proof. If char F  (G), by Corollary 5.15, F G is semisimple; then it is clear that
F is F G-injective.
Suppose that char F = p > 0, and F is F G-injective. Suppose that the
result is false, and let x be an element of G of order p. By Lemma 9.1, F is
F x-injective. Let L be the augmentation ideal of F x. We recall that L admits
94 Chapter 9. The injectivity of some simple modules

{x 1, x2 1, . . . , xp1 1} as F -basis. Dene the mapping : L F by

(1 (x1 1) + + p1 (xp1 1)) = 1 1 + + p1 (p 1),

and extend to F G. Then there exists some c F such that 1 = (x 1) =


c(x 1) = 0, a contradiction. 
Theorem 9.3 (G.O. Michler and O.E. Villamayor [193], O.E. Villamayor [278]).
Let G be a group. Then a eld F is F G-injective if and only if G is locally nite
and char F  (G).
Proof. Suppose that F is F G-injective but G is not locally nite. Then there exist
g1 , . . . , gn G such that G1 = g1 , . . . , gn  is innite. Put

 G 
R=F F G ,
n

and dene a mapping : F G R by

x = ((g1 1)x, . . . , (gn 1)x), x F G.

Clearly, is an F G-homomorphism. If z Ker , then

(gj 1)z = 0 or gj z = z for every j, 1 j n.

Therefore,
g1 z = = gn z = z,
and it follows that gz = z for every g G1 . Thus the coecients of z are constant
on the left cosets of G1 . Since Supp z is nite and G1 innite, we have z = 0. Hence
Ker = 0 , and is a monomorphism. Let : F G F be the augmentation
map of F G. Since F is F G-injective, there exists a mapping : R F such
that = . If 1 j n, let

j = (0, . . . , 0, 1, 0, . . . , 0) R

be the n-tuple with all zeros unless a unique 1 is in the place j, and dene j = j .
Then

1 = 1 = 1() = (1) = (g1 1, . . . , gn 1)


= ((g1 1)1 + + (gn 1)n ) = (g1 1)1 + + (gn 1)n = 0,

a contradiction. This contradiction shows that G is locally nite. The other con-
dition follows from Lemma 9.1 and Lemma 9.2.
Conversely, suppose that G is a locally nite group and char F  (G). Let
L be a non-zero right ideal of F G, and let : L F a non-zero mapping. Pick
u L such that u = 0. Since G is locally nite, H = Supp u is nite and
L F H = 0. By Lemma 9.2, the restriction of to L F H can be extended
Chapter 9. The injectivity of some simple modules 95

to F H; that is, there exists some F such that a = a = (a) for every
a LF H. Pick b L. There exists a nite subgroup S such that Supp b S and
H S. As above, there exists some F such that b = (b) and u = (u).
Then ( )(u) = 0. But (uw) = 0 since u = 0. Therefore = . Thus b = b
and can be extended. By the criterion of Baer (Theorem 7.1), F is F G-injective,
as required. 
In a natural way, Theorem 9.3 leads us to the following class of algebras. Let
A be an algebra over a eld F . Then A is said to be a locally Wedderburn algebra
if every nite set of elements in A generates a semisimple subalgebra having nite
dimension over F .
Now we study the conditions of injectivity of simple F G-modules. Let R be
a ring and A be a simple R-module, and put E =EndR (A). If EndR (A), then
Ker and Im are R-submodules of A. Therefore, if = 0, then Im = A and
Ker = 0. Hence, every non-zero endomorphism of a simple R-module A is an
automorphism. In other words, E is a division ring. Thus we can naturally assume
A to be in a vector space over E. On the other hand, let G be a group and let B
be an RG-module. If x RG, then the mapping x : b bx, b B is an R-
endomorphism, and the mapping : x x , x RG is a ring homomorphism
of RG in the ring EndR (B) of all R-endomorphisms of B.
Lemma 9.4 (D.R. Farkas and R.L. Snider [69]). Let W be a locally Wedderburn al-
gebra over a eld F . Suppose that A is a simple W -module and put E = EndW (A).
If dimE (A) is nite, then A is injective.
Proof. Suppose that the result is false. By the criterion of Baer (Theorem 7.1),
there exist a right ideal L of W and a W -homomorphism : L A, which
cannot be lifted to W . Given a subalgebra V of W , we put

D(V ) = {a A | ax = x for every x L V }.

Let M be the family of all nite dimensional semisimple subalgebras of W . Our


assumption yields 
D(V ) = .
V M

Since all modules over such an algebra are injective, D(V ) = , for each V
 We pick U M such that the dimension d of L/(L U ) is minimal. Since
M.
V M D(V ) = , there exists a subalgebra V1 M such that D(V1 ) does not
include D(U ). If V2 M contains a basis for U and V1 , then U V2 implies

= D(V2 ) D(U ) and D(V2 ) = D(U ).

Thus, if D(V2 ), then

D(V2 ) = + Ann A (L V2 ) + AnnA (L U )


96 Chapter 9. The injectivity of some simple modules

and
+ AnnA (L V2 ) = + Ann A (L U ) = D(U ),
contradicting the minimality of d. 
The next result is due to B. Hartley and leads us in a natural way to the
class of abelian-by-nite groups.
Lemma 9.5 (B. Hartley [107]). Let A be a simple F G-module, where G is an
abelian-by-nite group and F is a eld. If E = EndF G (A), then dimE (A) is nite.
Proof. Let K = (F G) be the image of F G in End F (A). Consider rst the case
in which G is abelian. Then x is an F G-endomorphism for every x F G. Given
E, we choose a xed non-zero element a of A. Since A is a simple F G-module,
there exists some z F G such that a = az. If b is an arbitrary element of A,
then there exists some x F G such that b = ax. Therefore

b = (ax) = (a)x = (az)x = a(zx) = a(xz) = (ax)z = bz,

and it follows that = z so that E = K. In particular, K is a eld.


In the general case, let H be an abelian normal subgroup of G of nite
index, and put R = (F H). By Theorem 5.5, there are nitely many simple
F H-submodules B1 , . . . , Bn such that

A = B1 Bn .

For each 1 j n, we put Lj =Ann R (Bj ). By the result shown in the above
paragraph, each R/Lj is a eld and clearly

L1 Ln = Ann R (A) = 0 .

Therefore R is a direct sum of nitely many elds; so, in particular, it is an artinian


ring. Let S be a transversal to H in G. Then K = xS xR and so, in particular,
K satises the minimal condition for R-submodules and a fortiori for right ideals.
Since K is a simple ring with the minimal condition for right ideals, the classical
WedderburnArtin structure theorem (see, for example, [66, Theorem 7.9]) shows
that there are a positive integer r and a division ring D such that K = Mr (D), the
ring of r r-matrices over D. Moreover, A is isomorphic to the natural Mr (D)-
module of r-row vectors with entries in D. Of course, this module has dimension
r over its endomorphism ring D, which proves the general case. 
Let F be a eld of characteristic p 0. Following B. Hartley [107], we say
that a locally nite p -group G is restricted if every simple F G-module has nite
dimension over its endomorphism ring (as usual, 0 denotes the set of all primes).
Our next goal is to prove here an important result due to B. Hartley, in which
it is shown that a restricted group is abelian-by-nite. To deal with this, some
technical results are needed.
Chapter 9. The injectivity of some simple modules 97

Lemma 9.6 ([69]). Let S be a ring, and let R be a subring of S. Suppose that
A is a simple R-module, B is a simple S-module, and : A B is an R-
monomorphism. Put E = EndR (A) and D = EndS (B). If X is a subset of A
linearly independent over E, then X is linearly independent over D.
Proof. Suppose that
(a1 )1 + + (an )n = 0,
where a1 , . . . , an are dierent elements of X and 1 , . . . , n D. Applying the
Jacobson-Chevalley Density Theorem (see, for example, C. Faith [67, Theorem
19.22]), for every 1 j n, there exists some uj R such that aj uj = aj ,
whereas ak uj = 0, provided k = j. Then

0 = ((a1 )1 ) + + (an )n )uk


= a1 uk 1 + + an uk n = ak k .

Since ak = 0 and D is a division ring, k = 0, establishing the result. 


Lemma 9.7 ([107]). Let F be a eld with char F = p, and suppose that G is
a restricted locally nite p -group. If V is a subgroup of G and U is a normal
subgroup of V , then V /U is likewise restricted.
Proof. Let A be a simple F (V /U )-module. We can think of A as a simple F V -
module by allowing U to act trivially; this does not aect the endomorphism
dimension. Then A = F V /M , for some maximal right ideal M of F V . Since
M (F G) is a proper right ideal of F G, there is a maximal right ideal L of F G
including M , and necessarily L F V = M . Thus we have an F V -monomorphism
of A = F V /M into the simple F G-module F G/L. By assumption, the latter
has nite endomorphism dimension. Lemma 9.6 implies that A also has nite
endomorphism dimension. 
Now we consider some needed facts about linear groups over division rings.
Lemma 9.8 ([107]). Let F be a eld, and suppose that D is a division algebra over
F and A is a vector space of nite dimension over D. Let G be a locally nite
p -subgroup of GL(D, A). Then:
(1) For each prime q, the Sylow q-subgroups of G are Chernikov.
(2) if for every subgroup H of nite index the DH-module A is indecomposable,
then every nite abelian normal subgroup of G is cyclic.
Proof. We think of F as naturally embedded in the ring of D-endomorphisms
of A. Let R be the subring of EndD (A) generated by F and G. Thus, R is a
homomorphic image of F G.
(1) Let Q be a nite elementary abelian q-subgroup of G, and let K be the
image of F Q in R. Then K is a nite dimensional F -algebra. We may write

1 = e1 + + et ,
98 Chapter 9. The injectivity of some simple modules

where e1 , . . . , et are orthogonal primitive idempotents in K. Then

A = Ae1 Aet ,

and we note that each summand is a non-zero D-subspace. Therefore t n =


dimD (A). Furthermore,
K = Ke1 Ket ,
and the summands are elds. We note that the projection of Q into each summand
has order 1 or q, and |Q| divides q t , q t |q n . In particular, every elementary abelian
q-subgroup of G is nite. By a result due to S. N. Chernikov [43, Theorem 4.1]
every Sylow q-subgroup of G is Chernikov.
(2) Suppose the contrary, and let U be a nite noncyclic abelian normal
subgroup of G. Denote by L the subring of EndD (A) generated by F and U . Then
L is not a eld, and so there exist non-zero orthogonal idempotents e1 , e2 L
such that 1 = e1 + e2 . It follows that

A = Ae1 Ae2

and the latter two summands are non-zero D-subspaces. Let C = CG (U ). Then
G/C is nite, and, since C commutes with e1 and e2 , Ae1 and Ae2 are C-invariant.
Thus A is a direct sum of two non-zero C-invariant subspaces. This contradiction
proves this result. 
Before dealing with the proof of the basis theorem, we require another result
about modules for normal subgroups. Let H be a normal subgroup of a group G
and let U be an F H-module. If g G, by denition, the conjugate F H-module
U g has U as underlying space with the H-action

(u, x) u(gxg 1 ), where u U and x H.

The stabilizer of U (or the inertia group of U ) is the set

S = {g G | U g
=F H U }.

Clearly, S is a subgroup of G and there exists a bijection between the isomorphism


classes of conjugates of U and the set of the right cosets Sg, g G. Indeed these
facts are all well known (see B. Huppert [115]).
Lemma 9.9 ([107]). Let F be a eld, and suppose that H is a normal subgroup of
a restricted group G. If U is a simple F H-module, then the stabilizer S of U in G
has nite index in G.
Proof. If |G : S| is innite, then U has innitely many pairwise non-isomorphic
conjugates under G. As in Lemma 9.7, U can be embedded in a simple F G-module
A. If g G, then U g =F H U g; so the F H-module A has innitely many homoge-
neous components. Each one of these is invariant under the ring E =End F G (A).
Thus, dimE (A) is innite, and G cannot be restricted. 
Chapter 9. The injectivity of some simple modules 99

The next proposition is the key step of the proof of the basis result. We show
it independently because the result is interesting in its own way.
Proposition 9.10 ([107]). Let F be a eld with charF = p. Suppose that G is
a locally nite p -group. If G is not abelian-by-nite, then there exists a simple
F G-module A that satises one of the following equivalent conditions:
(1) G/CG (A) is not abelian-by-nite.
(2) (F G) satises no non-trivial polynomial identity.
Proof. We rst show that (1) and (2) are equivalent. Indeed, if Q = G/CG (A)
is abelian-by-nite, then (see D.S. Passman [216, Theorem 5.1]) the algebra F Q
satises the standard identity; hence, so does the image of F G in EndF (A), because
this is a homomorphic image of F Q. Conversely, suppose that R = (F G) satises
a non-trivial polynomial identity. Since R is a primitive ring, by a result due to
Kaplansky (see D. S. Passman [216, Theorem 6.4]), R can be embedded in a ring
of matrices over a eld, which is a eld extension of F . By B.A.F. Wehrfritz [280,
Corollary 9.4], G/CG (A) is abelian-by-nite.
Thus, it suces to show (1). First of all, we note that G has a countable
subgroup H that is not abelian-by-nite (D.S. Passman [216, Lemma 13.2]). Ap-
plying the argument of Lemma 9.7, every simple F H-module B can be embedded
in a simple F G-module A. Moreover, if H/CH (B) is not abelian-by-nite, neither
is G/CG (A). Therefore, replacing G by H, we may assume that G is countable.
Then G has an ascending chain of nite subgroups

H1 H 2 Hn

such that 
G= Hn .
nN

We consider all the sequences

[S] S1 S2 Sn ,

in which each Sn is a minimal two-sided ideal of F Hn , and the mapping into Sn


is the monomorphism induced by the natural projection of F Hn onto Sn resulting
from writing F Hn as a direct sum of minimal ideals. Let rn be the smallest natural
number m such that Sn satises the standard identity sm . Clearly

r1 r2 rn ,

and the argument falls into two cases.


Case 1. There exists a sequence [S] for which rn . We note that [S]
induces
[S1 ] A1 A2 An ,
100 Chapter 9. The injectivity of some simple modules

in which every An is a minimal right ideal of Sn (and so of F Hn ), and the mapping


into An+1 is an F Hn -monomorphism. Having obtained An , we map it into Sn+1
through the sequence [S], and then project onto a suitable minimal right ideal
An+1 of Sn+1 . The injective limit of the sequence [S1 ] is a simple F G-module
A, and the image (F G) of F G in EndF (A) cannot satisfy a standard identity,
since it includes isomorphic copies of the Sn for each n N. If Q = G/CG (A) is
abelian-by-nite, then F Q satises an standard identity; hence, so does (F G),
a contradiction. This nishes Case 1.
Suppose now that for each j 1 and for a minimal (two-sided) ideal Uj of
F Hj , there exists some k j and a minimal ideal Uk of F Hk such that there is a
monomorphism Uj Uk and Uk fails to satisfy some standard identity satised
by Uj . Then we may clearly construct a sequence [S] to which Case 1 applies. In
the contrary case, by omitting some of the Hn and re-indexing the remainders, we
may suppose that H1 and S1 satisfy the following:
Case 2. For every sequence [S] beginning with S1 , we have rn = r for each
n N.
As indicated above, any sequence [S] induces a sequence [S1 ]. If n N, we put
En =End F Hn (An ). Clearly, En =End Sn (An ). Let dn be the dimension of An over
En . By Lemma 9.6, the image of an En -basis of An in An+1 is En+1 -independent,
and so
d1 d2 dn .
Applying the JacobsonChevalley Density Theorem (see, for example, C. Faith [67,
Theorem 19.22]), we get that Sn is isomorphic to a ring of dn dn matrices over
En , which includes a ring of dn dn matrices over F . Since each Sn satises the
standard identity sr , the AmitsurLevitzki theorem (D.S. Passman [216, Lemma
4.2]) shows that dn are bounded by a number depending only on r. By considering
a sequence [S] in which the least such bound is attained by omitting the rst few
terms and re-indexing the remainders, we may suppose
d1 = = dn = = d,
for every sequence [S] starting with S1 .
Consider now a sequence [S] and its induced sequence [S1 ]. Let e be a central
idempotent of F H1 generating S1 , that is, the identity of the ring S1 . If n 1, we
know that the image in An of an E1 -basis of A1 is an En -basis of An . Therefore
the projection of e in Sn xes by right multiplication an En -basis of An and it has
to be the identity of Sn . Write
F Hn = T 1 T s ,
where T1 , . . . , Ts are minimal ideals and the ordering is such that
e = a1 + + at ,
where t s and 0 = aj Tj for every 1 j t. Then each Tj can be taken as
the nth -term Sn of a sequence [S]. Therefore aj is the identity of each such Tj ,
Chapter 9. The injectivity of some simple modules 101

and each such Tj satises sr . Hence, e is central in F Hn and eF Hn satises sr . It


readily follows that e is a central idempotent of F G and so eF G satises sr .
From D.S. Passman [215], |G : F C(G)| is nite and [F C(G), F C(G)] is
nite. Since G is a p -group, F G is semisimple, and then for every 1 = g
[F C(G), F C(G)] there exists a simple F G-module Ag such that g  CG (Ag ). Let
Y be a nite subset such that [F C(G), F C(G)] = Y . If every group G/CG (Ag )
has an abelian subgroup Vg /CG (Ag ) of nite index, then putting

W = F C(G) ( Vg ),
gY

we obtain that
 
[W, W ] [F C(G), F C(G)] ( CG (Ag )) = 1 .
gY

Since W has nite index in G, we get a contradiction to the assumption that G


cannot be abelian-by-nite. Thus, one of the groups G/CG (Ag ) cannot be abelian-
by-nite, and all is done. 
Now we consider some particular cases of the mentioned theorem, which are
needed for its proof. The next result was originally proved by R. L. Snider in [267],
but we prefer to follow the proof given by B. Hartley in [107].
Proposition 9.11. Let F be a eld with char F = p, and suppose that G is a
restricted p -group. If G is locally nilpotent, then G is abelian-by-nite.
Proof. Suppose the contrary. By Lemma 9.7 and Proposition 9.10, we may assume
that there exists a simple F G-module A such that CG (A) = 1. Put E =End
F G (A). Since G is restricted, dimE (A) is nite. We have

G = Dr q(G) Gq ,

where Gq is the Sylow q-subgroup of G. By Lemma 9.8, each Gq is Cernikov; then,


in particular, it is abelian-by-nite. Therefore, there is an innite subset (G)
such that Gq is not abelian for every q . Since each one of these Gq has a nite
non-abelian subgroup, by Lemma 9.7 we may assume that

G = Dr q(G) Gq ,

where (G) is innite and each Sylow q-subgroup Gq is nite non-abelian. We


may even assume that 2  (G) and every proper subgroup and factor group of
Gq is abelian. If q (G), it follows that Gq is nilpotent of class 2, Zq = (Gq ) is
cyclic, and |[Gq , Gq]| = q. Moreover, Gq has a non-cyclic maximal abelian normal
subgroup Uq (see D. Gorenstein [91, Theorem 5.4.10]). Let Vq be a subgroup of
Uq , maximal under Vq Zq = 1. Then Uq /Vq is cyclic, Vq = 1. Moreover, if

U = Dr q(G) Uq and V = Dr q(G) Vq ,


102 Chapter 9. The injectivity of some simple modules

then U/V is locally cyclic. This means that there exists a simple F U -module B
such that V = CU (B) (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157,
Chapter 2]). Let H = NG (V ) so that
H = Dr q(G) H Gq .
Clearly, H contains the stabilizer S of B in G. Since Vq Zq = 1, Vq is not
normal in Gq ;hence, Gq = H Gq for all q (G). Therefore, |G : H| is innite,
and |G : S| is innite. By Lemma 9.9, G cannot be restricted, a contradiction. 
Proposition 9.12 ([107]). Let F be a eld with char F = p. Suppose that G is a
periodic restricted p -group. If G is metabelian, then G is abelian-by-nite.
Proof. Suppose the contrary. By Lemma 9.7 and Proposition 9.10, we assume that
there exists a simple F G-module A such that CG (A) = 1. Put E =End F G (A).
Since G is restricted, dimE (A) = n is nite and then G can be embedded into the
linear group GLn (E). We may suppose that n is minimal subject to the existence
of a counterexample, which can be embedded in this way. If G has a subgroup L
of nite index such that A = A1 A2 , where A1 and A2 are non-zero L-invariant
submodules, then by Lemma 9.7, L/CL (A1 ) and L/CL (A2 ) are restricted so that,
by the minimality of n, each L/CL (Aj ) is abelian-by-nite, 1 j 2. Therefore,
L is abelian-by-nite, and so is G; a contradiction that shows that this case cannot
occur.
By Lemma 9.8, the Sylow q-subgroups of G are Chernikov for every prime q.
It also follows that every nite abelian normal subgroup of G is cyclic. Let H be
the smallest normal subgroup of G such that G/H is locally nilpotent. Since G is
metabelian, H is certainly abelian. The Sylow q-subgroup Hq of H is Cernikov,
and m (Hq ) is nite, for every m N. Since the latter is normal in G, it has
to be cyclic. Therefore Hq , and hence H is locally cyclic. Then there exists a
monomorphism : H U (K), where K is the algebraic closure of F . Put
M = F [H]. Since every element of H is algebraic over F , M is a subeld of K.
We may extend by linearity over F to the whole group ring F H. If a M and
v F H, the action (a, v) a(v) makes M into an F H-module, which clearly
is simple (see, for example, L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157,
Chapter 2]). Let S be the stabilizer of the F H-module M in G. By Lemma 9.9,
|G : S| is nite. We are going to show how the F H-module structure of M can
be extended to an F S-module structure in such a way that the resulting module
has innite endomorphism dimension, then obtaining a contradiction. A crucial
consideration for this is the fact that G splits over H (B. Hartley [104]); that is,
there exists a subgroup B such that G = HB and H B = 1.
If g S, the conjugate module gM g 1 is F H-isomorphic to M . Thus, there
exists an F -linear isomorphism : M M such that
(av) = a(g 1 vg),
for every a M and v F H. Thus,
(av) = a(g 1 vg),
Chapter 9. The injectivity of some simple modules 103

if a M and v F H. Since the multiplication by any non-zero element of M


is an F H-automorphism of M , we may assume that 1 = 1. Putting a = 1, we
obtain that
v = (g 1 vg),
for every v F H; hence we have that

(av) = av,

for every a M and v F H. Since is surjective, it follows that is an element


of the Galois group of M over F . Further, is uniquely determined (in terms
of g) by
(av) = a(g 1 vg) and 1 = 1,
via v = (g 1 vg). We dene = g so that is a mapping from S into .
Since v = (g 1 vg), we readily obtain that is a homomorphism.
Let be the translation subgroup of the group of all F -linear automorphisms
of M . Thus consists of all maps y : a ay, where a M and 0 = y M .
If , then 1 y = (y) , and so normalizes . Since G splits over H,
S likewise splits over H. Actually, if B is a complement of H in G (as indicated
above), B S is a complement of H in S. Then the mapping : hb (h) b,
where h H and b B S, is a well-dened map of S into the group of
semilinear transformations of M . Let h H and b V . Then

hb = h b = h(bh b1 )(bb ).

Thus

(hbh b ) = (h) ((bh b1 )) (b)(b )


= (h) (b)((bh b1 ))b) (b ) = (h) (b)(h ) (b ).

Matching v = bh b1 and = b in v = (g 1 vg), we obtain

(h) (b)(h ) (b ) = ((hb))((h b )).

Consequently, is in fact a homomorphism; by means of it, we can view M as an


F S-module. The action of F H remains unchanged, and so M is certainly simple.
Let be any F S-endomorphism of M . If w F H, then

w = (1w)v = 1vw.

Since is surjective, is induced by the multiplication by the element y = 1 of


M . If b B S and a M , then

(ab)y = (ab) = (a)(b) = (ay)(b).

Putting a = 1, we nd that y is xed by the element b . Since this holds for


every b B S, we nd that is induced by the multiplication by an element of
104 Chapter 9. The injectivity of some simple modules

the xed eld P of (B S). It is easy to see that such multiplication induces an
F S-endomorphism of M, and so End F S (M ) can be identied with P . Thus, it
suces to show that M has innite dimension over P . This claim will follow if we
prove that (B S) is innite. To see this, pick b B S. Then b Ker if and
only if (w)(b) = w, for all w F H. Matching = b in w = (g 1 wg),
we see that the above assertion is equivalent to (b1 wb) = w, for all w F H,
or, since is F -linear, to the same condition with w H. Thus, b Ker if and
only if [w, b] = 1, for all w H. Since is a monomorphism of H, we conclude
that Ker ( |BC ) = CBC (H) = C. Since H is abelian and C is locally nilpotent,
HC = H C is likewise locally nilpotent. By Lemma 9.7, HC is restricted, and by
Proposition 9.11, HC is abelian-by-nite. If |(B S) : C| is nite, then |S : HC|
is nite. Since |G : S| is nite, |G : HC| is nite. This implies that G is abelian-
by-nite, which was assumed not to be so. Hence, C has innite index in B S;
so that (B S) is innite, and we are done. 
The next result was proved by R.L. Snider in [267], but we follow the proof
given by B. Hartley in [107].
Proposition 9.13. Let F be a eld with char F = p. Suppose that G is a restricted
p -group. If the conjugacy classes of G are nite (that is, G is an F C-group), then
G is abelian-by-nite.
Proof. Suppose the contrary. If Z = (G), recall that G/Z is residually nite (see
M.J. Tomkinson [272, Theorem 1.9]). If G/Z has an abelian subgroup U/Z of nite
index, then U is nilpotent of class at most 2 and restricted. By Proposition 9.11,
U is abelian-by-nite, and so is G. This contradiction shows that G/Z cannot
be abelian-by-nite. This and Lemma 9.7 assure that we may assume that G is
residually nite.
Let L1 be an arbitrary nite normal subgroup of G. Since G is residually
nite, there exists a normal subgroup H of G of nite index such that H L1 = 1.
By our assumption, H cannot be abelian and then H has a nite non-abelian G-
invariant subgroup L2 . Applying the above argument to L1 L2 and proceeding
in this way, we nd that G contains the direct product

L = L1 L2 Ln

of a countably innite number of nite non-abelian subgroups Ln , n N. By


Lemma 9.7, we may assume that

G = L = DrnN Ln ,

and, furthermore, that every proper subgroup of each Ln is abelian. Then Ln


is metabelian (see D.J.S. Robinson [242, 9.1.9]), so that we have a metabelian
counterexample G. It suces to apply Proposition 9.12 to get a contradiction. 
Lemma 9.14 ([107]). Let F be a eld with char F = p. If G is a restricted locally
nite p -group, and F C(G) is nite, then G is nite.
Chapter 9. The injectivity of some simple modules 105

Proof. Put Q = G/F C(G). Since F C(G) is nite, we actually have F C(Q) = 1.
We claim that Q is nite. Otherwise, it is easy to see that Q has a countably
innite subgroup H such that F C(H) = 1. By Lemma 9.7, H is restricted. On
the other hand, by a result due to E. Formanek and R.L. Snider [71], the group
ring F H is primitive. Applying the JacobsonChevalley Density Theorem (see C.
Faith [67, Theorem 19.22]) to a faithful simple F H-module (which necessarily has
nite dimension n over its endomorphism ring E), we nd that F H = Mn (E). In
particular, F H is simple. Consideration of the augmentation ideal leads us to the
contradiction H = 1. 

We are now in a position to show one of the relevant results of this chapter.
Theorem 9.15 ([107]). Let F be a eld with char F = p and suppose that G is
a locally nite p -group. Then every simple F G-module has nite endomorphism
dimension if and only if G is abelian-by-nite.
Proof. By Lemma 9.5, every periodic abelian-by-nite group is restricted.
Conversely, assume that G is restricted but G is not abelian-by-nite. Let
L be the locally nilpotent radical of G. By Lemma 9.7 and Proposition 9.11, L
includes an abelian normal subgroup L0 of nite index. Form the product R of
the abelian normal subgroups of nite index of L. Since |L : L0 | is nite, we have

R = L0 L1 Ln ,

for some normal abelian-by-nite subgroups L0 , L1 , . . . , Ln having nite index


in L. Clearly,
L1 Ln (L),
so that |L : (L)| is nite. Hence L has a characteristic abelian subgroup Z of nite
index. By our assumption, the factor-group G/Z is innite. By Lemma 9.7 and
Lemma 9.14, D/Z = F C(G/Z) is innite, and, by Lemma 9.7 and Proposition
9.13, D/Z includes an abelian normal subgroup U/Z of nite index. Then U/Z
is innite, and U is metabelian. By Lemma 9.7 and Proposition 9.12, U includes
an abelian normal subgroup V of nite index. Since L contains every subnormal
abelian subgroup of G, V L. Therefore, U/(U L) is nite. Since L/Z is nite,
so is (U L)/Z. It follows that U/Z is nite. This contradiction establishes the
result. 

Now we go back to the study of the injectivity of simple modules. As we


did above, we consider modules over locally Wedderburn algebras. Concretely, we
consider a locally Wedderburn algebra W of countable dimension over a eld F and
a simple W -module A with endomorphism ring E. By Lemma 9.4, A is injective if
dimE (A) is nite; however, we need to study the case in which dimE (A) is innite.
We suppose this is the case. Since A
=W W/M , where M is a maximal right ideal of
W , dimF (A) is countably innite. Let {wn | n N} and {an | n N} be the basis
of W and A, respectively. We recall the construction introduced by B. Hartley
106 Chapter 9. The injectivity of some simple modules

in [107]. We begin by constructing recursively quintuples (Wn , An , Bn , Cn , Dn )


satisfying the following conditions:
(a) Wn is a nite dimensional semisimple subalgebra of W such that Wn1 Wn
and {w1 , . . . , wn } Wn .
(b) An is a nite dimensional Wn -submodule of A such that An1 An , Dn1
An and {a1 , . . . , an } An .
(c) Bn and Cn are Wn -submodules of An such that Bn is simple and An =
Bn Cn .
(d) Dn is a simple Wn -submodule of A such that the E-span An E of An does not
include Dn , and the homogeneous component Dn of Dn in the Wn -module
A has innite E-dimension.
(e) If En =End Wn (Bn ) and n : An Bn is the projection associated with
the decomposition An = Bn Cn , then Bn1 n = 0 , and (An1 n )En
does not include Dn1 n .
(f) There exists a Wn1 -monomorphism of Dn1 into Dn .
In (d), as in other places, the homogeneous component of Dn means the sum
of all Wn -submodules of A isomorphic to Dn . Of course, this is an E-submodule.
We start the construction with n = 1. Let W1 be any nite dimensional
semisimple subalgebra of W containing w1 . Let A1 be any nite dimensional W1 -
submodule of A containing a1 . Let B1 be any simple submodule of A1 . Since W1
is nite dimensional and semisimple, there exists some W1 -submodule C1 of A1
such that A1 = B1 C1 . Note that A1 E is nite dimensional over E, and W1 -
invariant. Let X1 be the sum of A1 E, and all those homogeneous components of
the W1 -module A that are nite dimensional over E. Then dimE (X1 ) is nite, and
there exists a simple W1 -submodule D1 of A such that X1 does not include D1 .
Clearly, (d) holds for D1 , and we have just established (a)(d) in the case n = 1.
Now suppose that n 1, and we have already constructed (Wj , Aj , Bj , Cj , Dj )
for every j n satisfying the properties (a)-(f). By (d), there exists some d Dn ,
which is E-independent of An . By the JacobsonChevalley Density Theorem (see
C. Faith [67, Theorem 19.22]), there exists some w W such that An w = 0 and
0 = dw Bn . Pick a nite dimensional semisimple subalgebra Wn+1 of W such
that Wn Wn+1 , wn+1 Wn+1 and w Wn+1 . Let An+1 be a nite dimensional
Wn+1 -submodule of A such that An An+1 , Dn An+1 and an+1 An+1 . Then
(a) and (b) hold for j = n+1. We can clearly choose a direct decomposition An+1 =
Bn+1 Cn+1 such that Bn+1 is simple and the projection n+1 : An+1 Bn+1 is
non-zero. This gives (c). Since n+1 is a monomorphism, and w Wn+1 , we have
0 = (dw)n+1 = dn+1 w. Thus w does not annihilate Dn n+1 , whereas it clearly
annihilates An n+1 because it annihilates An itself. It follows that (An n+1 )En+1
does not include Dn n+1 , which proves (c).
Chapter 9. The injectivity of some simple modules 107

It remains to construct Dn+1 . As in the case n = 1, dimE (An+1 E) is nite,


and An+1 E is Wn+1 -invariant. Adding to it the sum of those homogeneous com-
ponents of the Wn+1 -module A that are nite dimensional over E, we obtain a
nite dimensional Wn+1 -invariant E-subspace Xn+1 of A. Since, by (d), Dn has
innite E-dimension, Xn+1 does not include Dn . Therefore Xn+1 does not include
some simple Wn -submodule S = Dn . We have A = Xn+1 T , where

T = T ,

and every T is a simple Wn+1 -submodule. It follows that there is some such
that the projection of S into T is non-zero. We put Dn+1 = T . Then (d) and (f)
hold for Dn+1 , and the construction is done.
It follows from (a) and (b) that
 
(g) W = nN Wn and A = nN An .
It is worth mentioning that condition (d) is only required to assure that the
induction may be carried out. In fact, we will not make use of it any longer.
Lemma 9.16 ([107]). Suppose that Bn+1 and Dn+1 are not isomorphic, and they
form the direct sum Yn+1 = Dn+1 An+1 . Then Yn+1 has a Wn -submodule Ln =
Dn such that Ln An+1 = 0 , and no Wn -complement to An in Ln + An lies in
a Wn+1 -complement to An+1 in Yn+1 .
Proof. It will be convenient to identify An+1 with the external direct sum Bn+1
Cn+1 (see (c)). In this case An is identied with the submodule
{(0, an+1 , an+1 ) | a An } Dn+1 Bn+1 Cn+1
where n+1 is the projection of An+1 onto Cn+1 associated with the decomposition
(c). First, we notice that each Wn+1 -complement to An+1 in Yn+1 is isomorphic to
Dn+1 , and, since Bn+1 and Dn+1 are not isomorphic, projects trivially into Dn+1 .
Next, we identify Dn in some way with a Wn -submodule of Dn+1 (which is possible
by (f)). Because of (e), there certainly exists a Wn -monomorphism of Dn into
Bn+1 such that An n+1 does not include Dn . Let Ln = {(a, a, 0) | a Dn }.
Clearly Ln is a Wn -submodule of Yn+1 such that Ln An+1 = 0. Choose an
element a Dn such that a  An n+1 . Then any complement Z to An in
Ln + An contains an element of the form
(a, a + an+1 , an+1 ) with a An .
By the choice of a, the second component of this element is non-zero. Therefore,
the projection of Z into Bn+1 is non-zero and cannot lie in a Wn+1 -complement
to An+1 in Yn+1 . 
Lemma 9.17 ([107]). Suppose that Bn
= Dn for every n N and form the direct
sum Yn+1 = Bn+1 An+1 . Then Yn+1 includes a Wn -submodule Mn = Bn such
that Mn An+1 = 0 , and no complement to An in Mn + An lies in a complement
to An+1 in Yn+1 .
108 Chapter 9. The injectivity of some simple modules

Proof. In this case, we will identify Yn+1 with the external direct sum Bn+1
Bn+1 Cn+1 , An+1 with the set of elements with zero as the rst coordinate, and
An with {(0, an+1 , an+1) | a An }. Since Bn
= Dn for every n N, and we
have the property (e), we obtain
(h) There is a Wn -monomorphism n+1 : Bn Bn+1 such that (An n+1 )En+1
does not include Bn n+1 .
Let
Mn = {(bn+1 , bn+1 , 0) | b Bn }.
By (e), n maps Bn monomorphically; thus Mn An+1 = 0 , and Mn is a Wn -
submodule isomorphic to Bn . Now any complement to An+1 in Yn+1 has the form

{(b, b, b ) | b Bn+1 },

where En+1 and is aWn+1 -homomorphism of Bn+1 in Cn+1 ). Choose


d Bn such that dn+1  (An n+1 )En+1 . The existence of d is assured by (h).
Then any complement to An in Mn + An contains an element of the form

(dn+1 , dn+1 + an+1 , an+1 )

with a An . If this element lies in a complement to An+1 in Yn+1 , then we


can identify it as an element of the form (b, b, b ), obtaining dn+1 = b and
dn+1 + an+1 = b. Substituting for b in the second equation and rearranging,
we get
dn+1 = dn+1 an+1 (An n+1 )En+1 .
But this contradicts the choice of d. Therefore, no complement to An in Mn + An
lies in a complement to An+1 in Yn+1 . 
Now we are ready to establish the second main result of this chapter.
Theorem 9.18 ([107]). Let W be a locally Wedderburn algebra of countable dimen-
sion over a eld F and let A be a simple W -module with endomorphism ring E.
Then exactly one of the following statements holds:
(1) dimE (A) is nite, and A is injective; or
(2) dimE (A) is innite, and A can be embedded in an indecomposable W -module
of composition length 2.
Proof. Consider
= {n N | Bn+1
= Dn+1 }.
We split the proof into two cases.
Case 1. The set N \ is innite.
In this case, we form the direct sum Yn = Dn An as in Lemma 9.16.
If Bn+1 is not isomorphic to Dn+1 , then we embed Yn as a Wn -submodule of
Chapter 9. The injectivity of some simple modules 109

Yn+1 embedding Dn as Ln and embedding An in An+1 by the natural inclusion.


Otherwise, using (f), we embed Dn in Dn+1 in any way and again embed An
in An+1 by the natural inclusion. Taking the direct limit of the direct system so
obtained and making suitable identications, we obtain a W -module Y that is the
union of the sequence
Y1 Y2 Yn ,
where each Yn is a Wn -submodule and Y contains A in such a way that A Yn =
An . Furthermore, Yn = Dn An , and for innitely many n N, no complement
to An in Yn lies in a complement to An+1 in Yn+1 . Now

Y /A = (A + Dn )/A,
nN
and
(A + Dn )/A
=Wn Dn /(A Dn )
=Wn Dn .

Since the Wn -module Dn is simple, Y /A is a simple W -module. Thus Y has com-


position length 2. Now let T be any non-zero W -submodule of Y, and choose n
such that Yn T = 0. If An T = 0, then T A. Otherwise, Yn T is a com-
plement to An in Yn . But we can arrange that no complement to An in Yn lies in a
complement to An+1 in Yn+1 . Therefore, the Wn+1 -submodule of An+1 generated
by Yn T meets An+1 non-trivially. It follows that A T = 0 and so T A.
Consequently, A is the unique simple submodule of Y , which is indecomposable.
If Case 1 does not hold, then by suppressing nitely many terms in our
sequence and re-indexing the rest, we reduce to
Case 2. Bn
= Dn for all n N.
The proof of the theorem in Case 2 now follows from Lemma 9.17, exactly
as it did with Case 1. 

The above result allows us to completely describe countable groups G, for


which every simple F G-module is injective.
Theorem 9.19 ([107]). Let F be a eld of characteristic p 0, and let G be a
countable group. Then every simple F G-module is injective if and only if G is a
periodic abelian-by-nite p -group.
Proof. Suppose that every simple F G-module is injective. If G is not a locally
nite p -group, then the F G-module F , which is clearly simple, is not injective
by Theorem 9.3. Thus we must consider only the case, when G is a locally nite
p -group. Suppose that G is not abelian-by-nite. By Theorem 9.15, there exists
a simple F G-module A that has innite endomorphism dimension. Clearly the
group algebra F G is a locally Wedderburn algebra of countable dimension over
F . By Theorem 9.18, A can be embedded in an indecomposable F G-module of
composition length 2. This shows that A cannot be injective, a contradiction.
110 Chapter 9. The injectivity of some simple modules

Conversely, if G is a periodic abelian-by-nite group, and A is a simple F G-


module, then by Lemma 9.5, A has nite endomorphism dimension. By Lemma
9.4, A is injective. 
Theorem 9.19 was formerly obtained by D.R. Farkas and R.L. Snider [69]
under the additional assumption that F contains all roots of identity if char F = 0.
Chapter 10

Direct decompositions in artinian


modules

In the rst chapter of this book we considered a celebrated result due to H. Fitting
known as the Fitting lemma (Proposition 1.5). This result raises very interesting
questions about artinian modules over group rings concerned with direct decom-
positions of modules that are determined by a group action. Since we may obtain
very fruitful results studying this topic and make them available in the scope of
this book, we develop them here. To deal with this, it is convenient to formulate
the Fitting lemma in the following way.

Fitting lemma. Let A be an RG-module of nite composition length, where R is a


ring and G is a nite nilpotent group. Then A = C E, where the RG-chief factors
U/V of C (respectively of E) satisfy G = CG (U/V ) (respectively, G = CG (U/V ).

This fact raises related issues such as the discovery of complements to the
upper RG-hypercenter (see the denition below) and the studying of certain exten-
sions of modules that are very near to modules of nite composition length. These
problems have found very important applications in the study both of groups and
modules with niteness conditions. Moreover, they are also related to the ques-
tion of the existence of complements for some residuals in groups. We quote some
papers dedicated to these problems: Z.Y. Duan [60, 61, 63], D.Y. Duan and M.J.
Tomkinson [64], B. Hartley and M.J. Tomkinson [111], M.L. Newell [204, 205],
D.J.S. Robinson [237, 238, 239, 240, 241, 243, 246, 247], E. Schenkmann [253],
M.J. Tomkinson [271, 273, 275], D.I. Zaitsev [301, 302, 304, 307, 308, 310]; see
also the survey L.S. Kazarin and L.A. Kurdachenko [132].

Now we formulate the necessary concepts in their more general form. In


Chapter 3 we dene the XC-center and the upper XC-hypercenter of a group and
associated constructions. Proceeding in a similar way, we can translate the above
concepts in those of the RG-center and the upper RG-hypercenter of modules over
a group ring RG that is connected with some formation of groups.
112 Chapter 10. Direct decompositions in artinian modules

Let X be a class of groups. The factor C/B of a group G is said to be X-central


(respectively, X-eccentric) if G/CG (C/B) X (respectively, G/CG (C/B)  X).
In general, the X-central and the X-eccentric factors of a group G can appear
randomly in a composition series of G. Thus, discovering the cases in which all the
X-central factors of a group or of a normal subgroup of it can be gathered in some
place, while all the X-eccentric factors can be gathered in another place, appears
to be a very interesting project. In this setting and for this purpose, R. Baer [8]
introduced two important subgroups in a nite group G to rule out the X-centrality
and the X-eccentrality with respect to G. These are the XC-hypercenter XC (G)
of G and the X-hypereccenter E (G) of XG. The XC-hypercenter was dened
in Chapter 3 (actually, in Chapter 3 it is denoted as the upper XC-hypercenter),
while the X-hypereccenter is dened in the following way. Every G-chief factor of
the normal subgroup XE (G) is X-eccentric and XE (G) is a maximal normal
subgroup under this property. Clearly XC (G) XE (G) = 1 always holds,
but the decomposition
G = XC (G) XE (G)
usually fails. Baer himself achieved an important result in this direction, which we
quote now (see R. Baer [8]).
Theorem. Suppose that X is a local formation of nite groups and A is a normal
subgroup of a nite group G such that Q = G/CG (A) is XC-nilpotent. Then
A = (A XC (G)) (A XE (G).
In passing, we mention that, under these assumptions, the stronger restriction
G X holds, since for a local formation X, the XC-nilpotency of G implies that
G X (see K. Doerk and T. O. Hawkes [56, Theorem IV.3.2]). We consider the
question of the existence of Baers decomposition for certain types of artinian
modules. Our results concern innite groups and artinian modules, all of which
are associated in some way to a specic formation of groups. To proceed to the
description, we need rst to develop the concepts involved. As we did before for
groups (see Chapter 3) we obtain a generalization of the RG-center and the upper
RG-hypercenter with respect to certain formations of groups.
Let R be a ring, G a group, X a class of groups and A an RG-module. As
above, if B C are RG-submodules of A, the factor C/B is said to be X-central
(respectively, X-eccentric) if G/CG (C/B) X (respectively, G/CG (C/B)  X).
To rule out these factors, we dene

XCRG (A) = {a A | G/CG (aRG) X}.

