Cyclic Quasinormal Subgroups of Arbitrary Groups (*).
STEWARTE. STONEHEWER(**) - GIOVANNIZACHER(***)
To Professor Guido Zappa on his 90th birthday
ABSTRACT - In recent years several papers have appeared showing how cyclic quasinormal subgroups are embedded in finite groups and many structure theorems have been proved. The purpose of the present work is twofold. First we show that, without exception, all of these theorems remain valid for finite cyclic quasinormal subgroups ofinfinitegroups. Secondly we obtain analogous results for infinite cyclic quasinormal subgroups, where the statements turn out to be even stronger.
1. Introduction and statement of results.
LetAbe a cyclicquasinormalsubgroup of a groupG. Thus for every subgroupXofG,AXXA hA;Xi. WhenGis finite, then the structure of the normal closureAGofAinGis quite well understood. IfAhas odd order, then [A;G] is abelian andAacts on it by conjugation as a group of power automorphisms ([2]). WhenAhas even order, then [A;G] is nilpo- tent of class at most 2. When dihedral actions are excluded in certain subgroups of productsAX, whereXis cyclic (see below), then [A;G]0has order at most 2 andAacts on [A;G]=[A;G]0as power automorphisms ([3]).
But always [A;G;A] is abelian and A acts on it again as power auto- morphisms ([4]).
Many of the arguments used in obtaining these results involved in- duction on group orders; and at the time, infinite groups did not appear to be the natural setting for the work. However, it has now transpired thatall
(*) 2000Mathematics Subject Classification: Primary: 20E15, 20E34.
(**) Indirizzo dell'A.: Mathematics Institute - University of Warwick, Coventry CV4 7AL (Gran Bretagna).
(***) Indirizzo dell'A.: Seminario Matematico - UniversitaÁ degli Studi di Padova, Via Belzoni, 7 - 35131 Padova.
the major results in [2], [3] and [4] are true without the hypothesis thatGis finite. A large part of the present work, therefore, is devoted to extending those earlier theorems to the case whereGis infinite. Many of the original proofs can be repeated almost verbatim, modulo some initial lemmas, and we shall endeavour to avoid tedious repetition. But of course when Gis infinite we have the new situation allowing the cyclic quasinormal sub- groupAto be infinite also. The arguments in this case are new and we give them in full. As one might intuitively expect, the structure ofAGis simpler whenAis infinite than whenAis finite.
Our work therefore divides naturally into three parts, dealing with the two cases when A is finite, corresponding to jAj odd and jAj arbitrary (analogous to the contents of [2], [3] and [4]); followed by the situation when Ais infinite. In general
all groups G are assumed to be arbitrary,
i.e. finite or infinite. For any setpof primes,Apdenotes thep-component of a cyclic groupA. Recall that all subgroups of a cyclic quasinormal subgroup are also quasinormal (see Lemma 2.3 below). Our main results are as follows.
THEOREM A. Let A be a finite cyclic quasinormal subgroup of odd order in a group G. Then[A;G]is abelian and A acts on it as a group of power automorphisms.
THEOREMB. Let A be a finite cyclic quasinormal subgroup of a group G. Suppose that each cyclic subgroup X of G is quasinormal in A2X, and let N[A;G]0. Then
(i) jNj42and N4A;
(ii) A acts on[A;G]=N as a group of power automorphisms;and (iii) [A;G]is abelian if and only if A acts on [A;G] as a group of power automorphisms.
In Example 5.2 of [3], a finite groupGof order 217is constructed, with a cyclic quasinormal subgroupAof order 27, in which [A;G]0has order 2 and the hypotheses of Theorem Bare satisfied.
THEOREMC. Let A be a finite cyclic quasinormal subgroup of a group G. Then
(i) [A;G][A2;G][A20;G];
(ii) [A20;G]is an abelian 20-group on which A acts as a group of power automorphisms;
(iii) [A2;G]is a2-group of class at most2;
(iv) [A2;G;A]is normal in G and lies in Z([A2;G]);and (v) A acts on[A2;G;A]as a group of power automorphisms.
