1 Introduction
Let G be a group.
A subgroup H of G is called permutable (in G) if Hβ’K=Kβ’H for all subgroups K of G.
Groups in which all proper subgroups are permutable, that is, quasihamiltonian groups, have been completely characterized (see [13]); thus the problem arose of understanding the structure of groups for which the set of non-permutable subgroups is small in some sense (see for instance [2, 3, 6, 4]).
A group is called metaquasihamiltonian if all its non-abelian subgroups are permutable.
Metaquasihamiltonian groups were introduced and studied by De Falco, de Giovanni, Musella and Schmidt in [2].
Of course, Tarski groups (that is, infinite simple groups whose proper non-trivial subgroups have prime order) are metaquasihamiltonian so that, in order to avoid pathological cases of this type, it is usual to work within the large universe of locally graded groups; here a group G is said to be locally graded if every finitely generated non-trivial subgroup of G contains a proper subgroup of finite index.
Under this assumption, it was proved that metaquasihamiltonian groups are finite-by-quasihamiltonian, i.e., they have a finite normal subgroup with quasihamiltonian quotient.
In particular, locally graded torsion-free metaquasihamiltonian groups are abelian.
The aim of this paper is to provide a further contribution to this topic by looking at metaquasihamiltonian groups in the general framework of group classes that can be obtained by iterating a restriction on non-permutable subgroups.
This approach has first been used in [7] to obtain a characterization of groups with a finite commutator subgroup (see also [8]).
Let π be a class of groups.
Put π1=π, and suppose by induction that a group class πk has been defined for some positive integer k; then we denote by πk+1 the class consisting of all groups in which every non-permutable subgroup belongs to πk.
Of course, a group class π is not in general contained in π2, as the consideration of the class of simple groups shows (see also [13, Theorem 6.4.1]), but this is certainly the case if π is closed with respect to forming subgroups.
Moreover, for each positive integer k, the class πk+1 is subgroup closed, and πk+1=(πk)2 so that πhβ€πk whenever 2β€hβ€k, and we put
πβ=βkβ₯1πk.
A similar construction was carried out in [7], where normality replaced permutability; the class corresponding to πk in that work will be here denoted by πΒ―k for all k in ββͺ{β}; clearly, in general, π1=πΒ―1=π and πΒ―iβπi for all iβ₯2.
If π is chosen to be the class π of abelian groups, we have that π2 is the usual class of metaquasihamiltonian groups and π3 consists of all groups whose non-permutable subgroups are metaquasihamiltonian.
Members of the class πk will be called k-quasihamiltonian groups (for any k).
Thus 1-quasihamiltonian and 2-quasihamiltonian groups are precisely the abelian and the metaquasihamiltonian groups, respectively.
In this paper, we will investigate the structure of k-quasihamiltonian locally graded groups for kβ₯3.
Let k be any positive integer, and let p1,β¦,pk be pairwise distinct odd prime numbers.
If A is a cyclic group of order p1β
β¦β
pk and x is the automorphism of A which inverts all elements, the semidirect product Ek=γxγβA is a (k+1)-quasihamiltonian group which is not k-quasihamiltonian.
Hence πk is a proper subclass of πk+1 for each positive integer k.
If Hk,q is a quasihamiltonian q-group with infinite commutator subgroup for a prime qβ p1,β¦,pk, then EkΓHk,q is in the same conditions as above, plus it is not finite-by-abelian, i.e., it has infinite commutator subgroup; in particular, πΒ―k is strictly contained in πk.
It is clear that πkβ€πk for all k whenever π and π are arbitrary group classes such that πβ€π.
Notice also that if π is the class of quasihamiltonian groups, then the classes πk and πk do not coincide as the consideration of the direct product of k copies of the quaternion group of order 8 shows.
However, it should be noticed that πkβπk+1 for all positive integers k, which implies that πβ=πβ; it is perhaps unnecessary to make the reader notice that a direct product of k coprime infinite extraspecial groups of prime exponent shows that the inclusion πkβπk+1 is strict.
The main result of the paper provides a characterization of the class πβ’π consisting of all finite-by-quasihamiltonian groups; in fact, we will show that we have πβ’π=πβ in the universe of locally graded groups.
Therefore, the class πβ’π is saturated by those of k-quasihamiltonian locally graded groups when k ranges over all positive integers.
Most of our notation is standard and can be found in [12].
For a full account of permutable subgroups and quasihamiltonian groups, we refer to [13].
2 Statements and proofs
As we pointed out in the introduction, the class πk is closed with respect to subgroups for every k>1, even if the class π=π1 is not subgroup closed.
We will begin this section by considering some closure properties that are inherited from π to the group classes πk.
First of all, notice that if π is a group class which is closed with respect to homomorphic images, the same property obviously holds also for every πk.
Recall that a group class π is local if a group G belongs to π whenever each of its finite subsets is contained in an π-subgroup.