Proposition 10.1. Let A be an RG-module, where R is a ring and G is a group. If


X is a formation of groups, then XCRG (A) is an RG-submodule of A.
Proof. If a1 , a2 XCRG (A), then

G/(CG (a1 RG) CG (a2 RG)) G/CG (a1 RG) G/CG (a2 RG),
Chapter 10. Direct decompositions in artinian modules 113

and
CG (a1 RG) CG (a2 RG) CG ((a1 a2 )RG).
It follows that G/CG ((a1 a2 )RG) X; that is, a1 a2 XCRG (A).
Now, suppose a XCRG (A) and x RG. Then a3 = ax aRG and so
a3 RG aRG. Thus CG (a3 RG) CG (aRG) and it follows that G/CG (a3 RG)
X, whence ax XCRG (A). 
The submodule XCRG (A) is called the XC-RG-center of A (in brief, the XC-
center of A). Proceeding in a manner similar to the one we used for groups, we
construct the upper XC-RG-central series of the module A as

0 = A0 A1 A A+1 A ,

where A1 = XCRG (A), A+1 /A = XCRG (A/A ), for all ordinals < and
XCRG (A/A ) = 0. The last term A of this series is called the upper XC-RG-

hypercenter of A (in short, the XC-hypercenter of A) and is denoted by XCRG (A);
the terms A are called the XC-RG-hypercenters of A. If A = A , then A is said
to be XC-RG-hypercentral ; if is nite, then A is called XC-RG-nilpotent .
We note that, if X = I is the class of all identity groups, then XCRG (A) =

RG (A) is called the RG-center of A and XCRG (A) = RG (A) is called the upper
RG-hypercenter of A. If X = F is the class of all nite groups, then XCRG (A) =

F CRG (A) is called the F C-center of A and XCRG (A) = F CRG (A) is called the
upper F C-hypercenter of A.
An RG-submodule C of A is said to be X-RG-hypereccentric if it has an
ascending series

0 = C0 C1 C C+1 C = C

of RG-submodules of A such that each factor C+1 /C is an X-eccentric simple


RG-module, for every < .
We say that the RG-module A has the Baer decomposition for the formation
X or A has the Baer X-RG-decomposition (in short, the Baer X-decomposition)),
if the following equality holds:

A = XCRG (A) XERG (A),

where XERG (A) is the maximal X-RG-hypereccentric RG-submodule of A. Note

that in this case, XERG (A) contains every X-RG-hypereccentric RG-submodule
and, in particular, it is unique. In fact, let B be an X-RG-hypereccentric RG-

submodule of A and put E = XERG (A). If (B + E)/E is non-zero, it has a
non-zero simple RG-submodule U/E. Since (B + E)/E = B/(B E), U/E is RG-
isomorphic to some simple RG-factor of B, and it follows that G/CG (U/E)  X.
On the other hand, (B + E)/E A/E = XCRG
(A); that is. G/CG (U/E) X.

This contradiction shows that B E. Hence XERG (A) contains every X-RG-
hypereccentric RG-submodule and, as we claimed, it is unique.
114 Chapter 10. Direct decompositions in artinian modules

If X = I, the decomposition is simply called the Z-decomposition, whereas


if X = F, we called it the F-decomposition. The rst results on the existence of
the Z-decomposition in innite modules were obtained by B. Hartley and M.J.
Tomkinson [111]. Later on, in the paper [301], D.I. Zaitsev proved that every
artinian ZG-module over a hypercentral group G has the Z-decomposition. The
next natural step is to consider the formation F of all nite groups. The existence
of the F-decomposition in artinian modules over F C-hypercentral groups has been
studied in the papers Z.Y. Duan [60, 62, 63], L.A. Kurdachenko, B.V. Petrenko
and I.Ya. Subbotin [159], and D.I. Zaitsev [307, 308]. The study of the existence of
the partial cases of the F-decomposition has played an important role in the study
of the general case of the existence of the X-decomposition. Actually, its solution
has allowed us to obtain solutions for many important formations X. To deal with
this, it is convenient to split the complete study into two cases:
(1) F X; and
(2) X is a proper formation of nite groups.
A formation X is said to be overnite if it satises the following conditions:
(i) If G X, and H is a normal subgroup of G of nite index, then H X.
(ii) If G is a group, H is a normal subgroup of nite index of G, and H X,
then G X.
(iii) I X.
Clearly, an overnite formation always contains F . The most important exam-
ples of these formations are polycyclic groups, Chernikov groups, soluble minimax
groups, soluble groups of nite special rank, and soluble groups of nite section
rank.
Lemma 10.2 (L.A. Kurdachenko, B.V. Petrenko and I.Ya. Subbotin [159]). Let X
be a formation of groups. Suppose that A is an RG-module, where R is a ring and

G is a group. If B is an RG-submodule of A, then XCRG (B) XCRG (A).
Proof. Let
0 = B0 B1 B B+1 B ,
and
0 = A0 A1 A A+1 A
be the upper XC-central series of B and A, respectively. It suces to show that
B A , for every ordinal , which will immediately imply the conclusion.
We proceed by transnite induction. Clearly B1 A1 . Suppose that > 1.
If is a limit ordinal, by induction,
 
B = B A = A ,
< <
Chapter 10. Direct decompositions in artinian modules 115

and we nish in this case. Thus, suppose that 1 exists. Then B1 A1 .


If B A1 , then B A and we are done. Therefore, we may assume that
A1 does not include B . Pick c B , put C = cRG and choose d C and g
CG ((C + B1 )/B1 ). Then there exists some bd B1 such that dg = d + bd .
Therefore,

(d + A1 )g = dg + A1 = d + bd + A1 = d + A1 .

It follows that

CG ((C + B1 )/B1 ) CG ((C + A1 )/A1 ),

and so

CG ((c + A1 )/A1 ) X and c + A1 XCRG (A/A1 ) = A /A1 .

Hence, B A , and we are done. 

Lemma 10.3 ([159]). Let X be a formation of groups. Suppose that A is an RG-


module, where R is a ring and G is a group. Let B and C be two RG-submodules of
A such that B C. If the non-zero RG-factors of both B and B/C are X-eccentric,
then every non-zero factor of C is X-eccentric.

Proof. Suppose that U and V are RG-submodules of C such that U V, and


U = V . If V B, then there is nothing to prove, so suppose that V + B = B. If
the factor (U + B)/(V + B) is non-zero, then G/CG ((U + B)/(V + B))  X. The
inclusion CG (U/V ) CG ((U +B)/(V +B)) implies that G/CG (U/V )  X. Finally,
let now U +B = V +B. Here (V +B)/V is non-zero and (V +B)/V =RG B/(BV ),
so that G/CG ((V + B)/V )  X. By the inclusion CG ((V + B)/V ) CG (U/V ),
we have G/CG (U/V )  X. 

Corollary 10.4 ([159]). Let X be a formation of groups. Suppose that A is an RG-


module, where R is a ring and G is a group. Let

0 = B0 B1 B B

be an ascending series of RG-submodules of A satisfying the following condition:

(*) If U and V are RG-submodules of A such that B V U B+1 for


some < , and U = V , then U/V is X-eccentric.

Then every non-zero RG-factor of B is X-eccentric.

Proof. We proceed by transnite induction on . If = 1, the assertion is trivial.


Suppose that > 1. If 1 exists, by Lemma 10.3, every non-zero RG-factor of
116 Chapter 10. Direct decompositions in artinian modules

B is X-eccentric. Thus, suppose that is a limit ordinal. Clearly,


 
U= (U B ), V = (V B ),
< <

and

U/V = ((U B ) + V )/V ).
<

Since U < V , there is an ordinal < such that U B = V B . By induction,


G/CG ((U B )/(V B )  X. Since

CG ((U/V ) CG ((U B )/(V B ),

G/G/CG (U/V )  X, as required. 


Corollary 10.5 ([159]). Let X be a formation of groups. Suppose that A is an RG-
module, where R is a ring and G is a group. Let {B | } be a family of
RG-submodules of A such that every non-zero RG-factor of B is X-eccentric, for
every . If B = B , then every non-zero RG-factor of B is X-eccentric.
Proof. Let be the least ordinal of cardinal ||. Then

{B | } = {B | < }.

If < , dene C = < B . Then

C+1 /C = (B+1 + C )/C


=RG B+1 /(B+1 C ).

In particular, every non-zero RG-factor of C+1 /C is X-eccentric, and it suces


to apply Corollary 10.4. 
Corollary 10.6 ([159]). Let X be a formation of groups. Suppose that A is an RG-
module, where R is a ring and G is a group. Let {B | } be a family of RG-
submodules of A such that each member B has the Baer X-RG-decomposition.
Then B = B has the Baer X-RG-decomposition.
Proof. If , by hypothesis, we have

B = C E ,

where C = XCRG (B ) and E = XERG (B ). Dene

C= C and E = E .


By Lemma 10.2, C XCRG (B). Further, every non-zero RG-factor of E is X-
eccentric by Corollary 10.5. It follows that C E = 0 , and thus B = C E.

Therefore C = XCRG (B) and E = XERG (B), as required. 
Chapter 10. Direct decompositions in artinian modules 117

Corollary 10.7 ([159]). Let X be a formation of groups, R be a ring, and G a


group.Then any RG-module A has a largest RG-submodule having the Baer X-
RG-decomposition.
Lemma 10.8 ([159]). Let A be an RG-module, where R is a ring and G is a group.
Suppose that H is a normal subgroup of G of nite index, and let B be an RH-
submodule of A such that A = BRG. If X is an overnite formation of groups and
B has the Baer X-RH-decomposition, then A has the Baer X-RG-decomposition.
Proof. Let {g1 , . . . , gn } be a transversal to H in G. Then
A = Bg1 + + Bgn .
If C and E are RH-submodules of B such that C E, put L = CH (C/E). If
g G, then CH (Cg/Eg) = g 1 Lg. Therefore
H/CH (Cg/Eg) = H/(g 1 Lg)
= H/L.
Consequently, if the RH-factor C/E is X-central (respectively X-eccentric), then
so is the RH-factor Cg/Eg. This means that the RH-module Bg has the Baer
X-RH-decomposition.
Let a XCRH (A) and put A0 = aRH and A1 = aRG. Then
A1 = A0 g1 + + A0 gn .
If U = CH (A0 ), then H/U X. It follows that H/CH (A0 g) X, for every g G
and so ag XCRH (A). In particular, XCRH (A) is an RG-submodule of A.
Furthermore,
CH (A1 ) = g11 U g1 gn1 U gn ,
hence H/CH (A1 ) X. Since X is overnite, G/CG (A1 ) X, and thus XCRH (A)
XCRG (A). As the converse inclusion is also valid, we get XCRH (A) = XCRG (A).

Applying transitive induction, we obtain that XCRH (A) = XCRG (A).

Suppose that B = B1 B2 , where B1 = XCRH (B) and B2 = XERH (B).
Then
B3 = B2 g1 + + B2 gn

is an RG-submodule of A. By Corollary 10.5, B3 XERH (A), whereas, by Lemma
10.2,

B1 g1 + + B1 gn XCRG (A)

and hence XERH (A) = B3 . In particular, XERH (A) is an RG-submodule of A.

Let U and V be RG-submodules of XERH (A) such that U V but U = V . Then

U/V is a non-zero RH-factor of XERH (A) and so H/CH (U/V )  X. Since X is

overnite, G/CG (U/V )  X. Thus, XERH (A) = XERG (A), as required. 
Lemma 10.9 ([159]). Let A be a nitely generated DG-module, where D is a
Dedekind domain and G is a policyclic-by-nite group. Suppose that A is mono-
lithic with monolith M, and let 1 = g (G) such that M (g 1) = 0. Then
there is some m N such that A(g 1)m = 0.
118 Chapter 10. Direct decompositions in artinian modules

Proof. Put P =Ann D (M ). First, we suppose that P = 0. Since M is a simple


DG-submodule of A, P Spec(D). Let T be the D-periodic part of A. Since A
is monolithic, T is the P -component of A. Since DG is a noetherian ring (see D.
S. Passman [217, Theorem 10.2.7]), A is a noetherian DG-module by Lemma 1.1.
According to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Lemma 1.9],
L =Ann D (T ) = 0 , and then there exists some d N such that L = P d . There
exists some D-submodule C such that A = T C (see I. Kaplansky [123]). In
particular, AL C and AL T = 0. Since M T , this means that A = T .
Put T1 = P,1 (T ) and think of T1 as an F G-module, where F = D/P is a
eld. Put R = F x, where x is an innite cyclic group and dene an action on
x on T1 by ax = ag, where a T1 . Then R is a principal ideal domain and T1
is an RG-module. Since M (g 1) = M (x 1) = 0, the (x 1)-component of
T1 is non-zero, and as above, we obtain that T1 is equal to its (x 1)-component.
Since the RG-module T1 is also noetherian, there must exist some t N such
that T1 (x 1)t = T1 (g 1)t = 0. If y P \ P 2 , then the mapping : a
ay, a P,2 (T ) = T2 is a DG-endomorphism of T2 and so Ker and Im are
DG-submodules of T2 . Since Im = T2 y T1 , (Im )(g 1)t = 0; that is,
T2 (g 1)t T1 and T2 (g 1)2t = 0. A simple inductive argument shows that
A(g 1)dt = 0.
Suppose now that P = 0. Since M is a simple DG-submodule, M is D-
divisible. Then we may consider M as a KG-module, where K is the eld of
fractions of D. Since A is monolithic, A is D-torsion-free. Let B = A D K so that
B is also D-torsion-free. Clearly B is a monolithic KG-module with KG-monolith
M . Put S = K y, where y is an innite cyclic group and dene an action of y
on B by bx = bg, b B. Obviously, B is a nitely generated KG-module, so that
B is a nitely generated SG-module. Since S is also a principal ideal domain, as
above, the ring SG is also noetherian (see D.S. Passman [127, Theorem 10.2.7]).
Again, M (x 1) = M (g 1) = 0 and, in particular, Ann S (M ) = 0. In this
case, we have already proved that there is some m N such that B(g 1)m = 0.
In particular, A(g 1)m = 0. 
Corollary 10.10 ([159]). Let A be a nitely generated DG-module, where D is a
Dedekind domain, G is a polycyclic-by-nite group, and CG (A) = 1. If 1 = g
(G) and CA (g) = 0, then A = A(g 1).
Proof. Pick 0 = a CA (g). Since g (G), CA (g) is a DG-submodule. Let Ba
be a maximal DG-submodule of A under a  Ba . Then A/Ba is a monolithic
DG-module with monolith (aDG + Ba )/Ba = M/Ba . Then (M/Ba )(g 1) = 0 ,
and it suces to apply Lemma 10.9. 
Lemma 10.11 ([159]). Let A be a nitely generated DG-module, where D is a
Dedekind domain, G is a locally polycyclic-by-nite group, and CG (A) = 1. If
1 = g (G) and CA (g) = 0, then A = A(g 1).
Proof. Let
A = a1 DG + + an DG
Chapter 10. Direct decompositions in artinian modules 119

and suppose that A = A(g 1). Then there exist some b1 , . . . , bn A such
that aj = bj (g 1), for every 1 j n. Pick 0 = c CA (g) and choose a
nitely generated subgroup H of G such that g H and c, b1 , . . . , bn B =
a1 DH + + an DH. If b B, then there are some x1 , . . . , xn DH such that

b = a1 x1 + + an xn = b1 (g 1)x1 + + bn (g 1)xn
= (b1 x1 + + bn xn )(g 1).

In other words, B = B(g 1). On the other hand, c CA (g) B = CB (g) and, in
particular, CB (g) = 0. By Corollary 10.10, B = B(g 1), a contradiction. 
Theorem 10.12. Let A be an artinian DG-module, where D is a ring and G is a
locally soluble F C-hypercentral group. If X is an overnite formation of groups,
then A has the Baer X-DG-decomposition.

Proof. If G X, then A = XCDG (A). Thus we may assume that G  X.
Suppose that the result is false. Let M be the family of DG-submodules B
of A such that B does not have the Baer X-DG-decomposition. Since A M,
M = . Since A is artinian, M has a minimal element C. By Corollary 10.7, C
contains the largest DG-submodule M of A having the Baer X-DG-decomposition.
By the choice of C, we deduce that M contains every proper DG-submodule of C.
In particular, M is a maximal DG-submodule of C. Decompose M = M1 M2 ,

where M1 = XCDG (M ) and M2 = XEDG (M ).
First, we suppose that G/CG (C/M )  X. We may replace C by C/M2 to

assume that M = XCDG (M ). We also may suppose that CG (C) = 1. Put
S = SocDG (C) so that S M . Since S is a direct sum of nitely many simple
DG-submodules, G/CG (S) X and, in particular, CG (S) = 1. By Lemma 3.15,
G
CG (S) F C(G) = 1. Pick some 1 = x CG (S) F C(G). By Corollary 3.4, x
is central-by-nite and the index |G : CG (xG )| is nite. Therefore either xG
G
includes a nite minimal G-invariant subgroup X or x includes a G-invariant
nitely generated torsion-free abelian subgroup Y . We note that if X is nite,
then X is also abelian since G is locally soluble. Dene Z = X in the rst case
and Z = Y in the second, and let H = CG (Z). Then Z (H), and |G : H| is
nite. By Theorem 5.5, there exist some n N such that we may express

C/M = (B/M )g1 (B/M )gn ,

where B/M is a simple DH-module and g1 , . . . , gn G.


We claim that H/CH (B/M )  X. Otherwise, since

H/CH ((B/M )g) = H/(g 1 CH (B/M )g)


= H/CH (B/M )
for every g G, we obtain H/CH (C/M ) X and it follows that G/CG (C/M ) X,
a contradiction. Thus our claim has just been proven. Since M does not include
B, BDG = C. Thus B does have the Baer X-DH-decomposition. Otherwise, C
would have the Baer X-DG-decomposition by Lemma 10.8.
120 Chapter 10. Direct decompositions in artinian modules

Let L be the family of those DH-submodules Q of C such that Q does not


have the Baer X-DH-decomposition. Since B L, L = . By Theorem 5.2,
C is an artinian DH-module and so L has a minimal element E. By Lemma
10.8, M cannot include E and so B = E + M . By Corollary 10.7, E contains
the largest DH-submodule E1 having the Baer X-DH-decomposition. By Lemma
10.8, E1 DG M and, in particular, E1 M . Moreover, E M = E1 since B/M
is a simple DH-module. Since

E/E1 = E/(E M )
=DH (E + M )/M = B/M,

we obtain that
H/CH (E/E1 )  X.
Let W =Soc DH (C). Since C is an artinian DH-module,

W = L1 Ls ,

where L1 , . . . , Ls are simple DH-submodules. For every 1 j n, since X


(H), Lj (DX) is a DH-submodule of Lj , and then either Lj (DX) = 0 or
Lj (DX) = Lj . Consequently,

W = CW (X) W (DX).

Since W is a DG-submodule of C and X is a normal subgroup of G, we have


that W (DX) is a DG-submodule of W . We claim W (DX) = 0. Otherwise,
W (DX)Soc DG (A) = 0. Since X CG (Soc DG (A)),

W (DX) Soc DG (A) W (DX) CW (X) = 0 ,

which is a contradiction. Thus we have just proven our claim; that is, W (DX) =
0 , so that W CC (X). It follows that E CC (X) = 0. If e E \ E1 , then E1
does not include eDH and so eDH = E. In particular, E is a nitely generated
DH-submodule. Since G is F C-hypercentral, G is locally (polycyclic-by-nite)
by Corollary 3.20. By Lemma 10.11, E(g 1) = E, for every g X, and thus

E(g 1) E1 . Since M = XCDG (M ), E1 = XCDH (E1 ) by Lemma 10.8. The
mapping
: a + E1 a(g 1) + E1 (g 1), a E, and g X,
is a DH-homomorphism. If E(g 1) = E1 (g 1), then the inclusion E(g 1) E1
implies that
H/CH (E(g 1)/E1 (g 1)) X.
Since E(g 1)/E1 (g 1) =DH E/E1 , CH (E/E1 ) = CH (E(g 1)/E1 (g 1)).
Since H/CH (E/E1 )  X, we get to a contradiction, which shows that E(g 1) =
E1 (g1), for every g X. In this case, it follows that E = CE (g)+E1 , and then, by
the choice of E, E = CE (g), because CE (g) is a DH-submodule. Since this is true
for every g X, we deduce E CC (X). In particular, M does not include CC (X).
Chapter 10. Direct decompositions in artinian modules 121

The latter is a DG-submodule, therefore CC (X) = C, i.e. X CG (C) = 1, a


contradiction.
Now we suppose G/CG (C/M ) X. Proceeding as above, we may assume

M = XEDG (M ) and CG (C) = 1. Since CG (C/M ) = 1, we may pick 1 = y
CG (C/M ) F C(G). Note that Y = yG is central-by-nite and |G : CG (Y )| is
nite. Put R = CG (Y ). We may suppose again that Y is abelian; that is, Y (R).
Let Q be the family of all DR-submodules Q of C of M does not include Q. Since
C Q, Q = . By Theorem 5.2, C is an artinian DR-submodule and so Q has a

minimal element U . By Lemma 10.8, M = XEDR (M ). For each g Y , U (g 1)
and CU (g) are DR-submodules, because g (R) and U (g 1) =DR U/CU (g).
By the choice of Y , U (g 1) M . If CU (g) M , then U/CU (g) has one non-zero
X-central DR-factor. On the other hand, since U (g 1) M , every non-zero
DR-factor of U (g 1) is X-eccentric. This contradiction shows that M does not
include CU (g). By the choice of U , we obtain that U = CU (g). Since this holds
for every g Y , U CC (Y ). In particular, M does not include CC (Y ). Since Y
is normal in G, CC (Y ) is a DG-submodule. Then C = CC (Y ), because M include
every proper DG-submodule of C. Hence Y CG (C) = 1, and we come to a
contradiction also in this case. 

We note that the case D = Z of this result was considered by L.A. Kur-
dachenko, B.V. Petrenko and I.Ya. Subbotin [160]. Also, it is worth mentioning
that, since every overnite formation X contains F, every F C-hypercentral group
is likewise X-hypercentral.
Corollary 10.13 ([159]). Let D be a Dedekind domain and G a locally soluble F C-
hypercentral group. If A is an artinian DG-module, then A has the F-decomposition.
Corollary 10.14 (Z.Y. Duan [60]). Let G be a locally soluble group, having an
ascending series of normal subgroups, every factor of which is nite or cyclic. If
A is an artinian ZG-module, then A has the F-decomposition.
Corollary 10.15 (D.I. Zaitsev [308]). Let G be a hypernite locally soluble group.
If A is an artinian ZG-module, then A has the F-decomposition.
Corollary 10.16. Let A be an artinian DG-module, where D is a Dedekind domain
and G is a locally soluble F C-hypercentral group. Then A has the Baer X-DG-
decomposition for the following formations X:
(1) X = P, the formation of all polycyclic groups.
(2) X = C, the formation of all Chernikov groups.
(3) X = S2 , the formation of all soluble minimax groups.
(4) X = S , the formation of all soluble groups of nite special rank.
(5) X = S0 , the formation of all soluble groups of nite section rank.
122 Chapter 10. Direct decompositions in artinian modules

In connection with Theorem 10.12, the following question naturally arises:


for what formations of nite groups does the Baer decomposition exist? Also note
that we did not even complete the case of the formation I of all identity groups.
In the next results, we will rst consider the existence of the Z-RH-decomposition
in artinian RG-modules when H is a hypercentral normal subgroup of G. As we
shall see, these results will play an important role in nding the answers for both
questions.
Lemma 10.17 (L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [158]). Let A be an
RG-module, where R is a ring and G is a group. Suppose that Y and H are normal
subgroups of G such that H Y and G/Y is nite. Suppose also that B/C is a
chief RG-factor of A such that RH (B/C) = 0. If U/V is a chief RY -factor such
that C V U B, then RH (U/V ) = 0.
Proof. By Theorem 5.5, the RG-submodule B has an RY -submodule E C such
that E/C is a simple RY -module and

B/C = (E/C)g1 (E/C)gn ,

for some g1 , . . . , gn . If CH (E/C) = H, then

CH ((E/C)gi ) = gi1 CH (E/C)gi = gi1 Hgi = H

implies H = CH (B/C), a contradiction. Since every (E/C)gi is a simple RY -


module and H is a normal subgroup of Y , RH ((E/C)gi ) is an RY -submodule,
hence, RH ((E/C)gi ) = 0 for every i, 1 i n. Since the RY -module B/C
is semisimple, the chief RY -factor U/V is RY -isomorphic with some (E/C)gi . It
follows that RH (U/V ) = RH ((E/C)gi ) = 0, as required. 
Lemma 10.18 ([158]). Let A be an RG-module, where R is a ring and G is a group.
Let X be a formation of groups, and suppose that H is a normal subgroup of G
such that A has an RH-submodule B satisfying the following conditions:
(1) Every non-zero RH-factor of B is X-eccentric.

(2) XCRG (A/B) = A/B.
Then B is an RG-submodule of A.
Proof. The proof of this lemma is rather obvious. 
Lemma 10.19 ([158]). Let A be an artinian RG-module, where R is a ring and G
is an F C-hypercentral group. Suppose that H is a normal hypercentral subgroup
of G and A has an RH-submodule B satisfying the following conditions:
(1) If U/V is a non-zero RH-factor of B, then CH (U/V ) = H.
(2) CH (A/B) = H.
Then B is an RG-submodule of A, and there exists an RG-submodule M such that
A = B M.
Chapter 10. Direct decompositions in artinian modules 123

Proof. By Lemma 10.18, B is an RG-submodule. Put

M = {C | C is an RG-submodule such that A = B + C}.

Since A M, M = . Since A is an artinian RG-module, M has a minimal


element M . Clearly, we may assume that CG (M ) = 1.
Suppose that M1 = M B = 0. By Corollary 3.16, (H) F C(G) = 1.
G
Pick 1 = x (H)F C(G), and dene X = x . Then the subgroup Y = CG (X)
has nite index in G, and H Y . In particular, X (Y ). Now put

M1 = {S | S is an RY -submodule such that M = M1 + S}.

Again, M1 = . By Theorem 5.2, A is an artinian RY -module so that M1 has


a minimal element U . Put U1 = M1 U . If z X, then z (H) and so the
mapping z : u u(z 1), u U is an RY -endomorphism; and, therefore,
U (z 1)/U1 (z 1) is an RY -epimorphic image of

U/U1 = U/(M1 U )
=RY (U + M1 )/M1
= M/M1 = M/(M B) =RY (M + B)/B = A/B.

It follows that
H = CH (U (z 1)/U1 (z 1)).
On the other hand, A(z 1) B by the condition (2). Since U (z 1) and U 1(z 1)
are RY -submodules, they are RH-submodules too. Thus, if U (z 1) = U1 (z 1),
by condition (1), H = CH (U (z1)/U1 (z1)). This contradiction shows U (z1) =
U1 (z 1). In this case, U = U1 + CU (z), hence, M = M1 + U = M1 + U1 + CU (z) =
M1 + CU (z), which implies CU (z) M1 . By the choice of U , U = CU (z). Since
this holds for every z X, U = CU (X) and, in particular, U CM (X). Note that
CM (X) is an RG-submodule since X is normal in G. Then

A = B + M = B + M1 + U = B + CM (X),

and, by the choice of M , M = CM (X). Hence, X CG (M ) = 1, a contradiction.


Therefore M B = 0 , hence, A = B M , as required. 
Lemma 10.20 ([158]). Let A an artinian DG-module, where D is a Dedekind do-
main and G is an F C-hypercentral group. Suppose that H is a normal hypercentral
subgroup of G and A has a DG-submodule B satisfying the following conditions:

(1) B DH (A).
(2) A/B is a simple DG-module.
(3) CH (A/B) = H.
Then there exists a DG-submodule M such that A = B M .
124 Chapter 10. Direct decompositions in artinian modules

Proof. Given a A \ B, put A1 = aDG. It suces to show A1 = (A1 B) M


for some DG-submodule M . For then

A = A1 + B = ((A1 B) M ) + B = B M,

as required. In other words, we may assume that A can be generated by any


element a A \ B. Put

M = {C | C is a DG-submodule such that A = B + C}.

Since A M, M = . Since A is an artinian DG-module, M has a minimal


element M . Clearly, we may assume that CG (M ) = 1. Suppose that M1 =
M B = 0. By Corollary 3.16, (H) F C(G) = 1. Let 1 = x (H) F C(G)
and dene X = < x >G . Then the subgroup Y = CG (X) has nite index in G
and H Y . In particular, X (Y ). Since

M/M1 = M/(M B)
= (M + B)/B = A/B

is a simple DG-module, by Theorem 5.5, M has a DY -submodule U M1 such


that U/M1 is a simple DY -module and

M/M1 = (U/M1 )g1 U/M1 )gn ,

for some g1 , . . . , gn . By Lemma 10.17, CH (U/M1 ) = H. Put

L = {S | S is a DY -submodule such that U = M1 + S}.

Obviously, L = . By Theorem 5.2, A is an artinian DY -module, and then L has


a minimal element Q. Put Q1 = M1 Q = M Q. Let R/S be a chief DY -factor
of M1 . Since H is a normal subgroup of G, DH (R/S) is a DY -submodule of R/S.

Since M1 ZDH (A), DH (R/S) = 0. It follows that CH (R/S) = H. Since

Q/Q1 = Q/(M1 Q)
=DY (Q + M1 )/M1 = U/M1 ,

we deduce that Q/Q1 is a simple DY -module. Then either DH (Q/Q1 ) = Q/Q1


or DH (Q/Q1 ) = 0. If H CY (Q/Q1 ), then H CY (U/M1 ) too. But we have
already proved that this is impossible. Thus CY (Q/Q1 ) does not include H. If
Q1 = 0, then we consider the DG-submodule QDG. In this case Q is a simple
DY -submodule, so QDG is a semisimple DY -submodule; that is,

QDG = Qw1 Qwm ,

for some w1 , . . . , wm G. If QDG B = 0, then it has a simple DY -submodule


V, and there exists an index 1 j m such that V =DY Qwj . In particular,
CY (V ) = CY (Qwj ). However, CY (Qwj ) = wj1 CY (Q)wj . Since Q =DY U/M1
and CH (U/M1 ) = H, we obtain that CH (Q) = H and

H = wj1 Hwj = wj1 CH (Q)wj . = CH (Qwj ) = CH (V ).


Chapter 10. Direct decompositions in artinian modules 125

Since V is a DY -submodule and H is normal in Y , CV (H) is a DY -submodule.



It follows that CV (H) = 0. On the other hand, B DH (A) and, in particular,
CV (H) = 0. This contradiction shows that QDG B = 0. Since A/B is a
simple DG-module, QDG + B = A. By the choice of M , QDG = M . Since we
supposed M B = 0, we may also assume that Q1 = 0. Note that Q can be
generated by every element a Q \ M1 . If z X, then z (Y ). By Lemma
10.11, Q(z 1) = Q. Since Q is a minimal element of L, Q(z 1)  L. This means
that Q(z 1) + M1 = U . It follows that Q(z 1) M1 because the DY -module
U/M1 is simple. Since z (Y ), the mapping z : u u(z 1), u U is a DY -
endomorphism. Thus the factor Q(z 1)/Q1 (z 1) is a DY -epimorphic image of
Q/Q1 . Since Q/Q1 is a simple DY -module, then either

Q(z 1)/Q1 (z 1)
=DY Q/Q1 or Q(z 1) = Q1 (z 1).

We have already proved that CY (Q/Q1 ) does not include H, so, in the rst case,
CY (Q(z 1)/Q1 (z 1)) does not either contains H. On the other hand, Q(z 1)
M1 and we have already shown that the centralizer of every chief DY -factor of
M1 contains H. Then the rst case is impossible, and so Q(z 1) = Q1 (z 1). In
this case Q = Q1 + CQ (z). Since z (Y ), CQ (z) is a DY -submodule of Q, then

U = M1 + Q = M1 + Q1 + CQ (z) = M1 + CQ (z),

which yields CQ (z) L. By the choice of Q, Q = CQ (z). Since this holds for every
z X, Q = CQ (X). In particular, Q CM (X). Since U = M1 , we deduce that
M1 cannot include CM (X). Since M/M1 is a simple DG-module and X is normal
in G, M1 + CM (X) = M . Therefore

A = B + M = B + M1 + CM (X) = B + CM (X)

and, by the choice of M , M = CM (X). Hence, X CG (M ) = 1, a contradiction,


which shows that M B = 0 against our assumption. Hence, A = B M . 

Theorem 10.21 ([158]). Let A be an artinian DG-module, where D is a Dedekind


domain and G is an F C-hypercentral group. If H is a normal hypercentral subgroup
of G, then A has the Z-DH-decomposition.
Proof. Suppose that the result is false. Let M be the family of all DG-submodules
B of A that does not have the Z-DH-decomposition. Clearly M = . Since A is an
artinian DG-module, M has a minimal element C. By Corollary 10.7, C contains
the largest DG-submodule M having the Z-DH-decomposition. By the choice of
C, M must include every proper DG-submodule. In particular, M is a maximal
DG-submodule of C.

Let M = M1 M2 , where M1 = DH (M ) and M2 = DH (M ). Since H is
a normal subgroup of G, DH (C/M ) is a DG-submodule of C/M . Then either
126 Chapter 10. Direct decompositions in artinian modules

DH (C/M ) = C/M or DH (C/M ) = 0. First, we suppose that DH (C/M ) =


C/M and consider the factor-module C/M1 . By Lemma 10.20, there exists a DG-
submodule M3 /M1 such that

C/M1 = M/M1 M3 /M1 .



In this case, M3 = DH (C) and so C = M3 M2 . Now suppose that DH (C/M ) =
0. By Lemma 10.20, there exists a DG-submodule M4 /M2 such that

C/M2 = M/M2 M4 /M2 ,



and we come to M4 = DH (C). Thus C = M1 M4 . 
Corollary 10.22. Let A be a DG-module, where D is a Dedekind domain and G is
a hypercentral group. If A is artinian, then A has the Z-decomposition.
Note that, for D = Z and G = H the above result was proven by D. I.
Zaitsev in [301]. Note that the result gives a positive answer for the question that
we raised above involving the formation I.
Consider now another question, namely, the existence of the Baer decompo-
sition for some natural formation of nite groups. Before dealing with it, we rst
notice that innite groups behave badly with respect to some properties automati-
cally satised by a nite group. For example, suppose that X is a formation of nite
groups and G is a group such that G RX. If G is nite, then G X. However,
if G is innite, the situation can be totally dierent. To avoid this complication,
we consider the following formation of nite groups.
A formation X of nite groups is said to be innitely hereditary concerning
a class of groups H if it satises the following condition:
(IH) Whenever an H-group G belongs to the class RX, then every nite factor-
group of G belongs to X.
It is worth mentioning that many formations of nite groups are innitely
hereditary concerning the class of F C-hypercentral groups; for example:
(1) A F, the nite abelian groups;
(2) Nc F, the nite nilpotent groups of class at most c;
(3) Sd F, the nite soluble groups of derived length at most d;
(4) S F, the nite soluble groups;
(5) B(n) F, the nite groups of exponent dividing n;
among others. Moreover, all these ve examples and
(6) N F, the nite nilpotent groups;
(7) U F, the nite supersoluble groups
Chapter 10. Direct decompositions in artinian modules 127

are innitely hereditary concerning both the classes of the F C-groups and hyper-
nite groups.
Theorem 10.23 ([158]). Let X be a formation of nite groups, and let A be an
artinian DG-module, where D is a Dedekind domain and G is an innite locally
soluble XC-hypercentral group. If X is innitely hereditary concerning the class of
F C-hypercentral groups, then A has the Baer X-DG-decomposition.
Proof. Since G is innite, G  X. Suppose that A does not have the Baer X-DG-
decomposition. Let M be the family of all DG-submodules B of A that does not
have the X-DG-decomposition. Then M = . Since A is an artinian DG-module,
M has a minimal element C. By Corollary 10.7, C includes the largest DG-
submodule M having the X-DG-decomposition. By the choice of C, M contains
every proper DG-submodule of C, and so M is a maximal DG-submodule of C.
Let
M = M1 M2 ,

where M1 = XCDG (M ) and M2 = XEDG (M ). Suppose that G/CG (C/M )  X.

Replacing C by C/M2 , we assume that M = XCDG (M ). By Corollary 10.13,

C = F CDG (C) FEDG (C).

Since C is indecomposable and M FDG (C), C = FDG (C). In particular, the
factor-module C/M is nite.
Let L be the set of normal subgroups H of G such that G/H X, and let
S be the intersection of the members of L. In other words, S is the X-residual
of G. If S = 1, then G RX. Since X is innitely hereditary concerning the
class of F C-hypercentral groups, the nite factor-group G/CG (C/M ) X. This
contradiction shows that this case cannot appear, and so S = 1. The group G
has an ascending series of normal subgroups
1 = G0 G1 G G+1 G = G
such that G/CG (G+1 /G ) X for every < . It follows that

CG (G+1 /G ) S.
<

By a result
due to Sh. S. Kemhadze [135, Theorem 1], the Baer radical of G
contains < CG (G+1 /G ). Note that the Baer radical of the group G is locally
nilpotent (see D. J. S. Robinson [234, Theorem 2.31]), and a locally nilpotent F C-
hypercentral group is hypercentral, so that, in particular, S is hypercentral. Since
G/S RX, CG (C/M )  S. Further, M has an ascending series of DG-submodules
0 = U0 U1 U U+1 U = M
such that G/CG (U+1 /U ) X, for every < . It follows that

CG (U+1 /U ) S,
<
128 Chapter 10. Direct decompositions in artinian modules

and then M is DS-hypercentral. We have already remarked that CS (C/M ) = S.



Therefore M = DH (C). By Theorem 10.21, C = M E, where E = JEDG (M ).

Since E =DG C/M , E = XEDG (A), a contradiction.
Suppose now that G/CG (C/M ) X. In this case, we replace C by C/M1 to

assume M = XEDG (M ). By Corollary 10.13,

C = F CDG (C) FEDG (C).

Since C is indecomposable and G/CG (C/M ) is nite, we obtain C = F CDG (C).
The DG-submodule M has an ascending series of DG-submodules

0 = U0 U1 U U+1 U = M

such that U+1 /U is a DG-chief factor and G/CG (U+1 /U )  X for every < .
It follows that CG (U+1 /U ) does not include S, and then DS (U+1 /U ) = 0

for every < , which means that M = DH (C). By Theorem 10.21, C = M E,
where E = DH (C). Since E

=DG C/M , E = XCDG
(A), another contradiction.

Corollary 10.24 ([158]). Let X be a formation of nite groups and let A be an
artinian DG-module, where D is a Dedekind domain and G is an innite locally
soluble XC-hypercentral group. Then A has the Baer X-DG-decomposition, pro-
vided X is one of the following formations: A F, Nc F, Sd F, S F, and
B(n) F.
Proof. It suces to show that the formation of all nite soluble groups is innitely
hereditary concerning the class of F C-hypercentral groups. To do this, it will suf-
ce to show that if an F C-hypercentral group G belongs to the class R(SF), then
it is locally soluble. So, G will be hyperabelian, and then its nite factor-groups
will be soluble. Let K a nitely generated subgroup of G. By Corollary 3.20, K is
nilpotent-by-nite. In particular, K satises Max, and so K has a maximal normal
soluble subgroup S. Assume that K = S. Then K has a subnormal subgroup L
such that L S and L/S is a nite non-abelian simple group. Since L R(S F),
L has a family {L | } of normal subgroups such that any L/L is nite
soluble and 
L = 1 .

If we assume that some L S = L, then L/L S must be a nite simple non-abelian


group. This contradiction shows that L S = L, for each . Then

L/L = L S/L
= S/(S L ).
It follows that each L/L is a nite soluble group of derived length at most d,
where d is the derived length of the soluble radical S. By Remaks theorem

L
L/L ,

Chapter 10. Direct decompositions in artinian modules 129

and so L is a soluble subgroup of derived length at most d. In particular, L = S,


a contradiction that shows K = S. Hence, K is soluble. 
Corollary 10.25 ([158]). Let X be a formation of nite groups, and let A be an
artinian DG-module, where D is a Dedekind domain, and G is an innite locally
soluble XC-hypercentral group. If G is an F C-group, then A has the Baer X-DG-
decomposition, provided X is one of the following formations: A F, Nc F, N F,
Sd F, S F, B(n) F, U F.
Corollary 10.26 ([158]). Let X be a formation of nite groups, and let A be an
artinian DG-module, where D is a Dedekind domain, and G is an innite locally
soluble XC-hypercentral group. If G is a hypernite group, then A has the Baer
X-DG-decomposition, provided X is one of the following formations: A F, Nc F,
N F, Sd F, S F, B(n) F, U F.
Corollary 10.27 ([158]). Let X be a formation of nite groups, and let A be an
artinian DG-module, where D is a Dedekind domain, and G is an innite XC-
hypercentral group. If G is a soluble Chernikov group, then A has the Baer X-DG-
decomposition.
Proof. Let R be the X-residual of the group G. Since G/R is residually nite and
G is Chernikov, G/R is nite. This means that any formation X of nite groups
is innitely hereditary concerning the class of Chernikov groups, and it suces to
apply Theorem 10.23. 
Consider now another important type of modules with niteness conditions,
namely, noetherian modules. In this case, the situation is not so satisfactory as it
is in the case of artinian modules. As a matter of fact, we can not ask about the
existence of X-decompositions in noetherian modules, even in the case X = I, as
the following elementary example shows. Let U = u v be a free abelian group
of rank 2, and let g be a cyclic group of order 3. We construct the semidirect
product G = U  g, in which g acts on U by ug = v and v g = u1 v 1 . Then
every non-identity G-invariant subgroup of U has nite index and, in particular,
the Z g-module U is noetherian. However, U is directly indecomposable and has
central G-chief factors and non-central G-chief factors.
Due to this, we may consider only a certain weakened version of the Baer
X-RG-decomposition, as the following one. Given a DG-module A, then we de-
0 +1
ne
DG (A) = Ag, DG (A) = (DG (A))g for every ordinal and DG (A) =


< DG (A) for a limit ordinal , where g = (DG) is the augmentation ideal
of DG. Thus we have just constructed the lower DG-central series of the module
A,
0 1
A = DG (A) DG (A) DG

(A) DG
+1
(A) DG (A).