In Theorems A, Band C, the power automorphisms areuniversal, i.e.
each element maps to thesamepower. Also one can easily check that if we form the product ofAwith any of the abelian subgroups on which we claim that Aacts as a group of power automorphisms, then that product is a modular group, i.e. its subgroups form a modular lattice.
THEOREM D. Let A be an infinite cyclic quasinormal subgroup of a group G. Then
(i) [A;G]is abelian and A acts on it as a group of power auto- morphisms;
(ii) [A;G]is periodic if and only if A\[A;G]1;and
(iii) when A\[A;G]61, i.e. when[A;G]is not periodic, then AGis abelian.
We shall see in Section 4 that the power automorphisms in Theorem D (i) are not universal in general. It follows from Lemma 2.5 below that in all the above theorems, the normal closureAGofAinGis locally nilpotent and so the periodic elements ofAGform a subgroupT, say. In [1], Theorem 2.2, Busetto shows that with the hypotheses of Theorem D, AGAT, T is abelian andT4Zv(G). Busetto also shows thatAacts onTas a group of power automorphisms and AGis a modular group. Clearly AGA[A;G]
and when A\[A;G]1, then [A;G]T. But in general [A;G]6T. In- deed Example 5 in Section 5 shows thatT64[A;G] in general.
Sections 2, 3 and 4 of our paper will establish the above results se- quentially. The final Section 5 contains examples relevant to the many questions that arise naturally throughout our discussion. The notation is standard. ThusAGandAGdenote, respectively, the normal closure and the core of a subgroup Ain a group G. When A is a quasinormal sub- group of G, then we shall write A qn G. Cyclic groups of order n and infinity will be denoted byCnandC1, respectively, and Q8 denotes the quaternion group of order 8. The centre ofGisZ(G) and, for any ordinal a, thea-th term of the upper central series isZa(G). Ifpis prime andAis ap-group, thenV(A) denotes the subgroup generated by the elements of orderp. Whenpis a set of primes, thenp0is the complementary set. Also Gp is a maximal p-subgroup of G and Gp0 a maximal p0-subgroup. The intersection of the normalisers ofallthe subgroups of a groupGis the
norm, writtennorm(G). Finally the subgroup lattice ofGis denoted by L(G) and the lattice of all subgroups of Gcontaining a subgroupH by L(G:H).
2. Finite cyclic quasinormal subgroups of odd order.
Whenp is an odd prime, Lemma 2.3 of [2] considers a finitep-group GAX, whereAandXare cyclic subgroups. Here we need the following generalisation.
LEMMA2.1. Let GAX be a nilpotent group, where A haiis a cyclic p-subgroup, with p an odd prime, and X hxiis also cyclic (possibly in- finite). Then
(i) G is metacyclic and G0is a p-group;
(ii) every subgroup of G is quasinormal;
(iii) G0 h[a;x]i;
(iv) for each integer i,h[ai;x]i h[a;xi]i h[a;x]ii; and (v) each element of G0has the form[a;g], for some g2X.
PROOF. (i) IfX is finite, thenXXpXp0. SoG(AXp)Xp0 and AXp is metacyclic, by [2], Lemma 2.3 (i). Thus G is metacyclic. Clearly G04AXp and so G0 is a p-group. If X is infinite, then A/G and both statements are immediate.
(ii) By (i) and Iwasawa's Theorem (see [11], Theorems 2.4.4 and 2.4.11), G is a modular group. Since G is also nilpotent, all its subgroups are subnormal and therefore quasinormal ([14], Proposition 1.7).
(iii) The easy argument is the same as in [2], Lemma 2.3 (iii).
(iv) WhenXis finite, then we argue as in (i) above and [2], Lemma 2.3 (iv). WhenXis infinite, thenA/GandX1X=CX(A)Cpn, for somen.
Again the result follows from [2], Lemma 2.3 (iv) applied to the split ex- tensionAjX1.
(v) The argument of [2], Lemma 2.3 (v) applies, making use of (i), (iii)
and (iv) above. p
We shall establish Theorem A in a somewhat circuitous way, as in [2]. It turns out that each element of [A;G] has a special form and we deduce this by considering first the caseA\[A;G]1. In fact with this hypothesis we can prove the strongest statement of all our results for a very general si- tuation.