Clearly, a subgroup closed group class π is local if and only if it contains all groups whose finitely generated subgroups belong to π.
Our first lemma shows that the property of being local is inherited by the classes πk.
Lemma 2.1.
Let X be a local group class.
Then, for each positive integer k, the class Xk is also local.
Proof.
Since π1=π, the statement is obvious if k=1.
Suppose now that the class πk is local for some positive integer k.
As πk+1 is subgroup closed, it is enough to prove that a group G belongs to πk+1 provided that all its finitely generated subgroups are πk+1-groups.
Let X be any subgroup of G which is not in πk, and let π²X be the set of all finitely generated subgroups of X which are not contained in an πk-subgroup of X.
Then π²X is not empty because πk is local.
If g is any element of G and Uβπ²X, the subgroup γg,Uγ belongs to πk+1, whence
γgγβ’U=Uβ’γgγ.
It follows that all elements of π²X are permutable in G.
Moreover, γx,Uγβπ²X for all xβX, and so X=γVβ£Vβπ²Xγ is likewise permutable in G.
Therefore, G belongs to πk+1, and hence πk+1 is a local class.
β
Since the class of abelian groups is obviously local, the choice π=π in the above statement gives of course the following interesting special case.
Corollary 2.2.
For each positive integer k, the class of k-quasihamiltonian groups is local.
The following result shows in particular that locally graded k-quasihamiltonian groups are locally (soluble-by-finite), i.e., every finitely generated subgroup has a soluble subgroup of finite index.
Lemma 2.3.
Let X be a class of locally (soluble-by-finite) groups.
Then all locally graded, finitely generated Xk-groups are soluble-by-finite for all kβ₯1.
Proof.
Working by induction on k, we may assume kβ₯2.
Let G be a locally graded, finitely generated πk-group.
If all proper subgroups of Gβ²β² have the πk-1-property, then Gβ²β² is either πk-1 itself or finitely generated (see Lemma 2.1).
In any case, being locally graded, it is soluble-by-finite by the induction hypothesis and hence even G is such.
By Lemma 2.1, it is now possible to take a finitely generated proper subgroup X of Gβ²β² which is not πk-1 (use Lemma 2.1).
Then X is permutable in G, and so |XG:X| is finite.
Clearly, G/XG is a quasihamiltonian group so that Gβ²β²=XG for all such subgroups X; in particular, Gβ²β²β²=Gβ²β², and Gβ²β² is perfect.
Thus the finite group XG/XXG cannot be non-trivial soluble, and hence XXG has the πk-1-property.
By induction, it is soluble-by-finite, along with Gβ²β² and G.
β
Before going on to the next result, we need to recall some definitions given in [7].
Let π be a group class.
A subgroup X of a group G is said to be compressed by π if it contains a normal subgroup N of G such that G/N is an π-group; in this case, such a subgroup N will be called an π-compressor for X in G.
Of course, if the class π is closed with respect to homomorphic images, the core XG of an π-compressed subgroup X is an π-compressor for X in G.
It is also clear that, in any group, the class of finite groups compresses all subgroups of finite index.
Lemma 2.4.
Let X be a group class, and let X,Y be subgroups of a group G such that X is not permutable in G and Yβ€X.
If Y is compressed in G by the class Xk for some integer k>1, then Y is compressed in X by Xk-1.
Proof.
Let N be an πk-compressor for Y in G.
Then X/N is a non-permutable subgroup of the πk-group G/N, and hence it belongs to πk-1, which means that Y is compressed by πk-1 in X.
β
We also need to recall the definition of a Robinson class.
We shall say that a subgroup closed group class π is a Robinson class if every finitely generated hyper-(abelian or finite) group, whose subgroups of finite index are compressed by π, belongs to π and is polycyclic-by-finite.
Recall here that a group is hyper-(abelian or finite) if it has an ascending normal series whose factors are either abelian or finite.
If the class π is closed also with respect to homomorphic images, we have that π is a Robinson class if and only if every finitely generated hyper-(abelian or finite) group whose finite homomorphic images belong to π is polycyclic-by-finite and belongs to π.
The most relevant group class of this type is that of nilpotent groups; a result that was proved by Robinson in [11].
On the other hand, although any polycyclic-by-finite group whose finite homomorphic images are supersoluble is likewise supersoluble (see [1]); it is easy to see that supersoluble groups do not form a Robinson class.
It follows easily from Robinsonβs theorem and from the fact that polycyclic groups are residually finite that also the class πc of all nilpotent groups of class at most c has the Robinson property; thus π is a Robinson class.
Our next theorem proves in particular that k-quasihamiltonian groups form a Robinson class for each positive integer k.
Theorem 2.5.
Let X be a Robinson class of groups.
Then, for each positive integer k, also the class Xk has the Robinson property.
Proof.
The statement is obvious if k=1.