D.J.S. Robinson employed the following form of the Z-decomposition for


notherian modules [236]
130 Chapter 10. Direct decompositions in artinian modules

Theorem 10.28. Let G be a nilpotent group, R a ring and let A be a noetherian


RG-module. Then there is some k 0 such that
k
RG (A) RG (A) = 0 .
k
In particular, RG (A) has no non-zero G-central factors. Moreover, RG (A) =
+1
RG (A).
The interested reader can nd the proof of this result in the book of D.J.S.
Robinson [242, 15.3]. In the paper [236], Robinson has been able to construct
an example showing that the discussed result cannot be extended to the case
of hypercentral groups having an upper central series of length + 1. For
hypercentral groups, D.I. Zaitsev [302] has been able to obtain the following even
weaker form of the Z-decomposition for noetherian modules.
Theorem 10.29. Let G be a hypercentral group and let A be a noetherian ZG-
module. Then A has a non-zero ZG-central factor if and only if A = DG (A).
In the paper [308] D. I. Zaitsev introduced the following weak form of the
F-decomposition for noetherian modules:
Theorem 10.30. Let G be a hypernite group and A a noetherian ZG-module. Then
A has a non-zero nite ZG-factor if and only if A has a proper ZG-submodule
having nite index.
Finally, Z.Y. Duan [61] proved that every noetherian ZG-module over a hy-
pernite locally soluble group G has the F-decomposition.
Chapter 11

On the countability of artinian


modules over F C-hypercentral groups

One of the rst important problems of the theory of artinian modules is the eval-
uation of their socular height. We initially discussed this question in Chapter 4,
studying the conditions under which this socle height is at most , where is the
rst innite ordinal. We showed that this is valid for artinian R-modules over a
commutative ring R (Theorem 4.12). For artinian modules over a more general
group ring, this question can be reformulated in the following way.
Let G be a group, D a Dedekind domain. What can be said about a group G if
the socular height of an arbitrary artinian DG-module (D is a Dedekind domain)
is at most ? Or more generally, what can be said about a countable group G if
an arbitrary artinian DG-module (D is a Dedekind domain) has a countable set
of generators as a D-module?
In his paper [105], B. Hartley has constructed several impressive examples of
uncountable artinian monolithic ZG-modules for dierent types of groups G in-
cluding some nilpotent groups (see Theorem 4.14). All these examples are uniserial
modules, so, in particular, they are monolithic. We note that in all these examples
the factor-group G/CG (ZG (A)) is not abelian-by-nite. Therefore the question
about the study of nilpotent groups G such that G/CG (ZG (A)) is abelian-by-
nite arises naturally. In Chapter 9 we realized that abelian-by-nite groups play
a special role in the theory of artinian modules over group rings, and this will be
supported by the results stated below. We consider the problem of countability
for a fairly wide generalization of nilpotent groups, namely, for F C-hypercentral
groups. These results have been obtained by L.A. Kurdachenko, N.N. Semko and
I.Ya. Subbotin [164].
We begin with the following theorem that has many applications.
Theorem 11.1. Let G be an F C-hypercentral group, D a Dedekind domain, A an
artinian monolithic DG-module, H = CG (DG (A)). Then A is a DH-hypercentral
module.
132 Chapter 11. On the countability of artinian modules

Proof. Since A is an artinian DG-module, A has an ascending series of DG-


submodules

0 = A0 A1 A A+1 A = A,

all factors of which are simple DG-modules. In particular, M = A1 is the monolith


of A.
Suppose that the result is false. Then there must exist an ordinal such
that CH (A+1 /A ) = H but the upper DH-hypercenter of A+1 contains A (and
hence, the upper DH-hypercenter of A+1 coincides with A ). Put C = A+1 ,
and B = A . There is no loss if we assume that CG (C) = 1. Let G1 be a normal
subgroup having nite index in G. By the HartleyZaitsev theorem (Theorem 5.5),

C/B = (L/B)x,
xS

where L/B is a simple DG1 -submodule, and S is a nite subset. Suppose that
CH ((L/B)g) = H for some element g S. If x S, then we have

CH ((L/B)x) = CH (((L/B)g)g 1 x) = (g 1 x)1 CH ((L/B)g)g 1 x


= (g 1 x)1 Hg 1 x = H,

which implies the equation H = CH (C/B). This is impossible. Consequently,


CH ((L/B)g) = H for every g S.
We claim that CH (B) = 1. Otherwise, by Lemma 3.15, F C(G) CH (B) =
G
1. Let 1 = h F C(G) CH (B), and put H1 = h , and G1 = CG (H1 ). Then
G1 is a normal subgroup of G having nite index. As we have seen above

C/B = (L/B)x,
xS

where L/B is a simple DG1 -submodule, and S is a nite subset of G. By the


choice of h, the mapping : c c(h 1), c C is a DG1 -endomorphism, so
that Ker = CC (h), and Im = C(h 1) are DG1 -submodules. The equation
CG (C) = 1 shows that Ker = C. Since B Ker , Im is a semisimple
DG1 -submodule of nite composition length. In particular, Im contains some
simple DG1 -submodule R. It happens that R must be isomorphic to some DG1 -
submodules (L/B)x, x S, and by the argument given above, CH (R) = H. The
niteness of the index |G : G1 | implies that

RDG = Rx1 Rxm

for some nite subset {x1 , . . . , xm } G. In particular, RDG is a semisimple


DG1 -submodule of nite composition length. If we suppose that RDG B = 0,
then this intersection contains a simple DG1 -submodule Q. As above, Q Is a
Chapter 11. On the countability of artinian modules 133

DG1 -submodule isomorphic to some Rxj , and so CH (Q) = CH (Rxj ) = H. On


the other hand, B is DH-hypercentral; therefore, the inclusion Q B together
with the DG1 -simplicity of Q imply the equation CH (Q) = H, a contradiction.
This contradiction yields that RDG B = 0. In particuar, RDG DG (A) =
0, which is impossible. This nal contradiction establishes our claim; that is,
CH (B) = 1.
Choose in H an arbitrary nitely generated subgroup K. Since K acts triv-
ially on all factors of the ascending series of B, K is residually nilpotent (see P.
Hall and B. Hartley [99, Theorem A2]). It follows that the set of all elements of
H having nite order is a (characteristic) locally nilpotent subgroup of H (see, for
example, [235, Theorem 6.14]). In particular, F C(H) is a locally soluble subgroup.
Let M be the set of all DG-submodule M such that B does not contain M . Since
A is artinian, M has a minimal element E. We note that E B = 0. Otherwise,
E B = 0 , and then the inclusion DG (A) B leads to a contradiction, be-
cause A is DG-monolithic. Hence E B = 0. Since C/B is a simple DG-module,
B + C = E, so that

C/B = (B + E)/B
=DG E/(E B)

is a simple DG-module, and CH (E/(E B)) = CH (C/B) = H. There is no loss


if we assume that CG (E) = 1. Let 1 = x F C(G) H, and put X = xG .
We noted above that F C(H) is locally soluble. Then X includes a non-identity
G-invariant abelian subgroup Y . By the choice of X, and Y , Z = CG (Y ) has nite
index in G, and Y (Z). By Theorem 5.5, we have

E/(E B) = (U/(E B))x,
xS

where U/(E B) is a simple DZ-submodule, and S is some nite subset of G.


Let now U be the set of all DZ-submodules M of U such that E B does not
contain M . By Theorem 5.2, U is an artinian DZ-module, and then U has a
minimal element V . We have now U = V + (B E), because U/(B E) is a simple
DZ-module, so that

U/(B E) = (V + (B E))/(B E)
=DZ V /(V B)

is also a simple DZ-module, and CH (U/(B E)) = CH (V /(V B)). As we demon-


strated above CH (V /(V B)) = H. If we assume that V B = 0; then, using
the arguments given above, we come to a contradiction. This contradiction shows
that W = V B = 0. Since W is a DH-hypercentral submodule, CW (y) = 0
for each element y Y , and thus CV (y) = 0. The submodule W cannot contain
any vDZ for every element v V \ W . The choice of V yields that vDZ = V , and,
in particular, V is a cyclic DZ-submodule. By Corollary 3.20, every nitely gen-
erated subgroup of an F C-hypercentral group is nilpotent-by-nite. Thus Lemma
10.11 implies that V (y 1) = V . By the choice of V , W contains every proper
134 Chapter 11. On the countability of artinian modules

DZ-submodule of V, so that, V (y 1) W . Since Y (Z), the mapping


: v v(y 1), v V is a DZ-homomorphism. It follows that V (y 1)/W (y 1)
is a DZ-homomorphic image of V /W . If V (y 1)/W (y 1) is non-zero, then by
the simplicity of V /W, we have V /W
=DZ V (y 1)/W (y 1). In particular,

CH (V (y 1)/W (y 1)) = CH (V /W ) = H.

On the other hand, W is a DZ-hypercentral submodule. Therefore every one of


its simple DZ-factors is H-central. Then CH (V (y 1)/W (y 1)) = H. This
contradiction shows that V (y 1) = W (y 1). It follows that for each element
v V, there exists an element w W such that v(y 1) = w(y 1); that is,
(v w)(y 1) = 0. Thus, we see that W cannot contain CV (y). Clearly, CV (y)
is a DZ-submodule, and, by the choice of B, we have CV (y) = V . This holds for
each element y Y , therefore CV (Y ) V . Moreover, CV (Y ) is a DG-submodule,
because Y is a normal subgroup of G, and B cannot contain CV (Y ). By the choice
of E, we obtain that CV (Y ) = E. In other words, Y CG (E) = 1. This nal
contradiction proves the theorem. 

Lemma 11.2. Let G be an abelian-by-nite group, D a Dedekind domain, and A


an artinian DG-module. If B is a nitely generated DG-submodule of A, then it
has nite DG-composition series. In particular, the socular height of A is at most
.
Proof. There is no loss if we assume that B is a cyclic DG-submodule; that is,
B = bDG, for some element b A. Let R be a normal abelian subgroup of G
having nite index in G. The DR-module A is artinian by Theorem 5.2. Put
C = bDR. Thus C =DR DR/Ann DR (b). Since R is abelian, the group ring DR
is commutative so that J =Ann DR (b) is a two-sided ideal of DR, and we may
consider the factor-ring DR/J, which is artinian. But an artinian commutative
ring is also noetherian ( see R. Sharp [263, Theorem 8.44]). So the DR-submodule
C is both artinian and noetherian; that is, C has nite DR-composition series.
Since R has nite index in G,

B = CDG = Cx1 Cxm

for some nite subset {x1 , . . . , xm } G. In particular, B has nite DR-compo-


sition length. Then B has nite composition length as DG-module. It follows that
every nitely generated DG-submodule of A lies in some term of the upper socular
series of A that corresponds to a nite ordinal. This means that the socular height
of A is at most . 

Lemma 11.3. Let G be a countable F C-hypercentral group, F a eld, and A an


artinian monolithic F G-module. Suppose that dimF A is uncountable, but every
proper F G-submodule of A has countable dimension, and CG (A) = 1. Then
CG (F G (A)) has no non-identity periodic G-invariant subgroups.
Chapter 11. On the countability of artinian modules 135

Proof. Put H = CG (F G (A)). Proceeding as we did in the proof of Theorem 11.1,


we notice that the set T of all elements of H having nite order is a (characteristic)
subgroup of H. Suppose the contrary. Let T = 1. Since G is an F C-hypercentral
group, T F C(G) = 1 by Lemma 3.15. Let 1 = x F C(G) T , and put
G
X = x . Then X is nite by Corollary 3.4. Furthermore, X has a non-identity
G-invariant abelian subgroup Y , because T is locally nilpotent. Then Z = CG (Y )
has nite index in G, and Y (Z). There is no loss if we assume that Y is a
q-subgroup, for some prime q. By Theorem 11.1, A is an F H-hypercentral module,
and hence A is likewise F Y -hypercentral. It follows that the natural semidirect
product A  Y is a hypercentral group.
Suppose rst that char F = 0. In a hypercentral group, all elements of nite
order form a (characteristic) subgroup, that is, in this case, B  Y = B Y , and
Y CG (B) = 1, a contradiction. Let now char F = p > 0. If we suppose that
q = p, then, by elementary properties of hypercentral groups, we obtain again
that A  Y = A Y , which leads to the contradiction Y CG (A) = 1. Suppose
that q = p. Let M be the set of all F Z-submodules U having uncountable F -
dimension. Since A is an artinian F Z-module by Theorem 5.2, M has a minimal
element C. Thus dimF C is uncountable but every proper F Z-submodule of C has
countable dimension over F . The inclusion Y (Z) implies that the mapping
: c c(y 1), c C is an F Z-homomorphism for every element y Y . It
follows that Ker = CC (y), and Im = C(y 1) are F Z-submodules of C, and
C(y 1) =F Z C/CC (y). If C(y 1) is a proper F Z-submodule of C, then it has
countable dimension. By the isomorphism C(y 1) =F Z C/CC (y), the submodule
CC (y) must have uncountable dimension. By the choice of C, C = CC (y). Since
this holds for each element y Y , CA (Y ) C. Since Y is a normal subgroup of
G, CA (Y ) is an F G-submodule. The inclusion CA (Y ) C shows that CA (Y ) has
uncountable dimension too. Under the conditions of this lemma, this implies that
CA (Y ) = A. In other words, Y CG (A) = 1. This contradiction shows that
C(y 1) = C.
Consider now the natural semidirect product C  y , y Y . Since char F =
p, the additive group of C is an elementary abelian p-subgroup. The niteness of
y implies that E = C  y is nilpotent ( see, for example, [235, Lemma 6.35]).
Clearly, [E, E] = C(y 1) = C, and, in particular, E/[E, E] is nite. This implies
that E is nite too (see, for example, [234, Theorem 2.2]). This nal contradiction
proves that T = 1. 
Proposition 11.4. Let D be a Dedekind domain, P a maximal ideal of D, y P \P 2 ,
and A a P -module over the ring D. Then A is D-divisible if and only if A = Ay.
Proof. If A is divisible, then clearly A = Ay. Conversely, suppose that A = Ay.
Let a A, and 0 = x D. If x P , then xD + P = D. Since A is a P -module,
Ann D (a) = P s for some s N. By Lemma 6.7, xD + P s = D, and so 1 = xu + y
where u D, and y P s . We have now

a = a 1 = a(xu + y) = axu + ay = aux = (au)x,


136 Chapter 11. On the countability of artinian modules

and thus A is D-divisible. Suppose that x P . Since A = Ay, A = Ay n for every


n N. By Theorem 6.3, Lemma 6.7 and Lemma 6.8, xD = P t Q, where P +Q = D.
It follows that x P t \P t+1 , and so x = x1 z, where x1 P t , and z Q. Moreover,
z P . From the equation A = Ay t we obtain that a = a1 y t , for some element
a1 A. Since Ann D (a) = P s , we have P s+t Ann D (a1 ). Proposition 6.13 yields
that P t = Dy t + P t+s , so x1 = y t u + v, u D, and v P t+s . From x  P t+1 it
follows that u  P . Since P is a prime ideal, u  P , z  P , and we can conclude
that uz  P . Hence, uzP + P = D. This implies that uzD + P t+s = D, so
1 = uzx2 + x3 , where x2 D, and x3 P t+s . Then

a1 = a1 1 = a1 (uzx2 + x3 ) = (a1 x2 )uz + a1 x3 = a2 uz,

where a2 = a1 x2 . Thus Ann D (a1 ) =Ann D (a2 ); that is, P t+s Ann D (a2 ). It
follows that

a = a1 y t = (a2 uz)y t = a2 uzy t + a2 vz = a2 (uzy t + vz) = a2 x,

and hence A = Ax. 


Lemma 11.5. Let G be a countable F C-hypercentral group, F a eld, and A an
artinian monolithic F G-module. Suppose that dimF A is uncountable, but every
proper F G-submodule of A has countable dimension over F . If CG (A) = 1, then
CG (F G (A)) = 1.
Proof. Put H = CG (F G (A)). Suppose that the result is false, that is H = 1.
Since G is F C-hypercentral, H F C(G) = 1 by Lemma 3.15. All elements of
an F C-group having nite order form a (characteristic) subgroup ( see Corollary
3.4). By Lemma 11.3, H has no non-identity periodic G-invariant subgroups, so
that H F C(G) is torsion-free. But a torsion-free F C-group is abelian (see again
Corollary 3.4), so H F C(G) is an abelian torsion-free subgroup.
G
Let 1 = x F C(G) H, and put X = x . The subgroup X is nitely
generated torsion-free abelian, and its centralizer Z = CG (X) has nite index in
G, and X (Z). Let M be the set of all F Z-submodules having uncountable
F -dimension. Since A is an artinian F Z-module by Theorem 5.2, M has a minimal
element C. Thus, dimF C has uncountable F -dimension, but every proper F Z-
submodule of C has countable dimension over F . The inclusion Y (Z) implies
that the mapping : c c(y 1), c C is an F Z-homomorphism for all elements
y Y . It follows that Ker = CC (y), and Im = C(y 1) are F Z-submodules
of C, and C(y 1) =F Z C/CC (y). If C(y 1) is a proper F Z-submodule of C,
then it has countable dimension. By the isomorphism C(y 1) =F Z C/CC (y),
the submodule CC (y) must have uncountable dimension. By the choice of C this
gives that C = CC (y). Since this holds for every element y Y , CA (Y ) C.
Since Y is a normal subgroup of G, CA (Y ) is an F G-submodule. The inclusion
CA (Y ) C shows that CA (Y ) has uncountable dimension too. By hypothesis, this
implies that CA (Y ) = A. In other words, Y CG (A) = 1. This contradiction
shows that C(y 1) = C. By Theorem 11.1, A is an F H-hypercentral module, so
Chapter 11. On the countability of artinian modules 137

that A is F X-hypercentral. It follows that the natural semidirect product A  X


is a hypercentral group. Let D = F x, where x is a non-identity element of X.
Since x has innite order, D is a principal ideal domain, in particular, D is a
Dedekind domain. Obviously, the ideal P = (x 1)F x is prime. Since C  x
is a hypercentral group, the D-module C coincides with its P -component. The
equation C(x 1) = C, together with Proposition 11.4 implies that C is a D-
divisible module. By Theorem 7.25,

C= A ,

where A is a Prufer P -module for every . Since dimF C is uncountable, then


the set of indexes cannot be nite. It follows that dimF C = dimF P,1 (C),
and, in particular, dimF P,1 (C) is uncountable. The inclusion X (Z) shows
that P,1 (C) is an F Z-submodule of C. By the choice of C, every proper F Z-
submodule of C has countable dimension. This contradiction gives the required
result. 
Theorem 11.6. Let G be a countable F C-hypercentral group, F a eld, and A an
artinian F G-module. If the factor-group G/CG (Soc F G (A)) is abelian-by-nite,
then dimF A is countable. In particular, if the eld F is countable, then A is
likewise countable.
Proof. Consider rst the case when the module A is monolithic. In this case Soc
F G (A) = F G (A). Put H = CG (F G (A)). Suppose again that dimF A is uncount-
able. Let M be the set of all F G-submodules having uncountable F -dimension.
Since A is an artinian F G-module, M has a minimal element B. In other words,
dimF B is uncountable, but every proper F G-submodule of B has countable di-
mension over F . By Lemma 11.5, H CG (B). Therefore, we can think of B as
an F (G/H)-module. The factor-group G/H is abelian-by-nite, and now Lemma
11.2 yields that the socle height of B is at most . Since A is artinian, every factor
of the upper socle series is decomposed into a direct sum of nitely many simple
submodules. Thus B has an ascending series of submodules
0 = B0 B1 Bn Bn+1 B
with simple factors, and . Each factor of this series is a cyclic F G-module,
so that it is an epimorphic image of the F G-module F G. Since G is countable,
dimF F G is countable. Thus, every factor of this series has nite or countable
dimension over F . Hence dimF B is countable, and we come to a contradiction.
Consider now the general case. Suppose that
Soc F G (A) = M1 Mk ,
where M1 , . . . , Mk are simple F G-modules. For each index j, 1 j k, let Uj be
a maximal F G-submodule under the following properties:
M1 , . . . , Mj1 , Mj+1 , . . . , Mk Uj ,
Uj Mj = 0 .
138 Chapter 11. On the countability of artinian modules

Then the factor-module A/Uj is monolithic, and with the monolith (Mj + Uj )/Uj .
Clearly, H CG ((Mj + Uj )/Uj ), so that G/CG ((Mj + Uj )/Uj ) is abelian-by-
nite. Applying the above paragraph, we obtain that every A/Uj has countable
dimension over F . But

U1 Uk Soc F G (A) = 0 .

Since A is artinian, this means that

U1 Uk = 0 .

Application of Remaks theorem gives the embedding

A
A/U1 A/Uk ,

which shows that dimF A is countable. 


Lemma 11.7. Let R be an integral domain, F the eld of fraction of R, G a group,
and A an artinian RG-module. If A is R-torsion-free, then E = A R F is an
artinian F G-module.
Proof. Let {En | n N} be a descending chain of F G-submodules of E. Then
{A En | n N} is a descending chain of RG-submodules of A. Since A is
artinian, there exists some k N such that A Ek = A Ek+n for every n N.
The R-module Ek /(A Ek ) is periodic, so that Ek /(A Ek+n ) is likewise periodic
for every n N. This means that the R-divisible envelope of every A Ek+n
coincides with Ek . On the other hand, that envelope is Ek+n . Hence Ek = Ek+n
for every n N. This proves that E is an artinian F G-module. 
Theorem 11.8. Let G be a countable F C-hypercentral group, D a Dedekind domain
and A an artinian DG-module. If G/CG (SocDG (A)) is abelian-by-nite, then |A|
|D|0 . In particular, if D is countable, then A is countable.
Proof. If D is a eld, then our assertion follows from Theorem 11.6. Therefore we
may suppose that D is not a eld. In particular, this implies that D is innite.
Let T be the D-periodic part of A. By Corollary 6.25,

T = AQ ,
Q

where =AssD (A), and AQ is the Q-component of A. Note also that AQ is a


DG-submodule of A. Since A is artinian, the set is nite. Put BQ = Q,1 (AQ ).
Then we may think of BQ as an FQ G-module, where FQ = D/Q is a eld. By
Theorem 11.6, BQ has countable dimension over FQ . Let y Q \ Q2 , and consider
the ideal yD + Qn . By Proposition 6.13,

yD + Qn = Qk
Chapter 11. On the countability of artinian modules 139

for some 1 k n. Since y  Q2 , k = 1. In other words,

yD + Qn = Q

for every n N. Put CQ = Q,2 (AQ ). The mapping : c cy, c CQ is a


D-endomorphism such hat Ker = BQ , and Im = (CQ )y = (CQ )Q BQ , so
that CQ /BQ is isomorphic to some D-submodule of BQ . It follows that

|CQ | |FQ |0 |D|0 = |D|.

Similar arguments give that |Q,n (AQ )| |D| for every n N, and, consequently,
|AQ || |D|. Since is nite, |T | |D|.
The factor-module A/T is D-torsion-free. By Lemma 11.7, E = (A/T ) R F
is an artinian F G-module, where F is the eld of fractions of the integral domain
D. Since D is innite, |F | = |D|. By Theorem 11.6,

|E| |F |0 |D|0 = |D|,

and hence |A| ||D|. 


With the help of some additional restrictions on G, one can obtain conditions
under which the socle height of any artinian DG-module A is at most .
Let G be a group. From now on, we denote by P (G) the maximal normal
periodic subgroup of G.
Lemma 11.9. Let G be an F C-hypercentral group with P (G) = 1. If G has a
nite 0-rank, then G has a normal nilpotent torsion-free subgroup of nite index.
Proof. The elements of A1 = F C(G) having nite order form a characteristic
subgroup (see Corollary 3.4), so that the equation P (G) = 1 implies that A1 is
torsion-free. Since it has a nite 0-rank, it follows that A1 has a nitely generated
subgroup B1 such that A1 /B1 is periodic. Put C1 = (B1 )G . The inclusion B1
F C(G) implies that C1 is a nitely generated subgroup, and the index |G : CG (C1 )|
is nite. Put E1 /C = P (G/C1 ) CG (C1 )/C1 so that C1 (E1 ), and E1 /C1 is
periodic. By Corollary 3.3, it follows that [E1 , E1 ] is also periodic. In particular the
elements of E1 having nite orders form a characteristic subgroup. It turns out that
E1 is an abelian torsion-free subgroup. Since the operation of taking roots in an
abelian torsion-free group is unique, CG (C1 ) = CG (E1 ) because E1 /C1 is periodic,
and hence the index |G : CG (E1 )| is nite. Put Z1 = CG (E1 ), and A2 /E1 =
F C(G/E1 ) Z1 /E1 . As above, it can be shown that A2 /E1 is an abelian torsion-
free group. Iterating the previous arguments, we nd a normal subgroup E2 such
that E1 E2 , E2 /E1 is abelian and torsion-free, and Z2 = Z1 CG (E2 /E1 ) has
nite index in G. Moreover, if G/E2 is innite, then A3 /E2 = F C(G/E2 ) Z2 /E2
is torsion-free abelian. Proceeding by ordinary induction, we continue the process
of construction of subgroups En . Since the 0-rank of G is nite, this process cannot
be innite; that is, after nitely many steps we come to a nilpotent torsion-free
subgroup En having nite index. 
140 Chapter 11. On the countability of artinian modules

Lemma 11.10. Let D be a Dedekind domain with innite set Spec (D), G be
a locally (polycyclic-by-nite) group of nite 0-rank, and A an artinian DG-
module. Then A is D-periodic. If AP is the P -component of A, P AssD (A) and
p =char(D/P ), then the additive group of AP is a p-group whenever p > 0, and
the additive group of AP is torsion-free whenever p = 0.
Proof. Since A is artinian, A has an ascending series of DG-submodules {A | <
}, every factor A+1 /A of which is a simple DG-module. According to L.A. Kur-
dachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.16], Ann D (A+1 /A )
Spec (D) for every index < . It is not hard to see now that A is D-periodic.
Let P Ass D (A), and suppose that p =char (D/P ). If p > 0, then the
additive group of D/P is an elementary abelian p-group. By Corollary 6.17,
P/P 2 =D D/P , and so the orders of the elements of D/P 2 divide p2 . By the same
logic, we obtain that the orders of the elements of D/P n divide pn . If a AP ,
then Ann D (a) = P n for some n N, so that aD =D D/P n . In particular, the
n
order of a divides p . This means that the additive group of AP is a p-group. If
p = 0, then the additive group of D/P is torsion-free. Using again the isomor-
phism P/P 2 =D D/P , we obtain that the additive group of D/P 2 is torsion-free.
And similarly we obtain that the additive group of D/P n is torsion-free for each
n N. If a AP , then Ann D (a) = P n for some n N, so that aD =D D/P n . In
particular, it follows that the order of a is innite. 
Corollary 11.11. Let D be a Dedekind domain with innite set Spec(D), G be a
locally (polycyclic-by-nite) group of nite 0-rank, and A an artinian monolithic
DG-module. Then P =Ann(DG (A)) = 0 and A is a P -module. Furthermore, if
p = char D/P > 0, then the additive group of AP is a p-group; and if p = 0, then
the additive group of A is torsion-free.
Proof. Indeed, by Lemma 11.10, A is D-periodic. By Corollary 6.25,

A= AQ ,
Q

where =Ass D (A). Since every primary component AQ is a DG-submodule, the


inclusion DG (A) AP implies that A = AP . 
Lemma 11.12. Let G be an F C-hypercentral group, D a Dedekind domain, A a
monolithic D-periodic DG-module, H = CG (DG (A)), CG (A) = 1 , and p =
char(D/Ann D (DG (A)). Suppose also that A is DH-hypercentral. If p > 0, and
the group G has nite section p-rank, then H has a normal nilpotent torsion-free
subgroup of nite index. If p = 0, and G has nite 0-rank, then P (H) = 1, and
H has a normal nilpotent torsion-free subgroup of nite index.
Proof. Using Lemma 2.10, it is not hard to see that if an F C-hypercentral group
has nite section p-rank for some prime p, then it has also nite 0-rank. In other
words, in each of these cases, the 0-rank of the group is nite. Let L be an arbi-
trary nitely generated subgroup of H. Since L acts trivially on the factors of an
Chapter 11. On the countability of artinian modules 141

ascending series of A, then L is residually nilpotent (see P. Hall and B. Hartley


[99, Theorem A2]. It follows that the set T of all elements of H having nite order
form a (characteristic) locally nilpotent subgroup of H (see, for example, [235,
Theorem 6.14]).
Suppose that p > 0, and let P be the Sylow p-subgroup of T , and Q the
Sylow p -subgroup of T . Since every abelian subgroup of P is Chernikov, P is
likewise Chernikov (see for example, [234, Theorem 3.32]). Let D be the divisible
part of P . Proceeding as we did in the proof of Corollary 11.11, we obtain that
the additive group of A is a p-group. The natural semidirect product A  T is a
hypercentral group, because A is F H-hypercentral. But, then Theorem 3.10 shows
that D Q CH (A), that is the periodic part of H is nite. Applying Lemma
11.9 to H/T we obtain that H/T is nilpotent-by-nite and almost torsion-free.
According to D.I. Zaitsev [291, Lemma 3], we obtain that H is nilpotent-by-nite
and almost torsion-free. Let now p = 0. Proceeding as we did in the proof of
Corollory 11.11, we obtain that the additive group of A is torsion-free. Since the
natural semidirect product A  T is a hypercentral group, and all elements of
hypercentral group having nite orders form a subgroup, then A  T = A T ,
that is, T CG (A) = 1. Applying Lemma 11.9 to H, we obtain that H/T is
nilpotent-by-nite and almost torsion-free. 
Corollary 11.13. Let G be an F C-hypercentral group, D a Dedekind domain with
innite set Spec(D), A an artinian monolithic DG-module, H = CG (DG (A)),
CG (A) = 1, and p = char(D/Ann D (DG (A)). If p > 0, and the group G has
nite section p-rank, then H has a normal nilpotent torsion-free subgroup of nite
index. If p = 0, and G has nite 0-rank, then P (H) = 1, and H has a normal
nilpotent torsion-free subgroup of nite index.
Proof. By Lemma 11.10, A is D-periodic. Theorem 11.1 yields that A is a DH-
hypercentral module. Thus all conditions of Lemma 11.12 are satised. 
Lemma 11.14. Let G be a nilpotent torsion-free group of nite 0-rank, and let p
be a prime. If H is a subgroup satisfying r0 (H) = r0 (G), then G has a subgroup
L with the following properties:
(1) H L and the index |L : H| is nite.
(2) There exists a subnormal series
L = L0  L1   Ln = G,
all factors of which are periodic p-divisible groups.
Proof. Since G is nilpotent, H is subnormal with defect at most k, where k is the
nilpotency class of G. Let
H = H 0  H1   H n = G
be a subnormal series. We proceed by induction on n. If H is normal in G, then
the equation r0 (H) = r0 (G) yields that G/H is periodic. Since G is a nilpotent
142 Chapter 11. On the countability of artinian modules

torsion-free group of nite 0-rank, then G has nite special rank. It follows that the
Sylow p-subgroup P/H of G/H is Chernikov (see, for example, [235, Corollary 1
to Theorem 6.36]). Denote by D/H the divisible part of P/H. Note that the center
of a nilpotent periodic group contains every divisible subgroup (see, for example,
[234, Lemma 3.13]), so that P/H = (D/H)(L/H) for some nite G-invariant
subgroup L/H. Clearly, G/L is p-divisible.
Let now n > 0, and suppose that our statement has been shown for subnormal
subgroups whose defect is less than n. In particular, there is a subgroup L1 H1
such that the index |L1 : H1 | is nite, and L1 is a member of a nite subnormal
series of G with p-divisible factors. There is no loss if we suppose that H1 is normal
in L1 . The family {H g | g L1 } is nite, that is,
{H g | g L1 } = {g11 Hg1 , . . . , gs1 Hgs }.
From H1 /H g = H1g /H g = H1 /H, we obtain that H1 /H g is periodic for each
element g L1 . By Remaks Theorem,
H1 /K
H1 /(g11 Hg1 ) H1 /(gs1 Hgs ),
where K = g11 Hg1 gs1 Hgs is an L1 -invariant subgroup. This embedding
shows that L1 /K is also periodic. Let be the set of all prime divisors of the
number |L1 : H1 |, and put = {p}. We have
L1 /K = U/K V /K,
where U/K is the Sylow -subgroup, and V /K is the Sylow -subgroup. By the
choice of , we obtain the inclusion V /K H1 /K, which proves that H is V -
invariant. Since is nite, U/K is a Chernikov group. Let D/K be the divisible
part of U/K. Since the center of a nilpotent periodic group includes every di-
visible subgroup (see, for example, [234, Lemma 3.13]), U/K = (W/K)(D/K)
for some nite L1 -invariant subgroup W/K. Put U1 /K = U/K H/K, and
V1 /K = V /K H/K. We have already noted that V1 /K is normal in V /K.
Since (U/K)/(W/K) is abelian, (U1 /K)(W/K) is normal in U/K. It follows that
L/K = (U1 /K)(W/K)(V1 /K) is normal in L1 /K. By the choice of L1 /L, we ob-
tain that the factor-group L1 /L is p-divisible so that L is a member of a nite
subnormal series of G with p-divisible factors. The niteness of W/K, and the
equation L/K = (H/K)(W/K) imply that |L : H| is nite; that is, conditions (1)
and (2) hold. 
Corollary 11.15. Let G be a nilpotent torsion-free group of nite 0-rank, and let p be
a prime. Then G has a nitely generated subgroup L with the following properties:
(1) r0 (L) = r0 (G).
(2) There exists a subnormal series
L = L0  L1   Ln = G,
all factors of which are periodic p-divisible groups.
Chapter 11. On the countability of artinian modules 143

Proof. Let
1 = Z0 Z1 Zn = G
be the upper central series of G. For every 1 j n, we choose elements

gj,1 , . . . , gj,k(j) Zj \ Zj1


 
such that Zj / gj,1 , . . . , gj,k(j) Zj1 is periodic, that is
 
r0 ( gj,1 , . . . , gj,k(j) Zj1 /Zj1 ) = r0 (Zj /Zj1 ).
 
Put now H = gj1 , . . . , gjk(j) |1 j n . The subgroup H is nitely generated,
and r0 (H) = r0 (G). It suces to apply Lemma 11.14 to H. 
Lemma 11.16. Let G be an F C-hypercentral group, F a eld of characteristic p, A
a monolithic F G-module, H = CG (F G (A)), and CG (A) = 1. Suppose also that
if p = 0, then G has nite 0-rank, and if p > 0, then G has nite section p-rank.
If A is an F H-hypercentral module, then the length of the upper F H-hypercentral
series of A is at most .
Proof. Corollary 11.13 yields that H has a normal nilpotent subgroup L having
nite index in H. Let |H/L| = k. Then H k L, and, in particular, H k is torsion-
free nilpotent. Clearly, H k is also G-invariant. Since the orders of the elements
of H/H k are bounded, and H has nite special rank, it follows that H/H k is
nite. In order to avoid additional notation we can assume that L is a G-invariant
subgroup of H. By Corollary 11.15, L has a nitely generated subgroup K such
that r0 (L) = r0 (K), and there exists a nite subnormal series

K = K0  K1   Kn = L,

whose factors are periodic, and p-divisible. Let a be an arbitrary element of A. The
natural semidirect product A  H is a hypercentral group, because A is an F H-
hypercentral module. Then the subgroup a, K is nitely generated and nilpotent.
It follows that a is contained in some F K-hypercenter of A corresponding to a
nite ordinal m. It follows that the length of the upper F K-central series of A
is at most . Put C = F K (A). Since K is normal in K1 , C is naturally an
F K1 -submodule.
Suppose rst that p > 0. Since K1 /K is p-divisible, Lemma 3.10 yields the
inclusion C CA (K1 ); this proves that the F K1 -center of the module A co-
incides with its F K-center. Similarly, we prove that every F K1 -hypercenter of
the module A coincides with its F K-hypercenter having the same number. Us-
ing the p-divisibility of the factors K2 /K1 , . . . , Kn /Kn1 , and ordinary induction,
we obtain at last that every F L-hypercenter of the module A coincides with its
F K-hypercenter having the same number. Since C is F H-hypercentral, from the
niteness of H/L, and reasoning as above, we can deduce that H/CH (C) is a nite
nilpotent group. It follows that E = C  H is nilpotent (see, for example, [235,
144 Chapter 11. On the countability of artinian modules

Lemma 6.34]). In other words, C has an F H-central series of nite length. The
same is valid for each factor of the upper F L-central series, which means that the
length of the F H-central series of the module A is at most .
Now let p = 0. In this case, the additive group of the module A is torsion-
free. Since the natural semidirect product A  K1 is a hypercentral group, and the
elements of a hypercentral group having nite order form a subgroup, A  K1 =
A K1 , that is C CA (K1 ), and hence the F K1 -center of the module A coincides
with its F K-center. As above, it follows that every F L-hypercenter of the module
A coincides with its F K-hypercenter having the same number. Since C is F H-
hypercentral, we obtain again that H/CH (C) is a nite nilpotent group. It turns
out that C CA (H), and hence the F H-center of A coincides with its F K-
center. Therefore, every F H-hypercenter of the module A coincides with its F K-
hypercenter having the same number. Thus, in this case, the length of the F H-
central series of the module A is at most . 
Lemma 11.17. Let G be a group, D be a Dedekind domain, and A a simple DG-
module. If Q = G/CG (A) has a normal abelian subgroup R of nite index, then
the periodic part T of R has nite special rank.
Proof. By Theorem 5.5, 
A= Cx,
xS

where C is a simple DR-submodule, and S is a nite set. Obviously, Cx is a simple


DR-submodule; therefore t(R/CR (Cx)) is a locally cyclic p -group, where p =char
(D/Ann D (A)) for every x S (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin
[157, Corollaries 2.4, 2.5]). The equation

CR (Cx) = CR (A) = 1 ,
xS

and Remaks Theorem together imply the embedding

R
DrxS R/CR (Cx),

which shows that t(R) has nite special rank. 


Lemma 11.18. Let G be a group of nite special rank. If [G, G] is nite, then
G/(G) is also nite.
Proof. We rst consider the case, when G is a periodic group. Put K = [G, G],
and C = CG (K) so that K C (C). Thus C is nilpotent. Put = (K), and
let P be the Sylow -subgroup of C, and Q the Sylow -subgroup of C. Clearly,
Q is G-invariant. Since Q K = 1, Q (G). Denote by D the divisible part of
P . Since G is an F C-group, D (G). By the niteness of the set , we obtain
that P is Chernikov so that P/D is nite. Then C/DQ is nite. Since K is nite,
G/C is nite. Then G/DQ is nite, and it follows that G/(G) is nite.
Chapter 11. On the countability of artinian modules 145

Consider now the general case. Let U be a maximal torsion-free subgroup of


(G). The factor-group G/U is periodic (see D. J.S. Robinson [234, Theorem 4.32]).
Applying the result shown in the above paragraph, Z/U = (G/U ) has nite index
in G/U . Let z Z, and g G so that [g, z] U, and then [g, z] U [G, G] = 1.
It follows that Z = (G), and hence G/(G) is nite as required. 
Proposition 11.19. Let G be an F C-hypercentral group, F a eld of characteristic
p, and A an artinian monolithic F G-module. Suppose further that if p > 0, then G
has nite section p-rank, and if p = 0, then G has nite 0-rank. If the factor-group
G/CG (F G (A)) is abelian-by-nite, then every nitely generated F G-submodule of
A has nite F G-composition series. In particular, the socular height of A is at most
.
Proof. Put H = CG (F G (A)). By Corollary 11.13, H has a normal torsion-free
subgroup L having nite index. We proceed by induction on r0 (L). If r0 (L) = 0,
then H is nite. By Lemma 11.17, and Lemma 11.18, G is abelian-by-nite, and it
suces to apply Lemma 11.2. Let now r0 (L) > 0. In this case (L) = 1, which,
by Corollary 3.16 gives that (L) F C(G) = 1. Pick 1 = x F C(G) (L), and
G
put X = x and Y = CG (X). Then X is a nitely generated abelian torsion-free
subgroup, Y is a normal subgroup of nite index, and X (Y ). The F Y -module
A is artinian by Theorem 5.2. Theorem 11.1 yields that A is F H-hypercentral.
By Theorem 11.6, the length of the upper F H-hypercentral series of A is at most
. In particular, every element b of A is contained in some F H-hypercenter of A
with a nite number m(b). Thus, the F Y -submodule B = bF Y is F H-nilpotent.
Then B is F g-nilpotent for each element g H so that B(g 1)m(b) = 0.
Since x (Y ), the mapping a a(x 1)n , a B, is an F Y -endomorphism for
every n N. Put B0 = B, and Bn+1 = Bn (x 1), n > 1. Then Bn is a nitely
generated F Y -submodule for every n N.
Suppose that A is a monolithic F Y -module. There is some t m(b) such that
B(g1)t+1 = 0 but B(g1)t = 0. Then x CL (Bt ), and hence r0 (L/CL (Bt ) <
r0 (L). It follows that Bt has a nite F Y -composition series. As we have already
noted above the mapping : a a(x 1), a Bt1 is an F Y -endomorphism,
and Im = Bt (x 1) = Bt , and Ker = Bt so that Bt1 /Bt =F Y Bt . It follows
that Bt1 /Bt , and hence Bt1 have nite F Y -composition series. Proceeding by
ordinary induction, after nitely many steps we obtain that B0 = B has nite
F Y -composition series. By the niteness of G/Y , the F G-submodule BF G has
nite F Y -composition series, thus BF G has also nite F G-composition series.
Consider now the general case. By Theorem 5.5,

M = F G (A) = Cx,
xS

where C is a simple F Y -submodule, and S is a nite subset. For every x S,


by Ux a maximal submodule among the F G-submodules that con-
we denote
tains x=vS Cv and meets Cx trivially. Then the factor-module A/U x is F Y -
monolithic with the monolith (Cx + U x)/U x. By the result shown in the above
146 Chapter 11. On the countability of artinian modules

paragraph, (B + U x)/U x has nite F Y -composition series for every x S. Fur-


thermore, 
Ux
xS

is an F G-submodule such that



( U x) M = 0 .
xS

Since A is a monolithic F G-module,



U x = 0 .
xS

Application of Remaks Theorem gives the embedding



A
A/U x.
xS

In this setting 
B
x S(B + U x)/U x,

so that B has nite F Y -composition series. Finally, by the niteness of G/Y , the
F G-submodule BF G has nite F Y -composition series; thus BF G has also nite
F G-composition series. 
Theorem 11.20. Let G be an F C-hypercentral group, D a Dedekind domain with
innite set Spec(D), A an artinian monolithic DG-module, and p the characteris-
tic of D/AnnD (DG (A)). Suppose further that if p > 0, then G has nite section
p-rank, and if p = 0, then G has nite 0-rank. If the factor-group G/CG (F G (A))
is abelian-by-nite, then every nitely generated DG-submodule of A has nite
DG-composition series. In particular, the socular height of A is at most .
Proof. Put H = CG (DG( A)). By Lemma 11.10, A is a P -module, where P =
Ann D (DG (A)). Pick b A, and put B = bDG. Then Ann D (b) = P n for some
n N, that is, B = P,n (B). We proceed by induction on n. If n = 1, then
b P,1 (A). Think of P,1 (A) as an F G-module, where F = D/P is a eld. By
Proposition 11.19, B has nite DG-composition series. Let now n > 1. Pick some
y P \ P 2 . By Proposition 6.13, yD + P m = P for every m N. The mapping
: c cy m , c B is a DG-endomorphism, and hence By m is a nitely generated
DG-submodule for every m N. In particular, By n1 is a nitely generated
DG-submodule. Since y AnnD (By n1 ), the equation P = yD + P n implies
that AnnD (By n1 ) = P . By the induction hypothesis, By n1 has nite DG-
composition series. Furthermore, Im = (By n2 )y = By n1 and Ker = By n1 ,
so that
By n2 /By n1
= By n1 .
Chapter 11. On the countability of artinian modules 147

It follows that By n2 /By n1 , and hence By n2 has nite DG-composition series.


Proceeding by ordinary induction, after nitely many steps we obtain that B has
nite DG-composition series. 
Corollary 11.21. Let G be an F C-hypercentral group of nite section rank, D a
Dedekind domain with innite set Spec(D), and A an artinian DG-module. If the
factor-group G/CG (Soc F G (A)) is abelian-by-nite, then every nitely generated
DG-submodule of A has nite DG-composition series. In particular, the socular
height of A is at most .
Proof. As in some of the above proofs, the general case can be reduced to the case
of a monolithic module, and for the latter it suces to apply Theorem 11.20. 
Lemma 11.22. Let F be a eld of characteristic p, G a periodic group, and A a
monolithic F G-module. Suppose further that the following conditions hold:
(1) G has a normal abelian subgroup U of nite index.
(2) If p > 0, then U is a p -subgroup.
Then A has nite FG-composition series.
Proof. Let M be the F G-monolith of A. Applying Theorem 7.28, we have that

A=M B

for some F U -submodule B. By Theorem 5.5,



M= Cx,
xS

where C is a simple F U -submodule, and S is a nite subset of G. In particular,


A/B has nite F U -composition series. This remains true for every factor-module
A/Bg, g G. Since G/U is nite, the set {Bg | g G} is nite. Let

{Bg | g G} = {Bg1 , . . . , Bgm }.