THEOREM2.2. Let A be any quasinormal subgroup of a group G and suppose that A\[A;G]1. Then[A;G]is abelian and A acts on it as a group of power automorphisms.
PROOF. LetXbe a subgroup of [A;G]. Then [A;X]4AX\[A;G]X.
SoAand hence alsoAGnormalise all subgroups of [A;G]. Therefore [A;G]
is a Dedekind group. If [A;G] is not abelian, we factorGby the 20-compo- nent of [A;G]. Then
[A;G]Q8C2C2
and there is a cyclic subgroupYof order 4 in [A;G]. IfAdoes not centralise Y, thenA=CA(Y) has order 2 andAY=CA(Y) is the dihedral group of order 8, withA=CA(Y) a non-central subgroup of order 2, contradicting the quasi- normality ofA. ThereforeAmust centralise [A;G] and thus so also doesAG.
But then [A;G] is abelian, a contradiction. p
Next we considera cyclic quasinormal subgroup A of odd prime-power order in a group G satisfying the hypotheses of Theorem 2.2, i.e.
A\[A;G]1. LetA hai. We will show that [A;G] f[a;g]jg2Gg:
1
In Lemma 2.4 of [2], this is proved for the case in which Gis a finitep- group, with p an odd prime, using the following two results (see [11], Lemma 5.2.11 and [1], respectively).
LEMMA2.3. If A is a cyclic quasinormal subgroup of a group G, then every subgroup of A is quasinormal in G.
LEMMA 2.4. Let A be a quasinormal subgroup of prime order p in a group G. Then AGis an elementary abelian p-group. Moreover, if A is not normal in G, then AG4Z2(G).
It is convenient to recall two more well-known results (see [9], [3] and [13], Lemma 2.1).
LEMMA 2.5. A quasinormal subgroup of a finite group is subnormal.
More generally, a quasinormal subgroup of any group is ascendant.
LEMMA2.6. Let A be a quasinormal subgroup of a group G and let X be an infinite cyclic subgroup such that A\X1. Then X normalises A.
The argument used to establish (1) for finitep-groupsGin Lemma 2.4 of [2] proceeds by induction on jGj. When G is infinite we can use the following (see [8], Lemma 7.1.9).
LEMMA 2.7. Suppose that a group G hA;g1; :::;gni, where A qn G.
ThenjAG:Ajis finite.
Thus in order to prove (1) whenGis infinite, clearly we may assume that Gis finitely generated. ThenjAG:Ajis finite and soAGis finite. Therefore suppose thatGis a counter-example withjAGjminimal. LetA1AG(<A), A2=A1V(A=A1) andA2 ha2i. Then by Lemmas 2.3 and 2.4
[A2;G]A1=A14Z(G=A1):
Hence [A2;G;G]4A1\[A;G]1 and so [A2;G]4Z(G). Therefore there is an elementxinGsuch that [a2;x]2Z(G) and
[a2;x]61:
2
We would like to write [a2;x] in the form [a;g], whereA hai. In fact this can be achieved by generalising a result due to Busetto (see [1]; also [11], Theorem 5.2.12).
LEMMA2.8. Let A be a cyclic quasinormal subgroup of order pn (p a prime) in a group G and let jAGj pm, where 04m<n. Then A4Z2n m(G).
PROOF. Busetto's theorem is the casem0. Suppose that the result is false and letG be a counter-example with jAj minimal. Thenm51. So BV(A)/Gand by choice ofG
A=B4Z2(n 1) (m 1)(G):
3
Letg2G. Ifjgj 1orjgj q`, whereqis a prime different fromp, then AgA. The first case follows from Lemma 2.6 and the second from Lemma 2.5. Thus [A;g]4Ap, by (3), and therefore [B;g]1. On the other hand, if jgj p`, thenB/hB;giimplies [B;g]1. It follows thatB4Z(G). Then
A4Z2n m(G), again by (3). p
Returning to (2) above, sinceA\[A;G]1,x does not normaliseA.