Suppose now by induction on k that πk is a Robinson class for some positive integer k, and let G be any finitely generated hyper-(abelian or finite) group in which all subgroups of finite index are compressed by πk+1.
If all subgroups of finite index of G are permutable, then every finite homomorphic image of G is nilpotent, and so it follows from Robinsonβs theorem that G itself is nilpotent, and hence also polycyclic.
On the other hand, if G contains a non-permutable subgroup X of finite index, Lemma 2.4 yields that every subgroup of finite index of X is compressed in X by the Robinson class πk; then, by definition, X must contain a polycyclic subgroup of finite index, and so G is polycyclic-by-finite in any case.
Let H be a subgroup of G which is not πk.
Then, by induction, there is a subgroup K of finite index in H that is not πk-compressed in H.
Now K=Hβ©L for some subgroup L of finite index in G.
If N is any normal subgroup of finite index of G, then H/Hβ©LGβ©N is not πk, and hence Hβ’(LGβ©N)/(LGβ©N) is permutable in G/(LGβ©N).
Now [10, Theorem A*] shows that H is permutable in G.
This shows that G is πk+1 and completes the proof of the theorem.
β
Let π be a class of groups.
A group is said to be minimal non-π if it is not an π-group but all its proper subgroups belong to π.
We shall say that a group class π is accessible if every locally graded group whose proper subgroups belong to π is either finite or an π-group, or equivalently if any locally graded minimal non-π group is finite.
In particular, a group class π containing all finite groups is accessible if and only if there are no minimal non-π groups in the universe of locally graded groups.
It is easy to show that abelian groups form an accessible group class, and π shares such a property with other relevant classes of groups.
Further information on accessible group classes can be found in [5].
Theorem 2.6.
Let X be a Robinson class made of locally (soluble-by-finite) groups which is local and closed with respect to homomorphic images.
If X is accessible, then the class Xk is accessible for all positive integers k.
Proof.
Assume for a contradiction that the statement is false, and let k be the smallest positive integer such that there exists an infinite locally graded group G which is not an πk-group while all its proper subgroups belong to πk.
Clearly, k>1 since π is accessible.
Moreover, G is finitely generated because πk is a local class by Lemma 2.1, and hence Lemma 2.3 shows that G is actually soluble-by-finite.
Now [9] implies that the Frattini subgroup Ξ¦β’(G) of G has infinite index in G.
Thus if N is any normal subgroup of finite index in G, then there is some maximal subgroup M such that Mβ’N=G, and hence all finite quotients of G are πk.
Since πk is a Robinson class (see Theorem 2.5), we get a contradiction.
β
It can be immediately checked that both the proofs of Lemma 2.3 and Theorem 2.6 can be used to prove similar results concerning the classes πΒ―k; in particular, the analogous of Theorem 2.6 generalizes [7, Theorem 6].
Corollary 2.7.
For each positive integer k, the class of k-quasihamiltonian is accessible.
Lemma 2.8.
Let X be a local group class which is closed by subgroups, homomorphic images and is made of soluble-by-finite groups.
Then every locally graded Xk-group is soluble-by-finite for all kβ₯1.
Proof.
Assume for a contradiction that the statement is false, and let k be the smallest positive integer such that the class πk contains a locally graded group G which is not soluble-by-finite; clearly k>1.
If Gβ²β² were not perfect, then Gβ²β²β² would either have the πk-1-property, and hence G would be soluble-by-finite by our choice of k, or G/Gβ²β²β² would be quasihamiltonian and hence metabelian, a contradiction.
Thus Gβ²β²=Gβ²β²β².
Let N be any proper normal subgroup of Gβ²β².
Then it must be an πk-1-group; otherwise, Gβ²β² would not be perfect, and by the choice of k, it is therefore soluble-by-finite.
Let SN be the largest normal soluble subgroup of N.
Then CGβ²β²β’(N/SN) has finite index in Gβ²β², which means that N/SN is abelian, and so N=SN is soluble.
By Lemma 2.1 and the fact that Gβ²β² is not an πk-1-group, it follows that there is a finitely generated subgroup F of Gβ²β² not having the πk-1-property.
Since every proper normal subgroup of Gβ²β² is πk-1, it follows that F is not contained in any proper normal subgroup of Gβ²β² and hence that Gβ²β² contains a maximal proper normal subgroup S (being also the product of two proper normal subgroups of Gβ²β² still proper), which is therefore soluble.
Being simple, Gβ²β²/S does not contain any proper non-trivial permutable subgroup, and this yields that all its proper subgroups have the πk-1-property.
However, Lemma 2.3 shows that Gβ²β²/S is locally graded, and hence it must be either finitely generated or πk-1 by Lemma 2.1.
In any case, the group G is soluble-by-finite.
β
It was proved in [2] that metaquasihamiltonian groups are soluble with derived length at most 4.