Since the intersection


C = Bg1 Bgm
is an F G-submodule, by its construction, C M = 0. It follows that C = 0.
By Remaks Theorem

A
A/Bg1 A/Bgm ,

which shows that A has nite F U -composition length. Then A has nite F G-
composition length. 
Proposition 11.23. Let G be an F C-hypercentral group, F a eld of characteristic
p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
148 Chapter 11. On the countability of artinian modules

(1) A is F H-nilpotent.
(2) If p > 0, then G has nite section p-rank.
(3) If p = 0, then G has nite 0-rank.
If G/H is abelian-by-nite, then A has nite F G-composition series.
Proof. By Lemma 11.12, H contains a normal torsion-free subgroup L of nite
index. We proceed by induction on r0 (L). If r0 (L) = 0, then H is nite. Moreover,
if p > 0, then H is a p-subgroup. By Lemma 11.17 and Lemma 11.18, G is periodic,
and abelian-by-nite. Let U/H be a normal abelian subgroup of G/H having nite
index. Put M = F G (A). By Theorem 5.5,

M= Cx,
xS

where C is a simple F U -submodule, and S is a nite subset of G. Clearly, Cx is a


simple F U -submodule for every x S, and therefore U/CU (Cx)) is a locally cyclic
p -group (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [158, Corollaries 2.4
and 2.5]). The equation

CU (Cx) = CU (A) = H
xS

and Remaks Theorem allow us to deduce the embedding

U/H
Dr xS U/CU (Cx),

which establishes that U/H is a p -group. By Lemma 11.17 and Lemma 11.18,
G is abelian-by-nite. Now it is not hard to obtain that G has a normal abelian
p -subgroup V of nite index. In this case, it suces to apply Lemma 11.22.
Let now r0 (L) > 0. In this case, it suces to repeat almost word for word
the arguments given in the proof of Proposition 11.19. 

Corollary 11.24. Let G be an F C-hypercentral group, F a eld of characteristic


p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
(1) A is F H-nilpotent.
(2) dimF (F G (A)) is nite.
(3) If p > 0, then G has a nite section p-rank.
(4) If p = 0, then G has nite 0-rank.
If G/H is periodic, then A has nite F G-composition series.
Chapter 11. On the countability of artinian modules 149

Proof. Put M = F G (A). The factor-group G/H is nilpotent-by-nite (see L.A.


Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary 6.13]). Let U be a normal
subgroup of G having nite index such that U/H is nilpotent. By Theorem 5.5,

M= Cx,
xS

where C is a simple F U -submodule, and S is a nite subset of G. Clearly, Cx is a


simple F U -submodule for every x S; therefore U/CU (Cx)) is abelian-by-nite
(see B.A.F. Wehrfritz [280, Lemma 3.5]). The equation

CU (Cx) = CU (A) = H
xS

and Remaks Theorem give the embedding

U/H
DrxS U/CU (Cx),

which establishes that U/H is abelian-by-nite. Now it suces to apply Proposi-


tion 11.23. 
Corollary 11.25. Let F be a locally nite eld of characteristic p, G an F C-
hypercentral group of nite section p-rank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-nilpotent, and G/H is abelian-by-nite, then A has nite
F G-composition series.
Proof. We recall that the niteness of the section p-rank implies the niteness of
the 0-rank. Also note that an F C-hypercentral group is locally polycyclic-by-nite.
Let U be a normal abelian subgroup having nite index in G. Put M = F G (A).
By Theorem 5.5, 
M= Cx,
xS

where C is a simple F U -submodule, and S is a nite subset of G. Clearly, Cx is


a simple F U -submodule for every x S, and therefore U/CU (Cx)) is a periodic
p -group (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.21]).
The equation 
CU (Cx) = CU (A) = H
xS

and Remaks Theorem give the embedding

U/H
DrxS U/CU (Cx),

which establishes that U/H is periodic. This implies that G/H is likewise periodic.
Now it suces to apply Proposition 11.23. 
Corollaries 11.24 and 11.25 at once give
150 Chapter 11. On the countability of artinian modules

Corollary 11.26. Let F be a locally nite eld of characteristic p, G an F C-


hypercentral group of nite section prank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-nilpotent, and dimF (F G (A)) is nite, then A has nite
F G-composition series.
Corollary 11.27. Let G be an F C-hypercentral group, F a eld of characteristic
p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
(1) A is F H-hypercentral.
(2) If p > 0, then G has nite section p-rank.
(3) If p = 0, then G has nite 0-rank.
If G/H is periodic abelian-by-nite, then each nitely generated F G-submodule of
A has nite F G-composition series. In particular, the socular height of A is nite.
Proof. Indeed, by Theorem 11.6, the length of the upper F H-central series of A
is at most . If a A, then a belongs to a term of the upper F H-central series
having a nite number. This means that aF G is F H-nilpotent, and it suces to
apply Proposition 11.23. 

Corollary 11.28. Let G be an F C-hypercentral group, F a eld of characteristic


p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
(1) A is F H-hypercentral.
(2) dimF (F G (A)) is nite.
(3) If p > 0, then G has a nite section p-rank.
(4) If p = 0, then G has nite 0-rank.
If G/H is periodic, then every nitely generated F G-submodule of A has nite
F G-composition series. In particular, the socular height of A is at most .
Corollary 11.29. Let F be a locally nite eld of characteristic p, G an F C-
hypercentral group of nite section p-rank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-hypercentral, and G/H is abelian-by-nite, then every
nitely generated F G-submodule of A has nite F G-composition series. In partic-
ular, the socular height of A is at most .
Corollary 11.30. Let F be a locally nite eld of characteristic p, G an F C-
hypercentral group of nite section p-rank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-hypercentral, and dimF (F G (A)) is nite, then every
nitely generated F G-submodule of A has nite F G-composition series. In partic-
ular, the socular height of A is at most .
Chapter 11. On the countability of artinian modules 151

Corollary 11.31. Let G be an F C-hypercentral group, F a eld of characteristic


p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
(1) A is F H-hypercentral.
(2) If p > 0, then G has nite section p-rank.
(3) If p = 0, then G has nite 0-rank.
If G/H is periodic abelian-by-nite, then every term of the upper socular series of
A has nite F G-composition series. In particular, A is an artinian F G-module.
Proof. In fact, by Corollary 11.27, the length of the upper socle series of A is at
most . Let
0 = S0 S1 Sn Sn+1
be the upper socular series of A. Since A is monolithic, S1 is the F G-monolith of
A. We have 
S2 /S1 = B /S1 ,

where B /S1 is a simple F G-submodule for every . By (1), B (h 1) S1


for every h H, and every index , so that S2 (h 1) S1 for every h H.
This means that S2 is F H-nilpotent. By Proposition 11.23, S2 has nite F G-
composition series. Using similar arguments, and applying ordinary induction, we
obtain that Sn has nite F G-composition series for every n N. Now it is not
hard to show that A is an artinian F G-module. 

Corollary 11.32. Let G be an F C-hypercentral group, F a eld of characteristic


p, A a monolithic F G-module, and H = CG (F G (A)). Suppose further that G
satises the following conditions:
(1) A is F H-hypercentral.
(2) dimF (F G (A)) is nite.
(3) If p > 0, then G has nite section p-rank.
(4) If p = 0, then G has nite 0-rank.
If G/H is periodic, then every term of the upper socular series of A has nite
F G-composition series. In particular, A is an artinian F G-module.
Corollary 11.33. Let F be a locally nite eld of characteristic p, G an F C-
hypercentral group of nite section p-rank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-hypercentral, and G/H is abelian-by-nite, then every
term of the upper socular series of A has nite F G-composition series. In partic-
ular, A is an artinian F G-module.
152 Chapter 11. On the countability of artinian modules

Corollary 11.34. Let F be a locally nite eld of characteristic p, G an F C-


hypercentral group of nite section p-rank, A a monolithic F G-module, and H =
CG (F G (A)). If A is F H-hypercentral, and dimF (F G (A)) is nite, then every
term of the upper socular series of A has nite F G-composition series. In partic-
ular, A is an artinian F G-module.
Theorem 11.35. Let G be an F C-hypercentral group, D a Dedekind domain with
innite set Spec(D), A a monolithic DG-module, H = CG (DG (A)), and p the
characteristic of D/AnnD (DG (A)). Suppose further that G satises the following
conditions:
(1) A is DH-hypercentral and D-periodic.
(2) If p > 0, then G has nite section p-rank.
(3) If p = 0, then G has nite 0-rank.
If G/H is periodic and abelian-by-nite, then A satises the following properties:
(i) Each nitely generated DG-submodule of A has nite DG-composition series
(ii) The socular height of A is at most .
(iii) A is an artinian DG-module.
Proof. Since A is monolithic and D-periodic, we have that A is a P -module, where
P =AnnD (DG (A)). Repeating almost word for word the arguments of the proof
of Theorem 11.20, we obtain (i). The statement (ii) is a direct consequence of
(i). Finally, by Corollary 11.31, P,1 (A) is artinian. Therefore, it suces to apply
Proposition 1.8. 
Theorem 11.36. Let G be an F C-hypercentral group of nite section rank, D a
Dedekind Z-domain, A a monolithic DG-module, and H = CG (DG (A)). Suppose
further that the following conditions hold:
(a) A is DH-hypercentral.
(b) A is D-periodic.
(c) G/H is abelian-by-nite.
Then A satises the following properties:
(i) Every nitely generated DG-submodule of A has nite DG-composition se-
ries.
(ii) The socular height of A is at most .
(iii) A is an artinian DG-module.
Chapter 12

Artinian modules over periodic


abelian groups

From now on, in this and other chapters, our goal is the consideration of artinian
modules over a group ring DG, where D is a Dedekind domain, and G is an abelian
group of nite special rank. The rst step here is the study of artinian modules over
periodic groups. B. Hartley and D. McDougall [110] considered artinian modules
over a ring having the form ZG, where G is an abelian Chernikov group. Their
results could be extended to the case of a ring of the form DG, where G is a
periodic abelian group of nite rank, and D is a Dedekind domain. As we shall
see in this investigation, the previously obtained results about the existence of
complements to submodules play a crucial role here.
We start with some general results.
Lemma 12.1 (L.A. Kurdachenko [150]). Let R be a commutative ring. Let A be an
artinian monolithic R-module with monolith M , and put L = AnnR (M ). Then L
annihilates every factor of any ascending composition series of A.
Proof. Assume the contrary. Then there exist two submodules B C of A such
that B/C is simple, and L annihilates all the factors of any ascending series of
C but AnnR (B/C) does not include L. Since B/C is simple, there exists some
b B such that B/C = (b + C)R. Put B1 = bR. Then B1 =R R/AnnR (b).
Since R is commutative, AnnR (b) is a two-sided ideal of R. Hence R/AnnR (b)
is an artinian ring, and then it is noetherian (see R. Sharp [263, Theorem 8.44]).
Therefore B1 has at least a nite composition series. If x L, by the Fitting lemma
(Proposition 1.5), there exists a decomposition B1 = U V , where x induces on
U an automorphism, and V xn = 1 for some n N. It follows that M V , and
so U = 1. Hence, B1 xn = 1. In particular, Bx C, and since this holds for
every x L, L AnnR (B/C), a contradiction. 
Corollary 12.2 ([150]). Let A be an artinian monolithic RG-module, where R is a
commutative ring, and G is an abelian group. Suppose that M is the RG-monolith
of A. If H = CG (M ), then A is an RH-hypercentral module.
154 Chapter 12. Periodic abelian groups

Proof. Indeed, if x H, then x1 AnnRG (M ). By Lemma 12.1, x1 annihilates


all the factors of every ascending composition series of A. 
Let A be an R-module, and pick y R. We say that A is a y-module (over
R) if A is an L-module, where L = Ry. In other words, A is a y-module if and
only if every element of A is annihilated by some element of the form y t , where
t N.
A Dedekind domain D is said to be a Dedekind Z-domain if D satises the
following conditions:
(1) Spec (D) is innite.
(2) If 0 = P Spec (D), then the eld D/P is locally nite.
Clearly, Z is a Dedekind Z-domain. Another important example for its ap-
plications in the theory of groups is the group algebra of an innite cyclic group
over a nite eld.
Theorem 12.3 ([150]). Let R be a commutative ring, G an abelian group, and A
an artinian RG-module. If S =Soc RG (A) and H = CG (S), then
(1) A is RH-hypercentral. In particular, A is an (x1)-module, for every x H.
(2) If R is a Dedekind Z-domain, and r0 (G) is nite, then G/H is a periodic
group of nite special rank.
Proof. We have
S = M1 Mn ,
where M1 , . . . , Mn are some simple RG-submodules of A. Choose submodules
B1 , . . . , Bn of A such that, for every 1 j n, every Bj is maximal under

(i) k=j Mk Bj ;

(ii) Mj Bj = 1.
Then every A/Bj is a monolithic RG-module with monolith (Mj + Bj )/Bj . Let
Hj = CG ((Mj + Bj )/Bj ). By Corollary 12.2, A/Bj is RHj -hypercentral. Let
x B1 Bn . Then a(x 1) Bj , for every a Mj . But a(x 1) Mj , so
that a(x 1) = 0, and x CG (Mj ). Thus H = H1 Hn .
Put
E = A/B1 A/Bn .
Clearly, E is an RH-hypercentral module. The mapping
: a (a + B1 , . . . , a + Bn ), a A
is an RG-homomorphism, hence Im is RH-hypercentral. Since

Ker S = (B1 Bn ) S = 0 ,

A
=RG Im , and thus A is RH-hypercentral.
Chapter 12. Periodic abelian groups 155

If x H, then the natural semidirect product A  a, x is a hypercentral


group. It follows that a, x is a nilpotent subgroup for every a A, so a(x 1)s =
0, for some s N. This means that A is an (x 1)-module for every x H.
To nish, we note that if R is a Dedekind Z-domain, and r0 (G) is nite, then
every G/Hj is a locally cyclic periodic group (see L. A. Kurdachenko, J. Otal and
I. Ya. Subbotin [157, Theorem 3.2]). By Remaks theorem
G/H
G/H1 G/Hn ,
therefore G/H is a periodic group of nite special rank, as required. 
Lemma 12.4 ([150]). Let A be an artinian DG-module, where D is a Dedekind
Z-domain, and G is an abelian group of nite 0-rank. Then A is a D-periodic
module. Moreover, if P AssD (A), and AP is the P -component of A, then the
additive group of Ap is a p-group where p = char(D/P ). In particular, A is also
Z-periodic.
Proof. Since A is artinian, A has an ascending series of DG-submodules

0 = A0 A1 A A+1 A

such that every factor A+1 /A of this series is a simple DG-module, < .
Acoording to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.16],
AnnD (A+1 /A ) Spec (D), for every < . This means that the additive group
of A is D-periodic.
Put P Ass D (A), and let AP be the P -component of A. Since D/P is a
locally nite eld, the additive group of D/P is a p-group, where p = char (D/P ).
By Corollary 6.17, P/P 2 =D D/P . It follows that the order of each element of
D/P 2 divides p2 . Similarly, we obtain that the order of each element of D/P n
divides pn . If a AP , then there is some n N such that Ann D (a) = P n , and
so aD =D D/P n . In particular, the additive order of a divides pn . Thus AP is a
p-group. Since AP is a DG-submodule, AssD (A) is nite. By Corollary 6.25,

A= AP ,
P

where = Ass D (A), then it follows that A is also Z-periodic. 


Proposition 12.5 ([150]). Let D be a Dedekind domain, and let G be an abelian
group of nite special rank. If P Spec D(A), let A be an artinian P -module over
DG, and put H = CG (Soc DG (A)). If H is periodic, then G has a subgroup Q of
nite index such that H Q CG (A).
Proof. Put p = char (D/P ). If p = 0, then the additive group of A is torsion-
free. By Theorem 12.3, A is a DH-hypercentral module. Put A1 = DH (A), and
A2 /A1 = DH (A/A1 ). Then H acts trivially on the factors of the series

0 = A0 A1 A2 .
156 Chapter 12. Periodic abelian groups

According to O.H. Kegel and B.A.F. Wehrfritz [134, 1.C.3], H/CH (A2 ) is iso-
morphic to some subgroup of HomZ (A2 /A1 , A1 ). Since A1 is Z-torsion-free,
HomZ (A2 /A1 , A1 ) is Z-torsion-free. As H is periodic, H = CH (A2 ), and hence
A2 = A1 . Then A = A1 , and so H = CH (A).
Suppose now that p > 0. By Lemma 12.4, the additive group of A is a
p-group. For every n N, put

An = p,n (A) = {a A | pn a = 0}.

We note that H = U V W , where U is a nite p-subgroup, V is a divis-


ible p-subgroup, and W is a p -subgroup. Put B1 = DV W (An ), and B2 /B1 =
DV W (A/An ). Then V W acts trivially on the factors of the series

0 = B0 B1 B2 ,

and then V W/CV W (B2 ) can be embedded in HomZ (B2 /B1 , B1 ). Since the order of
any element of B2 divides pn , the order of an element of HomZ (B2 /B1 , B1 ) divides
pn . But V W has no non-identity bounded p-factor-groups, and so CV W (B2 ) =
V W , that is B2 = B1 , and V W CG (An ). Since this holds for any n N,
V W CG (A). Put
SocDG (A) = M1 Ms ,
where M1 , . . . , Ms are simple DG-submodules. The periodic part of every factor
G/CG (Mj ) is a locally cyclic p -subgroup (see, for example, L.A. Kurdachenko, J.
Otal and I.Ya. Subbotin [157, Corollary 2.2]). Since

H = CG (M1 ) CG (Ms ),

by Remaks theorem
G/H
G/CG (M 1) G/CG (Ms ).
In particular, G/H is a p -group. This means that H/V W is the nite Sylow p-
subgroup of G/V W . Therefore, there exists some subgroup Q such that G/V W =
H/V W Q/V W . Hence, the index |G : Q| is nite, and H Q = V W CG (A).

If G is a group, and H is a normal subgroup of nite index of G, then an
artinian DH-module A is artinian as a DG-module. Conversely, an artinian DG-
module A is likewise an artinian DH-module by Theorem 5.2. This shows that the
study of artinian modules over a group ring can be carried out in subgroups of nite
index of the group in consideration. In particular, if G is a periodic abelian group
of nite rank, by Proposition 12.5, we may assume that CG (SocDG (A)) = CG (A),
and in this case, G/CG (A) is a p -group, where p =char (D/P ). We will focus
now on this case. The results we state below are slight variations on some results
due to B. Hartley and D. McDougall [110]. Before dealing with them, we need to
establish the following property of the injective envelopes.
Chapter 12. Periodic abelian groups 157

Lemma 12.6 (B. Hartley and D. McDougall [110]). Let G be a periodic central-
by-nite group, D a Dedekind domain, and P Spec(D). Suppose that A is an
artinian P -module over DG, and let E be the DG-injective envelope of A. If
char(D/P ) = p > 0, suppose further that p  (G). Then
(1) E is a P -module, and P,1 (E) = P,1 (A).
(2) A is DG-injective if and only if A is D-divisible.
Proof. By Corollary 7.30, P,1 (A) is a direct sum of nitely many simple DG-
submodules. Since E is a maximal essential extension of A, E is also a maximal
essential extension of P,1 (A). In particular, AssD (E) = {P }. According to D.W.
Sharpe and P. Vamos [265, Proposition 2.23], E is a direct sum of injective en-
velopes of direct summands of P,1 (A), and so we may assume that P,1 (A)
is a simple DG-module. If P,1 (A) = P,1 (E), by Corollary 7.30, there exists
a DG-submodule C such that P,1 (E) = P,1 (A) C, a contradiction. Thus
P,1 (E) = P,1 (A).
Let T be the D-periodic part of E. Then AssD (T ) = {P }. By Proposition
7.7, E is D-divisible, and so is T by Lemma 7.18. Moreover, T has a D-complement
in E. Thus T is a DG-monolithic submodule with monolith P,1 (A). By Theorem
7.28, T has a DG-complement in E, and because E is an essential extension of A,
we have a contradiction. Hence, E = T , and, in particular, E is a P -module.
If A is injective, by Proposition 7.7, A is D-divisible. Conversely, if A is D-
divisible, and E is a DG-injective envelope of A, then E is likewise D-divisible.
Since P,1 (A) = P,1 (E), we have E = A, and all the assertions of the result have
been proven. 
Lemma 12.7. Let D be a Dedekind domain, and let P Spec(D). Suppose that
A is a P -module over D such that P,n (A) = A for some n N . If B is a
D-submodule of A such that

A/B = (a + B)D,

where AnnD (a + B) = P n for every , then there exists some D-submodule


C such that A = B C.
Proof. Pick y P \ P 2 . By Proposition 6.13, P = Dy + P n . Put

C= a D

so that A = B + C. If b B C, then there are a nite subset 1 , and some


elements x D, 1 such that

b= a y k x .
1
158 Chapter 12. Periodic abelian groups

Thus

(a + B)y k x = B
1

and it follows that k = n for every 1 . But in this case b = 0. Hence,


B C = 0; that is, A = B C, as required. 
Let G be a periodic abelian group, D a Dedekind domain, P Spec (D), and
F = D/P . We recall that the simple F G-modules over periodic abelian group G
were already described by B. Hartley and D. McDougall (see, for example, [157,
Theorem 2.3]). Suppose that {A | } is a complete set of representatives
of the isomorphism types of simple F G-modules. We consider each A as a DG-
module, and dene E to be the DG-injective envelope of A . We write

P, (E ) = P,n (E ).
nN

Up to an isomorphism, the submodule P,n (E ) is determined by and n. If char


F = p > 0, by Lemma 12.6, A = P,1 (E ), and E = P, (E ). Therefore

P,n+1 (E )/P,n (E )
= P,1 (E ) = A

is a simple DG-module. It follows that the P,n (E ), where n N or n = are


the only submodules of E .
Theorem 12.8 ([110]). Let G be a periodic abelian group of nite special rank, D
a Dedekind domain, P Spec(D), A a P -module over DG, and if char(D/P ) =
p > 0, then p  (G). Then
(1) A module A is artinian if, and only if

A = A1 An ,

where each direct summand is isomorphic to certain P,r (E ), for suitable


, and r N or r = .
(2) If
A = B1 Bk
is another direct decomposition of A of such kind, then n = k, and there are
an automorphism of the module A, and a permutation Sn such that
Bj = Aj , for every 1 j n.
Proof. By Proposition 1.8, every DG-module P,n (E ) is artinian, for n N or
n = , and a direct sum of nitely many of such modules is artinian by Lemma
1.2.
Conversely, suppose that A is an artinian P -module over DG. Suppose that
A cannot be expressed as a direct sum. Then among the submodules of A that
cannot be expressed in such a way, we choose a minimal one. Replacing A by this,
Chapter 12. Periodic abelian groups 159

we may assume that every proper submodule of A admits such a decomposition


but A does not. In particular, A is indecomposable.
Let V be a maximal D-divisible submodule of A. Clearly, V is a DG-sub-
module. By Lemma 12.6, V is DG-injective. Then V has a DG-complement in
A, and then V = A. By Corollary 7.30, P,1 (E ) is a direct sum of nitely many
simple DG-submodules. According to D.W. Sharpe, and P. Vamos [268, Proposi-
tion 2.23], A is a direct sum of the injective envelopes of these simple submodules.
This contradiction shows that V = 1. Pick y P \ P 2 . By [157, Proposition 5.2],
A = Ay. Furthermore, by [157, Proposition 5.3] yields that AP s = Ay s , for every
s N. Note that the chain

A Ay Ay 2 Ay t

breaks after nitely many steps. In other words, there is some m N such that
Ay m = Ay m+1 . According to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157,
Proposition 5.2], Ay m is D-divisible. It follows that Ay m = 1. We may assume
that Ay m1 = 1. In particular, Ay m1 has a non-zero simple DG-submodule
U . There is an isomorphism of U into some A , and this may be extended to a
homomorphism of A into the injective module P, (E ). We have

(A)y m1 = (Ay m1 ) U = A ,

so that
P,n (A) = A, and P,n1 (A) = A.
It follows that A = P,m (E ). Put W = Ker so that

A/W
= Im = P,m (E )

is a direct sum of isomorphism copies D/P m . By Lemma 12.7, W has a D-


complement. Since W = A, W is a direct sum of nitely many monolithic sub-
modules of types P,1 (E ). By Corollary 7.29, W has a DG-complement, a con-
tradiction. Assertion (2) follows from Theorem 1.7. 
Suppose now that D is a Dedekind Z-domain, G is a periodic abelian group
of nite special rank, and A is an artinian DG-module. By Lemma 12.4, A is
D-periodic. By Corollary 6.25

A= AP ,
P

where AP is the P -component of A, and =Ass D (A) is nite. Thus the general
case reduces to the case of a P -module.
If char(D/P ) = p > 0, put

SocDG (A) = M1 Mn ,
160 Chapter 12. Periodic abelian groups

where M1 , . . . , Mn are simple,, and dene H = CG (SocDG (A)). Then

H = CG (M1 ) CG (Mn )

and every G/CG (Mj ) is a locally cyclic p -group (see, for example, L.A. Kur-
dachenko, J. Otal and I.Ya. Subbotin [157, Theorem 2.3]). Suppose that CG (A) =
1. By Proposition 12.5, the Sylow p-subgroup of G is nite or, that is to say,
the Sylow p -subgroup Q of G has nite index. By Theorem 5.2, A is an artinian
DQ-module. In other words, we may reduce this case to the case in which A is an
artinian P -module over DG, where char(D/P )  (G). Note that Theorem 12.8
gives a description of such modules.
Chapter 13

Nearly injective modules

In this chapter we continue our study of complementability of modules over group


rings. Starting from modules over rings with the form F G, where F is a eld,
we move toward modules over rings with the form RG, where R is a certain
commutative ring. In particular, we move toward modules over rings with the
form ZG. If A is a simple ZG-module, in many cases we have pA = 0 for some
prime p (the reader can nd such cases in L.A. Kurdachenko, J. Otal and I.Ya.
Subbotin [157, Chapter 1]). Let E be a ZG-injective envelope of A. Inside E
we consider the Z-injective envelope U of A. In connection with the results of the
previous chapter, the following question naturally arises: In which cases is E equal
to U ? This question was already considered by B. Hartley in [107]. This chapter
is devoted to Hartleys results.
Let R be a commutative ring whose underlying additive group is torsion-
free. Suppose that p is a prime, and A is an R-module such that pA = 0. Since
Z1 = Z, the elements of Z 1 are not zero divisors in R; so the Rinjective
envelope E of A is Z-divisible. Let T be the Z-periodic part of E, which is clearly
a divisible p-subgroup. Actually, T is an R-submodule of E. Consider 1 (T ) as
an R/pR-module. Of course, this submodule is an essential extension of A. Let
U and V be (R/pR)-modules such that U V , and let : U 1 (T ) be an
(R/pR)-homomorphism. Think of U and V as R-modules. Since E is R-injective,
can be extended to an R-homomorphism 1 : V E. Since p(Im 1 ) = 0,
Im 1 1 (T ). Thus, 1 is in fact an (R/pR)-homomorphism. This shows that
1 (T ) is (R/pR)-injective, and hence 1 (T ) contains an (R/pR)-injective envelope
B of A. By Proposition 7.2, there exists some (R/pR)-submodule C such that
1 (T ) = B C. But E is an essential extension of A, so that C = 0. Therefore
1 (T ) becomes an (R/pR)-injective envelope of A. Consequently,

1 (T ) = A if and only if A is (R/pR)-injective.

Let R be a commutative ring whose underlying additive group is torsion-free.


Suppose that p is a prime, and A is an R-module such that pA = 0. Following B.
162 Chapter 13. Nearly injective modules

Hartley, the R-module A is said to be nearly R-injective if the injective envelope


IER (A) of A is a p-group, and 1 (IER (A)) = A. Roughly speaking, this denition
means that A is nearly injective if its R-injective envelope is as small as possible.
In fact, this envelope is a Z-essential extension of A.
Lemma 13.1. Let R be a commutative ring whose underlying additive group is
torsion-free, and suppose that p is a prime, and A is an R-module such that pA =
0. Suppose further that we have a decomposition

A = A1 An

of A as a direct sum of some R-submodules A1 , . . . , An . Then A is nearly injective


if and only if A1 , . . . , An are nearly injective.
Proof. Indeed, according to D.W. Sharpe and P. Vamos [265, Proposition 2.23]
the injective envelope of a nite direct sum of modules is isomorphic to the direct
sum of the injective envelopes of the summands. 
Suppose now that A is a simple nearly injective R-module such that pA = 0,
for a certain prime p. Put E =IER (A). If j N, then the mapping j : a
pj a, a j+1 (E) is an R-homomorphism such that Im j = 1 (E), and Ker j =
j (E). Then j+1 (E)/j (E) = 1 (E), and, in particular, j+1 (E)/j (E) is a
simple R-module. Therefore the family of all R-submodules of E is exactly

{0 , 1 (E), 2 (E), . . . , j (E), . . .}



and E = jN j (E).
We set up the framework in which we are interested. Let p be a prime,
and let A1 , . . . , An a nite set of simple nearly injective R-modules such that
pA1 = = pAn = 0. If B1 IER (A1 ), . . . , Bn IER (An ) are (not necessarily
proper) R-submodules, then an R-submodule B having the form

B = B1 Bn

is said to be an H-module. A useful property of such modules is given by the next


result.
Lemma 13.2 (B. Hartley [107]). Let R be a commutative ring whose underlying
additive group is torsion-free, and let U be an R-module. Suppose that B is an
R-submodule of U such that there is a subgroup C of U satisfying U = B C. If
B is an H-submodule, then there is an R-submodule D of U such that U = B D.
Proof. As indicated above, we suppose

B = B1 Bn ,

where Bj is an R-submodule of IER (Aj ), and Aj is a simple nearly injective R-


module such that pAj = 0 (p a prime). We proceed by induction on n.
Chapter 13. Nearly injective modules 163

Suppose that n = 1. Then either there exists some j N such that B = j (E)
or B = E, where E =IER (A), and A is a simple nearly injective R-module. If
B = E, then B is injective, and so B is a direct summand of U (see Proposition
7.2). Suppose that B = j (E), for some j N. Since U = B C, pj U B = 0.
Let D be an R-submodule of U including pj U , and maximal under B D = 0.
If : U U/D is the canonical mapping, then U is an essential extension of
B, and so of A. Thus A = 1 (U ). Since pj (U ) = 0, B is a direct sum of
cyclic groups of order pj , and therefore B is additively a direct summand of U
(see I. Kaplansky [124, Theorem 7]). Since 1 (U ). = 1 (B), U = B . In other
words, U = B D, as required.
If n > 1, a straightforward induction yields the general case. For

U/B1 = B/B1 (C + B1 )/B1

and so, by induction, we may assume that U = B + V , for some R-submodule V


such that V B = B1 . Since

U = B1 Bn C = B1 W,

where
W = B2 Bn C,
we have V = B1 (W V ). Applying the case n = 1, there exists an R-submodule
D of V such that V = B1 D, and it readily follows that A = B D. 
An abstract description of H-modules is given in the next result.
Theorem 13.3 ([107]). Let R be a commutative ring whose additive group is torsion-
free, and let A be an R-module. Suppose that

A = A1 An ,

where A1 , . . . , An are simple nearly injective R-submodules. If S is an essential


extension of A, then:
(1) S is an artinian R-module;
(2) S is an H-module with indecomposable direct summands;
(3) if S = B1 Bn = C1 Cr are two direct decompositions of S, and
all the submodules B1 , . . . , Bn , C1 , . . . , Cr are indecomposable, then n = r
and there exist a permutation Sn and an automorphism of A such that
Cj = Bj for every 1 j n.
Proof. (1) By the denition of an injective envelope, the identity map on A can
be extended to an embedding of S in E =IER (A). By Lemma 13.1, A is nearly
injective, and we have 1 (E) = 1 (A) = 1 (S). By Lemma 1.2, 1 (A) satises
the minimal condition on R-submodules, and, by Corollary 1.9, so does S.
164 Chapter 13. Nearly injective modules

(2) Suppose the contrary. By (1), S has an R-submodule X, every proper


submodule of which is an H-module. Then X is an essential extension of X A,
which has the same form of A. Therefore, we may replace S by X and assume that
every proper submodule of S is an H-module. In this case, S is clearly indecom-
posable. Let D be a maximal divisible subgroup of S, and suppose that D = 0.
Actually, D is an R-submodule, and we claim D = S. Otherwise, by Theorem
7.16, there exists some subgroup L such that S = D L, and then by Lemma
13.2, D has an R-complement, which is impossible. Thus D = S as claimed, and
so S itself is divisible. Then it follows that the identity map on A can be extended
to an embedding

S IER (A) = IER (A1 ) IER (An ),

and, since the image is divisible, and contains A, this embedding is in fact an
isomorphism. Thus, the assumption D = 0 produces a contradiction, and so S
has no non-zero divisible subgroups. Since S is artinian, the descending chain

S pS pk S

has to break after nitely many steps, and, since the lower term of it is divisible,
it has to be zero. Therefore, there exists some m > 0 such that pm S = 0. We
choose m as small as possible. Actually m 1, and so pm1 S = 0. By (1),
pm1 S has a minimal R-submodule M , and by the denition of injectivity, the
identity mapping on M extends to an R-homomorphism : S IER (M ). Since
pm1 (S) = (pm1 S) = 0, S = m (IER (M )). Since Ker = S, Ker is an
H-module. Furthermore, pm S = 0, and S/ Ker is a direct sum of cyclic groups
of order pm precisely. Therefore Ker is additively a direct summand of S (see
I. Kaplansky [124, Theorem 7]). By Lemma 13.2, Ker has an R-complement. It
follows that Ker = 0, and so S =R m (IER (M )), a nal contradiction.
(3) Since S is artinian, the submodules B1 , . . . , Bn , C1 , . . . , Cr are monolithic,
and it suces to apply Theorem 1.7. 
Theorem 13.3 and Lemma 13.2 together give the next result.
Theorem 13.4 ([107]). Let R be a commutative ring whose additive group is torsion-
free, and suppose that
A = A1 An
is a nearly injective R-module such that pA = 0 for some prime p, and A1 , . . . , An
are simple. Let U be an R-module including an R-submodule S which is an essential
extension of A. If U = S C for some subgroup C, then U = S D for some
R-submodule D.
Suppose that G is a locally nite p -group, and let A be a simple Fp G-module,
which we view as a ZG-module in the obvious way. We now consider the following
question: when is A nearly injective?
Chapter 13. Nearly injective modules 165

Lemma 13.5 ([107]). Suppose that G is a nite group of order n, and let A be a
ZG-module whose additive group is divisible, and has no elements of order dividing
n. Then A is ZG-injective.
Proof. Denote by E the ZG-injective envelope of A. By Theorem 7.16, there exists
some subgroup L such that E = A L. By Corollary 5.11, there exists a ZG-
submodule U such that E = A U . Since E is an essential extension of A, E = A,
and, in particular, A is ZG-injective. 
Theorem 13.6 ([107]). Let G be a locally nite p -group, and let A be a simple
Fp G-module. If G/CG (A) is abelian-by-nite, then A is nearly ZG-injective.
Proof. Put E = IEZG(A), and let T be the Z-periodic part of E. By Lemma 9.5
and Lemma 9.4, A is Fp G-injective; the arguments given at the beginning of this
section yield that 1 (T ) = A. We are going to show that T = E by proving
that T itself is injective. Let be a ZG-endomorphism of A so that can be
extended to E since the latter is injective. Since T is a fully invariant subgroup
of E, this extension determines an endomorphism of T . Thus, if D = EndZG (A),
and Y = EndZG (T ), the above restriction determines an epimorphism from Y to
D. By Lemma 9.5, A is nite dimensional over D. In particular, A satises the
minimal condition on Y -submodules. By Theorem 1.7 so does T .
Let be an arbitrary ZG-homomorphism from a right ideal L of ZG into T .
It will suce to show that can be extended to the whole of ZG. In this case, by
Baers criterion (Theorem 7.1), T is ZG-injective; consequently, there must exist
an element v T such that x = vx for all x L. Suppose no such v exists. Given
a nite subgroup H of G, we put
S(H) = {v T | x = vx for all x L ZH}.
By Lemma 13.5, the restriction of on L ZH can be extended to ZH, and so
S(H) = . Let M be the family of all nite subgroups of G. Then our assumption
implies that 
S(H) = .
HM
Clearly, the annihilator
AnnT (L ZH) = {x T | x(L ZH) = 0}
is a Y -submodule of T ; since T satises Min-Y , there exists a nite subgroup H of
G such that AnnT (L ZH) is minimal among all these annihilators. Notice that
S(H) = v+AnnT (L ZH), for every v S(H). Since

S(H) = ,
HM

there exists a nite subgroup K of H such that S(K) S(H) S(H). Let F =
K, H. Then F is nite, and
= S(F ) S(K) S(H) S(H).
166 Chapter 13. Nearly injective modules

Let v S(F ). Then v+AnnT (L ZF ) v+AnnT (L ZH), whence AnnT (L


ZF ) < AnnT (L ZH), a contradiction. 
Using these arguments, Theorem 13.4, and the straightforward fact that a
direct sum of nitely many nearly injective modules annihilated by the same prime
is nearly injective (see Lemma 13.1), we obtain
Theorem 13.7 ([107]). Let G be a periodic abelian-by-nite p -group, and let U be a
ZG-module that includes a submodule S. Suppose that S is an essential extension
of a submodule A, which is a direct sum of nitely many simple submodules, and
satises pA = 0. If U = S C for some subgroup C, then there exists a ZG-
submodule D such that U = S D.
For the case of a ring Z, this theorem is a generalization of Corollary 7.29.
On the other hand, it seems natural to ask whether other restrictions on the
size of the ZG-injective envelope E of a simple Fp G-module A might be obtained
for suitable groups G. For example, when can we be sure that E is Z-periodic? This
seems to be a rather dicult question. In the paper [107], B. Hartley constructed
an example of a 2-group G whose commutator subgroup is nite, and a simple
Fp G-module A, where p is an odd prime, such that IEZG (A) is not Z-periodic.
This construction is far from the scope of this book, and therefore we will not give
more details here about it.
Our previous results, and other results that we do not mention here, show
that the theory of artinian modules over periodic abelian-by-nite groups is very
fruitful. The examples of uncountable artinian groups constructed by B. Hartley
[105] show that it is unlikely to expect signicant progress in the theory of artinian
DG-modules over such groups.
Chapter 14

Artinian modules over abelian groups


of nite section rank

The description of artinian modules over non-periodic groups is quite dierent from
that of the periodic case. If |G/Gp | p2 , then regarding representation theory, the
problem of description of A is wild (S.A. Kruglyak [140]). We can enlighten this
rough idea with the following analogy. The problem of the description of abelian
torsion-free groups of nite rank is also wild (A.V. Yakovlev [288]), but the
Z-injective envelopes of these modules are known to be direct sums of copies of
the additive group of the rational numbers. Therefore, the following two questions
arise: to describe the injective envelopes of artinian modules, and to describe
artinian modules over a group G such that |G/Gp | p. Both problems were
solved in the paper by L.A. Kurdachenko [150]. The results of this article formed
the base for this chapter. The paper [150] considered the artinian modules over
abelian groups of nite special rank. Here we will consider a generalized situation.
Let R be a ring, and suppose that A is an artinian R-module. Then
SocR (A) = M1 Mn ,
where M1 , . . . , Mn are minimal R-submodules of A. If 1 j n, let Uj be an
R-submodule of A maximal under
Mj Uj = 1 , and M1 Mj1 Mj+1 Mn Uj .
Then A/Uj is a monolithic R-module with R-monolith (Mj + Uj )/Uj . Put U =
U1 Un . Then U SocR (A) = 0, and so U = 0. By Remaks theorem,
A
A/U1 A/Un
and, by Lemma 1.2, the latter is an artinian R-module. This shows that the case
of monolithic artinian R-modules is fundamental for our interests.
Let D be a Dedekind Z-domain, and let G be an abelian group of nite
0-rank. Suppose that A is a monolithic artinian DG-module with DG-monolith
168 Chapter 14. Abelian groups of nite section rank

M . Then AnnD (M ) = P Spec(D) (see L.A. Kurdachenko, J. Otal and I.Ya.


Subbotin [157, Theorem 1.15]), and D/P is a locally nite eld. Put p =char D/P ,
and H = CG (M ) so that G/H is a periodic locally cyclic p -group (see L.A.
Kurdachenko, J. Otal and I.Ya. Subbotin [157, Theorem 2.3]). Suppose that G
has nite p-rank, and choose a nitely generated torsion-free subgroup L of H
such that G/L is periodic. Let S/L be the Sylow p-subgroup of G/L. Thus S/L is
a Chernikov subgroup, and, moreover, S H. Let U/L be the Sylow p -subgroup
of G/L. Then H U , and

G/L = U/L V /L W/L,

where V /L is a divisible Chernikov p-subgroup, and W/L is a nite p-subgroup.