Then Lemma 2.8 shows thathA;xiis nilpotent. Therefore, by Lemma 2.1 (v), there is an elementy2 hxisuch that [a2;x][a;y]. The argument of Lemma 2.4 in [2] now suffices to complete the proof of the following.
LEMMA 2.9. Let A hai be a cyclic quasinormal subgroup of odd prime-power order in a group G. Suppose that A\[A;G]1. Then each element of[A;G]has the form[a;g];for some element gin G.
We should point out that the hypothesis here thatAhasoddorder is necessary, even whenGis finite. This is illustrated in Example 2 in Sec- tion 5.
The next step is to remove the hypothesisA\[A;G]1 in Lemma 2.9.
LEMMA 2.10. Let A hai be a cyclic quasinormal subgroup of odd prime-power order in a group G and suppose that A is not normal in G.
Then each element of[A;G]has the form[a;g];for some g2G.
We have to exclude A/Ghere, as can be seen from the non-abelian group of order 6. In [2], Theorem 2.5, Lemma 2.10 is proved in the case whenGis ap-group. We wish to use the same argument here. SinceAis not normal in G, it follows from Lemma 2.8 thatA4Zm(G) for somem.
Then it is convenient for us to prove the following result.
Let A hai be a cyclic quasinormal subgroup of odd prime-power order in a group G and suppose that A4Zm(G), for some finite m. Then each element of[A;G]has the form[a;g];for some element gin G.
Now we can prove this stronger form of Lemma 2.10 by means of the argument in Theorem 2.5 of [2]. There a counter-example was chosen with jAjminimal. Here we assume the above statement is false and so there is a finitely generated counter-example G. Then jAGj is finite and we can chooseGwithjAGjminimal. LetjAj pn. Then we need to know that the statement holds withAreplaced byAp. But this must be the case, because j(Ap)Gj<jAGj is an immediate consequence of the following elementary result.
LEMMA2.11. Let A be a cyclic quasinormal subgroup of finite order k in a group G. Then AGhas exponent k.
PROOF. Ifk1, the Lemma is trivial. Thus we proceed by induction on kand assume the usual induction hypothesis. Letpbe a prime divisor ofk and let Bbe the subgroup of order p in A. Then B qn G and BG is ele- mentary abelian, by Lemma 2.4. By inductionAG=BGhas exponentk=pand
henceAGhas exponentk. p
Thus our argument above can continue as in Theorem 2.5 of [2] and so Lemma 2.10 follows. As a consequence we obtain
THEOREM2.12. Let A be a cyclic quasinormal subgroup of odd prime- power order pnin a group G. Then[A;G]is an abelian p-group of exponent dividingjAjand A acts on[A;G]as a group of power automorphisms.
The proof is exactly the same as that of Theorem 2.1 in [2], using Lemmas 2.10, 2.1 (iii) and 2.11 above. The power automorphisms here are universal, by [10], 13.4.3 (ii).
We can now extend Theorem 2.12 to the case whereAhas finite odd order, i.e. Theorem A. If AP1P2 Ps is the primary decom- position ofA, then [A;G][P1;G][P2;G] [Ps;G] and eachPiis quasinormal inG. Theorem A follows from Theorem 2.12 by the easy di- vision algorithm argument which was used to establish Theorem 1.1 from Theorem 2.1 in [2]. The only difference here is thatGmay be infinite, but each step in the proof remains valid using the lemmas above.
3. Arbitrary finite cyclic quasinormal subgroups.
It suffices to prove Theorem Bfor the case whereAis a 2-group. For, if APQ, withPthe 2-component andQthe 20-component ofA, then P andQare quasinormal inG. Also [A;G][P;G][Q;G]. By Theorem A, [Q;G] is abelian andQacts on it as a group of power automorphisms. By Lemma 2.5,Pcentralises [Q;G] andQcentralises [P;G]. Thus if Theorem Bholds for quasinormal 2-subgroups (Phere), then it holds generally. The statements referring to power automorphisms follow as in Theorem A.