It follows from Lemma 2.8 that locally soluble πk-groups are soluble, and we can now show that their derived length is bounded in terms of k.
Here πn denotes the class of soluble groups of derived length at most n.
Lemma 2.9.
Let G be a soluble group in the class (An)k, where n and k are positive integers.
Then G has derived length at most n+3β’(k-1).
Proof.
Working by induction on k, we may assume kβ₯2.
Obviously, if Gβ²β²β² were not in (πn)k-1, then Gβ²β²=Gβ²β²β²={1} and we would be done.
Thus Gβ²β²β² is in (πn)k-1, and by induction, it is soluble of derived length at most n+3β’(k-2).
It follows that the derived length of G is at most n+3β’(k-2)+3=n+3β’(k-1), and the result is proved.
β
Corollary 2.10.
Let k be a positive integer, and let G be a locally soluble k-quasihamiltonian group.
Then G is soluble of derived length at most 3β’k-2.
The next result shows that a statement similar to that of Lemma 2.3 holds for locally (nilpotent-by-finite) groups, that is, groups whose finitely generated subgroups have a nilpotent subgroup of finite index.
It shows in particular that the class of k-quasihamiltonian groups locally satisfies the maximal condition on subgroups for all kβ₯1.
Lemma 2.11.
Let X be a class of locally (nilpotent-by-finite) groups.
Then all locally graded, finitely generated Xk-groups are nilpotent-by-finite for all positive integers k.
Proof.
Working by induction on k, let G be any finitely generated locally graded πk-group for some k>1; clearly G is soluble-by-finite by Lemma 2.3.
If all subgroups of G having finite index are permutable, then Robinsonβs theorem shows that G is nilpotent.
Thus let H be a subgroup of finite index which is not permutable in G.
Then H has the πk-1-property, and it is therefore nilpotent-by-finite, as well as G.
Induction completes the proof.
β
Our final result characterizes finite-by-quasihamiltonian groups as those locally graded groups that belong to some πk.
To prove this result, we first deal with periodic and torsion-free cases separately.
Lemma 2.12.
Let G be a periodic locally graded Ak for some positive integer k.
Then there is a finite normal subgroup N such that the factor group G/N is quasihamiltonian.
Proof.
Suppose the result is false, and let k be the smallest positive integer for which there is a periodic locally graded πk-group G not satisfying the statement; clearly k>1.
By Lemma 2.1, there is a finitely generated subgroup E which is not πk-1, and hence it is permutable in G; Lemma 2.8 yields that E is finite.
Let X be any subgroup of G; then the product XE is permutable in G, and the index |Xβ’E:X| is finite.
Therefore, every subgroup of G has finite index in a permutable subgroup, and hence G contains a finite normal subgroup N such that G/N is quasihamiltonian (see [3]).
β
Lemma 2.13.
Let G be a torsion-free locally graded k-quasihamiltonian group for some positive integer k.
Then G is abelian.
Proof.
Let k be the smallest positive integer for which there is a torsion-free πk-group G which is not abelian.
Then all non-permutable subgroups are abelian so that G is metaquasihamiltonian, and being torsion-free, it is abelian by [2, Lemma 2.5].
β
Recall now that a soluble-by-finite group G is said to have finite 0-rank r if there is a finite series of subgroups
{1}=N0β’β΄β’N1β’β΄β’β―β’β΄β’Nm=G
such that there are precisely r factors of the series which are infinite cyclic, the others being periodic; if there is no such an integer r, then G will be said to have infinite 0-rank.
It is clear that, for torsion-free abelian groups, the concept of 0-rank and the usual concept of rank coincide.
Recall also that the periodic radical of a group is its largest normal periodic subgroup.
Lemma 2.14.
Let G be a locally graded k-quasihamiltonian group for some positive integer k.
Then the set of all periodic elements is a subgroup.
Proof.
Suppose the result is false, and let k be the smallest positive integer such that there exists a counterexample G in πk; clearly kβ₯2.
By Lemma 2.11, G is locally (nilpotent-by-finite), so we can find two periodic elements x,y in G such that γx,yγ is not finite and the 0-rank of γx,yγ is smallest possible.
If T is the periodic radical of γx,yγ, then we can clearly assume G (=γx,yγ/T) has a normal torsion-free nilpotent subgroup N of finite index, and Lemma 2.13 yields that N is abelian.
If [Nm,x]=1=[Nn,y] for two positive integers n,m, then Nnβ’mβ€Zβ’(G) so that G would be center-by-finite and hence finite-by-abelian, a contradiction.
We may thus assume [Nm,x]β 1 for all integers m.
If Nmβ’γxγ belonged to πk-1 for some m, then we would get [Nm,x]=1, a contradiction.
Therefore, γxγβ’Nm/Nm is permutable in γxγβ’N/Nm for all m, and if we chose m such that (m,oβ‘(x))=1, then γxγβ’N/Nm would even be abelian, being locally nilpotent.