Put S = U V . As a consequence of the Theorem of Wilson (Chapter 5), the study
of artinian modules over group rings can be reduced to subgroups of nite index,
so that A is an artinian DS-module. In other words, we can further assume that
H = CG (M ) has a nitely generated torsion-free subgroup L such that H/L has
no subgroups of index p.
Further reductions can be derived from the following result.
Lemma 14.1 (L.A. Kurdachenko [144]). Let G be a hypercentral group, and let
H be a normal subgroup of G such that G/H has no subgroups of index p, (p a
prime). Let T be a G-invariant p-subgroup of H. Then every H-invariant subgroup
of T is G-invariant.
Proof. Let L be an H-invariant subgroup of T . We may assume that L = 1.
Since H is hypercentral, L (H) = 1. Note that L (H) T (H) = U .
Clearly, U is a G-invariant subgroup, and CG (U ) H. In particular, G/CG (U )
contains no subgroups of index p. By Lemma 3.10, U (G), and, in particular,
L (H) (G). It suces to apply transnite induction and similar arguments
to obtain that L is G-invariant, as required. 
Lemma 14.2. Let D be a Dedekind Z-domain, and let G be an abelian group of
nite section rank. Let A be a monolithic artinian DG-module with monolith M .
Suppose that H = CG (M ) includes a nitely generated torsion-free subgroup L
such that H/L contains no subgroups of index p, where p = char(D/AnnD (M )).
Then G has a subgroup K L such that K/L is a periodic p -group of special
rank 1, and every DK-submodule of A is likewise a DG-submodule. In particular,
the DK-module A is artinian.
Proof. Note that G/H is a periodic p -group of special rank 1 (see L.A. Kur-
dachenko, J. Otal and I.Ya. Subbotin [157, Theorem 2.3]). Therefore, we have a
decomposition
G/L = S/L K/L Q/L,
where S/L is the Sylow p-subgroup of G/L, Q H, K/L has nite special rank
1, (K/L) = (G/H), and KH = G. Let P =AnnD (M ). By Lemma 12.4, A is a
Chapter 14. Abelian groups of nite section rank 169

D-periodic module, and therefore a P -module. Also, the additive group of A is a


p-group. Since H/L contains no subgroups of index p, S/L is a divisible Chernikov
group. By Corollary 12.2, the DH-module A is DH-hypercentral. In particular,
the natural semidirect product A  H is a hypercentral group. Let B be a DK-
submodule of A. Since SQ/L contains no subgroups of index p, every L-invariant
subgroup of A is SQ-invariant. In particular, B is a D(SQ)-submodule of A. Since
KSQ = G, B is a DG-submodule. 
This allows us to assume that G has a nitely generated torsion-free subgroup
L CG (M ) such that G/L is a p -group of nite special rank. In particular, we
may reduce the study to the case in which G has nite special rank.
Lemma 14.3 (L.A. Kurdachenko [150]). Let D be a Dedekind Z-domain, and let
G be an abelian group of nite section rank. If A is an artinian DG-module, B a
nitely generated DG-submodule of A, and H = CG (SocDG (A)), then there exists
some n N such that the nth term of the DH-central series of A contains B. In
particular, B is DH-nilpotent.
Proof. By Lemma 12.4, the additive group of A is periodic. Since A is artinian,
AssZ (A) = (A) is nite. There exists a nitely generated subgroup X of H
such that H/X is periodic, and p-divisible for every p (A). Suppose that
b1 , . . . , bk B satisfy
B = b1 DG + + bk DG.
Put
B1 = b1 Z + + bk Z = b1 , . . . , bk  .
By Theorem 12.3, A is DH-hypercentral, and so is B1 . In particular, B1 is ZX-
hypercentral. Then the natural semidirect product B1  X is a nitely generated
hypercentral group; hence, it is nilpotent. Its periodic part is nite, and contains
1 -ZX-hypercenter of A for some n1 N.
B1 . It follows that B1 is included in the nth
Then the nth -DX-hypercenter of A contains B2 = B1 D. Since each term of the
upper DX-central series of A is a DG-submodule, it follows that there exists some
n2 N such that the nth 2 -DX-hypercenter of A contains B = B2 DG.
To nish we are going to show that the upper DX-central series, and the
upper DH-central series of A coincide. It suces to prove that A1 = DX (A)
DH (A) or that DH (A) contains every nitely generated DG-submodule C of
A1 . Suppose that the latter is false, that is, C1 = DH (C) = C. Put C2 /C1 =
DH (C/C1 ). Then C2 = C1 by our assumption. Every element of H/CH (C2 ) acts
trivially on the factors of series
0 = C0 C1 C2 .
According to O.H. Kegel and B.A.F. Wehrfritz [134, 1.C.3], H/CH (C2 ) can be
embedded in HomZ (C2 /C1 , C1 ). Since C is nitely generated, the orders of the
elements of C are bounded. Since C A1 the choice of X, implies that H/CH (C2 )
is a periodic p-divisible group for every p (A). This contradiction shows that
DH (A) = DH (A), as required. 
170 Chapter 14. Abelian groups of nite section rank

Corollary 14.4 ([150]). If the conditions of Lemma 14.3 hold, then the DH-hyper-
central length of A is at most .
Corollary 14.5. If the conditions of Lemma 14.3 hold, then the DG-socular height
of A is at most .
Proof. Pick b B, and put B = bDG. By Lemma 14.3, B is DH-nilpotent, and
there exists some n N such that B = P,n (B). We claim that B has a nite
composition DG-series, and proceed by induction on n. Suppose that n = 1. Let

0 = B0 B1 Bc = B

be the upper DH-central series of B. If 1 j c, each term Bj is clearly a


DG-submodule of B. Since CG (Bj+1 /Bj ) H, G/CG (Bj+1 /Bj ) is a periodic
p -group. By Corollary 7.30, Bj+1 /Bj is a direct sum of nitely many simple DG-
submodules, and so Bj+1 /Bj has a nite DG-composition series. Since this holds
for every j, B has a nite DG-composition series. Suppose now that n > 1. Put
E = BP . Then E is DH-nilpotent, and E = P,n1 (E). By induction, E has
a nite DG-composition series. This is also true for B/E, hence B has a nite
DG-composition series, as claimed. 
Lemma 14.6 ([150]). Let R be a commutative ring, G an abelian group, and A
an RG-module whose additive group is a p-group, for some prime p. If H is a
subgroup of G such that A is RH-hypercentral, and B is an RG-submodule of A,
then H/CH (B) has no non-identity p-divisible subgroups. In particular, if G/H is
a periodic p -group, and G/CG (B) is periodic, then G/CG (B) has no non-identity
divisible p-subgroups.
Proof. Let P/CH (B) be a p-divisible subgroup of H/CH (B). For every n N, put
Bn = P,n (B), C1 = RP (Bn ), and C2 /C1 = RP (Bn /C1 ). Then P/CP (C2 ) acts
trivially on the factors of the series

0 = C0 C1 C2 .

According to O.H. Kegel and B.A.F. Wehrfritz [134, 1.C.3], P/CP (C2 ) can be
embedded in HomZ (C2 /C1 , C1 ). Since the order of each element of C2 divides pn ,
the order of each element of HomZ (C2 /C1 , C1 ) divides pn . But P/CP (C2 ) has no
bounded p-factor-groups so that P = CP (C2 ), and hence C2 = C1 . It follows that
P CG (Bn ). Since this holds for all n N, P CG (B). 
Before we continue further considerations, we need to make the following
remark. Let R be a ring, A be an R-module. Put L = AnnR (A). We think of A as
a K-module, where K = R/L. Then we may look at the consideration of injective
envelopes from two points of view. First we consider a K-injective envelope E of
A, and think of E as an R-module such that L = AnnR (E). This seems to be very
clever, because L has no inuence on the structure of A. Alternatively, we could
consider a full injective R-envelope I. In this case AnnR (I) could not include L
Chapter 14. Abelian groups of nite section rank 171

itself, and therefore the structure of I depends on some elements of the ideal L.
This case was explicitly considered above, so that we will look at the rst one.
Actually, this was the way chosen by B. Hartley and D. McDougall (see Chapter
12). Given a Chernikov abelian group G and an artinian ZG-module A, Hartley
and McDougall were able to reduce their study to the case in which the additive
group of A is a p-group, where p is a prime. Then G/CG (A) has a nite Sylow
p-subgroup, and, by the theorem of Wilson, it can be supposed that G/CG (A) is
a p -group; this supposition enables us to consider injective envelopes. Below we
will follow the mentioned approach of Hartley and McDougall.
Theorem 14.7. Let D be a Dedekind Z-domain, and let G be an abelian group of
nite section rank. If A is an artinian DG-module, I = AnnDG (A), Q = DG/I . If
E is a Q -injective envelope of A, then E is an artinian DG-module.
Proof. Suppose that SocDG (A) = M1 Mt , where M1 , . . . , Mt are simple
DG-submodules. According to D.W. Sharpe and P. Vamos [265, Proposition 2.23],
E = E1 Et ,
where each Ej is the DG-injective envelope of Mj . Hence, without loss of gen-
erality, it can be assumed that SocDG (A) = M1 is a simple DG-submodule, i.e.
A is a monolithic module with the monolith M = M1 . Put L = SocDG (M ), and
H = CG (M ). In particular DG/L is a eld. By Lemma 12.4, A is D-periodic and
then A is a P -module, where P = AnnD (M ) Spec(D). Since A is an artinian
DG-module, the upper socular series of A reaches A and every factor of this series
can be decomposed into a direct sum of nitely many simple modules. This and
Corollary 14.5 shows that A has an ascending series of submodules
0 = A0 A1 = M A2 An An+1 A = A,
every factor of which is a simple DG-module. Suppose rst that A has a nite
composition series. Then A is nitely generated, say A = a1 DG + + am DG.
Since each summand aj DG = DG/AnnDG (aj ), A has a nite composition series,
and, in particular, A is noetherian. Since
AnnDG (A) = AnnDG (a1 ) AnnDG (am ),
Remaks theorem gives the embedding
DG/AnnDG (A)
DG/AnnDG (a1 ) DG/AnnDG (am ).
By Lemma 1.1, the factor-ring Q = DG/AnnDG (A) is noetherian. By a result due
to E. Matlis [183], IEQ (A) is likewise an artinian DG-module.
We suppose now that the ascending DG-composition series of A are innite.
By Lemma 12.1, L annihilates every factor of any ascending composition series of
A.
It follows that Ln AnnDG (An ), for each n N. It turns out that AnnDG (A) =
n
nN L . Let
R = lim {DG/Ln | n N}.
172 Chapter 14. Abelian groups of nite section rank

By Remaks theorem, Q = DG/AnnDG(A)


R. According to R. Sharp [264,
Theorem 3.2], R is a noetherian local ring, and A is an artinian R-module. Further,
IER (A) is again an artinian R-module (E. Matlis [183]). We claim that IER (A)
is in fact a Q-injective module. For, suppose that V is a Q-module such that
IER (A) V . Put V1 = L Q R. Then V1 contains IER (A) Q R =IER (A). Since
IER (A) is R-injective, there exists an R-submodule U such that V1 = IER (A) U .
But IER (A) V V1 , and so V = IER (A) (U L), and hence U L is
a Q-submodule. Thus IER (A) is a direct summand of every Q-module including
IER (A). By Proposition 7.2, this means that IER (A) is an injective Q-module, and
our claim has been proven. It follows that E = IEQ (A) IER (A). Since IER (A) is
an R-essential extension of M , IER (A) is an artinian monolithic R-module with
the monolith M . By Lemma 12.1, L = AnnR (M ) annihilates every factor of an
ascending composition series of IER (A). Obviously,

L =
lim {L/Ln | n N},

and, therefore, L/AnnDG (A) L , so that P annihilates every factor of an as-


cending composition series of IER (A). Given x H = CG (M ), x 1 L L .
In particular, IER (A) is an F y1 -hypercentral module. Let p = char F . Then the
additive group of IER (A) is a p-group. Since G has nite section rank, H has a
nitely generated subgroup X such that H/X is a periodic p-divisible group.
Let
X = x1  xn 

for some n N. If m N, we put Bm = P,m (E) so that E = mN Bm . Look at
B1 as an F G-module, where F = D/P . We may think of B1 as an F y1 -module,
where y1 is an innite cyclic group, and y1 acts on B1 by ay1 = ax1 , a B1 . Since
IER (A) is an (x1 1)-module, B1 = mN S,m (B1 ), where S = (y1 1)F y1 .
Put
C1 = B1 , and C2 = S,1 (B1 ) = B1 CE (x1 , x2 ).
In this case we think of C2 as an F y2 -module, where y2  is an innite cyclic
group, and y2 acts on C2 by ay2 = ax2 , a C2 . Since IER (A) is an (x2 1)-
module, C2 = mN U,m (C2 ), where U = (y2 1)F y2 . Proceeding in this way,
we put

C3 = B1 CE (x1 , x2 , x3 ), . . . , Cn1 = B1 CE (x1 , . . . , xn1 )

and
Cn = B1 CE (x1 , . . . , xn ).
Since CG (Cn ) X, and G/H is a p -group (see L.A. Kurdachenko, J. Otal and
I.Ya. Subbotin [157, Theorem 2.3]), G/CG (Cn ) is a periodic p -group by Lemma
14.6. By Corollary 7.30, Cn is a semisimple F G-module, and so, by Corollary 4.3,
there exists some F G-submodule W such that Cn = M W . Since E is an essential
extension of M , W = 0, and so Cn = M , ando so Cn1 = mN V,m (Cn1 ),
Chapter 14. Abelian groups of nite section rank 173

where V = (yn 1)F yn . By Proposition 1.8, Cn1 is an artinian F G-module.


Iterating the argument, we obtain that B1 is an artinian F G-module. Since E =

mN Bm , it suces to apply Proposition 1.8 to obtain that E is an artinian
DG-module in this case. 
Lemma 14.8. Let F be a locally nite eld of characteristic p, and let G be an
abelian group of nite special rank. Suppose that A is a monolithic artinian F G-
module with monolith M such that CG (A) = 1, and put H = CG (M ). Suppose
further that H has an innite cyclic subgroup g such that G/ g is a periodic
p -group. If E is the F G-injective envelope of A, then E has an ascending chain
of submodules

M = M0 M1 Mk Mk = E
kN

such that
Mk+1 /Mk
= M and Mk+1 (g 1) = Mk
for every k N.
Proof. We claim that E = E(g 1). For, suppose the contrary, that is, E is not
(g1)-divisible. By Theorem 14.7, E is an artinian F G-module, and so there exists
some m N such that B = E(g 1)m = E(g 1)m+1 . In other words, B is (g 1)-
divisible. Assume that B = 0. Since E is artinian, E(g 1)m1/B has a non-zero
simple F G-submodule, lets say (aF G+B)/B. Since a(g 1) B = B(g 1), there
exists some b B such that a(g1) = b(g1). It follows that (ab)(g1) = 0, that
is, a b CE (g). Hence G/CG ((a b)F G) is a periodic p -group. By Corollary
7.30, (a b)F G is a semisimple F G-module. Since E is a monolithic F G-module,
M (a b)F G. By the choice of a we have a b  B, and so a b  M . By
Corollary 4.3, there exists an F G-submodule W such that (a b)F G = M W ,
and, since E is an essential extension of M , W = 0, a contradiction. Hence
B = E(g 1)m = 0. In this case g t CG (A) = 1, where t = pm1 , and
we come to another contradiction, proving our claim. It follows that the mapping
: e e(g 1), e E, is an epimorphism. Put

M1 = M 1 , Mn+1 = Mn 1 , and C = Mn , n N.
nN

By construction, C = C(g 1). We must have C = E,otherwise, proceeding as


above, we would nd another contradiction. Hence E = nN Mn , as required. 
Lemma 14.9. Let F be a locally nite eld of characteristic p, and let G be an
abelian group of nite special rank. Suppose that A is a monolithic artinian F G-
module with monolith M such that CG (A) = 1, and put H = CG (M ). Suppose
further that H has a torsion-free nitely generated subgroup

L = DrsS < gs >


174 Chapter 14. Abelian groups of nite section rank

such that G/L is a periodic p -group. Let E be the injective envelope of A. If S1


is a non-empty subset of S, then CE (DrsS\S1 gs ) is (gt 1)-divisible for every
element t S1 .
Proof. We proceed by induction on s1 = |S1 |. Suppose rst that s1 = 1, that is,
S1 = {t} , for some t S. Put

C = CE (DriS\S1 gi ),

and suppose that C is not (gt 1)-divisible. By Theorem 14.7, E is an artinian F G-


module, and so there exists some m N such that B = C(gt 1)m = C(gt 1)m+1 .
As in the proof of Lemma 14.8, B = 0. Let

Q = G/(DriS\S1 gi ),

and think of C as an F Q-module. Let E1 be the F Q-injective envelope of C. By


Theorem 14.7, E1 is artinian, and, by Lemma 14.8, E1 = E1 (gt 1). We now think
of E1 as an F G-module such that DriS\S1 gi  CG (E1 ). Suppose we are given
a monomorphism : C E1 , and put U = E E1 and V = {a a | a C}.
Obviously, V is an F G-submodule of U , and V E = 0 = V E1 . Therefore,

(E + V )/V
= E, and (E1 + V )/V
= E1 ,

i.e. E and E1 can be embedded in U/V . By the denition of V , (C + V )/V


(E1 + V )/V . Put
C1 /V = CU/V (DriS\S1 gi ).
Since (E1 + V )/V C1 /V ,

C1 /V = ((E1 + V )/V ) + ((E + V )/V ) (C1 /V )).

If a E and a + V ((E + V )/V ) (C1 /V ), then a(gi 1) V , for each i S.


Since E is an F G-submodule of U , a(gr 1) V E = 0, for every r S. Thus

((E + V )/V ) (C1 /B) (C + V )/V (E1 + V )/V

and hence
C1 /V = (E1 + V )/V
= E1 /(E1 V )
= E1 .
We have already noted that E1 = E1 (gt 1). Let W/V be an F G-submodule of
U/V , and pick some V = w + V W/V . Then w + V = (a1 + V ) + (a2 + V ), for
suitable a1 E and a2 E1 . If a2 = 0, then w + V (E + V )/V . Suppose that
a2 = 0. Since E1 is an essential extension of C, a2 F G C = 0, and so there
exists some x F G such that 0 = a2 x C. Then there exists some a3 C such
that a2 x = a3 . But a3 a3 V whence a3 + V = a3 + V = a2 x + V . Thus,

(w + V )x = (a1 x + V ) + (a2 x + V ) = a1 x + a3 + V (E + V )/V.


Chapter 14. Abelian groups of nite section rank 175

Thus, (W/V ) ((E + V )/V ) = 0, and hence U/V is an essential extension of
(E + V )/V
= E. Since E is F G-injective, every essential extension of E is equal to
E. Then (E +V )/V = U/V and E = U/V . It follows that C is (gt 1)-divisible, as
required. The general result can be deduced by applying induction, and proceeding
in the same way. 
We are now in a position to describe the injective envelopes.
Lemma 14.10. Let F be a locally nite eld of characteristic p, and let G be an
abelian group of nite special rank. Suppose that A is a monolithic artinian F G-
module with monolith M such that CG (A) = 1, and put H = CG (M ). Suppose
further that H has a torsion-free nitely generated subgroup L = g1  gn 
such that G/L is a periodic p -group. If E is the F G-injective envelope of A, then
E has a nite series of submodules

E = E0 E1 En1 En = M

satisfying the following properties:


(1) E1 = CE (g1 ), E2 = CE (g1 , g2 ), . . . , En1 = CE (g1 , . . . , gn1 ).
(2) Any term Ej has an ascending series of submodules

Ej+1 = Mj+1,0 Mj+1,1 Mj+1,k Mj+1,k = Ej
kN

such that

Mj+1,k+1 /Mj+1,k
= Ej+1 , and Mj+1,k+1 (gj+1 1) = Mj+1,k

for every k N.
Proof. By Lemma 14.9, En1 = En1 (xn 1). Thus the mapping 1 : a
a(xn 1), a En1 is an epimorphism. Put

Mn,0 = M, Mn,1 = Mn,0 1 1


1 , Mn,k+1 = Mn,k 1 , k N.

Clearly,
Ker 1 = En1 AnnE (xn 1) = CE (x1 , . . . , xn ).
Therefore G/CG (Ker 1 ) is a periodic p -group. By Corollary 7.30, there exists an
F G-submodule U such that Ker 1 = M U . Since E is an essential extension
of A, U = 0, and so Ker 1 = M . Thus Mn,1 (xn 1) = Mn,0 = M , and
Mn,1 /Mn,0 = M . Since En1 / Ker 1 = En1 , we proceed in the same way to
get M
n,k+1 (xn 1) = Mn,k , and Mn,k+1 /Mn,k = M , for every k N. Dene
V = kN Mn,k so that V (xn 1) = V . Suppose that V = En1 . Proceeding
in the same way as in the proof of Lemma 14.8, we see that there exists some
F G-submodule W such that En1 = V W , and, since M V , we conclude that
W = 0, and so En1 = kN Mn,k . Consider the submodule En2 . By Lemma
176 Chapter 14. Abelian groups of nite section rank

14.9, En2 = En2 (xn1 1). As above, the mapping 2 : a a(xn1 1), a
En2 is an epimorphism. Note that

Ker 2 = En2 CE (xn1 ) = CE (x1 , . . . , xn2 ) CE (xn1 )


= CE (x1 , . . . , xn1 ) = En1 .

Put
Mn1,0 = En1 , Mn1,k+1 = Mn1,k 1
2 , k N.

Dene U = kN Mn1,k ,, and suppose that U = En2 . By Theorem 14.7, E is an
artinian F G-module, and so En2 /U has a non-zero simple F G-submodule, lets
say (aF G + U )/U . By Corollary 12.2, a(xn1 1) U , and, by construction, U =
U (xn1 1). Therefore, there exists some u U such that a(xn1 1) = u(xn1
1). It follows that (a u)(xn1 1) = 0, that is, a u Ker 2 = En1 U .
Thus a U , which contradicts the choice of a. Then U = En2 . By construction,
we also have that Mn1,k+1 (xn1 1) = Mn1,k , and Mn1,k+1 /Mn1,k = En1 ,
for every k N. Applying similar arguments to E1 , . . . , En2 , we construct the
submodules Mj,k , 1 j n 3, and k N in an analogous way. 
Theorem 14.11. Let D be a Dedekind Z-domain, and let G be an abelian group
of nite section rank. Suppose that A is a monolithic artinian DG-module with
monolith M , and let P = AnnD (M ). If E is a DG-injective envelope of A, then:
(1) E is an artinian P -module over DG.
(2) P,1 (E) is the F G-injective envelope of P,1 (A), where F = D/P .
(3) E is the D-divisible envelope of P,1 (E).
Proof. (1) follows from the proof of Lemma 14.8.
(2) We claim that P,1 (E) is F G-injective. To show this, put L =AnnDG (A),
T =AnnF G (A), R = DG/L, and Q = F G/T = DG/S, where S is the preimage
of T in DG under the canonical map. Clearly S L. Let B, and C be a pair of
Q-modules such that B C, and let : B P,1 (E) be a Q-homomorphism.
We may think of B, and C as DG-modules such that AnnDG (B) AnnDG (C) S.
Then can be extended to a DG-homomorphism : C E. Since (Im )P =
0, Im P,1 (E), and so is in fact an F G-homomorphism, which shows
our claim. Hence P,1 (E) has an F G-injective envelope U of P,1 (A). Then there
exists some F G-submodule V such that P,1 (E) = U V , and it follows that
U = 0 since E is an essential extension of A. Therefore, P,1 (E) is an F G-
injective envelope of P,1 (A).
(3) By Proposition 7.5, E is D-divisible so that E is a D-divisible envelope
of P,1 (E). 
Corollary 14.12. Let D be a Dedekind Z-domain, and let G be an abelian group
of nite section rank. Suppose that A is a monolithic artinian DG-module with
monolith M , and put P = AnnD (M ). Then A is DG-injective if and only if A is
D-divisible, and P,1 (A) is F G-injective, where F = D/P .
Chapter 14. Abelian groups of nite section rank 177

Proof. By Theorem 14.11, the conditions are necessary. Conversely, assume that
A is D-divisible, and P,1 (A) is F G-injective. Let E be the DG-injective envelope
of A. Since P,1 (A) is F G-injective, there exists some F G-submodule B such that
P,1 (E) = P,1 (E) B. It follows that B = 0 since E is an essential extension
of A. Hence P,1 (E) = P,1 (A). Since E and A are D-divisible, E = A, and we
are done. 
The description of the injective envelope of a monolithic DG-module can be
obtained in a similar way to that of the case of a eld. We simply state here that
result without a proof.
Theorem 14.13. Let D be a Dedekind Z-domain, and let G be an abelian group
of nite special rank. Suppose that A is a monolithic artinian DG-module with
monolith M such that CG (A) = 1, and put H = CG (M ). Let P = AnnD (A),
and let p be the characteristic of the eld D/P . Suppose further that H has a
torsion-free nitely generated subgroup L = g1  gn  such that G/L is a
periodic p -group. If E is the DG-injective envelope of A, then E has a nite series
of submodules
E = E0 E1 En1 En = M
satisfying the following properties:
(1) E1 = CE (g1 ), E2 = CE (g1 , g2 ), . . . , En1 = CE (g1 , . . . , gn1 ), and En is
the DQ-injective envelope of SocDG (A), where Q = G/CG (SocDG (A)).
(2) Any term Ej has an ascending series of submodules

Ej+1 = Mj+1,0 Mj+1,1 Mj+1,k Mj+1,k = Ej
kN

such that

Mj+1,k+1 /Mj+1,k
= Ej+1 , and Mj+1,k+1 (gj+1 1) = Mj+1,k

for every k N.
Let A be an artinian P -module over DG, where P Spec(D), and suppose
that p =char (D/P ). As we mentioned above, the problem of the description of
A can be wild if |G/Gp | p2 . If |G/Gp | p, the corresponding description is
contained in L.A. Kurdachenko [150]. Note that some of the results of this chapter
can be translated to artinian modules over a ring with the form F G, where char
F = 0 (see L.A. Kurdachenko [148]).
Chapter 15

The injective envelopes of simple


modules over group rings

As we have already mentioned, one of the most important aspects of the study of
artinian modules is to obtain information about their injective envelopes. In par-
ticular, the following question is very important: when is an injective envelope of
an artinian module likewise artinian? The presence of this property would allow
us to obtain the desired description in the best possible way. In this chapter, we
will obtain some classes of group rings RG having the property that the injective
envelope of every artinian RG-module also is artinian. Actually, the injective enve-
lope of an artinian module is equal to the injective envelope of its socle. Moreover,
injective envelopes work well with nite direct sums, that is,

IER (M1 Mt ) = IER (M1 ) IER (Mt ).

These remarks allow us to reduce our investigation to the consideration of the


injective envelope of a simple module. For commutative rings an important result
was obtained by E. Matlis [183], who proved that if R is a noetherian commutative
ring, the injective envelope of each artinian R-module is likewise artinian. The
group rings that appear to be closer to noetherian commutative rings are group
rings of the form RG, in which R is noetherian, and G is a nitely generated
nilpotent group. In fact, R.L. Snider [269] proved that if G is a nitely generated
nilpotent group, then the injective ZG-envelope of a simple module is artinian.
In this chapter, we consider a more general situation: the case of a group ring
having the form DG, where G is a nitely generated nilpotent group, and D is
a Dedekind domain. The reader will also be able to nd proofs that are dierent
from Sniders.
Lemma 15.1 (P. Hall [96]). Let G be a nitely generated nilpotent group. If (G)
is nite, then G is nite.
Proof. Let
1 = C0 C1 Cn = G
180 Chapter 15. The injective envelopes of simple modules

be the upper central series of G, and suppose that |C1 | = m. We show by induction
on n that the subgroups C2 , . . . , Cn are all nite. Suppose that G = g1 , . . . , gk .
Given g G, the mapping g : x [x, g], x C2 is a homomorphism such that
Ker g = CG (g) C2 , and Im g = [G, g] (G). Therefore, |C2 / Ker g | m.
Since
(CG (g1 ) C2 ) (CG (gk ) C2 ) = (G),
|C2 /(G)| mk by Remaks theorem. Thus C2 is nite.
Suppose that n > 2, and that we have already proved that Cn1 is nite.
Since (G/Cn2 ) = Cn1 /Cn2 , repeating the arguments given in the above para-
graph, we obtain that G/Cn2 is nite. Hence G is nite, as required. 
Proposition 15.2 ([96]). Let A be a simple F G-module, where F is a locally nite
eld, and G is a nitely generated nilpotent group. Then dimF (A) is nite.
Proof. By L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.21],
(G/CG (A)) is periodic. Since a nitely generated nilpotent group satises Max,
it follows that (G/CG (A)) is nite. By Lemma 15.1, G/CG (A) is nite. Therefore
dimF (A) is nite. 
As the following result shows, in the case of nitely generated abelian-by-
nite groups the restrictions on the eld could be removed.
Proposition 15.3 ([96]). Let A be a simple F G-module, where F is an arbitrary
eld, and G is a nitely generated abelian-by-nite group. Then dimF (A) is nite.
Proof. We suppose rst that G is abelian. We have A = aF G, for each 0 = a A,
so that A
= F G/L, where L =AnnF G (a). Since A is simple, L is a maximal ideal
of F G. Put G = g1 , . . . , gn . We look at F G as a homomorphic image of the
polynomial ring F [X1 , . . . , X2n ] in 2n variables with coecients in F , where the
homomorphism is the extension of the correspondence

Xj aj , 1 j n, and Xj a1
j , n + 1 j 2n.

In other words, A = F [X1 , . . . , X2n ]/S, where S is a maximal ideal of


F [X1 , . . . , X2n ]. In particular, S = 0, and hence the eld F [X1 , . . . , X2n ]/S is
a nitely generated algebraic extension of F , that is, nite dimensional over F .
Thus dimF (A) is nite.
Suppose now that G has a normal abelian subgroup U of nite index. By
Theorem 5.5, A contains a simple F U -submodule B, and there exists a nite subset
S such that A = gS Bg. By Lemma 5.4, each Bg is a simple F U -submodule.
So, dimF (Bg) is nite by the result shown in the above paragraph. Since S is
nite, dimF (A) is nite. 
However, we have
Lemma 15.4 ([96]). Let F be a eld, and let G be a group. Suppose that G has a
normal subgroup H such that there exists a simple F H-module B with dimF (B)
innite. Then there exists a simple F G-module A such that dimF (A) is innite.
Chapter 15. The injective envelopes of simple modules 181

Proof. We have B =F H F H/Q for some maximal right ideal Q of F H. Let S be


a transversal to H in G. Since H is a normal subgroup of G,


P = Qg = Qg
gS gS

is a right ideal of F G. Since P F H = Q, P = F G. Hence


there is a maximal
right ideal P1 of F G such that P P1 . Since F G = gS F Hg, P1 does not
include F H. Now F H P1 contains F H P = Q, and F H P1 is a right ideal
of F H. Then F H P1 = Q since Q is a maximal right ideal of F H. Hence,

(F H + P1 )/P1
= F H/(F H P1 ) = F H/Q,

which is innite dimensional over F . Thus F G/P1 is innite dimensional over F .


Since P1 is a maximal right ideal of F G, the F G-module A = F G/P1 is simple,
as required. 
Proposition 15.5 ([96]). Suppose that F is a non locally nite eld, and let G be a
polycyclic group which has no abelian subgroups of nite index. Then there exists
a simple F G-module A such that dimF (A) is innite.
Proof. The group G has a nite series

1 = G0  G1   Gn = G

in which every factor Gj /Gj1 is cyclic. We are arguing by induction on n. Put


H = Gn1 so that G = H, y for some y G. We claim that we may assume
that H is abelian-by-nite. For, otherwise, by the induction hypothesis, there
exists a simple F H-module B such that dimF (B) is innite, and it suces to
apply Lemma 15.4. Suppose then that H has a normal abelian subgroup U of
nite index. In this case G/H has to be innite. Put m = |H/U |. Then H m is a
characteristic subgroup of H, and so H m is G-invariant. Since H/H m is a nitely
generated periodic soluble group, it is nite. On the other hand, H m U , so that
H m is abelian. Since H m is nitely generated, the periodic part of H m is nite,
and so there exists some t N such that H mt is torsion-free, and once more,
H/H mt is nite. In other words, without loss of generality, we may suppose that
U is G-invariant and torsion-free. Since H/U is nite, G/CG (H/U ) is nite. Put
k = |CG (H/U )|, and dene z = y k and L = U, z. Then L is a normal subgroup
of G, and L/U is innite. We claim that U = CL (U ). For, otherwise CL (U ) is
a normal abelian subgroup of L having nite index; so, since G/L is nite, G is
abelian-by-nite, a contradiction, which shows our claim: CL (U ) = U .
Suppose that
U = u1  ur  ,
and for each 1 s r, put

Us = u1  us1  us+1  ur  .


182 Chapter 15. The injective envelopes of simple modules

Suppose, if possible, that NL (Us ) = U for each 1 s r. Then there exists some
q N such that
z q NL (U1 ) NL (Ur ).
Take an index 1 j r. Then

z q uj z q U1 Uj1 Uj+1 Ur = uj  .

Since uj has innite order, [uj , z 2q ] = 1. Hence, [U, z 2q ] = 1, a contradiction.


Thus, there exists some 1 s r such that NL (Us ) = U . For this value s, we
write u = us and V = Us , so that U = V, u and U/V is innite cyclic.
Since F is not locally nite, the multiplicative group U (F ) = F \ 0 contains
an element having innite order. If j Z, we dene the mapping j : U F
in this way:
for any w U , we put wj = a(j,w) , where a(j, w) is an integer such that
z j wz j ua(j,w) V .
This denition gives immediately that a(j, ww1 ) = a(j, w) + a(j, w1 ), and hence,
(ww1 )j = (wj )(w1 j ) for all w, w1 U . In other words, j is a homomorphism
of U in U (F ). Suppose that j = i for some j < i. Since has innite order in
U (F ), a(j, w) = a(i, w) for all w U . Thus [w, z ij ] V for all w U , and hence,
z ij NL (V ), which is impossible by the construction of V . Thus, we have just
shown that j = i if j = i.
Now let A be a vector space over F with a basis {aj | j Z}, and dene the
action of L on A by
aj z = aj+1 , and aj w = aj (wj ) for any w U , j Z.
It is easy to see that this action makes A into an F L-module. Since j = i for
all j = i, the F U -submodules of A are only the direct sum of the one-dimensional
subspaces aj F . Hence, if B is a non-zero F L-submodule of A, B must contain
some aj . Then B contains ai = aj z ij for all i, that is B = A. Hence A is a simple
F L-module such that dimF (A) is innite, and it suces to apply Lemma 15.4. 
In connection with Propositions 15.2, 15.3 and 15.5, the following question
naturally arises: Let F be a eld. Determine groups G for which every simple F G-
module has nite dimension over F . The following interesting papers should be
mentioned in connection with this problem: B. Hartley [108], R.L. Snider [270],
B.A.F. Wehrfritz [281, 282, 283].
Lemma 15.6. Let D be a Dedekind domain, and let G be a polycyclic-by-nite
group. Suppose that A is a nitely generated monolithic DG-module with monolith
M . If AnnD (M ) = P = 0, then AnnD (A) = P n , for some n N.
Proof. Since M is a simple DG-submodule of A, P Spec(D). Let T be the D-
periodic part of A. Since A is monolithic, T is equal to the P -component of A.
But DG is a noetherian ring (see D.S. Passman [217, Theorem 10.2.7]), A is
Chapter 15. The injective envelopes of simple modules 183

a noetherian DG-module by Lemma 1.1. Then there is some n N such that


P,n (A) = P,n+m (A) for each m N. This means that T P n = 0, and it follows
that there exists a D-submodule C such that A = T C (see I. Kaplansky [123]).
Thus AP n C, and AP n T = 0. Since M T , this means that A = T . 
Lemma 15.7. Let F be a locally nite eld, and let G be a nitely generated nilpo-
tent group. If A is a nitely generated monolithic F G-module, then G/CG (A) is
nite, and hence, dimF (A) is nite.
Proof. Let M be the F G-monolith of A. There is no loss if we assume that
CG (A) = 1. According to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157,
Corollary 1.21], (G/CG (M )) is periodic. It follows that (G/CG (M )) is nite,
and, by Lemma 15.1, so is G/CG (M ). We claim that (G) is periodic. Otherwise,
there exists some z (G) CG (M ) of innite order since G/CG (M ) is nite.
Put J = F z so that J is a principal ideal domain, and we can think of A as
a JG-module. By the choice of z, AnnJ (M ) = (z 1)J. By Lemma 15.6, there
exists some n N such that A(z 1)n = 0. For every k N we have

(z 1)k = z k C1k z k1 + C2k z k2 + (1)k ,

where C1k , C2k ,. . . are the binomial coecients. Let p = char F , and choose s N
such that t = ps n. If a A, then 0 = a(z 1)t = az t + (1)t a, and so
az t = (1)t+1 a. If p = 2, then a = a, and hence, az t = a. If p > 2, then again
az t = a. In any case, z t CG (A) = 1, a contradiction. Hence, (G) is periodic.
Then it follows that (G) is nite, and it suces to apply Lemma 15.1 to conclude
that G is nite, as required. 
Proposition 15.8. Let D be a Dedekind Z-domain, G a locally (polycyclic-by-nite)
group of nite 0-rank. Suppose that A is a nitely generated DG-module such that
CG (A) = 1. If the set of all maximal DG-submodules of A is nite, then A is
D-periodic, and (G) is periodic.
Proof. Suppose that
A = a1 DG + + an DG.
According to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Theorem 1.15],
A is D-periodic, and, then AnnD (aj ) = 0 for every index 1 j n. Since D is
an integral domain,

L = AnnD (A) = AnnD (a1 ) AnnD (an ) = 0 .

By Theorem 6.3, L = P1k1 Ptkt , where P1 , . . . , Pt Spec(D) are pairwise dierent


prime ideals, and k1 , . . . , kt N. We proceed by induction on k(L) = k1 + + kt .
If k(L) = 1, then L is a maximal ideal of D, so D/L is a locally nite eld. By L.A.
Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.21], (G) is periodic.
Suppose that k(L) > 1, and we have already proved our assertion for mod-
ules B such that k(AnnD (B)) < k(L). Put Z = (G), and B = AP1 . Then
184 Chapter 15. The injective envelopes of simple modules

AnnD (B) = P1k1 1 Ptkt , so that k(AnnD (B)) < k(L). Moreover, AnnD (A/B) =
P1 , so that k(AnnD (A/B)) < k(L). Applying induction, (G/CG (B)), and
(G/CG (A/B)) are periodic, and so are Z/(Z CG (B)), and Z/(Z CG (A/B)).
If U = CG (B) CG (A/B) Z, then, by Remaks theorem, Z/U is periodic. Put

A0 = A, A1 = AP1 , A2 = AP12 , . . . , Ak1 = AP1k1 , Ak1 +1 = AP1k1 P2 , . . . ,

Ak1 +k2 = AP1k1 P2k2 , . . . , Ak1 ++kt = AP1k1 Ptkt = 0 .


We have just constructed a descending chain of submodules

A = A0 A1 Ak1 ++kt = 0 .

Given any factor Aj /Aj+1 of this series, we put Pm =AnnD (Aj /Aj+1 ) (m = m(j)).
Then we may consider Aj /Aj+1 as an Fm G-module, where Fm = D/Pm . Since
Fm is a locally nite eld, the additive group of Aj /Aj+1 is an elementary abelian
pm -group, where pm =char Fm . Hence every factor of the above series {Aj | 0
j k(L)} is elementary abelian, and it readily follows that the additive group of A
is periodic. Actually, this group is bounded, that is sA = 0, where s = pk11 pkt t .
Pick z U , and a A. Then there exists some b B such that az = a + b. If
n N , then az n = a + nb, and so, in particular, az s = a + sb = a. This means
that U is periodic. Then Z is periodic, because Z/U is periodic. 
Corollary 15.9. Let D be a Dedekind Z-domain, and let G be a nitely gen-
erated nilpotent group. Suppose that A is a nitely generated monolithic DG-
module with the DG-monolith M . Then AnnD (A) = P n for some n N, where
P = AnnD (DG (A)) Spec(D), and G/CG (A) is nite. In particular, A has nite
D-composition series.
Proof. There is no loss if we suppose that CG (A) = 1. According to L.A. Kur-
dachenko, J. Otal and I.Ya. Subbotin [157, Corollary 1.16], P =AnnD (M )
Spec(D). By Lemma 15.6, AnnD (A) = P n , for some n N. Put Aj = P,j (A), for
j N. Pick y P \ P 2 . The mapping : a ay, a A2 is a DG-endomorphism
of A2 and so Ker and Im are DG-submodules. Clearly, Im = A2 y A1 and
Ker = {a A2 | ay = 0} A1 . By Proposition 6.13, P = yD + P 2 , and so
Ker = {a A2 | aP = 0} = A1 . Hence, we may deduce A1 = A2 /A1 . By
Lemma 15.7, dimF (A1 ) is nite so that dimF (A2 /A1 ) is likewise nite. It follows
that A2 has nite D-composition series. Since An = A, we may apply induction to
obtain that A has nite D-composition series. Then A has nite DG-composition
series. By Proposition 15.8, (G) is periodic, and hence nite. It suces to apply
Lemma 15.1 to obtain that G is nite, as required. 
Corollary 15.10. Let D be a Dedekind Z-domain, and let G be a nitely generated
nilpotent group. Suppose that A is a monolithic DG-module with the DG-monolith
M . Then A satises the following conditions:
(1) A is a P -module, where P = AnnD (M ) Spec(D).
Chapter 15. The injective envelopes of simple modules 185

(2) A is the last term of its upper socular series.


(3) The socular height of A is at most , the rst innite ordinal.
Proof. According to L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Corollary
1.16], P = AnnD (M ) Spec(D). Let T be the D-periodic part of A. Since A is
monolithic, T is equal to the P -component of A. Suppose that B is a nitely
generated DG-submodule of A. By Corollary 15.9, B is D-periodic. It follows that
T = A, and thus A is a P -module. Since B has nite composition DG-series, it
suces to apply Lemma 4.9. 
Lemma 15.11. Let D be a Dedekind Z-domain, and let G be a nitely generated
nilpotent group. Suppose that A is a monolithic DG-module with DG-monolith M ,
and put H = CG (M ). Let

0 = A0 A1 = M A2 An An = A
nN

be an ascending DG-composition series of A. Then H CG (An /An1 ) for every


n N.
Proof. Suppose that there exists some m N such that CG (Am /Am1 ) does not
include H. Let b An \ An1 , and put B = bDG, and C = B An1 . Then
G/CG (B/C) does not include H. By Corollary 15.9, B has nite DG-composition
series. By Theorem 10.21, B has the Z DK-decomposition, that is, B = U Z,

where Z = DH (B) is the upper hypercenter of B, and U = DH (B) is the unique
maximal hypereccentric DG-submodule of B. Since H is a normal subgroup of G,
U , and Z are DG-submodules of B. By the choice of H, we have U = 0. On
the other hand, H = CG (M ), so that M Z, and hence, U = 0 since A is
monolithic, a contradiction. 
Lemma 15.12. Let F be a locally nite eld, and let G be a nite nilpotent group.
If A is a monolithic F G-module, then A is an artinian F G-module.
Proof. Let M be the F G-monolith of A, and put H = CG (M ). Then G/H is a
p -group, where p =charF (see, for example, L.A. Kurdachenko, J. Otal and I.Ya.
Subbotin [157, Theorem 3.1]). By Corollary 15.10, A has an ascending series of
F G-submodules

0 = A0 A1 = M A2 An An = A
nN

with simple F G-factors. By Lemma 15.11, H CG (Aj /Aj1 ), for every j N. We


proceed by induction on |H|. If H = 1, by Corollary 5.15, A = M , and so A is a
simple F G-module. Suppose now that H = 1, and pick 1 = z H (G). Since
A is F H-hypercentral, A is likewise F z-hypercentral; furthermore, the length of
an upper F z-central series of A is at most . Let

0 = C0 C1 C2 Cn Cn = C
nN
186 Chapter 15. The injective envelopes of simple modules

be the upper F z-central series of A. Since z (G), every term Cn is an F G-


submodule of A. The F G-submodule C1 = AnnA (z 1) is monolithic, and z
H CG (C1 ). In particular, |H CG (C1 )| < |H|. By induction, C1 is an artinian
F G-module. Hence, by Proposition 1.8, the F G-module A is artinian. 
Lemma 15.13. Let F be a locally nite eld, and let G be a nitely generated nilpo-
tent group. If A is a monolithic F G-module, then A is an artinian F G-module.
Proof. Let M be the F G-monolith of A, and put H = CG (M ). Then G/H is a p -
group, where p =char F (see L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157,
Theorem 3.1]). By Corollary 15.10, A has an ascending series of F G-submodules

0 = A0 A1 = M A2 An An = A
nN

with simple F G-factors. By Lemma 15.11, H CG (Aj /Aj1 ), for every j N. We


proceed by induction on r0 (H). If r0 (H) = 0, then H is nite, and our assertion
follows from Lemma 15.12. Suppose that r0 (H) > 0. By Lemma 15.1, there exists
some z H (G) having innite order. Since A is F H-hypercentral, A is likewise
F z-hypercentral; furthermore, the length of an upper F z-central series of A
is at most . Let

0 = C0 C1 C2 Cn Cn = C
nN

be the upper F z-central series of A. Since z (G), every term Cn is an F G-


submodule of A. The F G-submodule C1 = AnnA (z 1) is monolithic, and z
H CG (C1 ). In particular, r0 (H CG (C1 )) < r0 (H). By induction, C1 is an
artinian F G-module, and so is A by Proposition 1.8. 
Theorem 15.14. Let D be a Dedekind Z-domain, and let G be a nitely generated
nilpotent group. If A is a monolithic DG-module, then A is an artinian DG-
module.
Proof. If M is the DG-monolith of A, according to L.A. Kurdachenko, J. Otal
and I.Ya. Subbotin [157, Corollary 1.16], P = AnnD (M ) Spec(D). By Corollary
15.10, A is a P -module. Put B = P,1 (A), and look at B as an F G-module, where
F = D/P is a locally nite eld. By Lemma 15.13, B is an artinian F G-module,
and so A is an artinian DG-module by Proposition 1.8. 
Corollary 15.15. Let D be a Dedekind Z-domain, and let G be a nitely generated
nilpotent-by-nite group. If A is a monolithic DG-module, then A is an artinian
DG-module.
Proof. Let M be the DG-monolith of A, and let H be a nilpotent normal subgroup
of G having nite index. Then H is nitely generated (see, for example, [234,
Theorem 1.41]). By Theorem 5.5, there exist
a simple DH-submodule B of M ,
and a nite subset S of G such that M = gS Bg. Let U be a DH-submodule
Chapter 15. The injective envelopes of simple modules 187

of A which is maximal under B U = 0. Then A/U is a monolithic DH-module


with DH-monolith (B + U )/U . Since H is a nitely generated nilpotent subgroup,
by Theorem 15.14,
A/U is an artinian DH-module. Let T be a transversal to H
in G. Put V = xT U x so that V is a DG-submodule. If V = 0, then M V ,
and so B V . On the other hand, V U , and U B = 0. This contradiction
shows that V = 0. By Remaks theorem,

A
A1 = A/U x.
xT

Further, for every x T , A/U x = Ax/U x


= A/U is an artinian DH-module.
By Lemma 1.2, A1 and A are artinian DH-modules. Thus, A is an artinian DG-
module, as required. 