Thus we may assume thatAin Theorem Bis a 2-group and we have the hypothesis
() X is quasinormal in AX for all cyclic subgroups X of G.
The key to making progress is again the understanding of the structure of the productsAX, whereXis a cyclic subgroup ofG. We need the ana- logue of Lemma 2.1 for the casep2.
LEMMA 3.1. Let GAX, where A hai is a cyclic quasinormal 2- subgroup of G and X hxiis a cyclic subgroup. Then
(i) G is metacyclic and G0is a2-group;
(ii) G0 h[a;x]i;
(iii) for each integer i,h[ai;x]i h[a;x]ii;
(iv) a conjugates[a;x]to a power congruent to1modulo4; and (v) each element of G0has the form[ai;x];for some integer i.
If in addition()holds, then
(iii)0 for each integer i,h[a;xi]i h[a;x]ii;
(iv)0 x conjugates[a;x]to a power congruent to1modulo4; and (v)0 each element of G0has the form[a;xi];for some integer i.
The proofs proceed by considering the cases X finite and X infinite separately. In the former,G(AX2)X20; and in the latter,A/G. The arguments are straightforward, following the pattern established in Lemma 2.3 of [2], Lemma 2.3 of [3], Lemma 2.1 of [4] and Lemma 2.1 above, and may safely be omitted.
From (iii) we obtain
COROLLARY3.2. LetA haibe a cyclic quasinormal2-subgroup of a groupG. Then[A; G] h[a; g]jg2Gi.
We have already reduced the proof of Theorem Bto the case where A haiis a 2-group. WhenGis also a finite 2-group, then Theorem Bis contained in Theorems 2.1 and 2.2 of [3]. The arguments given in Sections 2, 3 and 4 of [3], upto and including the proofs of Theorems 2.1 and 2.2, use the finiteness of G only for the purpose of applying induction. In the present context, following on from Lemma 3.1, whenever this is necessary, we may assume thatGis finitely generated. Then, as in Lemma 2.7 above, we deduce thatAGis finite and this suffices for the induction arguments of [3] to apply here. The presence of elements of infinite order inGis not a problem, essentially because they always normaliseA. Thus we can safely omit the remaining details of the proof of Theorem B, and thereby spare the reader much tedium.
In the same way, we can proceed as in Corollary 4.1 of [3] to obtain the following consequence of Theorem B.
COROLLARY3.3. LetA be a finite cyclic quasinormal subgroup of a groupGsatisfying(). ThenA\[A; G]4Z([A; G]).
We now pass to the proof of Theorem C. HereGis any group andAis a finite cyclic quasinormal subgroup of G. Our arguments reduce easily to the case whereAis a 2-group. Then we have the following.
THEOREM3.4. Let A haibe a finite cyclic2-subgroup, quasinormal in an arbitrary group G. Then
(i) [A;G;A]/G; and
(ii) [A;G;A] f[u;a]ju2[A;G]g.
THEOREM3.5. Let A be a finite cyclic2-subgroup, quasinormal in an arbitrary group G, and put B[A;G;A]. Then
(i) B4Z([A;G]);
(ii) A centralises[A;G]=B;
(iii) A acts by conjugation on B as a group of power automorphisms;
and
(iv) [A;G]is nilpotent of class at most2.
REMARK. As in Theorem 2.12, the power automorphisms here are universal.
WhenGis a finite 2-group, Theorem 3.5 is Theorem 1.3 of [4]. Using Lemma 3.1, Corollary 3.2 and Theorem 3.4 above, the argument of [4]
applies here, without modification, to establish Theorem 3.5.
PROOF OFTHEOREMC. We have
[A;G][A2A20;G][A2;G][A20;G];
because both factors are normal in G and are a 2-group and a 20-group respectively. So (i) follows.
By Theorem A, [A20;G] is abelian andA20acts on it as a group of power automorphisms. Since (A2)G is a 2-group, A2[A20;G]A2[A20;G].
Therefore (ii) is true. Part (iii) follows from Theorem 3.5 (iv). For part (iv), [A2;G;A][A2;G;A2]B, say, sinceA20centralises [A2;G]. By Theorem 3.4 (i),B/G; and by Theorem 3.5 (i),B4Z([A;G]).