As γxγβ’N is residually abelian, it is also abelian, the final contradiction.
β
Now we deal with the case of 0-rank 1, and in order to do so, the following technical lemma will be very useful, but first define a non-permutability chain of length n in a group G as a chain of subgroups H1<H2<β―<Hn+1 such that Hi is not permutable in Hi+1 for all i=1,β¦,n; it is clear that a group having a non-permutability chain of length n cannot be n-quasihamiltonian.
Lemma 2.15.
Let p be a prime, and let G=γgγβH be a semidirect product of an infinite cyclic subgroup γgγ and a quasihamiltonian p-group H with infinite commutator subgroup.
Then GβAβ.
Proof.
Suppose Gβπk for the minimal possible integer k (clearly kβ₯2).
By the structure of locally finite quasihamiltonian groups, the subgroup H has finite exponent; moreover, there exist an element hβH and an abelian subgroup K such that H=Kβ’γhγ and h acts like a universal power automorphism on K.
By our choice of k, there is a finitely generated subgroup F which is not (k-1)-quasihamiltonian and hence is permutable.
Since G locally satisfies the maximal condition, we can assume F=γgmγβ’S for some integer m and some finite subgroup S of H.
Since H has finite exponent, there exists an infinite subgroup T of K such that Tβ©S={1} and γhγβ’T has infinite commutator subgroup.
Let tβT, and note that, since Fβ’γtγ=γgmγβ’Sβ’γtγ and Sβ’γtγ are subgroups (actually γtγ is normal in Sβ’γtγ), then g1=gm normalizes Sβ’γtγ.
But it normalizes S, and hence tg1=sβ’tm1 for some sβS and some positive integer m1 which is prime to the order of t.
Now there is a power g2 of g1 centralizing S, and since the exponent of H is finite, it is clear by the above argument that there is a power g3 of g2 centralizing T.
Finally, as G locally satisfies the maximal condition on subgroups, there is a power of g3 centralizing h.
All we have proved so far allows us to assume G to be of the form γgγΓγhγβ’K.
Let A and B be subgroups of K such that both γhγβ’A and γhγβ’B have infinite commutator subgroups and Aβ©B={1}=γhγβ©Aβ’B.
Take a in A not commuting with h.
We show that the subgroup
L=γgoβ‘(a)γΓ(γhγβ’B)
does not permute with γaβ’gγ.
Indeed, the only way in which the element
hβ
aβ’gβLβ’γaβ’gγ
could possibly belong to γaβ’gγβ’L is by taking a power a1=am of a such that a1h=a and m<oβ‘(a), and looking at products of the type
a1β’gm+oβ‘(a)β
k1β
(goβ‘(a))k2β
hβ
b
for some integers k1,k2 and bβB; but mβ’1(modoβ‘(a)), and so the exponent m+oβ‘(a)β
(k1+k2) of g can never be equal to 1.
The subgroup L is subject to the same conditions as G, and this allows us to construct non-permutability chains of arbitrarily large length, a contradiction proving the statement. β
Proposition 2.16.
Let G be a locally graded k-quasihamiltonian group for some positive integer k.
If the 0-rank of G is 1, then G is finite-by-quasihamiltonian.
Proof.
Let k be the smallest positive integer for which there is a counterexample G to the statement; clearly kβ₯3.
Lemma 2.14 implies that the set of all periodic elements of G is a subgroup, say T, while G/T is abelian by Lemma 2.13.
Furthermore, there is a finite normal subgroup N of T such that T/N is quasihamiltonian (see Lemma 2.12).
In particular, T/N is locally nilpotent, and its primary components are either abelian or of finite exponent.
The same can then be said about T/NG so that we may suppose T has those properties, replacing G with G/NG (however, note that quasihamiltonianity could not be preserved).
Our next aim is to show that Tβ² is finite.
Having this in mind, suppose first that there is some p-primary component Tp of T with infinite commutator subgroup.
Let g be any aperiodic element of G, and consider the section γg,Tpγβ’Nγgγ/Nγgγ to which an application of Lemma 2.15 gives a contradiction.
Now suppose there are infinitely many non-abelian primary components in T, and let Tp1,Tp2,β¦,Tpn be different non-abelian primary components of T such that pi does not divide |N| for all i=1,β¦,n.
Let gn be some aperiodic element, and put S0={1}, Si=Tp1β’β¦β’Tpi for all i=1,β¦,n.
Suppose an aperiodic element gi+1 has been defined for some i in {0,1,β¦,n-1}.
The group γgi+1γβ’Si+1/Si contains an infinite cyclic subgroup γgi+1mi+1β’si+1β’Siγ
(si+1βSi+1) which is not permutable in γgi+1γβ’Si+1/Si, and we let gi=gi+1mi+1β’si+1.