Corollary 15.16. Let D be a Dedekind Z-domain, and let G be a nitely gener-


ated nilpotent-by-nite group. If M is a simple DG-module, then IEDG (M ) is an
artinian DG-module.
Corollary 15.17. Let D be a Dedekind Z-domain, and let G be a nitely generated
nilpotent-by-nite group. If A is an artinian DG-module, then IEDG (A) is an
artinian DG-module.
As we mentioned at the beginning of this chapter, in the case D = Z, these
two last results were shown by R.L. Snider [269] and extended to polycyclic-by-
nite groups by I.M. Musson [195], who proved the following facts.
Theorem 15.18. The following results hold:
(1) Let G be a polycyclic-by-nite group, and suppose that A is a nitely generated
ZG-module such that |G/CG (A)| is nite, and pA = 0 for some prime p
(in particular, this happens if A is simple). Then IEZG (A) is an artinian
ZG-module.
(2) Let F be a eld of characteristic p > 0, and let G be a polycyclic-by-nite
group. If A is a nite dimensional F G-module (that is, dimF (A) is nite),
then IEF G (A) is an artinian F G-module.
(3) Let F be a non-locally nite eld, and let G be a polycyclic-by-nite group.
Then the injective envelope of every simple F G-module is artinian if, and
only if G is abelian-by-nite.
In connection with some parts of the above result, S. Donkin [58] proved:
Theorem 15.19. The following results hold:
(1) Let F be a eld, and let G be a polycyclic-by-nite group. Suppose that A is a
locally nite dimensional F G-module (that is, dimF (aF G) is nite for every
a A). Then every essential extension of A is locally nite dimensional.
188 Chapter 15. The injective envelopes of simple modules

(2) Let F be a eld of characteristic 0, and let G be a polycyclic-by-nite group.


If A is a nite dimensional F G-module, then every essential extension of A
is artinian, and EndF G (IEF G (A)) is a noetherian ring.
The proofs of these results depend on applications from both the theory of
polycyclic group rings and the theory of Hopf algebras, and we omit them here.
To nish, it is worth mentioning the following generalization of a result ob-
tained by I.M. Musson [199].
Theorem 15.20. Let F be a eld of characteristic 0, and let G be a soluble-by-
nite torsion-free group of nite special rank. If A is a nite dimensional F G-
module, then every essential extension of A is artinian, and EndF G (IEF G (A)) is
a noetherian ring.
Chapter 16

Quasinite modules

As we have seen in previous chapters, there are not too many cases in which
artinian modules can be satisfactorily described, although many problems require
the investigation of some specic artinian modules. We have not dealt with this
question in this book in full. In this chapter, however, we consider one of the
most important types of artinian modules, namely, the quasinite modules. These
modules appear in the following way. Suppose that A is an artinian DG-module,
and let U be the family of all innite submodules of A. Choose a minimal element
M of U. Then either M is a minimal (innite), and hence, a simple submodule
of A or M is innite and not simple, but every proper submodule of M is nite.
D.I. Zaitsev introduced this type of modules in connection with the study of the
complementability of normal subgroups [296]. These modules also appeared in
other group-theoretical researches, for example, in the study of groups with the
weak maximal or minimal conditions for normal subgroups (L.A. Kurdachenko
[145, 147], D.I. Zaitsev, L.A. Kurdachenko and A.V. Tushev [312]). In [299], D.I.
Zaitsev initiated the investigation of modules over integral group rings in which
all proper submodules are nite. Besides these modules, there are many types of
Dedekind domains for which the unique nite module is the zero module. Therefore
the condition of being a nite submodule is a reasonable change to the condition
of being a nitely generated submodule. In other words, we come to a module A
over a group ring DG with the property:
every proper DG-submodule of A is nitely generated as a D-submodule.
In this setting, the two following situations appear:
(1) A has a proper DG-submodule B such that A/B is a simple DG-module;
(2) every proper DG-submodule of A is nitely generated as a D-submodule,
and A is the union of its proper DG-submodules.
The rst case can be reduced to the case of nitely generated D-modules and
simple DG-modules. Thus, the second case remains more interesting. To study it,
we consider the following concept. Let R be a ring, and let G be a group. An
190 Chapter 16. Quasinite modules

RG-module A is said to be a quasinite RG-module if A satises the following


conditions:
(1) A is not nitely generated as a D-module.
(2) If B is a proper RG-submodule of A, then B is nitely generated as a D-
submodule.
(3) A is the union of its proper DG-submodules.
The case in which R = F is a nite eld was considered by D.I. Zaitsev
[299] while the case in which R = F is an arbitrary eld was considered by L.A.
Kurdachenko and I.Ya. Subbotin in [172]; the case in which R = D is a Dedekind
domain was considered in the paper L.A. Kurdachenko [152]. In this chapter, we
collect the basic results of these papers.
If R = Z and G = 1, then a Pr ufer p-group gives us an example of a
quasinite module. Let F be a eld, G = x be an innite cyclic group D = F x.
Then a Pr ufer P -module is an example of a quasinite F G-module for every
maximal ideal P of D. As we will see later, the Pr ufer P -modules play a very
important role in the structure of quasinite modules.
Lemma 16.1. Let A be a quasinite RG-group, where R is a ring, and G is a group.
Then A cannot be decomposed as a direct sum of two proper RG-submodules. In
particular, SocR (A) is a proper submodule of A.
Proof. This lemma is obvious. 
Corollary 16.2. Let D be a Dedekind domain, G a group, and A a quasinite
DG-module. Then SocDG (A) is a proper submodule of A.
Lemma 16.3 (L.A. Kurdachenko [152]). Let A be a quasinite RG-module, where
R is an integral domain, and G is a group. Then either A is R-divisible or there
exists some 0 = x R such that Ax = 0.
Proof. If Ax = A for every element 0 = x R, then A is R-divisible. Therefore
suppose that there exists some 0 = x R such that Ax = A. The mapping
: a ax, a A is an RG-endomorphism of A such that Im = Ax and
Ker =AnnA (x). We claim that Ker = A. For, otherwise Ker is proper, and
so it is nitely generated as an R-module. Since Im = Ax = A, Im is likewise
a nitely generated R-module. Since

A/ Ker
= Im ,

A is a nitely generated R-module. This contradiction shows that Ker = A so


that Ax = 0, as required. 
Further, we need to establish a result on the structure of nitely generated
modules over Dedekind domains. Before doing so, we need to establish the unique-
ness of the direct decompositions there involved.
Chapter 16. Quasinite modules 191

Theorem 16.4. Let P be a maximal ideal of a Dedekind domain D, and suppose


that A is a P -module over D. Given two direct decompositions of A as direct sum
of cyclic submodules,  
A= a D = b D,

then there are an automorphism of A, and a bijection : such that


b = a , for every .
Proof. For every n N, we put

n = { | P n = AnnD (a )}, and


n = { | P n = AnnD (b }.

It suces to show that |n | = |n |, for every n N. Dene Cn = P,1 (A)AP n1 .


We have  
AP n1 = ( a P n1 ) ( a P n1 )
n \

and  
AP n1 = ( b P n1 ) ( b P n1 ),
n \

where = 1 n , and = 1 n . Therefore,


   
Cn = ( a P n1 ) Cn+1 = ( b P n1 ) Cn+1 .
n n

Hence, |n | = |n | = dimD/P (Cn+1 /Cn ). 


Lemma 16.5. Let R be an integral domain, and suppose that A is a nitely gener-
ated R-module. If A is R-periodic, then AnnR (A) = 0.
Proof. Let M be a nite subset of A such that A = M R. Obviously,
 
AnnR (a) AnnR (a) AnnR (A).
aM aM

Since R has no zero-divisors, aM AnnR (a) = 0. Hence, AnnR (a) = 0, as
required. 
Proposition 16.6. Let D be a Dedekind domain, and suppose that A is a nitely
generated periodic D-module. Then:
(1) A is a direct sum of nitely many cyclic submodules.

(2) If A = a D = b D, where the sets and are nite, and for
s
every and we have AnnD (a D) = Pk , and AnnD (b D) = P
for certain maximal ideals P and P of D, then there exist an automorphism
of A, and a bijection : such that b = a , for every .
192 Chapter 16. Quasinite modules

Proof. By Lemma 16.5, AnnD (A) = 0. Then A is a direct sum of a nitely many
cyclic submodules (D.W. Sharpe and P. Vamos [265, Theorem 6.14]), which gives
(1). The assertion (2) follows from Corollary 6.25 and Theorem 16.4. 

Suppose that A is a nitely generated module over a Dedekind domain D, and


put T = tD (A). By Lemma 1.1, A is a noetherian D-module, and, in particular, T
is nitely generated. As the nitely generated factor-module A/T is D-torsion-free,
our next step is the consideration of nitely generated torsion-free modules.

Lemma 16.7. If L is an ideal of an integral domain R, there is a Z-isomorphism


: L1 HomR (L, R) such that a(x) = xa, for every x L1 and a L.

Proof. Pick x L1 , and dene a mapping x : L R by ax = xa, for every


a L. Obviously,

(a + b)x = x(a + b) = xa + xb = ax + bx ,

and
(ca)x = x(ca) = c(xa) = c(ax )

so that x is an R-homomorphism. This denes a map by the rule x = x , for


every x L1 . Since

ax+y = (x + y)a = xa + ya = ax + ay = a(x + y ),

we have (x + y) = x + y, and, therefore, is a Z-homomorphism. If x Ker ,


ax = 0 for each a A. Since R has no zero-divisors, Lx = 0 implies x = 0.
Thus, is a monomorphism.
Finally, given HomR (L, R), let 0 = a, b L. Then

(a)b = (ab) = (ba) = (b)a,

so that (a)a1 = (b)b1 . It follows that (a)a1 is constant when a runs through
L\0. Put (a)a1 = x. Thus, a = xa provided 0 = a L. Since 0 = 0 = x0,
= x . Furthermore, a R, i.e. ax R, for every a A. Therefore, x L1 ,
which shows that = (x ). Consequently, is a Z-isomorphism. 

Recall that a module A over a ring R is said to be hereditary if A, and


all of its submodules are projective. It is obvious that submodules of hereditary
modules are hereditary. A ring R is said to be hereditary if all of its right ideals
are projective modules.

Proposition 16.8. Let L be a non-zero ideal of an integral domain R. Then the


R-module L is projective if and only if L is an invertible ideal. In particular, R is
a heredity ring if and only if every non-zero ideal of R is invertible.
Chapter 16. Quasinite modules 193

Proof. Suppose that L is invertible. Then there are some a1 , . . . , am L, and


x1 , . . . , xm L1 such that
1 = a1 x1 + + am xm .
Let C be a free R-module with basis {c1 , . . . , cm }. There is an R-homomorphism
: C L such that cj = aj , for every 1 j m (see D.S. Passman [219,
Thoerem 2.4]). Dene : L C by
a = c1 (ax1 ) + + cm (axm )
for each a L. Since R is commutative, is an R-homomorphism. Further, if
a L we have
a() = (a) = (c1 ax1 + + cm axm )
= (c1 )ax1 + + (cm )axm
= a1 ax1 + + am axm = a(a1 x1 + + am xm ) = a 1 = a,
and so a = L . By Lemma 7.15, C = Im Ker , and L = Im . It follows
that L is a projective R-module (D.S. Passman [219, Thoerem 2.8]).
Conversely, suppose that L is a projective R-module. Then there exists a free
R-module C1 such that C1 = L A, for some R-submodule A (see D. S. Passman
[219, Theorem 2.8]). Let {c | } be a free basis of C1 . If a L, then

a= c (a),

for some (a) R, . If b is another element of L, then




a+b= c (a) + c (b) = c ( (a) + (b)).


Since

a+b= c (a + b)

and {c | } is a free basis of C1 ,


(a + b) = (a) + (b),
for every . Similarly, if d R,
(da) = d (a),
for every . Hence, the mapping a (a) is an R-homomorphism for every
. By Lemma 16.7, for each a L, there are elements y L1 such that
(a) = y a, for every . In other words

a= c (y a).

194 Chapter 16. Quasinite modules

Let 0 = b L, and put b = Supp(( (b)) ). Then b is nite, say b =


{1 , . . . , k }. In particular, (b) = 0, for every \ b . This means that
y b = 0, for every \ b , and therefore y = 0, for every \ b . Hence,

b= c (y b).

Let be the projection mapping of C1 on L. Since is an R-homomorphism,



b = b = ( c (y b)) = b (y b).

Multiplying both sides of this equation by y 1 (R is a subring of its eld of


fractions), we obtain, then

1= b y .

Thus, R = LL1 . 
Lemma 16.9. Let A be a nitely generated D-module, where D is a Dedekind
domain. Suppose that there exists some a A such that AnnD (a) = 0. Then A
has a D-submodule B such that a  B and that satises the following conditions:
(1) There exists some D-submodule C such that A = B C.
(2) There exists an ideal L of D such that C
=D L.
Proof. We dene a mapping : D A by x = ax, for every x D. Obvi-
ously, is a D-homomorphism, and, since Ker = AnnD (a) = 0, is in fact a
monomorphism. Let F be the eld of fractions of D. Since F is D-divisible, F is
an injective D-module by Theorem 7.9. Therefore, if : D F is the canonical
embedding, then there exists a D-homomorphism : A F such that = .
Then
1 = 1 = 1() = (1) = a
and so is a non-zero homomorphism.
Moreover, A is a fractional ideal of D. It follows that there is an ideal L of
D such that A =D L. Since D is a Dedekind domain, the ideal L is invertible,
and so L is a projective D-module by Theorem 7.9. It follows that A = B C,
where B = Ker and C =D L. Since a = 0, we have a  B, and we are done. 
Corollary 16.10. Let A be a nitely generated D-module, where D is a Dedekind
domain. If A is D-torsion-free, then

A = C1 Cm ,

where C1 , . . . , Cm are D-isomorphic to ideals of D. In particular, every nitely


generated D-torsion-free D-module is projective.
Chapter 16. Quasinite modules 195

Proof. Let M be a maximal D-free subset of A, and put



E= aD.
aM

By Lemma 1.1, E is nitely generated, and, in particular, M is nite, let us say


M = {a1 , . . . , am }. By Lemma 16.9, A = C1 B1 , where a1  B1 , and C1 is
D-isomorphic to some ideal of D. Since A/B1 is D-torsion-free, a1 D B1 = 0. If
M1 is a maximal D-free subset of B1 , then M1 {a1 } is a maximal D-free subset
of A. Since D is an integral domain, |M1 | = m 1, so that M1 = {b1 , . . . , bm1 }.
Applying Lemma 16.9 to B1 , we obtain a decomposition B1 = C2 B2 , where C2
is D-isomorphic to some ideal of D, and b1  B2 . Since A/B2 is D-torsion-free,
(a1 Db1 D)B2 = 0. Proceeding in the same way, we nally get a decomposition

A = C1 Cm E1 ,

where C1 , . . . , Cm are isomorphic to ideals of D. Since r0 (C1 Cm ) = m =


r0 (A), we have that E1 = 0; hence,

A = C1 Cm ,

as required. 
Corollary 16.11. Let A be a nitely generated D-torsion-free D-module, where D
is a Dedekind domain. Then there exists a non-zero ideal L of D such that, if
m = r0 (A),
A=D D D L.
 
m1

Moreover, L is determined up to multiplication by a non-zero constant in the eld


of fractions of D.
Proof. This follows at once from Corollary 16.10 and D.S. Passman [219, Theorem
7.7]. 
Corollary 16.12. Let A be a nitely generated D-module, where D is a Dedekind
domain. Then A = T C, where T = tD (A), and C is a nitely generated D-
torsion-free submodule.
Proof. This follows at once from Corollary 16.10 and the standard properties of
projective modules. 
We are now in a position to prove the basis structure theorem of nitely
generated modules over Dedekind domains.
Theorem 16.13. Let A be a nitely generated D-module, where D is a Dedekind
domain. Then the following assertions hold:
(1) A = T C, where T = tD (A) is the D-periodic part of A and C is a nitely
generated D-torsion-free submodule.
196 Chapter 16. Quasinite modules

(2) T = A1 Ak , where {A1 , . . . , Ak } are primary cyclic submodules.


(3) If T = B1 Bt where every Bj is a primary cyclic submodule, then
k = t, and there are a permutation Sk , and an automorphism of T
such that Bj = Aj , for every 1 j k.
(4) If m = r0 (C), then there exists a non-zero ideal L of D such that

C
=D D D L.
 
m1

Moreover, L is determined up to multiplication by a non-zero constant in the


eld of fractions of D.
Proof. Apply Corollary 16.12, Proposition 16.6, and Corollary 16.11. 
Now, we may continue the study of quasinite modules.
Lemma 16.14 ([152]). Let A be a quasinite DG-module, where D is a Dedekind
domain and G is a group. If A is D-periodic, then there exists some P Spec(D)
such that A is a P -module, and either AP = 0 or

A = C1 Cn ,

where C1 , . . . , Cn are Pr
ufer P -modules.
Proof. Since A is D-periodic, 
A= AP ,
P

where =AssD (A). Since every AP is a DG-submodule, by Lemma 16.1, there


exists some P Spec(D) such that A = AP . Consider A1 = P,1 (A). If A = A1 ,
then AP = 0. Otherwise A1 is nitely generated over D. In particular, dimD/P A1
is nite and L.A. Kurdachenko, J. Otal and I.Ya. Subbotin [157, Lemma 5.6] yields
that A is D-artinian. By Lemma 16.3, A is D-divisible and L.A. Kurdachenko, J.
Otal and I.Ya. Subbotin [157, Theorem 5.7] shows that A is a direct sum of nitely
many Pr ufer P -modules. 
Lemma 16.15 ([152]). Suppose that D is a Dedekind domain that is not a eld.
Let G be a group, and let A be a quasinite DG-module. Then we have:
(1) If D is not a local ring, then A is D -periodic.
(2) If D is a local ring, and A is not D-periodic, then A is D-torsion-free and
divisible. Moreover, if F is the eld of fractions for D, then A as an F G-
module is simple, and dimF A is nite.
Proof. Put T = tD (A). Suppose rst that T = 0. By Lemma 16.3, either Ax =
0 for some 0 = x D or A is D-divisible. In the rst case A = T . In the
second one, A is D-divisible so that T is also D-divisible by Lemma 7.18. Also,
Chapter 16. Quasinite modules 197

by Theorem 16.13, T is not nitely generated over D, and it follows that T = A.


Suppose now that T = 0. By (QF 3), A has a proper DG-submodule B, which
a fortiori is nitely generated over D. In fact, B can be chosen such that B has
minimal D-rank. Let C be a D-pure envelope of B so that C is a D-divisible
DG-submodule of A. By Theorem 7.25, C can not be nitely generated as D-
submodule so that C = A. Since A/B is D-periodic and D-divisible, by Theorem
7.25 
A/B = AP /B,
P Spec(D)

where AP /B is a a direct sum of Prufer P -modules. If D is not a local ring, A/B


cannot be a P -module for some P Spec(D), and we obtain a contradiction with
Lemma 16.14.
Let now D be a local ring. Since A is D-divisible and D-torsion-free, we can
consider A as an F G-module. Since A is a pure envelope of a nitely generated
D-submodule, dimF A is nite.
Finally, let U be a non-zero F G-submodule of A. Then B U is a non-zero
DG-submodule of B. By the choice of B we have r0 (B) = r0 (B U ). This means
that a pure envelope of B U coincides with the pure envelope of B, that is, with
A. In other words, A is a simple F G-module. 
The structure of simple F G-modules A of nite dimension can be studied
in a satisfactory way. Remark that the most interesting cases appear when D is
a eld, and when D is a not a eld but A is D-periodic. Let D be a Dedekind
domain. If P Spec(D), and n N, we denote by n : D/P n D/P n+1 the
canonical epimorphism, and we put

D(P ) =
lim {D/P n , n | n N}.

Proposition 16.16 (L.A. Kurdachenko and H. Smith [168]). Let D be a Dedekind


domain. If P Spec(D), and A is a Pr ufer P -module, then the ring of endo-
morphisms of A is isomorphic to D(P ). Moreover, D(P ) is a principal ideal
domain, and the set of all non-zero ideals of D(P ) is {D(P )P n | n N}.
Proof. Let y P \ P 2 , and consider the mapping a ay, a A. For n N, put
An = P,n (A) so that An+1 y = An . Since A1 is simple, there exists some 0 = a1
such that A1 = a1 D and AnnD (a1 ) = P . Since A2 y = A1 , there exists some a2
such that a1 = a2 y. Thus, a2  A1 , and therefore a2 D = A2 and AnnD (a2 ) = P 2 .
Similarly, we choose elements {an | n N} such that

A= an D, An = an D an+1 D = An+1 ,
nN

a1 y = 0, an+1 y = an , and AnnD (an ) = P n .


In particular, an A
= D/P n , for every n N.
198 Chapter 16. Quasinite modules

Pick AnnD (A). Then (an )P n = 0, and so an an D. It follows that


there exists some n D such that an = an n . If n D and an n = an n ,
then an (n n ) = 0, that is n n AnnD (an ) = P n . In particular, n + P n =
n + P n , for every n N. Dene a mapping

: EndD (A) D/DP n
nN

by = (an + P n )nN . We have

an n = an = (an+1 y) = (an+1 )y = (an+1 n+1 )y = an n+1 .

It follows that an (n+1 n ) = 0, that is n+1 n AnnD (an ) = P n , which


gives n + P n = n+1 + P n . It follows that D(P ). Let EndD (A), and
suppose that = (n + P n )nN . One easily obtains that

an ( + ) = an (n + n ), and an () = an (n n ).

It follows that

( + ) = + , and () =

and so is a homomorphism. If (n +P n )nN D(P ), then we dene a mapping


by an = an n , for every n N. It is easy to see that is an endomorphism of
A, and = (n + P n )nN . Hence, Im = D(P ). Obviously, Ker = 0, and
so it follows that
EndD (A) = Im = D(P ).
Suppose that (n + P n )nN U (D(P )). Then there is (n + P n )nN

D(P ) such that

(n + P n )nN (n + P n )nN = (1 + P n )nN .

Thus, n n + P n = 1 + P n , for every n N, and, in particular 1  P .


Conversely, suppose we are given some (n +P n )nN D(P ) such that 1 
P . Since P is a maximal ideal of D, P + 1 D = D. We have n + P m = m + P m ,
for every pair m n, and then n P for every n N. Again P + n D = D. By
Lemma 6.7, n D + P n = D. Hence, there exist some n D, and n P n such
that n n + n = 1. Therefore,

1 + P n = n n + n + P n
= n n + P n = (n + P n )(n + P n ).

Since the inverse element is unique, n + P m = m + P m , for every pair m n,


and so (n + P n )nN D(P ). Hence,

U (D(P )) = {(n + P n )nN D(P ) | 1  P }.


Chapter 16. Quasinite modules 199

Suppose now we are given a non-zero element (n + P n )nN D(P ), and


some t N such that 1 , . . . , t P but t+1  P . Note that D(P )P t =
D(P )y t . Indeed, let (n + P n )nN D(P ), x P t . By Proposition 6.13,
P t = Dy t + P t+m , for every m N, and then there exist some zt,m D, and
wt,m P t+m that x = y t zt,m + wt,m . Then
((n + P n )nN )x = ((n + P n )x)nN
= (n x + P n )nN = (n (y t zt,m + wt,m ) + P n )nN
= (0, . . . , 0, t+1 (y t zt,1 + wt,1 ) + P t+1 , . . . , t+m (y t zt,m + wt,m ) + P t+m , . . .)
= (0, . . . , 0, t+1 y t zt,1 + P t+1 , . . . , t+m y t zt,m + P t+m , . . .)
because wt,m P t+m , for every m N. Since D(P )P t = D(P )y t ,
(n + P n )nN = ((n + P n )nN )y t ,
where (n + P n )nN U (D(P )). Since the D-module D(P ) is torsion-free (see
D.G. Northcott [207, 9.10, Proposition 14]), the expression
(n + P n )nN = ((n + P n )nN )y t ,
where (n + P n )nN U (D(P )), is unique. If (n + P n )nN is another non-zero
element of D(P ), again
(n + P n )nN = ((n + P n )nN )y m ,
where (n + P n )nN U (D(P )). Thus,
(n + P n )nN (n + P n )nN = ((n + P n )nN )(n + P n )nN y t+m = 0.
Consequently, D(P ) is an integral domain.
Finally, let I be a non-zero ideal of D(P ), and pick 0 = (n + P n )nN I.
Then
(n + P n )nN = ((n + P n )nN )y t ,
where (n + P n )nN U (D(P )). It follows that D(P )y t I. Note that
D(P )y t = D(P )P t = lim {D/P t+m | m N}

and D(P )/D(P )P t


= D/P t . By Proposition 6.13, the set of all ideals of D/P t
is
{D/P t , P/P t , . . . , P t1 /P t , 0},
hence, I/D(P )P t is isomorphic to P tk /P t for some k N. Therefore I =
D(P )P tk = D(P )y tk . This means that D(P ) is a principal ideal domain,
and the set of all non-zero ideals of D(P ) is equal to
{D(P )P n | n N},
and all has been proved. 
200 Chapter 16. Quasinite modules

Corollary 16.17 ([152]). Let D be a Dedekind domain that is not a eld, G a


group, and A a quasinite DG-module. If A is D-periodic, then G/CG (A) can be
embedded in GLn (D(P )) where n = dimD/P P,1 (A), and {P } = AssD (A).
Proof. By Lemma 16.14,
A = C1 Cn ,
ufer P -modules. It suces to apply Proposition 16.16. 
where C1 , . . . , Cn are Pr
Proposition 16.18 ([152]). Let D be a Dedekind domain, G a group, H a nor-
mal subgroup of G of nite index, {g1 , . . . , gt } a transversal to H in G, and A a
quasinite DG-module such that CG (A) = 1. If A is D-periodic, then A has a
DH-submodule B satisfying the following properties:
(i) Bgj is a quasinite DH-submodule for each 1 j t.
(ii) A = Bg1 + + Bgt .
1 1
(iii) H
H /(g 1 CH (B )g 1 ) H /(g t CH (B )g t ).
Proof. Let M be the set of DH-submodules E of A that are not nitely generated
as D-submodules. If U is a proper DG-submodule, then U is nitely generated
over D. Since U is D-periodic, by Theorem 16.13, U has a nite D-composition
series so that U certainly has a nite DG-composition series. It follows that A is an
artinian DG-module. By Theorem 5.2, A is an artinian DH-module as well. Hence,
M has a minimal element B. By the choice of B, every proper DH-submodule of
B is nitely generated over D. We claim that B is a quasinite DH-submodule.
Otherwise, B has a proper DH-submodule C such that B/C is a simple DH-
module. Put
C1 = CDG = Cg1 + + Cgt .
Thus, C1 is nitely generated as a D-submodule. Since A is a quasinite DG-
module, A has an ascending series of proper DG-submodules

C1 C2 Cn

such that 
A= Cn .
nN

It follows that 
B= (B Cn ).
nN

Since B/C is a simple DH-module, either (B/C) (Cn /C) = B/C or this inter-
section is zero. Since B is not nitely generated over D, the same is true for B/C.
Since each Cn is nitely generated as a D-submodule, (B/C) (Cn /C) = 0,
that is B Cn = C, which gives B = C, a contradiction. This contradiction
shows our claim. For every 1 i t, each mapping a agi , a A is a D-
isomorphism that applies isomorphically the DH-submodules among themselves.
Chapter 16. Quasinite modules 201

Therefore Bg1 , . . . , Bgt are quasinite DH-submodules. Since Bg1 + + Bgt is


a DG-submodule, which is not nitely generated over D,

A = Bg1 + + Bgt .

Since every CH (Bgi ) = gi1 CH (B)gi ,

g11 CH (B)g1 gt1 CH (B)gt = CH (A) = 1 ,

and, therefore, the embedding (3) is guaranteed by Remaks theorem. 


Lemma 16.19 ([152]). Let D be a Dedekind domain that is not a eld, G a locally
soluble group, and A a quasinite DG-module such that CG (A) = 1. If A is
D-periodic, then G is abelian-by-nite.
Proof. By Corollary 16.17, AssD (A) = {P }, where P Spec(D), and

A = C1 Cn ,

ufer P -module, 1 j n. Let R = D(P ), K be a eld


where C1 , . . . , Cn are Pr
of fraction for R, and let R 0 denote the R-module K/R. Put A = HomR (A, R0 ).
Then A is a free R-module, and
 r0 (A ) = n (B. Hartley [106, Lemma 1.2]).

By [106, Lemma 2.1] B = (A ) R K is a simple KG-module. By Zassenhauss
theorem (see, for example, B.A.F. Wehrfritz [280, Theorem 3.7]) G is soluble, and
Maltsevs theorem (see, for example, B.A.F. Wehrfritz [280, Lemma 3.5]), G is
abelian-by-nite. 
Lemma 16.20 ([152]). Let D be a Dedekind domain that is not a eld, suppose that
G is a group having an abelian normal subgroup U of nite index, and let A be
a quasinite DG-module with CG (A) = 1. If A is D-periodic, then the periodic
part T of U has nite special rank.
Proof. By Corollary 16.17, AssD (A) = {P }, where P Spec(D), and

A = C1 Cn ,

ufer P -modules, 1 j n. By Proposition 16.18, A


where C1 , . . . , Cn are Pr
includes a quasinite DU -submodule B satisfying the conditions (i)-(iii). Let
R = D(P ), K be the eld of fractions for R, and let R0 denote the R-module
K/R. Put B* = HomR (B, R0 ). According to B. Hartley [106, Lemma 2.1], C =
(B*) R K is a simple KU -module, dimK C = r0 (B) = n. Then the periodic part
of U/CU (B) is locally cyclic (see, for example, L.A. Kurdachenko, J. Otal and
I.Ya. Subbotin [157, Theorem 2.3]). By Proposition 16.18,

U
U/(g11 CU (B)g1 ) U/(gt1 CU (B)gt ),

where {g1 , . . . , gt } is a transversal to U in G, so that, T is a subgroup of nite


special rank. 
202 Chapter 16. Quasinite modules

Lemma 16.21 ([152]). Let R be a ring, G a group, and A a quasinite RG-module


such that CG (A) = 1. If 1 = x (G), then A = A(x 1).
Proof. Since the mapping : a a(x 1), a A is an RG-endomorphism of
A, Im = A(x 1), and Ker = CA (x) are RG-submodules. Since x  CG (A),
CA (x) = A. It follows that CA (x) is nitely generated over R. Since

A(x 1)
= A/CA (x),

A(x 1) cannot be nitely generated over R. This means that A = A(x 1), as
required. 
Corollary 16.22 ([152]). Let D be a Dedekind domain of characteristic p > 0, G
a group, and A a quasinite DG-module such that CG (A) = 1. Then G has no
non-identity normal nite p-subgroups.
Proof. Suppose the contrary, and let P a non-identity normal nite p-subgroup
of G. Let S be a minimal G-invariant subgroup of P . Then S is abelian. If H =
CG (S), then |G : H| is nite, and S (H). By Proposition 16.18, A has a
quasinite DH-submodule B satisfying the conditions (1)(3) of that statement.
We claim that CH (B) does not include S. For, otherwise S CH (B), so S =
g 1 Sg g 1 CH (B)g = CH (Bg) for every g G, and, by above (2), S CG (A) =
1, a contradiction that shows our claim. Pick x S \CH (B) of prime order. Since
char D = p, the additive group of B is an elementary abelian p-subgroup, and
it follows that the natural semidirect product B  x is a nilpotent p-group. In
particular, [B, x] = B(x 1) = B. By Lemma 16.21, B(x 1) = B, and, then we
have just found a nal contradiction that proves the result. 
Theorem 16.23 ([152]). Let D be a Dedekind domain that is not a eld, G a locally
soluble group, and A a quasinite DG-module such that CG (A) = 1. Suppose that
A is D-periodic, then the following assertions hold:
(1) AssD (A) = {P }, where P Spec(D).
(2) A = C1 Cn , where C1 , . . . , Cn are Pr
ufer P -modules.
(3) G has an abelian normal subgroup U of nite index.
(4) The periodic part of U has nite special rank.
(5) If char D = p > 0, then Op (G) = 1.
Proof. (1) and (2) follow from Lemma 16.14, (3) from Lemma 16.19 and (4) from
Lemma 16.20. Suppose that Op (G) = 1. By (3) and (4), Op (G) has nite special
rank, hence, Op (G) has a non-identity nite G-invariant subgroup. But this is a
contradiction to Corollary 16.22. 
Suppose that A is a quasinite DG-module. As we mentioned above, the
study of the structure of A falls into two cases:
Chapter 16. Quasinite modules 203

(1) either A is D-periodic and D-divisible, or


(2) there exists some P Spec(D) such that AnnD (A) = P .
Note that we already dealt with case (1) in Theorem 16.23. Therefore, we
focus on the case (2), that is we consider a quasinite DG-module A such that
AnnD (A) = P is a maximal ideal. If F = D/P is the corresponding residue eld,
A can be naturally made into an F G-module; so we need to study quasinite
F G-modules, where F is a eld.
Lemma 16.24 (L.A. Kurdachenko and I.Ya. Subbotin [172]). Let A be a quasinite
F G-module, where F is a eld, and G is a group. Then SocF G (A) is a non-zero
proper submodule of A.
Proof. Since A is not simple, A has a non-zero proper F G-submodule B. Then
dimF B is nite, and therefore B has a non-zero simple F G-submodule S. It follows
that SocF G (A) = 0. It suces to apply Corollary 16.2. 

Lemma 16.25 ([172]). Let F be a eld, G a group, and A a quasinite DG-module


such that CG (A) = 1. If H is a non-identity nite normal subgroup of G, then
SocF H (A) = A.
Proof. Suppose the contrary, that is, B =SocF H (A) = A. Since H is normal, B
is an F G-submodule. It follows that dimF B is nite. Then there exists a vector
F -subspace C such that
A = B C.
Dene 
C0 = Ch.
hH

Since
A/Ch = Ah/Ch
= A/C,
dimF (A/Ch) = dimF (A/C) = dimF B is nite. It follows that C0 has nite
codimension, and, in particular, C0 = 0. Pick 0 = c1 C0 , and put C1 = c1 F H.
Since H is nite, dimF C1 is nite. Therefore C1 has a non-zero simple F H-
submodule C2 . We have C2 SocF H (A) = B, and, on other hand, C2 C0 and
C0 B = 0, a contradiction. 

Corollary 16.26 ([172]). Let F be a eld, G a group, and A a quasinite DG-


module such that CG (A) = 1. If H is a non-identity nite normal subgroup of
G, then CA (H) = 0, and A(F H) = A.
Proof. By Lemma 16.25,

A= Cn ,
nN
204 Chapter 16. Quasinite modules

where every Cn is a simple F H-submodule. Since Cn (F H) is an F H-submodule


of Cn , either Cn (F H) = Cn or Cn (F H) = 0. Therefore

A = CA (H) A(F H).

By Lemma 16.1, CA (H) = 0, and so A(F H) = A. 


Corollary 16.27 ([172]). Let F be a eld, G a group, and A a quasinite F G-
module such that CG (A) = 1. If H is a non-identity nite normal subgroup of
G, and B is a non-zero proper F G-submodule of A, then CH (B) = 1.
Corollary 16.28 ([172]). Let F be a eld, G a group, and A a quasinite F G-
module such that CG (A) = 1. Suppose that H is a non-identity normal subgroup
of G that has an ascending series of G-invariant subgroups

1 = H0 H1 H H+1 H = H

with nite factors. If B is a non-zero proper F G-submodule of A, then CH (B) =


1.
Proof. We proceed by induction on . If = 1, then it suces to apply Corollary
16.27. Let > 1, and suppose we have H CH (B) = 1, for every ordinal < .
Put C = H CH (B). If is a limit ordinal, then

H = H ,
<

so that  
C = (C H ) = C = 1 .
< <

Suppose now that is not a limit. Put L = H1 , and assume that C = 1.
Then L C = 1, and so

= C /(L C )
C = (L + C )/L H /L.

Thus, C is nite, which contradicts Corollary 16.26 since B CA (C ). Hence,


C = 1. By induction, matching = , we obtain CH (B) = 1, as required. 
Let G be a group. A normal subgroup H of G is said to be the hypernite
radical of G if it satises the following conditions:
(1) H possesses an ascending series of G-invariant subgroups

1 = H0 H1 H H+1 H = H,

every factor of which is nite.


(2) G/H has no non-identity normal nite subgroups.
Chapter 16. Quasinite modules 205

We denote the hypernite radical by HF (G).


Corollary 16.29 ([172]). Let F be a eld, G a group, and A a quasinite DG-
module such that CG (A) = 1. Then HF (G) is an abelian-by-nite subgroup of
nite special rank. Moreover, if char F = p > 0, then Op (HF (G)) = 1.
Proof. Put H = HF (G). Let B be a proper non-zero F G-submodule of A. By
Corollary 16.28, CH (B) = 1 so that we can think of H as a subgroup of
GL(n, F ), where n = dimF B. By a result due to Kargapolov (see B.A.F. Wehrfritz
[280, Corollary 9.31]), H is soluble-by-nite. By Maltsevs theorem (see B.A.F.
Wehrfritz [280, Theorem 3.6]) H has a normal subgroup S of nite index such
that g 1 Sg T (n, F1 ), where F1 is a nite eld extension of F . Put

U = (g 1 Sg) U T (n, F1 ), V = gU g 1 ,

so that V is normal in S. If char F = p > 0, then U T (n, F1 ) is a bounded nilpotent


p-subgroup. Suppose that U = 1. Then Op (H) = 1. By the denition of H, this
means that H has a nite non-identity G-invariant p-subgroup in contradiction
with Corollary 16.22. If char F = 0, then U T (n, F1 ) is a torsion-free nilpotent
subgroup, and hence, U = 1 in this case. Further,

T (n, F1 )/U T (n, F1 )


= U (F1 ) U (F1 )
 
n

Since the periodic subgroups of U (F1 ) are locally cyclic (see G. Karpilovsky [131,
Proposition 4.4.1]), then S is an abelian subgroup of nite special rank. 

Corollary 16.30 ([172]). Let F be a eld, G a hypercentral group, and A a quasi-


nite F G-module such that CG (A) = 1. Then the periodic part T of G is an
abelian-by-nite p -subgroup of nite special rank, where p = char F .
Proof. Indeed, in a hypercentral group, the periodic part is known to be equal to
the hypernite radical. 

Put S = SocF G (A). By Lemma 16.24, S is a proper non-zero F G-submodule


of A. In particular, dimF S = s is nite. If CG (S) = 1, then we think of G as a
subgroup of GLs (F ) (in fact, it is a completely reducible subgroup of GLs (F )).
Lemma 16.31 ([152]). Let F be a eld, G a locally soluble group, and A a quasinite
F G-module such that CG (A) = 1. If CG (SocF H (A)) = 1, then G is abelian-by-
nite.
Proof. As we mentioned above, G is a completely reducible subgroup of GLs (F ),
where s = dimF SocF G (A)). Again, by Zassenhauss theorem (see B.A.F. Wehrfritz
[280, Theorem 3.7]), G is soluble, and, by Maltsevs theorem (see B.A.F. Wehrfritz
[280, Lemma 3.5]), G is abelian-by-nite. 
206 Chapter 16. Quasinite modules

Lemma 16.32 ([152]). Let F be a eld, G an abelian group, and A a quasinite


F G-module such that CG (A) = 1. Given 1 = x G, we consider A as a D(x)-
module, where D(x) = F tx  is the group algebra of an innite cyclic group tx 
over F whose action is induced by atx = ax, for every a A. Then there exists
some x G such that AnnD(x) (A) = 0.
Proof. We recall that D(x) is a principal ideal domain. By Lemma 16.15, A is
a D(x)-periodic module, and, by Lemma 16.14, either AnnD(x) (A) = 0 or A is
D(x)-divisible. Suppose that AnnD(x) (A) = 0 for every x G. Put

S1 = SocF G (A), and S2 /S1 = SocF G (A/S1 ).

If 0 = a S2 , then aF G S1 = 0. Hence, there exists some u F G such that


0 = au S1 . We write
u = a1 x1 + + am xm ,
where a1 , . . . , am F , and x1 , . . . , xm G. By our assumption, there exists P1
Spec(D(x1 )) such that AP1 = 0. Put F1 = D(x1 ))/P1 , and think of A as an F1 G-
module. Since A has a non-zero annihilator in D(x1 ), A has a non-zero annihilator
P2 in F1 t2 , where t2 = tx2 . By Lemma 16.14, P2 Spec(F1 t2 ), which means
that the F x1 , x2 -module S2 is semisimple, and homogeneous. Using the same
arguments, after nitely many steps we obtain that the F x1 , . . . , xm -module
S2 is semisimple and homogeneous. But in this case aF x1 , . . . , xm  is a simple
submodule so that aF x1 , . . . , xm  S1 = 0, which contradicts the choice of a.
This contradiction proves the result. 

Theorem 16.33 ([152]). Let F be a eld, G a locally soluble group, and A a quasi-
nite F G-module such that CG (A) = 1. Put S =SocF G (A). If CG (S) = 1, then
the following assertions hold:
(1) G has an abelian normal subgroup U of nite index.
(2) The periodic part of U has nite special rank
(3) If char F = p > 0, then Op (G) = 1.
(4) U contains an element x of innite order, and A has a quasinite F U -
submodule B such that AssF
x (B) = {P }, for some P Spec(F x), and

B = C1 Cn ,

where C1 , . . . , Cn are Pr
ufer P -modules.
(5) A = B Bg1 Bgt , where {1, g1 , . . . , gt } is a transversal to U in G.
Proof. The assertion (1) follows from Lemma 16.31, while (2) and (3) follow from
Corollary 16.29. Let B be a quasinite F U -submodule satisfying the conditions
(1)-(3) of Proposition 16.18. By Lemma 16.32, U \ CU (B) contains an element
Chapter 16. Quasinite modules 207

x such that AnnF


x (B) = 0. Clearly |x| is innite. By Lemma 16.14, B is a
P -module, for some P Spec(F x), and

B = C1 Cn ,

where C1 , . . . , Cn are Pr
ufer P-modules. Finally, (5) follows from Proposition
16.18. 
Lemma 16.34 ([152]). Let F be a eld, G a hypercentral group, and A a quasinite
F G-module such that CG (A) = 1. If CG (SocF G (A)) = 1, then CG (SocF G (A))
(G) contains an element x of innite order such that A is F x-periodic, and
AssF
x (A) = {(x 1)F x}.
Proof. Put S1 = SocF G (A), and Z = CG (S1 ) (G). Then Z = 1, and, by
Corollary 16.27, Z is torsion-free. Pick 1 = x Z. Then the mapping : a
a(x 1), a A is an F G-endomorphism of A, and so Ker = CA (x), and
Im = A(x 1) are F G-submodules. Put S2 /S1 = SocF G (A/S1 ). By Lemma
16.3, S2 /S1 is a proper F G-submodule, and it follows that

S2 /S1 = E1 /S1 Ek /S1 ,

where E1 /S1 , . . . , Ek /S1 are simple F G-submodules. For every 1 j k, since

Ej
= Ej / Ker = Ej /S1 ,

Ej is simple so that Ej S1 . Thus, S2 (x 1) S1 . Put

S3 /S2 = SocF G (A/S2 ), . . . , Sn+1 /Sn = SocF G (A/Sn ), . . . .