Finally (v) follows from Theorem 3.5 (iii), again sinceA20centralisesB.
p It remains to prove Theorem 3.4. When Gis a finite 2-group, this is Theorem 1.2 of [4]. For part (i) in [4], we were able to assume thatGis generated byAand at most 2 other elements. That argument did not re- quireGto be a 2-group or even finite, and so again we may assume that
G hA;x;yi:
4
By Lemma 2.7,AGis finite. Thus we suppose that Theorem 3.4 (i) is false and choose a counter-example (4) with jAGj minimal. Let A hai and
B[A;G;A]. ThenBG1 andB61. Two cases must be distinguished.
Case(i). Suppose that h[a;x]i \ h[a;y]i 61. Ifa commutes with [a;x]
and [a;y], then, as in [4], it would follow thatB1. (This is Lemma 4.2 below.) Thus we may assume that [a;x;a]61. Then
V(h[a;x]i)V(h[a;y]i)4B\Z(G);
contradictingBG1.
Case (ii). Suppose that h[a;x]i \ h[a;y]i 1. For each integer i50, writeAiA2i. Chooseisuch that aacts non-trivially on [Ai;G] and tri- vially on [Ai1;G]. Then following the detailed analysis of the analogous situation in [4], we can show that a acts as a power automorphism on [Ai;G]. Thus suppose thataconjugates each element of [Ai;G] to itsn-th power and letL[Ai;G]. Then
Ln 14[L;A]4Ln 1
and so [L;A]Ln 1/G. But 16[L;A]4B, again contradictingBG1.
This completes the proof of Theorem 3.4 (i).
For Theorem 3.4 (ii), againA haiis a finite cyclic 2-subgroup, qua- sinormal in an arbitrary groupG. We must show that
[A;G;A] f[u;a]ju2[A;G]g:
5
WhenGis a finite 2-group, we argued in [4] by induction onjAjand we do the same here. When jAj44, then j[a;g]j44, for all g2G, by Lemma 2.11. Thusacentralises all [a;g], by Lemma 3.1 (iv), so [A;G;A]1 and (5) is true. Therefore we suppose thatjAj58 and assume the usual induction hypothesis. This means that
[A2;G;A2] f[v;a2]jv2[A2;G]g K;
say. By Theorem 3.4 (i),K/G. Also by Lemma 3.1 (ii),Anormalises each cyclic subgroup ofKand henceAGdoes the same. ThusK is abeliananda acts as a universal power automorphism on K. As in Theorem 3.5 (i), it follows that
[A;G;K]1:
6
Following the analogous argument in Theorem 1.2 (ii) of [4] (without modification), we deduce from (6) that each element of K has the form [u;a], with u2[A;G]. Recall that by Lemma 2.8, eitherA/G (in which case (5) is trivial) orA4Zm(G), for some integerm. Thus we may assume the latter and so there is a central series of Gbetween 1 andK. Then a
simple induction on the length of this series allows us to assume that K1:
Again as in Theorem 1.2 (ii) of [4], we can now deduce from Lemma 3.1 that
[A;G;A2] h[a;g;a2]jg2Gi L;
say. In turn this leads toL f[y;a]jy2[A2;G]g [A2;G;A]. But using Lemma 3.1, we can show thatL[A;G;A]2/G, and so the Three Sub- group Lemma gives
[A2;G;A]4L:
Thus we have equality here and in exactly the same way as we were able to assume thatK1, we may also assume that
L1:
Using Lemma 3.1 (iv), this leads to [A;G;A;[A;G]]1, from which we deduce (5). This completes the proof of Theorem 3.4 (ii). p REMARKS. IfAis aperiodic locally cyclicquasinormal subgroup of a groupG, then, for each primep, thep-componentSofAis quasinormal in G(see [1], Proposition 1.6, or [11], Lemma 6.2.16). Moreover ifSpS, then S/G, as a consequence of Lemma 2.7. Then applying Theorems A, Band C to the p-components of Amakes the structure of AG quite transparent.