In particular, the subgroup γgiγβ’Si is not permutable in γgi+1γβ’Si+1.
This construction leads to a non-permutability chain of arbitrary length, which contradicts G being a k-quasihamiltonian group.
Thus we have reduced so far to the case in which the non-abelian primary components of T have finite commutator subgroup and are finite in number so that we can factor Tβ² out and get T abelian.
Let F=γgγβ’S be a non-(k-1)-quasihamiltonian (and hence permutable) subgroup of G, for some gβGβT and some finite subgroup S of T; moreover, write
G=βnββγhnγβ’T,
where h1=g and γhnγβ’Tβ€γhn+1γβ’T for all nββ.
If n is any positive integer, we have g=hnknβ’un, where kn is an integer and unβT, so that the subgroup γhnknγβ’S is permutable in G, being isomorphic to F.
Therefore, γhnγβ’S is a subgroup so that hn normalizes S.
The arbitrariness of n shows that S is normal in G.
Being finite, we may assume S={1}, and hence all subgroups γhnγ are permutable in G.
In particular, each element of infinite order acts as a power automorphism on T, and all elements of prime order and all elements of order 4 of T are central.
Theorem 2.4.11 of [13] now implies that G is quasihamiltonian.
β
Some preliminary work is needed for larger 0-ranks.
Lemma 2.17.
Let G be a group and P a normal subgroup of G such that G/P is free abelian of finite 0-rank rβ₯2 and P is an abelian p-group of infinite exponent for some prime p.
If there is some element hβG acting as an automorphism of infinite order on P, then GβAβ.
Proof.
It is enough to show the result in the case r=2 and G=γgγβ’γhγβ’P for some element gβG of infinite order such that γgγβ©Pβ’γhγ={1}.
Suppose by way of contradiction that G is k-quasihamiltonian for some positive integer k.
Then Proposition 2.16 shows that Pβ’γhγ is finite-by-quasihamiltonian, so there is a finite normal subgroup N such that Pβ’γhγ/N is quasihamiltonian.
Factoring out NG, which is finite, we reduce ourselves to the case in which h acts like a (locally universal) power automorphism on P.
In a similar way, we may assume g acting like a (locally universal) power automorphism on P.
Finally, hg=hβ’t for some tβP, and we even factor out γtγG so that [h,g]=1.
Let a be an element of P not commuting with h, and let ga be a power of g centralizing a.
Then the subgroup γgaoβ‘(a),hγ does not permute with γaβ’gaγ since the element hβ
aβ’gaβγgaoβ‘(a),hγβ’γaβ’gaγ does not belong to γaβ’gaγβ’γgaoβ‘(a),hγ.
Let now an be an element of P having order (n+1)β
oβ‘(a) for nβ₯1.
Then the subgroup Li=γgano(a)i,h,ano(a)iγ is not permutable in
Li-1=γgano(a)i-1,h,ano(a)i-1γ
for all i=1,β¦,n+1 as its quotient by γano(a)iγ is not such in Li-1/γano(a)iγ for the above reasons.
Thus we may construct non-permutability chains of arbitrarily large lengths, and this is absurd.
β
Lemma 2.18.
Let G be a locally graded k-quasihamiltonian group for some positive integer k.
Let N be a normal abelian subgroup of G such that the torsion part T of G is contained in N and G/N is locally cyclic and periodic.
Suppose also that γzγβ’T has a finite commutator subgroup for all aperiodic elements z of G.
If the 0-rank of G is finite and β₯2, then G is finite-by-abelian.
Proof.
Let HΒ―=G/N, take an aperiodic element z of N, and put
GΒ―=(HΒ―βT)Γγzγ,
with HΒ― naturally acting onto T.
It is easy to see that GΒ― is locally isomorphic to a section of G, and hence it must be k-quasihamiltonian.
Indeed, let hΒ―=gβ’NβHΒ―, and take g1βGβT such that
γgγβ©γg1γ={1}βandβ[γgγβ’T,g1]=1,
which is possible by hypothesis and the fact that G/T is abelian; then the section γg,g1γβ’T/γgoβ‘(hΒ―)γ is isomorphic to (γhΒ―γβT)Γγzγ, and the arbitrariness of hΒ― proves our claim.
Then, by Proposition 2.16, the group GΒ― must be finite-by-quasihamiltonian and hence finite-by-abelian.
Thus there is a normal subgroup M of T such that HΒ―βT/M is abelian so that T/M is central in G/M, and we may assume T is central in G factoring out M.
Using induction on k, we may now assume G is a minimal counterexample with respect to k and kβ₯2.
Thus there is a finitely generated subgroup F of G which is not (k-1)-quasihamiltonian and is such that G/Fβ’T is periodic.
We factor out the periodic part of F and assume therefore F being torsion-free and permutable in G; put M=Tβ’F, and note that G/M is abelian of finite rank (in the usual sense of abelian groups).