Similarly, Sn+1 (x 1) Sn , for every n N. Since dimF ( nN Sn ) is innite, we
deduce 
Sn = A.
nN

It follows that A is F x-periodic. Moreover, AssF


x (B) = {(x 1)F x}. 
Theorem 16.35 ([152]). Let F be a eld, G a hypercentral group, and A a quasinite
F G-module such that CG (A) = 1. Put S =SocF G (A). If CG (S) = 1, then the
following assertions hold:
(1) G is abelian-by-nite.
(2) The periodic part T of a group G is a p -group of nite special rank where
p = char F .
(3) T (G) is locally cyclic.
(4) CG (S) (G) contains an element x of innite order such that A is F x-
periodic, and AssF
x (A) = {P } where P = (x 1)F x.
208 Chapter 16. Quasinite modules

(5) A = C1 Cn , where C1 , . . . , Cn are Pr


ufer P -modules.
Proof. By Lemma 16.34, CG (S) (G) contains an element x of innite order
such that A is F x-periodic, and AssF
x (B) = {P }, where P = (x 1)F x. If
D = F x, then D is a principal ideal domain, and it suces to apply Theorem
16.23. 
We note that this result can be extended to F C-hypercentral groups.
To nish, we are dealing with results of D.I. Zaitsev that concern quasinite
F G-modules, that are carried out when F is a nite eld. In this case, it is possible
to obtain more information concerning the periodic subgroups of the group G
considered.
Lemma 16.36 (D.I. Zaitsev [299]). Let F be a nite eld, G a group, and A a
quasinite F G-module such that CG (A) = 1. If (G) is innite, then G has a
nitely generated subgroup H such that A is a quasinite F H-module.
Proof. Let T be the periodic part of Z = (G), and suppose that B is a proper
F G-submodule of A. Since F is nite, B is likewise nite. By Corollary 16.28,
CT (B) = 1. If follows that T is nite, and hence, (G) contains an element x of
innite order. Think of A as a DG-module, where D = F x. By Lemma 16.15
and Lemma 16.14,
A = A1 An ,
where A1 , . . . , An are Pr  for some P Spec(F x), 1 j n. Put
ufer P -modules
Cm = P,m (A), m N. Then A = mN Cm . Besides, Cm is nite for any m.
Let G be the family of all nitely generated subgroups, containing the element
x. Since A is an artinian F x-module, then A is an artinian F K-module for every
subgroup K G. Therefore A includes a quasinite F K-submodule A(K). Clearly,
C1 A(K) = 0. If B is an F -subspace of A, then put

G(B) = {K G | C1 A(K) = B}.

Since C1 is nite, there is a nite F -subspace B1 such that G(B1 ) is a local system
for G. Since A is a P -module, C2 A(K) = B1 , K G(B1 ). There is a nite F -
subspace B2 such that G(B2 ) is a local system for G. Proceeding in this way, we
construct a strictly ascending chain of F -subspaces

B1 < B2 < < Bj < ,

and a descending chain of local systems for G

G1 G2 Gj

that Gj = G(Bj ) (that is, Bj = Cj A(K) for every K Gj , j N). Let


such 
B = jN Bj , and b B, g G. Then b Bj for some i N. Since Gj is a local
system for G, g K for some subgroup K Gj . Since A(K) is an F K-submodule,
Chapter 16. Quasinite modules 209

Bj = Cj A(K) is an F K-submodule too. It follows that bg Bj B, i.e. B is


an F G - submodule. Since B is innite, then B = A. Thus, C1 Bj for some j
N. If K G j , then

C1 A(K) = (C1 Cj ) A(K)


= C1 (Cj A(K)) = C1 Bj = C1 .

On the other hand, K Gj G1 . Thus, C1 A(K) = B1 ; that is, C1 B1 . Hence

P,1 (A) = C1 = B1 A(K)

for every K G1 . By Lemmas 16.15 and 16.14 A(K) is a D-divisible P -module,


so that the equation P,1 (A) = P,1 (A(K)) implies A(K) = A. Thus, we may
choose every subgroup from the family G1 as a candidate for H . 
Theorem 16.37 ([299]). Let F be a nite eld, G a group, and A a quasinite F G-
module such that CG (A) = 1. If (G) is innite, then the following assertions
hold:
(1) If S is a periodic normal subgroup of G, then S is nite.
(2) If S is a periodic subgroup of G, then S has a bounded nilpotent p-subgroup
of nite index, where p = char F .
Proof. As in the previous result, we deduce that there exists some x (G) of
innite order such that
A = C1 Cn ,
ufer P -modules for some P Spec(F x). Together
where C1 , . . . , Cn are Pr
with Proposition 16.16 this implies that G is isomorphic with some subgroups of
GLn (D(P )). Since D(P ) is a principal ideal domain, there exists the eld of
fractions K for D(P ), i.e. G can be embedded in GL(n, K). By Schurs theorem
(see, for example, B.A.F. Wehrfritz [280, Theorem 9.1]) every periodic subgroup
of G is locally nite.
Let S be a periodic normal subgroup of G, and dene H = S x. Since
A is an artinian F x-module, A is an artinian F H-module, and so A has a
quasinite F H-submodule B. Suppose that xt CH (B) for some t = 0. Then
B CA (xt ), and, in particular, CA (xt ) is innite. Note that CA (xt ) is an F G-
submodule, because xt (G). This means that A = CA (xt ), and then xt
CG (A) = 1. Since |x| is innite, this is a contradiction that shows x CH (B) =
1. Thus, CH (B) S. Since CH (B) is normal in H, H = H/CH (B) is innite.
By Lemma 16.36, there exists a nitely generated subgroup Q such that x Q,
and B is a quasinite F (Q )-module, where Q = QCH (B)/CH (B). Thus, Q/ x
is a nitely generated locally nite group, and so it is nite. Thus, Q = x U ,
where U = Q S is nite. Obviously,

B1 = CB (x) = 0 .
210 Chapter 16. Quasinite modules

Since U = U CH (B)/CH (B) is a nite normal subgroup of Q , we may apply


Corollary 16.27 to the quasinite F (Q )-module B, and its submodule B1 to
obtain
U CH (B1 ) = U CQ (B1 ) = 1 .
In other words,

CH (B) = U CH (B) CH (B1 ) = CH (B)(U CH (B1 )) = CH (B)CU (B1 ),

so that CU (B1 ) CH (B). Since this remains true for every nite subgroup V U ,
CU (B1 ) CH (B). Since B1 is a nite F H-submodule, S/CS (B1 ) is nite, and
the inclusion CH (B) S, gives further that S/CH (B) is nite as well. By the
niteness of P,1 (A), and P,1 (B), there are elements g1 , . . . , gt such that

P,1 (A) = P,1 (B)g1 + + P,1 (B)gt .

Thus,
E = Bg1 + + Bgt
is an F x-divisible P -module since so is P , and we have P,1 (E) = P,1 (A).
Since A is an F x-divisible P -module, E = A. Since S is a normal subgroup of
G, |S : CH (B)| = |S : CH (Bgj )| for every 1 j n. It follows that

|S : (CH (Bg1 ) CH (Bgt ))|

is nite. Since
CH (Bg1 ) CH(Bgt ) CG (A) = 1 ,
S is nite.
Now let S be a periodic subgroup of G. Again we put H = S x. Let B1 be
a quasinite F H-submodule of A, B2 /B1 a quasinite F H-submodule of A/B1 ,
and so on. Proceeding as above, we show that the series

0 = B0 B1 B2 Bn .

must be nite, that is there exists some n N such that Bn = A. For every
1 j n, we dene Hj = CH (Bj /Bj1 ). As above, we nd that Hj S. By (1),
every periodic normal subgroup of H/Hj is nite. In particular, S/Hj is nite.
Dene
S 0 = H1 Hn
so that S/S0 is nite, and moreover S0 acts trivially on all factors of the series

0 = B0 B1 Bn = A.

It suces to apply, for example, O.H. Kegel and B.A.F. Wehrfritz [134, Theorem
1.C.1], to get that S0 is a bounded nilpotent p-subgroup, as required. 
Chapter 17

Some applications: splitting over the


locally nilpotent residual

We recall that if G is a group and N is a normal subgroup of G, it is said that


G splits over N if there exists a subgroup H of G such that G = N  H (that is,
G = N H, and N H = 1). Such an H is said to be a complement to N in G. If
all complements to N are conjugate, then it is said that G conjugately splits over
N ; the results involving this are known in general as splitting theorems.
In the theory of nite soluble groups, splitting theorems play a very signicant
role. We mention here the result of W. Gasch utz [83] and E. Schenkman [253],
asserting that if the nilpotent residual L of the nite group G is abelian, then
G conjugately splits over L, is one of the most important achievements in this
direction. Here it is decisive that the niteness of G implies the existence of the

Z-decomposition in L, that is, L = Z E, where Z = ZG (L), and E = IE ZG (L).
By the choice of L, we have L = IE ZG (L), and, in particular, L = [L, G], and
L (G) = 1. The latter two conditions are the most common restrictions
involved in almost all splitting theorems.
Since the condition Min-G appears as an extension of the niteness, we are
naturally drawn to the following situation: the locally nilpotent residual L of the
group G is abelian and is an artinian ZG-module. The theorems about splitting
of a group over its abelian generalized nilpotent radical are very useful in many
dierent investigations. Among these results we choose a theorem proven by D.J.S.
Robinson [246]. In [246] the authors proof is based on homological methods. In
this chapter we decided to give a group theoretical proof of this theorem. The
following proof is based on the ideas and results of D.I. Zaitsev [304, 307, 308] and
M.J. Tomkinson [275]. This fact aptly illustrates the eectiveness of the artinian
condition.
Let G be a group. Suppose that H1 , . . . , Hn are normal subgroups of G and
put
H = H1 Hn .
212 Chapter 17. Splitting over the locally nilpotent residual

By Remaks theorem, the mapping g (gH1 , . . . , gHn ), g G, denes a homo-


morphism
: G
G/H1 G/Hn .
In particular, G/H is isomorphic to some subgroup of G/H1 G/Hn . If
Im = G/H1 G/Hn , we will write

G/H = G/H1 G/Hn .

Lemma 17.1 (M.J. Tomkinson [275]). Let G be a group, H a normal subgroup of G


having nite index and T = {g1 , . . . , gn } a transversal to H in G. Suppose that A
is a ZG-module and B is a proper ZH-submodule of A. For each set J of subsets
of T , form

B(J ) = ( Bg)
SJ gS

and choose B(J0 ) to be a maximal member of the set

{B(J ) | B B(J ) and B(J ) = A}.

Then 
B0 = B(J0 )g
gG

is a ZG-submodule of A and

A/B0 = A/B(J0 )g,
gS

for some subset S of T .


Proof. Clearly, we may write

B0 = B(J0 )g1 B(J0 )gn .

We show, by induction on m, that

A/(B(J0 )g1 B(J0 )gm )

is a direct sum of certain A/B(J0 )g. So we assume that



A/(B(J0 )g1 B(J0 )gm1 ) = A/B(J0 )g
gM

for some subset M of T . If B(J0 )gm B(J0 )g1 B(J0 )gm1 , then

A/(B(J0 )g1 B(J0 )gm1 ) = A/B(J0 )g.
gM
Chapter 17. Splitting over the locally nilpotent residual 213

Otherwise, by maximality of B(J0 ) and, consequently, of B(J0 )gm , we have

B(J0 )gm + (B(J0 )g1 B(J0 )gm1 ) = A,

and so 
A/B0 = A/B(J0 )g,
gS

where S = M {gm }. 
Lemma 17.2 ([275]). Let G be a group, H a non-identity normal subgroup of G
and A a non-zero ZG-module. Assume further that the following conditions hold:
(i) HCG (A)/CG (A) F C (G/CG (A)).
(ii) A is an artinian ZG-module.
(iii) A = A(ZH).
Then G has a normal subgroup K and A has a proper ZG-submodule Q satisfying
the following conditions:
(1) K/CK (A/Q) is nitely generated.
(2) K/CK (A/Q) F C(G/CK (A/Q)) HCK (A/Q)/CK (A/Q).
(3) A/Q = (A/Q)(ZK).
Moreover, if H is locally soluble, then K/CK (A/Q) is the normal closure of a
single element of G/CK (A/Q), and is either a nite elementary abelian p-group
for some prime p or a free abelian group of nite 0-rank. Further, if H is locally
nilpotent, we can choose K so that K/CK (A/Q) (HCK (A/Q)/CK (A/Q)).
Proof. Since A is an artinian ZG-module, it suces to deduce the general case
from the case when the result holds for every proper ZG-submodule of A; that
is, we may assume that the latter is true. There is no loss if we assume that
CG (A) = 1. By Lemma 3.15, H F C(G) = 1. Pick 1 = x H F C(G), and
G
put F = x . Then F is a nitely generated subgroup; that is, F = x1 , . . . , xn .
If A = A(ZF ), then it suces to choose K = F and B = 0.
Therefore, we may suppose that A1 = A(ZF ) = A. Since CG (A) = 1,
A1 = 0. Put U = CG (F ) so that G/F is nite. By Theorem 5.2, A is an artinian
ZU -module. Let V = H U = CH (F ).
Suppose rst that A1 (ZV ) = A1 . Let : A A/A1 (ZV ) be the canon-
ical epimorphism. Since V CG (A1 ), A(y 1) A(ZF ) A1 for ev-
ery y F . Thus, V CG (A(y 1)). From the isomorphism A(y 1) =ZV
A/CA (y) we obtain the inclusion V CG (A/CA (y)) for every yF , and
therefore, V CG (A/CA (F )).
If we suppose that CA (F ) = A, then A(ZF ) A1 (ZV ), and so A1 =
A(ZF ) = A1 (ZV ) contrary to our assumption. Hence, the factor-module A =
A/CA (F )) is non-zero. The equation A = A(ZH) implies that A = A(ZV ),
214 Chapter 17. Splitting over the locally nilpotent residual

and A = A (ZV ). Also V CG (A ), and so H/CH (A ) is nite. Therefore, in


this case, we may choose K = H.
Suppose now that A1 (ZV ) = A1 . Since A1 is a proper ZG-submodule, by
our initial assumption, G has a normal subgroup L, and A1 has a proper ZG-
submodule C such that A1 /C = (A1 /C)(ZL), and L/CL (A1 /C) is a nitely
generated subgroup of F C(G/CL (A1 /C)) (V CL (A1 /C)/CL (A1 /C)). Let

D1 = A(x1 1), . . . , Dn = A(xn 1).

Then every Dj is a ZU -submodule, and

A1 = A(ZF ) = D1 + + Dn .

Let : A A/C be the canonical epimorphism. Then

A1 = D1 + + Dn .

Let k be the largest integer such that

A1 = Dk + + Dn .

Then the ZU -module A1 /(Dk+1 + + Dn ) is non-zero, and

A1 /(Dk+1 + + Dn )
=ZU Dk /(Dk (Dk+1 + + Dn ).

Since A1 = A1 (ZL),

(Dk /(Dk (Dk+1 + + Dn ))(ZL)


= Dk /(Dk (Dk+1 + + Dn ).

This means that Dk has a non-zero ZU -image Dk such that Dk (ZL) = Dk .


Using the isomorphism
Dk = A(xk 1)
=ZU A/CA (xk )
we see that A has a proper ZU -submodule B such that Dk
=ZU A/B, and so
(A/B)(ZL) = A/B. Now we form the ZG-submodule B0 as we did in Lemma
17.1. Then

B0 = B(J0 )g, B(J0 ) B, and
gG

A/B0 = A/B(J0 )g,
gS

where S is a subset of a transversal T to U in G. Since (A/B)(ZL) = A/B,


A(ZL) + B = A, and so A(ZL) + Bg = A for every g G. From B(J0 )g
Bg, (A/B(J0 )g)ZL) = A/B(J0 )g, and, since A/B0 is the direct sum of certain
Chapter 17. Splitting over the locally nilpotent residual 215

modules A/B(J0 )g, we have A/B0 = (A/B0 )(ZL). Put Q = B0 so that A/Q =
(A/Q)(ZL).
Let : A A/Q be the canonical epimorphism. Since CK (A1 ) acts triv-
ially on A1 , so does it on Dk . From the isomorphism Dk
=ZU A/B we see that
CK (A1 ) acts trivially on A/B. It follows that CK (A1 ) acts trivially on every
A/B(J0 )g, and so on A = A/B0 .
Since CL (A) = CL (A/B0 ) CL (A1 ), and L/CL (A1 ) is a nitely gener-
ated subgroup of F C(G/CL (A1 ))(V CL (A1 )/CL (A1 )), we see that L/CL (A)
is a nitely generated subgroup of F C(G/CL (A))(V CL (A)/CL (A)). The sub-
group L has a nite series of G-invariant subgroups

CL (A) = L0 L1 Ls = L

such that Lj+1 /Lj is the normal closure of a single element, and is either a nite
abelian group or a free abelian group of nite 0-rank, 1 j s 1. If H is locally
nilpotent, we will choose a series like this in such a way that every factor Lj+1 /Lj
is H-central. Choose m minimal such that A/Q = (A/Q)(ZLm ). Let

: A/Q (A/Q)/(A/Q)(ZLm )

be the canonical epimorphism, and let be the composition of and . Put R =


Lm . If R/CR (A) is innite, then there exists an integer t such that K/CR (A) =
(R/CR (A))t is free abelian. If A = (A)(ZK), then we are done. If A =
(A)(ZK), then we consider (A)/(A)(ZK). We have

(A)/(A)(ZK) = ((A)/(A)(ZK))(ZR), and


CR ((A)/(A)(ZK)) K,

so that R/CR ((A)/(A)(ZK)) is nite. In other words, it suces to consider


the following case: G has a normal subgroup L, and A has a proper ZG-submodule
Q satisfying that L/CL (A/Q) is a nite abelian group, and A/Q = (A/Q)(ZL).
In this case, L has a nite series of G-invariant subgroups

CL (A) = L0 L1 Lr = L

such that Lj+1 /Lj is the normal closure of a single element, and is an elemen-
tary abelian group, 1 j r 1. Choosing an  minimal such that A/Q =
(A/Q)(ZL ) we can dene (A/Q)/(A/Q)(ZL ) to be the required ZG-image of
A, and put K = L . 
Lemma 17.3. Let G be a locally nilpotent group, and let A be a ZG-module. Suppose
that for every nite subset M of A, and every nitely generated subgroup K of G
we have that every K-chief factor of M ZK is K-central. Then CA (K) = 0.
Proof. Let R be a ZG-chief series of A, and consider an arbitrary factor C of
this series. Then C is a simple ZG-module, and so it follows that either C is
216 Chapter 17. Splitting over the locally nilpotent residual

an elementary abelian p-group for some prime p or C is Z-divisible. We consider


the natural semidirect product L = C  K. If B = M ZK, then D = B  K
is nitely generated soluble. Let H be an arbitrary normal subgroup of D of
nite index. Since every D-chief factor of BH/H is central in D, and D/BH is
nilpotent, D/H is nilpotent. By a result due to D.J.S. Robinson (see [235, Theorem
10.51]), D is likewise nilpotent. Elementary properties of nilpotent groups give that
B (D) = 1. It follows that CA (K) = 0. 
Lemma 17.4. Let G be a group, H a normal subgroup of G, and A a ZG-module.
Assume that the following conditions hold:
(i) The natural semidirect product A  H is locally nilpotent.
(ii) HCG (A)/CG (A) F C (G/CG (A)).
(iii) H is locally nilpotent.
Then every chief ZG-factor of A is H-central.
Proof. We suppose the contrary; that is, there exists a ZG-chief factor W of A
such that W is H-eccentric. Let

: A A/CG (W )

be the canonical epimorphism. By Lemma 3.15, (H) F C(G) = 1. Pick


G
1 = x (H) F C(G), and put Q = xv . Then Q is a nitely generated
abelian subgroup. By the choice of Q, the index |G : CG (Q)| is nite. Put
X = CG (Q). Then Q (G). By Theorem 5.5,

W = W1 Wm

for some simple Z(X)-submodules W1 , . . . , Wm . Since Q is nitely generated, by


(i), it is not hard to see that CW1 (Q), . . . , CWm (Q) = 1. Since every CWj (Q)
is a Z(X)-submodule of Wj , CWj (Q) = Wj . This holds for every 1 j m, so
that we conclude that either CW (Q) = W or Q CG (W ), a contradiction. This
contradiction shows that H CG (W ), as required. 
Lemma 17.5. Let G be a group, H and L normal subgroups of G, and A a ZG-
module. Suppose that H L, and A has a ZL-submodule B such that the natural
semidirect product B  H is a locally nilpotent
group. Then for every nite subset
X of G the natural semidirect product ( xX Bx)  H is locally nilpotent.
Proof. Since L is normal in G, Bx is a ZG-submodule of G. Let F be an arbi-
trary nitely generated subgroup of H. Put K = F x | x, x1 X , so that K
is likewise nitely generated. Let M be an arbitrary nite subset of B, and put
D = M ZK. Since B  H is locally nilpotent, D has a nite upper ZK-central
series
0 = D0 D1 Dk = D.
Chapter 17. Splitting over the locally nilpotent residual 217

By the construction of K, the series

0 = D0 x D1 x Dk x = Dx

consists of ZK-submodules. Let g K. Then xgx1 = y K; that is, xg = yx. If


1 j k, we have

Dj x(g 1) = Dj (xg x) = Dj (yx x) = Dj (y 1)x Dj1 x.

This shows that {Dj x | 0 j k} is an upper ZK-central series of Dx. Moreover,

0 = (D0 + D0 x) (D1 + D1 x) (Dk + Dk x) = D + Dx

is an upper ZK-central
series of D + Dx. Proceeding
by induction on |X|, we
obtain that ( xX Dx) is ZK-nilpotent. Then ( xX Dx) is ZF -nilpotent, and
it follows that ( xX Bx)  H is locally nilpotent. 
Lemma 17.6. Let G be a group, H a normal subgroup of G, and A a ZG-module.
Assume that the following conditions hold:
(i) A has an artinian ZG-submodule B.
(ii) Every chief ZG-factor of B is H-eccentric.
(iii) The natural semidirect product (A/B)  H is a locally nilpotent group.
(iv) H F C (G).
(v) CG (A) = 1.
Then A includes a ZG-submodule C such that A = B C.
Proof. Suppose the contrary; that is, every ZG-submodule X such that A = B +X
satises that B X = 0. Choose in A a ZG-submodule M maximal under
B M = 0. Replacing M by A/M , we may assume that M = 0. This means
that every non-zero ZG-submodule of A has a non-zero intersection with B. Put
M = {U | U is a ZG-submodule such that B does not include U },
S = {U B | U M}.
Clearly, S = . Since B is an artinian ZG-module, S has a minimal element L.
Let D be a ZG-submodule of A such that D B = L. For every non-zero ZG-
submodule E of D such that E M we have B E = 0. By the choice of D,
it follows that E B = D B. If we suppose that E = (E B) + V for some
proper non-zero ZG-submodule V of E with the property V M, then, from the
inclusion V B D B, we obtain V B = D B; that is, V = E.
Clearly, there is no loss if we assume that CG (D) = 1. Suppose that
CH (L) = 1. By Lemma 3.15, CH (L) (H) F C(G) = 1. Pick 1 = x
G
CH (L) (H) F C(G), and put Q = x . Then Q is a nitely generated sub-
group. By the choice of Q, the index |G : CG (Q)| is nite. Put X = CG (Q). Then
218 Chapter 17. Splitting over the locally nilpotent residual

the mapping : a a(x 1), a D, is a ZX-endomorphism of A. By the choice


of x, we have L Ker , so that D(x 1) =ZX D/ Ker . The inclusion H X
implies that the natural semidirect product (D/ Ker )  H is a locally nilpotent
group. Then the natural semidirect product D(x 1)  H is likewise a locally
nilpotent
group. Let T be a transversal to X in G. Since T is nite, by Lemma
17.5, ( tT D(x 1)t)  H is locally nilpotent. Note that tT D(x 1)t is a
ZG-submodule of D. By Lemma 17.4, every ZG-chief factor of tT D(x 1)t is
H-central. By (ii),

( A(x 1)t) B = 0 .


tT

By our assumption this means that 0 = tT D(x1)t, in particular, D(x1) =
0, and x CG (D) = 1. Thus, CH (L) = 1.
By Lemma 3.15, (H) F C(G) = 1. Pick 1 = y (H) F C(G), and
G
put Y = y . Then Y is a nitely generated abelian subgroup; that is, Y =
g1 , . . . , g2 . By the choice of Q, the index |G : CG (Y )| is nite. Put R = CG (Y ).
Since the natural semidirect product (D/L)  H is a locally nilpotent group,
every ZY -chief factor of an arbitrary nitely generated ZY -submodule of (D/L)
is Y -central. By Lemma 17.3, P/L = CD/L (Y ) = 0. We note that P is a ZG-
submodule, because Y is a normal subgroup of G. The mapping

: a a(g1 1), a P,

is a ZR-endomorphism of A. If P (g1 1)/L(g1 1) is non-zero, since it is a ZR-


image of D/L, the natural semidirect product P (g1 1)/L(g1 1)  H is a locally
nilpotent group (note also that H R). By Lemma 17.4, every ZR-chief factor
of P (g1 1)/L(g1 1) is H-central. On the other hand, P (g1 1) L B
by the choice of g1 . Since the index |G : R| is nite, it is not hard to see that
every ZR-chief factor of B is H-eccentric. In particular, every ZR-chief factor of
P (g1 1), hence, of P (g1 1)/L(g1 1), is H-eccentric. This contradiction shows
the equality P (g1 1) = L(g1 1). In other words,
P = CP (g1 ) + L.
Applying similar arguments to the ZR-submodule S = CP (g1 ), we obtain the
equality S = CS (g2 ) + (S L), which implies that
P = CP (g1 , g2 ) + L.
Proceeding in this way, after nitely many steps, we obtain

P = CP (Y ) + L.

But in a beginning of the proof we have already demonstrated that in this case,
P = CP (Y ). Since L P , CH (L) = 1, a contradiction. This nal contradiction
proves the required result. 
Chapter 17. Splitting over the locally nilpotent residual 219

Lemma 17.7. Let G be a group, and let H, B and A be normal subgroups of G.


Assume that the following conditions hold:
(i) A is abelian, and A B CH (A).
(ii) H/B F C (G/B).
(iii) H/A is locally nilpotent.
(iv) A is an artinian ZG-module.
(v) Every chief ZG-factor of A is H-eccentric.
Then G has a G-invariant subgroup L such that B = A L. Moreover, if S is a
subgroup of G such that G = AS, then L S.
Proof. Since A (B), B is locally nilpotent. Choose in H an arbitrary nitely
generated subgroup K, and put D = A K, and E = B K. Then
K/D = K/(A K)
= KA/A
is nitely generated nilpotent. Let U/V be a ZK-chief factor of D [E, E]. The
factor-group K/V is nitely generated abelian-by-nilpotent. By a result due to
P. Hall (see [235, Theorem 9.51]), K/V is residually nite. It follows that K
has a normal subgroup F of nite index such that F U = V . Consider now
the nite group K/F . Since (K/F )/(DF/F ) is locally nilpotent, DF/F contains
the nilpotent residual R/F of K/F . In particular, R/F is abelian. We have above
noticed that in this case, K/F = R/F S/F , for some nilpotent subgroup S/F (W.
Gasch utz [83], E. Schenkman [253]). Since R/F DF/F , and DF/F (EF/F ),
EF/F = R/F (S/F EF/F ).
Clearly, (S/F EF/F ) is a normal subgroup of K/F . The inclusion U [E, E]
implies that U F/F (S/F EF/F ). In particular, [R/F, U F/F ] = 1. Since
U F/F is a chief factor of K/F , the equation K/F = (R/F )(S/F ) implies that
U F/F is a chief factor of S/F . From the nilpotency of S/F , we obtain that U F/F
is central in S/F ; hence, in K/F . Therefore,
[K, U ] F U = V.
It follows that every ZK-chief factor of D [E, E] is K-central.
Let W be a ZG-chief factor of A [B, B], and suppose that W is not H-
central. Let
: A A/CG (W )
be the canonical epimorphism. By Lemma 3.15, HF C(G) = 1. Pick 1 = x
H F C(G), and put Q = xvG . Then Q is a nitely generated subgroup;
that is, Q = x1 v, . . . , xn v. By the choice of Q, the index |G : CG (Q)| is
nite. Put X = CG (Q). By Theorem 5.5,
W = W1 Wm
220 Chapter 17. Splitting over the locally nilpotent residual

for some simple Z(X)-submodules W1 , . . . , Wm . Since Q is nitely generated,


by the above paragraph, every Z(Q)-chief factor of Wj is Q-central, 1 j m.
By Lemma 17.3, CWj (Q) = 1. But CWj (Q) is a Z(X)-submodule of Wj , and
therefore, CWj (Q) = Wj . This holds for 1 j m, and so CW (Q) = W ; that
is, Q CG (W ), a contradiction. This contradiction shows that H CG (W );
hence, it establishes that every ZG-chief factor of A [B, B] is H-central. If we
suppose now that A [B, B] = 1, then we obtain a contradiction with condition
(v).
Consider now the factor-group G/[B, B]. Its normal subgroup B/[B, B] is
abelian. Since H/A[B, B] is locally nilpotent, by Lemma 17.6, B has a G-invariant
subgroup L [B, B] such that
B/[B, B] = (A[B, B]/[B, B]) L/[B, B].
This and A [B, B] = 1 give that B = A L.
Let S be a subgroup of G such that G = AS. Since A B, B = A(S B).
The subgroup B is normal in G, so that S B is normal in S. The inclusion
B CH (A), and the equation G = AS imply that S B is normal in G. Since
B/(B S) = A(B S)/(B S) = A/(A S)
every G-chief factor of B/(B S) is H-eccentric. On the other hand, we have seen
above that every G-chief factor of L is H-central. Since L B, it follows that
L(B S)/(B S) = 1; that is, L B S. 
Lemma 17.8 ([275]). Let G be a group, A a ZG-module. Suppose that (G) contains
an element g such that A is not Z g-nilpotent. Then A has a ZG-submodule B
such that (A/B)(g 1) = A/B, and CA (g) B.
Proof. There is a descending chain of ZG-submodules
A A(g 1) A(g 1)2 A(g 1)n .
Since A is artinian, there exists some m such that
A(g 1)m1 = A(g 1)m = A(g 1)m+1 = .
If j N, put Aj = A(g 1)j . Then
Am1 (g 1) = Am = Am (g 1).
It follows that Am1 CA (g) + Am , and therefore,
(Am1 + CA (g))/CA (g) = (Am + CA (g))/CA (g).
Since g (G), CA (g) is a ZG-submodule of A. Proceeding by ordinary induction,
after nitely many steps we nd a ZG-submodule B such that CA (g) B, and
A = B + Am . Thus,
(A/B)(g 1) = ((B + Am )/B)(g 1) = ((B + Am (g 1))/B
= ((B + Am )/B) = A/B,
as required. 
Chapter 17. Splitting over the locally nilpotent residual 221

Lemma 17.9. Let G be a group, and let A be a ZG-module. Suppose that (G)
contains an element g such that A = A(g 1). Let E be any extension of A by G.
If B is a ZG-submodule of A, and CA (g) B, then E/B splits over A/B.
Proof. Let y E such that g = yA. In this proof, we will denote A multiplicatively;
in particular, instead of A(y 1) we will write [A, y]. Let x be an arbitrary element
of E. We have y x = ay for some a A. Since A = [A, y], A = [A, y 1 ], and so
there exists some b A such that a = [b, y 1 ]. We have
y x = ay = [b, y 1 ]y = b1 yby 1 y = y b .
Therefore, xb1 CE (y), and x ACE (y). Since CE (y)A = CA (y) = CA (g), we
see that CE (y)/CA (y) is a complement to A/CA (y) in E/CA (y). But B CA (y);
so it is clear, that BCE (y)/B is a complement to A/B in E/B. 
Lemma 17.10 ([275]). Let G be a group, and let A be a normal subgroup of G.
Suppose that there exist G-invariant subgroups A1 , . . . , An of A such that
A1 An = 1 , and A = A/A1 A/An .
Then we have:
(1) If every factor A/Aj has a complement Sj /Aj in G/Aj , then A has a com-
plement S in G.
(2) If S and R are complements to A in G such that SAj is conjugate to RAj
for every 1 j n, then S and R are conjugate in G.
Proof. Put A1 At1 = B, and assume, inductively, that A/B has a com-
plement C/B in G/B. Consider C St . Certainly C St A = B At . Also,
(C St )A = (C St )BAt = (C St B)At
= (C St A)At = CAt = CBAt = CA = G.
Therefore, (C St )/(At B) is a complement of A/(A1 At ) in G/(A1
At ). Assume now inductively that SB and RB are conjugate. Replacing R
by an appropriate conjugate we may assume that SB = RB. There is an element
a A such that S a At = RAt . Since A = BAt we may assume that a B, and so
S a B = RB. Now
R(B At ) = R((R A)At B) = R(RAt B) = RAt RB.
So
(RAt B) = S a At SB = S a (At B);
hence R/(A1 At ) is conjugate to S/(A1 At ). 
Lemma 17.11 ([275]). Let G be a group, and let A be a ZG-module. Suppose that
(G) contains an element g such that A = A(g 1). Let E be a split extension of
A by G. If K is another complement to A in E, and B is a ZG-submodule of A
such that CA (g) B, then GB and KB are conjugate in E.
222 Chapter 17. Splitting over the locally nilpotent residual

Proof. It is clear that A g G = g, and A g K = ag for some element
a A. Thus, K CE (ag), and CE (ag) = KCA (g). Also CE (g) = GCA (g).
Since A = [A, g], A = [A, g 1 ], and so a = [b, g 1 ] for some element b A,
and we have ag = [b, g 1 ]g = g b . It follows that

KCA (g) = CE (ag) = CE (g b ) = (GB)b ,

as required. 
Theorem 17.12 (D.J.S. Robinson [246]). Let G be a group, H a subgroup of G,
and A a ZG-module. Suppose that the following conditions hold:
(i) A is an artinian ZG-module.
(ii) H is a normal subgroup of G.
(iii) H is locally nilpotent.
(iv) HCG (A)/CG (A) F C (G/CG (A)).
(v) CA (H) = 1.
Then every extension E of A by G splits over A, and all complements of A in
such an extension are conjugate.
Proof. We note that the condition (iv) implies that HCG (A)/CG (A) is hyper-
central. Repeating word by word the proof of Theorem 10.21, we obtain that A
has the Z-ZH-decomposition; that is,

A = ZH (A) ZH (A).

Condition (v) implies that ZH (A) = 0. Hence, every non-zero ZG-factor of A is
H-eccentric. In particular A = A(ZH). Since we consider A not only as a module
but as a normal subgroup of a group E, we will can write A multiplicatively. In
particular, instead of A(ZH) we will write [A, H].
Existence of complements
Suppose the contrary; that is, for every subgroup X such that G = AX,
A X = 0. Note that A X is a G-invariant subgroup of A. Let M be the set
of subgroups X such that G = AX, and put S = {X A | X M}. Clearly,
S = . Since A is an artinian ZG-module, S has a minimal element M . Let Y be a
subgroup of A such that A Y = M . If we suppose that Y = (A Y )Y1 for some
subgroup Y1 of Y , then G = AY1 , and therefore, 1 = A Y1 A Y , so, by the
choice of Y , A Y1 = A Y . This means that Y = Y1 . In other words, for every
proper subgroup Y1 of Y we have G = AY1 . Without loss of generality, we obtain
that E = Y , and A = M . By Lemma 17.2, G has a normal subgroup K, and A
has a proper G-invariant subgroup Q such that the following conditions hold:
(i) The factor-group (K/Q)/CK/Q (A/Q) is the normal closure of a single ele-
ment of (G/Q)/CK/Q (A/Q).
Chapter 17. Splitting over the locally nilpotent residual 223

(ii) (K/Q)/CK/Q (A/Q) is either a nite elementary abelian p-group for some
prime p or a free abelian group of nite 0-rank.
(iii) (K/Q)/CK/Q (A/Q) lies in the intersection of F C((G/Q)/CK/Q (A/Q)), and
((H/Q)CK/Q (A/Q)/CK/Q (A/Q)).
(iv) A/Q = [A/Q, K/Q].
Replacing E by E/Q, we may assume that Q = 1. By Lemma 17.7,
CK (A) = A B for some E-invariant subgroup B. We may replace E by E/B to
assume B = 1. Thus, CK (A) = A. Put
E
K = x = x1 , . . . , xn  .

It is not hard to see that if A is Z xj -nilpotent for every 1 j n, then A is


ZK-nilpotent. But in this case, A = [A, K]. Therefore, there exists some index j
such that A is not Z xj -nilpotent. Let U/A = CE/A (K/A). By Lemma 17.8, A
has a U -invariant subgroup B such that

CA (g) B, and [A/B, gB] = A/B.

Now we form the G-invariant subgroup B0 just as in Lemma 17.1. Then



B0 = B(J )g , B(J0 ) B CA (g),
gG

and A/B0 = DrgS A/B(J0 )g

for some subset S of a transversal T to U in G. Note also that

B(J0 )g B g CA (g g ) = CA (g).

By Lemma 17.9, U/B(J0 )g splits over A/B(J0 )g . By Lemma 17.10, there is a


complement L/B0 to A/B0 in U/B0 .
Since B(J0 ) B, we have [A/B(J0 ), x] = A/B. Denote by C/B(J0 ) the
centralizer of x in A/B(J0 ). Then C/B(J0 ) = A/B(J0 ). Note that the centralizer
of xg in A/B(J0 )g is C g /B(J0 )g , so that C g /B0 CA/ B0 (xg ). Let

C0 = Cg.
gG

Then C0 is a ZE-submodule, and

A/C0 = DrgS A/C g .

Let L/B0 and L1 /B0 be two complements to A/B0 in U/B0 . By Lemma 17.11,
L/C g and L1 /C g are conjugate in U . It now follows from Lemma 17.10 that
LC0 and L1 C0 are conjugate in U . Application of the Frattini argument gives
224 Chapter 17. Splitting over the locally nilpotent residual

that E = ANE (LC0 ). Since A has no proper supplement in E, we must have


E = NE (LC0 ), i.e. LC0 is normal in E. Hence,
[A, LC0 ] A LC0 = (A L)C0 = C0 = A,
contrary to [A, K] = A.
This completes the proof that E splits over A.
Conjugacy of complements
Let P1 and P2 be two complements to A in E. Let C be the set of all E-invariant
subgroups W of A such that W P1 and W P2 are conjugate. Clearly, A C, so that
C = . Since A is an artinian ZG-module, C has a minimal element M . Suppose
that M = 1. Then BP1 and BP2 are not conjugate for every proper G-invariant
subgroup B of M . Without loss of generality we may assume that M P1 = M P2 ,
A = M , and E = M P1 , so that we have AP1 = AP2 = E, and BP1 and BP2 are
not conjugate for every proper E-invariant subgroup B of A. By Lemma 17.2, E
has a normal subgroup K, and A has a proper E-invariant subgroup Q satisfying
the following conditions:
(i) The factor-group (K/Q)/CK/Q (A/Q) is the normal closure of a single ele-
ment of (G/Q)/CK/Q (A/Q).
(ii) (K/Q)/CK/Q (A/Q) is either a nite elementary abelian p-group for some
prime p or a free abelian group of nite 0-rank.
(iii) (K/Q)/CK/Q (A/Q) lies in
F C((G/Q)/CK/Q (A/Q)) ((H/Q)CK/Q (A/Q)/CK/Q (A/Q)).

(iv) A/Q = [A/Q, K/Q].


We may replace E by E/Q to assume Q = 1. By Lemma 17.7,
CK (A) = A B
for some E-invariant subgroup B. Replacing E by the factor-group E/B, we may
assume B = 1. Then CK (A) = A. Put
E
K = x = x1 , . . . , xn  .
Proceeding as we did in the existence part of this proof, we can choose a U -
invariant subgroup B such that
CA (g) B, and [A/B, gB] = A/B.
Again let U/A = CE/A (K/A). We form the E-invariant subgroup B0 as in Lemma
17.1. Then

B0 = B(J0 )g , B(J0 ) B CA (g),
gG

and A/B0 = DrgS A/B(J0 )g ,


Chapter 17. Splitting over the locally nilpotent residual 225

where S is a subset of a transversal T to U in G. Note also that

B(J0 )g B g CA (g g ) = CA (g).

By Lemma 17.11, (U P1 )B(J0 )g is conjugate to (U P2 )B(J0 )g in U . By Lemma


17.10, (U P1 )B0 is conjugate to (U P2 )B0 in U . But

(U P1 )B0 = U P1 B0 , and Pj B0 NE (U Pj B0 ), j {1, 2}.

Therefore,
NE (U Pj B0 ) = Pj NA (U Pj )B0 ), j {1, 2}.
However, NA (U Pj )B0 ) = CA/B0 (U ). The inclusion H U implies that A has
a non-zero H-central ZG-section. We have already said above that every non-
zero ZG-factor of A is H-eccentric. It follows that NA (U Pj )B0 ) = B0 , and so
NE (U Pj B0 ) = Pj B0 . Therefore, P1 B0 and P2 B0 are conjugate. This contradic-
tion completes the proof of the theorem. 
Bibliography

[1] J. Alc
azar, J. Otal, Sylow subgroups of groups with Chernikov conjugacy
classes. J. Algebra 110 (1987), 507513.
[2] F. Anderson, K. Fuller, Rings and categories of modules. Springer, 1974.
[3] M.F. Atiyah, I.G. MacDonald, Introduction to commutative algebra.
Addison-Wesley, 1969.
[4] R. Baer, Finiteness properties of groups. Duke Math. J. 15 (1948), 1021
1032.
[5] R. Baer, Groups with descending chain condition for normal subgroups.
Duke Math. J. 16 (1949), 122.
[6] R. Baer, Au osbare Gruppen mit Maximalbedingung. Math. Ann. 129
(1955), 139173.
[7] R. Baer, Polyminimaxgruppen. Math. Ann. 175 (1968), 143.
[8] R. Baer, Durch Formationen bestimmte Zerlegungen von Normalteilern
endlicher Gruppen. J. Algebra 20 (1972), 3856.
[9] R. Baer, H. Heineken, Radical groups of nite abelian subgroup rank. Illinois
J. Math. 16 (1972), 533580.
[10] S.D. Berman, The group algebras of countable abelian p-groups. Pub. Math.
14 (1967), 365405.
[11] S.D. Berman, A.G. Podgorny, On the representations of a group of type
p over one class of the elds. Ukrain. Math. J. 29 (1977), 632637.
[12] S.D. Berman, N.I. Vishnyakova, On the representations of a group of type
p over a eld of characteristic p. Proc. Moskow Math. Soc. 29 (1973),
87100.
[13] S.D. Berman, N.I. Vishnyakova, Locally cyclic modules over a valuation
rings. Representation theory, group rings and coding theory, Contem-
porary Math. Amer. Math. Soc. 1989, 85103.
[14] N. Bourbaki, Algebre (Polynomes et fractions rationelles; Corps commu-
tatifs). Hermann, 1959.
228 Bibliography

[15] N. Bourbaki, Algebre commutative (Modules Plats, Localisation). Her-


mann, 1961.
[16] N. Bourbaki, Algebre commutative (Graduations, ltrations et topologies;
Ideaux premier associes et decomposition primaire). Hermann, 1961.
[17] N. Bourbaki, Algebre (Algebre lineaire). Hermann, 1962.
[18] N. Bourbaki, Algebre commutative (Entiers, valuations). Hermann, 1964.
[19] N. Bourbaki, Algebre (Groupes et corps ordonnes; Modules sur les anneaux
principaux). Hermann, 1964.
[20] N. Bourbaki, Algebre commutative (Diviseurs). Hermann, 1965.
[21] Z.I. Borevich, A.R. Shafarevich, Number theory. Academic Press, 1966.
[22] A. A. Bovdi, The multiplicative group of an integral group rings. University
Of Uzhgorod, 1987.
[23] A. A. Bovdi, Group rings. University of Uzhgorod, 1988.
[24] D.C. Brewster, J.C. Lennox, Subnormality and quasinormality in soluble
minimax groups. J. Algebra 48 (1977), 368381.
[25] C.J.B. Brookes, Ideals in group rings of soluble groups of nite rank. Math.
Proc. Cambridge Phil. Soc. 97 (1985), 2749.
[26] C.J.B. Brookes, Abelian subgroups and Engel elements of soluble groups.
J. London Math. Soc. 32 (1985), 467472.
[27] C.J.B. Brookes, Engel elements of soluble groups. Bull. London Math. Soc.
16 (1986),, 710.
[28] C.J.B. Brookes, Modules over polycyclic groups. Proc. London Math. Soc.
57 (1988), 88108.
[29] C.J.B. Brookes, J.R.J. Groves, Modules over nilpotent group rings. J. Lon-
don Math. Soc. 52 (1995), 467481.
[30] K.A. Brown, The structure of modules over polycyclic groups. Math. Proc.
Cambridge Phil. Soc. 89 (1981), 257283.
[31] K.A. Brown, The Nullstellensatz for certain group rings. J. London Math.
Soc. 26 (1982), 425434.
[32] S.U. Chase, Direct products of modules. Trans. Amer. Math. Soc. 97 (1960),
437473
[33] V.S. Charin, A remark on the minimal condition for subgroups. Doklady
AN SSSR 66 (1949), 575576.
[34] V.S. Charin, On the minimal condition for normal subgroups in locally
soluble groups. Mat.Sbornik 33 (1953), 2736.
Bibliography 229

[35] V.S. Charin, On the automorphism groups of certain classes of soluble


groups. Ukrain. Math. J. 5 (1953), 363369.
[36] V.S. Charin, On the automorphism groups of nilpotent groups. Ukrain.
Math. J. 6 (1954), 295304.
[37] V.S. Charin, On locally soluble groups of nite rank. Mat. Sb. 41 (1957),
3748.
[38] C.S. Charin, On soluble groups of type A4. Mat. Sb. 52 (1960), 895914.
[39] V.S. Charin, On soluble groups of type A3. Mat. Sb. 54 (1961), 489499.
[40] V.S. Charin, D.I. Zaitsev, On groups with niteness conditions and other
restrictions on subgroups. Ukrain. Math. J. 40 (1988), 277287.
[41] A.W. Chatters, C.R. Hajarnavis, Rings with chain condition. Pitman,
1980.
[42] S.N. Chernikov, The niteness conditions in general group theory. Uspekhi
Mat. nauk 14 (1959), 4596.
[43] S.N. Chernikov, The groups with prescribed properties of systems of sub-
groups. Nauka Moskow, 1980.
[44] G.H. Cli, S.K. Sehgal, Group rings whose units form an F C-group. Math.
Z. 161 (1978), 163168.
[45] A.H. Cliord, Representations induced in an invariant subgroups. Annals
Math. 38 (1937), 533550.
[46] J. Cornick, P.H. Kropholler, Homological niteness conditions for modules
over group algebras. J. London Math. Soc. 58 (1998), 4962.
[47] C.W. Curtis, I. Reiner, Representation theory of nite groups and associa-
tive algebras. John Wiley, 1962.
[48] C.W. Curtis, I. Reiner, Methods of representation theory, Vol 1. John Wi-
ley, 1981.
[49] C.W. Curtis, I. Reiner, Methods of representation theory, Vol 2. John Wi-
ley, 1981.
[50] G. Cutolo, L.A. Kurdachenko, Weak chain conditions for non-almost nor-
mal subgroups. London Math. Soc. Lecture Notes Series 211 (1995), 120
130.

[51] R. Dedekind, Uber die Theorie der ganzen algebraischen Zahlen. Supple-
ment XI to Dirichlets Vorlesungen uber Zahlentheorie, 2nd ed., 1871.
[52] M.R. Dixon, Sylow theory, formations and Fitting classes in locally nite
groups. World Scientic, 1994.
230 Bibliography

[53] M.R. Dixon, M.J. Evans, H. Smith, Locally (soluble-by-nite) groups of


nite rank. J. Algebra 182 (1996), 756769.
[54] M.R. Dixon, M.J. Evans, H. Smith, Locally (soluble-by-nite) groups with
all proper insoluble subgroups of nite rank. Arch. Math. 68 (1997), 100
109.
[55] M.R. Dixon, L.A. Kurdachenko, N.V. Polyakov, Locally generalized radical
groups satisfying certain rank conditions. To appear.
[56] K. Doerk, T. Hawkes, Finite soluble groups. Walter de Gruyter, 1992.
[57] S. Donkin, Polycyclic groups, Lie algebras and algebraic groups. J. reine
angew. Math. 326 (1981), 104123.
[58] S. Donkin, Locally nite representations of polycyclic-by-nite groups.
Proc. London Math. Soc 44 (1982), 333348.
[59] Z.Y. Duan, The structure of noetherian modules over hypernite groups.
Math. Proc. Cambridge Phil. Soc. 112 (1992), 2128.
[60] Z.Y. Duan, Extensions of abelian-by-hyper(cyclic or nite) groups. Comm.
Algebra 20 (1992), 23052321.
[61] Z.Y. Duan, The decomposition of noetherian modules over hypernite
groups. Ricerche Mat. 44 (1995), 6589.
[62] Z.Y. Duan, The F-decomposition of artinian modules over hypernite
groups. Proc. Edinburgh Math. Soc. 38 (1995), 117120.
[63] Z.Y. Duan, Modules over hypernite groups. Rings, groups and algebras,
Marcel Dekker, 1996, 5161.
[64] Z.Y. Duan, M.J. Tomkinson, The decomposition of minimax modules over
hypernite groups. Arch. Math. 81 (1993), 340343.
[65] Z.Y. Duan, L.A. Kurdachenko, I.Ya. Subbotin, Artinian modules over hy-
pernite groups. South. Asian Bull. Math. 22 (1998), 273286
[66] C. Faith, Algebra: Rings, Modules and Categories I. Springer, 1973.
[67] C. Faith, Algebra: Rings, Modules and Categories II. Springer, 1976.
[68] D.R. Farkas, Noetherian group rings: an exercise in creating folklore and
induction. Noetherian rings and their applications, Amer.Math. Soc.,
Math. Surveys and monographs 24 (1987), 89118.
[69] D.R. Farkas, R.L. Snider, Group algebras whose simple modules are injec-
tive. Trans. Amer. Math. Soc. 194 (1974), 241248.
[70] I. Fleischner, Modules of nite rank over Pr
ufer rings. Annals of Math. 65
(1957), 250254.
Bibliography 231

[71] E. Formanek, R.L. Snider, Primitive group rings. Proc. Amer. Math. Soc.
36 (1972), 357360.
[72] S. Franciosi, F. de Giovanni, On maximal subgroups of minimax groups.
Atti. Accad. Naz. Lincei. Mat. Applic. 9 (1995), 2327.
[73] S. Franciosi, F. de Giovanni, L.A. Kurdachenko, The Schur property and
groups with uniform conjugace classes. J. Algebra 174 (1995), 823847.
[74] S. Franciosi, F. de Giovanni, L.A. Kurdachenko, On groups with many
almost normal subgroups. Annali Mat. 169 (1995), 3565.
[75] S. Franciosi, F. de Giovanni, L.A. Kurdachenko, Groups whose proper quo-
tients are F C-groups. J. Algebra 186 (1996), 544577.
[76] S. Franciosi, F. de Giovanni, M.J. Tomkinson, Groups with polycyclic-by-
nite conjugacy classes. Boll. Unione Mat. Italiana 4B (1990), 3555.
[77] S. Franciosi, F. de Giovanni, M.J. Tomkinson, Groups with Chernikov con-
jugacy classes. J. Austral. Math. Soc.(series A) 50 (1991), 114.
[78] L. Fuchs, Innite abelian groups, Vol. 1. Academic Press, 1970.
[79] L. Fuchs, Innite abelian groups, Vol. 2. Academic Press, 1973.
[80] L. Fuchs, S.B. Lee, Primary decompositions over domains. Glasgow Math.
J. 38 (1996), 321326
[81] L. Fuchs, L. Salce, Modules over valuation domains. Marcel Dekker, 1985.
[82] A.D. Gardiner, B. Hartley, M.J. Tomkinson, Saturated formations and Sy-
low structure in locally nite groups. J. Algebra 17 (1971), 177211.
[83] W. Gasch
utz, Zur Erweiterungstheorie der endlichen Gruppen. J. Reine
Angew. Math. 190 (1952), 93107.
[84] R. Gilmer, Multiplicative ideal theory. Marcel Dekker, 1972.
[85] J.S. Golan, T. Head, Modules and structure of rings. Marcel Dekker, 1991.
[86] M. Gonz
alez, J. Otal, P. Halls covering group and embedding of countable
CC-groups. Comm. Algebra 18 (1990), 34053412.
[87] M. Gonz alez, J. Otal, Embedding theorems for residually Chernikov CC-
groups. Proc. Amer. Math. Soc. 123 (1995), 23832332.
[88] M. Gonzalez, J. Otal, The extension of results due to Gorchakov and
Tomkinson from F C-groups to CC-groups. J. Algebra 185 (1996), 314
328.
[89] M. Gonz
alez, J. Otal, J.M. Pe
na, CC-groups with periodic central factors.
Manuscripta Math. 69 (1990), 93103.
232 Bibliography

[90] Yu.M. Gorchakov, The groups with nite classes of conjugacy elements.
Nauka Moskow, 1978.
[91] D. Gorenstein, Finite groups. Harper Row, 1968.
[92] R.I. Grigorchuk, Burnsides problem on periodic groups. Funktsional Analis
i prilozh. 14 (1980), 5354.
[93] R.I. Grigorchuk, Solved and unsolved problems around one group. Geo-
metric, Combinatorial and Dynamical Aspects . Progress in Mathematics
Series Vol. 248, Birh
auser, pp. 117218.
[94] K.W. Gruenberg, Ring theoretic methods and niteness conditions in in-
nite soluble group theory. Lecture Notes Math. 319 (1974), 7584.
[95] P. Hall, Finiteness conditions for soluble groups. Proc. London Math. Soc.
4 (1954), 419436.
[96] P. Hall, On the niteness of certain soluble groups. Proc. London Math.
Soc. 9 (1959), 595632.
[97] P. Hall, Nilpotent groups. Queen Mary College Mathematics Notes, 1969.
[98] P. Hall, The Frattini subgroup of nitely generated groups. Proc. London
Math. Soc. 11 (1961), 327352.
[99] P. Hall, B. Hartley, The stability group of a series of subgroups. Proc.
London Math. Soc. 16 (1966), 139.
[100] B. Hartley, A note on the normalizer conditions. Proc. Cambridge Phil.
Soc. 74 (1973), 1115.
[101] B. Hartley, A class of modules over locally nite groups I. J. Austral. Math.
Soc.(series A) 16 (1973), 431442.
[102] B. Hartley, A class of modules over locally nite groups II . J. Austral.
Math. Soc.(series A) 19 (1975), 437469.
[103] B. Hartley, A class of modules over locally nite groups III. Bull. Austral.
Math. Soc. 14 (1976), 95110.
[104] B. Hartley, Splitting over the locally nilpotent residual of locally nite
groups. Quart. Journal Math. 27 (1976), 395400.
[105] B. Hartley, Uncountable artinian modules and uncountable soluble groups
satisfying Min-n. Proc. London Math. Soc. 35 (1977), 5575.
[106] B. Hartley, A dual approach to Chernikov modules. Math. Proc. Cambridge
Phil. Soc. 82 (1977), 215239.
[107] B. Hartley, Injective modules over group rings. Quart. Journal Math. 28
(1977), 129.
Bibliography 233

[108] B. Hartley, Locally nite groups whose irreducible modules are nite di-
mensional. Rocky Mountain Math. J. 13 (1983), 255263.
[109] B. Hartley, T.O. Hawkes, Rings, modules and linear algebra. Chapman &
Hall, 1974.
[110] B. Hartley, D. McDougal, Injective modules and soluble groups satisfying
the minimal condition for normal subgroups. Bull. Austral. Math. Soc. 4
(1971), 113135.
[111] B. Hartley, M.J. Tomkinson, Splitting over nilpotent and hypercentral resid-
uals. Math. Proc. Cambridge Phil. Soc. 78 (1975), 215226.
[112] H. Heineken, I.J. Mohamed, Groups with normalizer condition. Math. An-
nalen 198 (1972), 178187.
[113] I.N. Herstein, Conjugates in division rings. Proc. Amer. Math. Soc. 7
(1956), 10211022.
[114] T.W. Hungerford, Algebra, 2nd edition. Springer, 1996
[115] B. Huppert, Endliche Gruppen I. Springer, 1967.
[116] B. Huppert, N. Blackburn, Finite groups II. Springer, 1982.
[117] B. Huppert, N. Blackburn, Finite groups III. Springer, 1982.
[118] A.V. Jategaonkar, Integral group rings of polycyclic-by-nite groups. J.
pure applied Algebra 4 (1974), 337343.
[119] A.V. Jategaonkar, Certain injectives are artinian. Lecture Notes Math.
545 (1976), 128139.
[120] A.V. Jategaonkar, Localization in noetherian rings. Cambridge University
Press, 1986.
[121] I. Kaplansky, Modules over Dedekind and valuation rings. Trans.Amer.
Math. Soc. 72 (1952), 327340.
[122] I. Kaplansky, Projective modules. Annals Math. 68 (1958), 372377.
[123] I. Kaplansky, The splitting of modules over integral domains. Arch. Math.
13 (1962), 341343
[124] I. Kaplansky, Innite abelian groups. Univ. Michigan Press, 1969.
[125] I. Kaplansky, Topics in commutative ring theory. Univ. Chicago Notes
from Dept. Math., 1974.
[126] M. Karbe, Unendliche Gruppen mit schwachen Kettenbedingungen f
ur
endlich erzeugte Untergruppen. Arch. Math. 45 (1985), 97110.
[127] M. Karbe, L.A. Kurdachenko, Just innite modules over locally soluble
groups. Arch. Math. 51 (1988), 401411.
234 Bibliography

[128] M.I. Kargapolov, On soluble groups of nite rank. Algebra i logika 1 (1962),
3744.
[129] M.I. Kargapolov, Yu. I. Merzlyakov, Foundations of the theory of groups.
Springer, 1979.
[130] G. Karpilovsky, Commutative group algebras. Marcel Dekker, 1983.
[131] G. Karpilovsky, Field theory. Marcel Dekker, 1988.
[132] L.S. Kazarin, L.A. Kurdachenko, The niteness conditions and the factor-
izations in innite groups. Russian Math. Surveys 47 (1992), 81126.
[133] L.S. Kazarin, L.A. Kurdachenko, I.Ya. Subbotin, On groups saturated
with abelian subgroups. International Journal Algebra and Computation
8 (1998), 443466.
[134] O.H. Kegel, B.A.F. Wehrfritz, Locally nite groups. North Holland, 1973.
[135] Sh.S. Kemhadze, On stable groups of automorphisms. Doklady AN USSR
158 (1964), 510512.
[136] A. Kertesz, Lectures on artinian rings. Akademiai Kiado, 1987.
[137] L.G. Kovacs, M.F. Newman, Direct complementation in group with opera-
tors. Arch. Math. 13 (1963), 427433.
[138] P.H. Kropholler, On nitely generated soluble groups with no large wreath
product section. Proc. London Math. Soc. 49 (1984), 155169.
[139] P.H. Kropholler, P.A. Linnel, J.A. Moody, Applications of a new K-
theoretic theorem to soluble group rings. Proc. Amer. Math. Soc. 104
(1988), 675684.
[140] S.A. Kruglyak, On representations of groups (p, p) over a eld of charac-
teristic p. Doklady AN SSSR 153 (1963), 12531256.
[141] L.A. Kurdachenko, The groups satisfying the weak minimal and maximal
conditions for normal subgroups. Sibir. Math. J. 20 (1979), 10681075.
[142] L.A. Kurdachenko, The groups satisfying the weak minimal and maximal
conditions for subnormal subgroups. Math. Notes 29 (1981), 1930.
[143] L.A. Kurdachenko, On some embeddings conditions of F C-groups in the
direct product of nite groups and torsion-free abelian groups. Math. USSR
Sbornik 42 (1982), 499514.
[144] L.A. Kurdachenko, Locally nilpotent groups with the weak minimal condi-
tion for normal subgroups. Sibir. Math. J. 25 (1984), 589594.
[145] L.A. Kurdachenko, Locally nilpotent groups with the weak minimal and
maximal comditions for normal subgroups. Doklady AN Ukrain. SSR, 8
(1985), 912.
Bibliography 235

[146] L.A. Kurdachenko, The locally nilpotent groups with condition Min-oo-n.
Ukrain. Math. J. 42 (1990), 303307.
[147] L.A. Kurdachenko, On some classes of groups with the weak minimal and
maximal conditions for normal subgroups. Ukrain. Math. J. 42 (1990),
10501056,
[148] L.A. Kurdachenko, Artinian modules over groups of nite rank and some
niteness conditions. Innite groups and adjoining algebraic structures,
Naukova Dumka Kiev, 1993, 144159.
[149] L.A. Kurdachenko, On groups with minimax conjugacy classes. Innite
groups and adjoining algebraic structures, Naukova Dumka Kiev, 1993,
160177.
[150] L.A. Kurdachenko, Artinian modules over groups of nite rank and the
weak minimal condition for normal subgroups. Ricerche Mat. 44 (1995),
303335.
[151] L.A. Kurdachenko, On normal closures of elements in generalized F C-
groups. Innite groups 1994 (Ravello 1994). Walter de Gruyter, 1996,
141151.
[152] L.A. Kurdachenko, Modules over the group rings with some niteness con-
ditions. Ukrain. Math. J. 54 (2002), 931940.
[153] L.A. Kurdachenko, V.E. Goretsky, On groups with the weak minimal and
maximal conditions for non-normal subgroups. Ukrain. Math. J. 41 (1989),
17051709.
[154] L.A. Kurdachenko, J. Otal, Some noetherian modules and non-monolithic
just non-CC-groups. J. Group Th. 2 (1999), 5364.
[155] L.A. Kurdachenko, J. Otal, Simple modules over CC-groups and mono-
lithic just non-CC-groups. Boll. Unione Mat. Italiana 8 (2001), 381390.
[156] L.A. Kurdachenko, J. Otal, Frattini properties of groups with minimax
conjugacy classes. Quaderni di Matematica 8 (2001), 221237.
[157] L.A. Kurdachenko, J. Otal, I.Ya. Subbotin, Groups with prescribed quotient
groups and associated module theory. World Sci. Publ., 2002.
[158] L.A. Kurdachenko, J. Otal, I.Ya. Subbotin, Baer like decomposition of
modules. Illinois Journal Math. 47 (2003), 329343.
[159] L.A. Kurdachenko, B.V. Petrenko, I.Ya. Subbotin, On generalized hyper-
centers in artinian modules. Comm. Algebra 25 (1997), 10231046.
[160] L.A. Kurdachenko, B.V. Petrenko, I.Ya. Subbotin, Direct decompositions
in artinian modules over F C-hypercentral groups. Matematica contempo-
ranea 14 (1998), 8999.
236 Bibliography

[161] L.A. Kurdachenko, N.V. Polyakov, I.Ya. Subbotin, On some generalization


of a theorem of B. H. Neumann. Matematica contemporanea 21 (2001),
131145.
[162] L.A. Kurdachenko, V.V. Pylaev, On groups with the minimax factor-
groups. Ukrain. Math. J. 42 (1990), 550553.
[163] L.A. Kurdachenko, V.V. Pylaev, V.E. Goretsky, The groups with the same
systems of minimax factor-groups. Doklady AN USSR 3A (1988), 1720.
[164] L.A. Kurdachenko, N.N. Semko, I.Ya. Subbotin, Artinian modules over
F C-hypercentral groups. Proceedings of the National Academy of Sciences
of Ukraine, 2005, no 2, 2932.
[165] L.A. Kurdachenko, H. Smith, Groups with the maximal condition on non-
subnormal subgroups. Boll. Unione Mat. Italiana 10B (1996), 441460.
[166] L.A. Kurdachenko, H. Smith, Groups with the weak minimal condition for
non-subnormal subgroups. Annali Mat. 173 (1997), 299312
[167] L.A. Kurdachenko, H. Smith, The nilpotency of some groups with all sub-
groups subnormal. Pub. Matematiques 42 (1998), 411421.
[168] L.A. Kurdachenko, H. Smith, Groups with the weak maximal condition for
non-subnormal subgroups. Ricerche Math. 47 (1998), 2949.
[169] L.A. Kurdachenko, I.Ya. Subbotin, Groups with restrictions for cocentral-
izers of elements. Comm. Algebra 24 (1996), 11731187.
[170] L.A. Kurdachenko, I.Ya. Subbotin, Groups whose proper quotients are hy-
percentral. J. Austral. Math. Soc., series A 65 (1998), 224237.
[171] L.A. Kurdachenko, I.Ya. Subbotin, Groups with many periodic factor-
groups. Comm. Algebra 28 (2000), 15931602.
[172] L.A. Kurdachenko, I.Ya. Subbotin, On minimal artinian modules and min-
imal artinian linear groups. International Journal of Mathematics and
mathematical Sciences 27 (2001), 704714.
[173] L.A. Kurdachenko, I.Ya. Subbotin, On some innite dimensional linear
groups. South. Asian Bulletin of Math. 26 (2002), 773787
[174] L.A. Kurdachenko, I.Ya. Subbotin, On some types of Noetherian modules.
J. Algebra and Its Applications 3 (2004), 169179.
[175] L.A. Kurdachenko, A.V. Tushev, D.I. Zaitsev, Noetherian modules over
nilpotent groups of nite rank. Arch. Math. 56 (1991), 433436.
[176] A.G. Kurosh, The theory of groups. NAUKA Moskow, 1967.
[177] A.G. Kurosh, Lectures on general algebra. NAUKA Moskow, 1973.
[178] T.Y. Lam, A rst course in noncommutative rings. Springer, 1991.
Bibliography 237

[179] S. Lang, Algebra. AddisonWesley, 1965.


[180] M.D. Larsen, P.J. McCarthy, Multiplicative theory of ideals. Academic
Press, 1971.
[181] A.I. Maltsev, On groups of nite rank. Mat. Sbornik 22 (1948), 351352.
[182] A.I. Maltsev, On certain classes of innite soluble groups. Mat. Sbornik
28 (1951), 567588. English translation: Amer. Math. Soc. Translations 2
(1956), 121.
[183] E. Matlis, Injective modules over noetherian rings. Pacic J. Math. 8
(1958), 511528.
[184] E. Matlis, Injective modules over Pr
ufer rings. Nagoya Math. J. 15 (1959),
5769.
[185] E. Matlis, Divisible modules. Proc. Amer. Math. Soc. 11(1960), 385391.
[186] E. Matlis, Cotorsion modules. Mem. Amer. Math. Soc. 49 (1964).
[187] E. Matlis, Torsion-free modules. The University Chicago Press, 1972.
[188] E. Matlis, Generalizations of divisible abelian groups to the theory of mod-
ules. Symp. Mat.Ist. Naz. Alta Mat. 23 (1979), 241250
[189] D.J. McCaughan, Subnormality in soluble minimax groups. J. Austral.
Math. Soc.(series A) 17 (1974), 113128.
[190] J.C. McConnel, J.C. Robson, Noncommutative noetherian rings. John Wi-
ley, 1987.
[191] D. McDougall, Soluble minimax groups with the subnormal intersection
property. Math. Z. 114 (1970), 241244.
[192] D. McDougall, Soluble groups with the minimum condition for normal sub-
groups. Math. Z. 118 (1970), 157167.
[193] G.O. Michler, O.E. Villamayor, On rings whose simple modules are infec-
tive. J. Algebra 25 (1973), 185201.
[194] I. Musson, Injective modules for group algebras of locally nite groups.
Math. Proc. Cambridge Phil. Soc. 84 (1978), 247262.
[195] I. Musson, Injective modules for group rings of polycyclic groups. Quart.
J. Math. 31 (1980), 429466.
[196] I. Musson, Irreducible modules for polycyclic group algebras. Canad.
J.Math. 33 (1981), 901914.
[197] I. Musson, Some examples of modules over noetherian rings. Glasgow
Math. J. 23 (1982), 913.
238 Bibliography

[198] I. Musson, Representations of innite soluble groups. Glasgow Math. J. 24


(1983), 4352.
[199] I. Musson, On the structure of certain injective modules over group algebras
of soluble groups of nite rank. J. Algebra 85 (1983), 5175.
[200] I. Musson, Injective modules, localization and completion in group algebras.
J. pure applied Algebra 57 (1989), 265279.
[201] W. Narkiewicz, Elementary and analithic theory of algebraic numbers.
Springer, 1989.
[202] M.L. Newell, On soluble Min-by-Max groups. Math. Annalen 186 (1970),
282286.
[203] M.L. Newell, A subgroup characterization of soluble Min-by-Max groups.
Arch. Math. 21 (1970), 128131.
[204] M.L. Newell, Supplement in abelian-by-nilpotent groups. J. London Math.
Soc. 11 (1975), 7480.
[205] M.L. Newell, Some splitting theorem for innite supersoluble groups. Math.
Z. 144 (1975), 265275.
[206] D.G. Northcott, Ideal theory. Cambridge Univ. Press, 1960.
[207] D.G. Northcott, Lessons on rings, modules and multiplicities. Cambridge
Univ. Press, 1968.
[208] A.Yu. Olshanskii, Geometry of dening relations in groups. Kluwer, 1991.
[209] J. Otal, J.M. Pe
na, Minimal non-CC-groups. Comm. Algebra 16 (1988),
12311242.
[210] J. Otal, J.M. Pena, Groups in which every proper subgroups is Chernikov-
by-nilpotent or nilpotent-by-Chernikov. Arch. Math. 51 (1988), 193197.
[211] J. Otal, J.M. Pe na, Groups with minimal condition related to niteness
properties of conjugate classes. Rend. Semin. Mat. Univ.Padova 81 (1989),
7984.
[212] J. Otal, J.M. Pe
na, Characterizations of the conjugacy of Sylow p-subgroups
of CC-groups. Proc. Amer. Math. Soc. 106 (1989), 605610.
[213] J. Otal, J.M. Pe
na, Sylow theory of CC-groups: a survey. London Math.
Soc. Lecture Notes Ser. 160 (1991), 400407.
[214] J. Otal, J.M. Pe
na, M.J. Tomkinson, Locally inner automorphisms of CC-
groups. J. Algebra 141 (1991), 382398.
[215] D.S. Passman, Group rings satisfying a polynomial identity II. Pacic J.
Math. 39 (1971), 425438.
Bibliography 239

[216] D.S. Passman, Innite group rings. Marcel Dekker, 1971.


[217] D.S. Passman, The algebraic structure of group rings. John Wiley, 1977.
[218] D.S. Passman, Group rings of polycyclic groups. Group theory: essays for
Philip Hall, Academic Press, 1984, 207256.
[219] D.S. Passman, A course in ring theory. Wadsworh & Brooks, 1991.
[220] R.S. Pierce, Associative Algebras. Springer, 1982.
[221] B.I. Plotkin, Automorphism groups of the algebraic systems. Nauka
Moskow, 1966.
[222] C. Polcino-Milies, Group rings whose units form an F C-groups. Arch.
Math. 30 (1978), 380384; corrigendum: Arch. Math. 31 (1978), 528.
[223] C. Polcino-Milies, Units in group rings: A short survey. London Math. Soc.
Lecture Notes Ser. 71 (1982), 281297.
[224] C. Polcino-Milies, S.K. Sehgal, F C-elements in a group ring. Comm. Al-
gebra 9 (1981), 12851293.
[225] Ya.D. Polovicky, Groups with extremal classes of conjugate elements. Sibir.
Math. J. 5 (1964), 891895.

[226] R. Remak, Uber minimale invariante Untergruppen in der Theorie der
endlichen Gruppen. J. reine angew. Math. 192 (1930), 116.
[227] D.J.S. Robinson, Groups in which normality is a transitive relation. Proc.
Cambridge Philos. Soc. 60 (1964), 2138.
[228] D.J.S. Robinson, On nitely generated soluble groups. Proc. London Math.
Soc. 18 (1965), 508516.
[229] D.J.S. Robinson, On soluble minimax groups. Math. Z. 101 (1967), 1340.
[230] D.J.S. Robinson, Innite soluble and nilpotent groups. Queen Mary College
Mathematics Notes, 1968.
[231] D.J.S. Robinson, Residual properties of some classes of innite soluble
groups. Proc. London Math. Soc. 18 (1968), 495520.
[232] D.J.S. Robinson, A note on groups of nite rank. Compositio Math. 31
(1969), 240246.
[233] D.J.S. Robinson, A theorem of nitely generated hyperabelian groups. In-
vent. Math. 10 (1970), 3843.
[234] D.J.S. Robinson, Finiteness conditions and generalized soluble groups, part
1. Springer, 1972.
[235] D.J.S. Robinson, Finiteness conditions and generalized soluble groups, part
2. Springer, 1972.
240 Bibliography

[236] D.J.S. Robinson, Hypercentral ideals, noetherian modules and theorem of


Stroud. J. Algebra 32 (1974), 234239.
[237] D.J.S. Robinson, Splitting theorems for innite groups. Sympos. Mat. Ist.
Naz. alta Mat. 17 (1975), 441470.
[238] D.J.S. Robinson, On the cohomology of soluble groups of nite rank. J.
pure applied Algebra 6 (1975), 155164.
[239] D.J.S. Robinson, A new treatment of soluble groups with niteness condi-
tions on their abelian subgroups. Bull. London Math. Soc. 8 (1976), 113
129.
[240] D.J.S. Robinson, The vanishing of certain homology and cohomology group.
J. pure applied Algebra 7 (1976), 145167.
[241] D.J.S. Robinson, On the homology of hypercentral groups. Arch. Math. 32
(1979), 223226.
[242] D.J.S. Robinson, A course in the theory of groups. Springer, 1982.
[243] D.J.S. Robinson, Applications of cohomology to the theory of groups. Lon-
don Math. Soc. Lecture Notes Ser. 71 (1982), 4680.
[244] D.J.S. Robinson, Finiteness, solubility and nilpotence. Group theory: es-
says for Philip Hall, Academic Press, 1984, 159206.
[245] D.J.S. Robinson, Soluble products of nilpotent groups. J. Algebra 98 (1986),
183196.
[246] D.J.S. Robinson, Cohomology of locally nilpotent groups. J. pure applied
Algebra 48 (1987), 281300.
[247] D.J.S. Robinson, Homology and cohomology of locally supersoluble groups.
Math. Proc. Cambridge Philos. Soc. 102 (1987), 233250.
[248] J.E. Roseblade, Group rings of polycyclic groups. J. pure applied Algebra
3 (1973), 307328.
[249] J.E. Roseblade, Polycyclic group rings and the Nullstellesatz. Lecture
Notes Math. 319 (1973), 156167.
[250] J.E. Roseblade, Application of the ArtinRees lemma to group rings. Sym-
pos. Mat. 17 (1976), 471478.
[251] J.E. Roseblade, Prime ideals in group rings of polycyclic groups. Proc.
London Math. Soc. 36 (1978), 385447; Corrigendum: Proc. London Math.
Soc. 39 (1979), 216218.
[252] J.E. Roseblade, Five lectures on group rings. London Math. Soc. Lecture
Notes Ser. 121 (1985), 93109.
Bibliography 241

[253] E. Schenkman, The splitting of certain solvable groups. Proc. Amer. Math.
Soc. 6 (1935), 286290.
[254] O.F.G. Schilling, The theory of valuations. Amer Math. Soc. Math. Surveys
Vol. 4, 1950.
[255] D. Segal, Polycyclic groups. Cambridge Univ. Press, 1983.
[256] D. Segal, On the group rings of abelian minimax groups. J. Algebra 237
(2001), 6494.
[257] D. Segal, On modules of nite upper rank. Trans. Amer. Math. Soc. 353
(2001), 391410.
[258] S.K. Sehgal, Topics in group rings. Marcel Dekker, 1978.
[259] S.K. Sehgal, Units in integral group rings; a survey. Proc. First. Sympo-
sium in Algebra and Number Theory, Worls Sci. Publ., 1990, 255268.
[260] S.K. Sehgal, H.J. Zassenhaus, Group rings whose units form an F C-group.
Math. Z. 153 (1977), 2935.
[261] S.K. Sehgal, H.J. Zassenhaus, On the supercenter of a group and its ring
theoretic generalizations. Lecture Notes. Math. 882 (1981), 117144.
[262] H.H.G. Seneviratne, On permutable subgroups of soluble minimax groups.
Arch. Math. 44 (1985), 481484.
[263] R. Sharp, Steps in commutative algebra. Cambridge Univ. Press, 1990.
[264] R. Sharp, Artinian modules over commutative rings. Math. Proc. Cam-
bridge Phil. Soc. 111 (1992), 2533.
[265] D.W. Sharpe, P. Vamos, Injective modules. Cambridge Univ. Press, 1972.
[266] H. Smith, Groups with few non-nilpotent subgroups. Glasgow Math.J. 39
(1997), 141151.
[267] R.L. Snider, Group algebras whose simple modules are nite dimensional
over their commutative rings. Comm. Algebra 2 (1974), 1525.
[268] R.L. Snider, Primitive ideal in group rings of polycyclic groups. Proc.
Amer. Math. Soc. 51 (1976), 810.
[269] R.L. Snider, Injective hulls of simple modules over group rings. Ring the-
ory, Proc. Ohio conf., 1976, 223226.
[270] R.L. Snider, Solvable groups whose irreducible modules are nite dimen-
sional. Comm. Algebra 10 (1982), 14771485.
[271] M.J. Tomkinson, Splitting theorems in abelian-by-hypercentral groups. J.
Austral. Math.Soc.(series A) 25 (1978), 7191.
[272] M.J. Tomkinson, F C-groups. Pitman, 1984.
242 Bibliography

[273] M.J. Tomkinson, Abelian-by-F C-hypercentral groups. Ukrain Math. J. 43


(1991), 10381042
[274] M.J. Tomkinson, F C-groups: recent progress. Innite groups 1994 (Rav-
ello 1994), Walter de Gruyter, 1996, 271285.
[275] M.J. Tomkinson, A splitting theorem for abelian-by-F C-hypercentral
groups. Glasgow Math. J. 42 (2000), 5565.
[276] A.V. Tushev, On the primitivity of group algebras of certain classes of
soluble groups of nite rank. Sbornik Math. 186 (1994), 447463.
[277] A.V. Tushev, Spectra of conjugated ideals in group algebras of abelian
groups of nite rank and control theorems. Glasgow Math. J. 38 (1996),
309320.
[278] O.E. Villamayor, On weak dimension of algebras. Pacic J. Math. 9 (1959),
941951.
[279] J.M. Wedderburn, On hypercomplex numbers. Proc. London Math. Soc. 6
(1908), 77118.
[280] B.A.F. Wehrfritz, Innite linear groups. Springer, 1973.
[281] B.A.F. Wehrfritz, Groups whose irreducible representations have nite de-
gree. Math. Proc. Cambridge Phil.Soc. 90 (1981), 411421.
[282] B.A.F. Wehrfritz, Groups whose irreducible representations have nite de-
gree. Proc. Edinburgh Math. Soc. 25 (1982), 237243.
[283] B.A.F. Wehrfritz, Groups whose irreducible representations have nite de-
gree. Math. Proc. Cambridge Phil. Soc. 91 (1982), 397406.
[284] J.S. Wilson, Some properties of groups inherited by normal subgroups of
nite index. Math.Z. 114 (1970), 1921.
[285] J.S. Wilson Locally soluble groups satisfying the minimal condition for nor-
mal subgroups. Lecture Notes Math. 573 (1977),130142.
[286] J.S. Wilson, Soluble products of minimax groups and nearly surjective
derivations. J. pure applied Algebra 53 (1988), 297318.
[287] J.S. Wilson, Soluble groups which are products of minimax groups. Arch.
Math. 50 (1988), 1921.
[288] A.V. Yakovlev, To the problem of classication of torsion-free abelian
groups of nite rank. Zapiski nauchn. semin. Lomi 57 (1976), 171175.
[289] D.I. Zaitsev, Groups satisfying the weak minimal condition. Ukrain. Math.
J. 20 (1968), 472482.
[290] D.I. Zaitsev, On groups satisfying the weak minimal condition. Mat. Sb.
78 (1969), 323331.
Bibliography 243

[291] D.I. Zaitsev, To the theory of minimax groups. Ukrain. Math. J. 23 (1971),
652660.
[292] D.I. Zaitsev, Groups satisfying the weak minimal condition for non-abelian
subgroups. Ukrain. Math. J. 23 (1971), 661665.
[293] D.I. Zaitsev, On soluble groups of nite rank. Groups with restrictions on
subgroups, Naukova Dumka Kiev 1971, 115130.
[294] D.I. Zaitsev, On complementations of subgroups in extremal groups. Inves-
tigations of groups with prescribed properties of subgroups, Math. Inst.
Kiev, 1974, 72130.
[295] D.I. Zaitsev, On the index of minimality of groups. Investigations of
groups with prescribed properties of subgroups, Math. Inst. Kiev, 1974,
72130.
[296] D.I. Zaitsev, Groups with complemented normal subgroups. Some prob-
lems in Group Theory, Math. Inst. Kiev, 1975, 3074.
[297] D.I. Zaitsev, On existence of direct complements in groups with operators.
Investigations in Group Theory, Math. Inst. Kiev, 1976, 2644.
[298] D.I. Zaitsev, On soluble groups of nite rank. Algebra i logika 16 (1977),
300312.
[299] D.I. Zaitsev, Innitely irreducible normal subgroups. Structure of groups
and properties of its subgroups, Math. Inst. Kiev, 1978, 1738.
[300] D.I. Zaitsev, On locally soluble groups of nite rank. Doklady AN SSSR
240 (1978), 257259.
[301] D.I. Zaitsev, Hypercyclic extensions of abelian groups. Groups dened by
the properties of systems of subgroups, Math. Inst. Kiev, 1979, 1637.
[302] D.I. Zaitsev, On extensions of abelian groups. A constructive description
of groups with prescribed properties of subgroups, Math. Inst. Kiev, 1980,
1640.
[303] D.I. Zaitsev, Products of abelian groups. Algebra i logika 19 (1980), 94106.
[304] D.I. Zaitsev, On splitting of extensions of abelian groups. Investigations
of groups with the prescribed properties of systems of subgroups, Math.
Inst. Kiev, 1981, 1425.
[305] D.I. Zaitsev, The residual nilpotence of metabelian groups. Algebra i logika
20 (1981), 638653.
[306] D.I. Zaitsev, On properties of groups inherited by its normal subgroups.
Ukrain. Math. J. 38 (1986), 707713.
[307] D.I. Zaitsev, Splitting extensions of abelian groups. The structure of
groups and properties of its subgroups, Math. Inst. Kiev, 1986, 2231.
244 Bibliography

[308] D.I. Zaitsev, Hypernite extensions of abelian groups. Investigations of


groups with restrictions on subgroups, Math. Inst. Kiev 1988, 1726.
[309] D.I. Zaitsev, On direct decompositions of abelian groups with operators.
Ukrain. Math. J. 40 (1988), 303309.
[310] D.I. Zaitsev, On locally supersoluble extensions of abelian groups. Ukrain.
Math. J. 42 (1990), 908912.
[311] D.I. Zaitsev, M.I. Kargapolov, V.S. Charin, Innite groups with prescribed
properties of systems of subgroups. Ukrain. Math. J. 24 (1972), 618633.
[312] D.I. Zaitsev, L.A. Kurdachenko, A.V. Tushev, Modules over nilpotent
groups of nite rank. Algebra and logic 24 (1985), 412436.
[313] D.I. Zaitsev, V.A. Maznichenko, On the direct decomposition of artinian
modules over hypercyclic groups. Ukrain. Math. J. 43 (1991), 930934.
Index

(G), 17 th -XC-hypercenter, 27
s(G), 19 X-group, 25
t(G), 18 XC-group, 27
XC-hypercenter, 27, 112
algebra XC-hypercentral, 27
locally Wedderburn algebra, 95 XC-hyperecccenter, 112
annihilator XC-nilpotent, 27
of a module, 3
abelian U-group, 15
of a subset of a module, 2 abelian A1 -group, 16
of a subset of a ring, 59 abelian A0 -group, 15
augmentation artinian, 8
ideal augmentation, 51 central direct product, 26
unit augmentation, 51 Charin group, 44
Chernikov, 8
chain condition
complement, 213
ascending, 1
factor
descending, 1
X-central, 112
chain conditions
X-eccentric, 112
Max-, 8
generalized radical, 22
Max-G, 8
hypernite radical, 206
Max-n, 8
just innite, 9
Min-, 8
layer of a group, 9
Min-G, 8
layers of a p-group, 11
Min-n, 8
lower layer of a p-group, 11
cocentralizer, 25
conjugacy classes, 25 minimax, 16
noetherian, 8
formation, 26 Prufer, 8
innitely hereditary, 126 quasinite, 9
overnite, 114 restricted, 96

soluble U-group, 15
group soluble S1 -group, 16
CC-group, 26 soluble A0 -group, 15
F C-group, 25 soluble A1 -group, 14
-invariant, 7 soluble A2 -group, 16
-operator group, 7 soluble A3 -group, 16
246 Index

Tarsky Monster, 9 maximal, 67


upper CC-hypercenter, 27 factor
upper F C-hypercenter, 27 X-central, 112
upper hypercenter, 27 X-eccentric, 112
Frattini submodule of a module,
idempotent 90
orthogonal, 84 hereditary, 194
primitive, 84 indecomposable, 6
injective, 66
maximal condition, 1 injective envelope of a module,
minimal condition, 1 66
module monolith of a module, 6
F C-center, 113 monolithic, 6
I-module, 60 nearly injective, 164
RG-center, 113 noetherian, 2
H-module, 164 periodic, 59
Mc -module, 89 periodic part of a module, 59
X-RG-hypereccentric, 113 Prufer P -module, 72
XC-RG-hypercentral, 113 primary, 60
XC-RG-nilpotent, 113 primary decomposition, 62
XC-RG-center of a module, 113 quasinite, 192
XC-RG-hypercenters, 113 semisimple, 38
XC-center of a module, 113 series
artinian, 2 XC-RG-central series, 113
assasinator of a module, 60 lower DG-central series, 130
complemented submodule, 65 simple, 6
component of a module, 60 socle of a module, 37
composition length of a module, stabilizer, 98
39 strong locally semisimple, 91
conjugate, 98 the upper XC-RG-hypercenter of
decomposable, 6 a module, 113
decompositions the upper XC-hypercenter of a
Z-decomposition, 114 module, 113
F-decomposition, 114 torsion-free, 59
Baer X-decomposition, 113 uniserial, 43
divisible length, 43
R-divisible, 66 upper F C-hypercenter, 113
x-divisible, 66 upper RG-hypercenter, 113
divisible part of a module, 70
envelope divisible, 69 rank
minimal divisible, 69 P -rank of a module, 74
element-module, 156 torsion-free rank, 74
essential extension of a submod- rank of a group
ule, 67 0-rank, 13
Index 247

p-rank, 15 splitting, 213


p-rank of an abelian group , 14 conjugately splitting, 213
Hirsch number, 13
index of minimality, 16
minimax rank, 16
PruferMaltsev rank, 15
reduced rank, 15
section rank, 15
special rank, 15
torsion-free rank, 13
total rank, 16
rationally irreducible, 20
ring
artinian, 2
Dedekind
Dedekind Z-domain, 156
Dedekind domain, 54
element integral over a ring, 54
hereditary, 194
ideal
divisor, 55
greatest common divisor, 55
least common multiple, 56
relatively prime, 56
ideal fractional, 54
invertible, 54
product, 54
integral closure, 54
integrally closed, 54
Jacobson radical of a ring, 84
noetherian, 2
prime spectrum of a ring, 54

series, 87
composition series, 87
factor, 87
renement, 87
term, 87
upper XC-central series, 27
series of a module
ascending Loewy series, 40
composition series, 39
upper socular series, 40
socular height, 40