Now let fβF and gβG be aperiodic elements such that [g,f]β 1, in particular fg=fβ’t for some 1β tβT.
Clearly, |γgγ:Cγgγ(f)|=o(t).
The group γgγ/Cγgγβ’(f)β(TΓγfγ) is isomorphic to a section of G since Tβ’γfγβ©γgγ=1; otherwise, the subgroup γf,gγ would be abelian.
If g1 is another element of G such that some power of it is equal to gβ’m1 for some m1βM, then γgγ/Cγgγβ’(f) embeds in γg1γ/Cγg1γβ’(f).
This shows that the group
G/CGβ’(f)β(TΓγfγ),
is locally k-quasihamiltonian and hence k-quasihamiltonian.
Suppose G/CGβ’(f) is infinite.
Then, by Proposition 2.16, the above group is finite-by-quasihamiltonian, and T contains elements of the same orders of those contained in G/CGβ’(f).
But T is central, and hence this is a contradiction.
Thus G/CGβ’(f) is finite, and f has finitely many conjugates in G.
It follows that each f in F has finitely many conjugates, and hence it is possible to find a finite subgroup of T such that Tβ’γfγ is normal in G.
Therefore, there exists a finite subgroup U of T such that Fβ’U/U is central in G/U.
We factor out this subgroup.
If f is any element of G, then γfγ is central in G is finite-by-quasihamiltonian.
If it is finite-by-abelian, then Gβ² would be finite, so, by induction on the 0-rank, we may assume the 0-rank of G being 2.
In this case, suppose first that the commutator subgroup of G has elements of infinitely many pairwise coprime orders, and let p1,β¦,pn be n of these prime orders.
Then there are two infinite cyclic subgroups, say γaγ and γbγ, such that ab=aβ’t with oβ‘(t)=p1β
β¦β
pn.
Clearly, Tβ’γaγβ©γbγ={1}, and the quotient
Tβ’γa,bγ/γaoβ‘(t),boβ‘(t)γ
is a direct product of n non-quasihamiltonian groups, which means it contains a non-permutability chain of length n.
The arbitrariness of n gives a contradiction.
The index |FG:F| is finite, but then G/FG is periodic and hence even finite-by-quasihamiltonian.
Thus there is a finite normal subgroup N/FG of G/FG such that G/N is quasihamiltonian and periodic.
The non-abelian primary components of G/N have finite exponent, which means they are center-by-finite since G/Tβ’F has finite rank (in the usual sense of abelian groups) and hence finite commutator subgroup by Schurβs theorem.
But they are also in a finite number by what we have shown above, and hence Gβ²β’N/N is finite.
Therefore, Gβ² is finitely generated and even finite.
β
Before moving on to the next lemma, we recall that, in any group G, the set of all elements admitting only finitely many conjugates is a subgroup which is denoted by FCβ‘(G).
Lemma 2.19.
Let G be a locally graded m-quasihamiltonian group for some positive integer m.
If N is any normal abelian subgroup of G such that the torsion part T of G is contained in N and G/N is locally cyclic and periodic, then Tβ€Fβ’Cβ’(G).
Proof.
Suppose that Tβ°Fβ’Cβ’(G), and let xβTβFβ’Cβ’(G).
For all nβ₯1, there are aperiodic elements g1,β¦,gnβG such that
γxγγg1γ<γxγγg1,g2γ<β―<γxγγg1,β¦,gnγ.
The hypotheses show that there is gβG with
γxγγg1γ=γxγγgk1γ<γxγγg1,g2γ=γxγγgk2γ<β―<γxγγg1,β¦,gnγ=γxγγgknγ
and kn=1,kn-1|kn-2,β¦,k2|k1.
Let iβ{1,β¦,n-1}.
If γxγγgkiγβ’γgkiγ permutes with γgki+1γ, then γxγγgkiγ would be normalized by γgki+1γ, which is not possible.
The arbitrariness of n contradicts the assumption that G is m-quasihamiltonian.
β
We now deal with the case in which the 0-rank is strictly greater than 1.
Proposition 2.20.
Let G be a locally graded k-quasihamiltonian group for some positive integer k.
If the 0-rank of G is strictly greater than 1, then G is finite-by-quasihamiltonian.
Proof.
Let T be the periodic part of G, and let x be any element of infinite order.
Proposition 2.16 shows that the subgroup Tβ’γxγ is finite-by-quasihamiltonian so that T is finite-by-abelian (see [13, Theorem 2.4.11]), and hence we may even assume T abelian, replacing G by G/Tβ².
We now let kβ₯2, working by induction on k, and let E be a finitely generated subgroup of G which is not (k-1)-quasihamiltonian.
If G/T has infinite 0-rank, then there is an element of infinite order modulo E so that E is normal in G by [2, Lemma 2.6].
It follows that G/E is abelian (see [13, Theorem 2.4.11]), which means Gβ² is finitely generated and hence even finite, being locally finite.
It is therefore possible to consider only the case in which G/T has finite 0-rank.
Suppose now there is an aperiodic element g not commuting with infinitely many primary components of T.
Let Tp be a p-primary component of T such that
[Tp,g]β {1}.
Lemma 2.17, Proposition 2.16 and the fact that the 0-rank of G is β₯2 yield the existence of an infinite cyclic subgroup
H=γhγβwithβ’[H,Tp]={1}β’andβ’Hβ©Tpβ’γgγ={1};
moreover, it is even possible to choose Tp such that [H,γgγ]β©Tp={1}.
Factoring out Tpβ² for the moment being, we obtain [h,g]=1.
Now γg,hoβ‘(tp)γ does not permute with γtpβ’hγ, where tp is any element of Tp not commuting with g.
We can repeat the process in the subgroup Tpβ²β’γg,hoβ‘(tp)γ getting a contradiction to the fact that G is πk.
This shows that any aperiodic element of G does not commute with finitely many primary components of T.
Let Tp be any p-component of T for some prime p, and assume first that it has finite exponent.
Let xβG be an aperiodic element such that Lx=γxγβ’Tp has infinite commutator subgroup.
Again, we factor out Tpβ² and take yβGβT with Lxβ©γyγ={1}.
Since Lx is finite-by-quasihamiltonian by Proposition 2.16, there is a finite normal subgroup Nx of Lx such that Lx/Nx is quasihamiltonian.
Let C=Cγxγβ’(Lx/Nx).
Then, applying Lemma 2.15 to γyγβ’Lx/γC,Nxγγy,xγ, we reach a contradiction.
If Tp has infinite exponent, then Lemma 2.17 shows that Tpβ’γzγ has finite commutator subgroup for any aperiodic element z of G.
Summing up, the commutator subgroup of Tβ’γzγ is finite for any zβG.
Now let F be any non-(k-1)-quasihamiltonian subgroup of G with maximal 0-rank.
Factoring out the commutator subgroup of the normal subgroup FT (here we use Lemma 2.19 and what we have just proved) makes it possible to apply Lemma 2.18 to each subgroup L of G such that L/Tβ’F is periodic and locally cyclic, and hence L is finite-by-abelian.
Since G/T has finite rank, we can find a finite normal subgroup R of G such that, factoring it out, we get TF central in G.
If the 0-rank of G is strictly greater than 2, by induction on the 0-rank, it follows that G/γfγ is finite-by-abelian for every non-trivial element f of F.
Thus Gβ² is finitely generated and so finite.
We are leaving with the case 0-rank 2.
In this case, suppose first that the commutator subgroup of G has elements of infinitely many pairwise coprime orders, and let p1,β¦,pn be n of these prime orders.
Then there are two infinite cyclic subgroups, say γaγ and γbγ, such that ab=aβ’t with oβ‘(t)=p1β
β¦β
pn.
Clearly, Tβ’γaγβ©γbγ={1} and the quotient
Tβ’γa,bγ/γaoβ‘(t),boβ‘(t)γ
is a direct product of n non-quasihamiltonian groups, which means it contains a non-permutability chain of length n.
The arbitrariness of n gives a contradiction.
Therefore, Gβ² has finitely many non-trivial primary components.
Now G/F is quasihamiltonian and periodic so that it contains only finitely many non-abelian primary components.
These primary components must have finite exponents, and since G/Tβ’F has finite rank (in the usual sense of abelian groups), these non-abelian primary components must even be center-by-finite (recall that T is central) and hence have a finite commutator subgroup by Schurβs theorem.
Thus G/F has a finite commutator subgroup, which means that Gβ² is finitely generated and so finite.
β
Theorem 2.21.
A locally graded group G is finite-by-quasihamiltonian if and only if it belongs to the class Aβ.
Proof.
If G belongs to the class πβ, then Proposition 2.16 and Proposition 2.20 show that it is finite-by-quasihamiltonian.
Conversely, suppose G has a finite normal subgroup N such that G/N is quasihamiltonian.
Clearly, there is some non-negative integer m (depending only on the order of N) such that if β―β€Hn+1β€Hnβ€β―β€H1β€H0 is a descending chain of subgroups of G with either Hi+1=Hi or Hi+1 not permutable in Hi, then Hjβ©N=Hj+1β©N for all jβ₯m.
Thus
Hj+1/Nβ©Hj+1=Hj+1/Nβ©Hjβ€Hj/Nβ©HjβHjβ’N/N
for all jβ₯m.
As Hjβ’N/N is quasihamiltonian, it follows that Hj+1 is permutable in Hj, and hence Hj=Hj+1 for all jβ₯m.
This shows that Hm is quasihamiltonian, and hence G is (m+1)-quasihamiltonian.
β