Also all the subgroups ofAare quasinormal inG, by Lemma 2.3; and ifNis the join of all the normalp-components ofA, then AG=N4Zv(G=N), by Lemma 2.8.
4. Infinite cyclic quasinormal subgroups.
We move on now to consider an infinite cyclic quasinormal subgroupA of a groupG. As for finite cyclic quasinormal subgroups, the structure of products AX, where X is also cyclic, is of fundamental importance (see Lemmas 2.1 and 3.1 above). Once again it turns out that these subgroups are metacyclic.
LEMMA 4.1. Let GAX, where A haiis an infinite cyclic quasi- normal subgroup of G and X hxiis also cyclic. Then G is metacyclic and G0 h[a;x]i.
PROOF. Clearly the second statement follows from the first. To show thatGis metacyclic, we consider five cases.
Case1.Suppose that XC1 and A\X1. Then by Lemma 2.6, A/Gand there is nothing to prove.
Case2.Suppose that XC1 and A\X61. We may assume that A6/G. LetNAG. ThenN5A\X. Since xmust either invert or cen- tralise N, we can see that the latter must apply. So N4Z(G). Let p be prime and let P=N be any non-trivial p-component of A=N. Then P=N qn G=Nand hence
P=N4Zm(G=N);
for some finite integerm, by Lemma 2.8. ThereforeP4Zm1(G). Since this applies for all primesp, we haveA4Zn(G), for some finite integern, and so Gis nilpotent. The Hirsch lengthh(G)1 and the torsion subgroupTofG is finite. Moreover
L(T) L(AT:A) L(G:A) L(X:A\X);
a distributive lattice, soT is cyclic, by [9] (see also [11], Corollary 1.2.4).
Finally, sinceG=Tis torsion-free, it must be cyclic and soGis metacyclic. It is easy to see thatGis always nilpotent in this case.
Case3.Suppose that XCpn, where p is an odd prime.We claim that X/ G and G is nilpotent:
7
To see this, we argue by induction onjXj. IfjXj p, thenA/G, by Lemma 2.5. SoGis abelian and (7) is true. Therefore assume thatjXj5p2and let X1V(X). Then by the same argument, AX1 is abelian and hence X14Z(G). Now by inductionX=X1/G=X1, i.e.X/G. AlsoG=X1is nilpo- tent and soGis nilpotent, establishing (7).
REMARK. The groupGhere does not have to be a finite cyclic extension of an infinite cyclic group. For example, this is not the case whenxaxp1. ButGis a modular group and all its subgroups are quasinormal, by [7] (see also [11], Theorem 2.4.11).
Case4.Suppose that XC2n. Here we claim that X2/G and AX2is nilpotent:
8
Consider first the casejXj 4. We know that all the subgroups ofAare quasinormal inG. So there are two possibilities.
(a)Suppose that axxia, for some i. ThenX/GandGis abelian. So (8) is true.
(b)Suppose that axxia 1, for some i. Ifi1, thenA/Gandaxa 1. Sox22Z(G) and again (8) is true. On the other hand, ifi 1, then (ax)21.
ButGAhaxiand this would imply thatjG:Aj 2, a contradiction.
We have shown that whenjXj 4,
either GAX or A/G and x inverts A:
9
Now we suppose thatjXj58 and argue by induction onjXjto prove (8).
Again let X1V(X). Then X14Z(G), by (9). Therefore by induction X2=X1/G=X1and soX2/G. AlsoAX2=X1 is nilpotent and soAX2 is nil- potent. Thus (8) is true.
To prove thatGis metacyclic, we may assume thatjXj58, by (9). Since X2/G, we also haveX4/Gand so (a) or (b) above applies toG=X4. If (a) holds, thenX/Gand we are finished. Therefore we may assume that
axxia 1mod X4:
Herei 1. But ifi 1, then moduloX4we get the same contradiction as in (b). Thereforei1 and so we have
axx14ja 1;
for some integer j. Then x 1axx4ja 1 and x 2ax2x4jax 4j. Hence x24j commutes withaand so [A;X2]1. Therefore
(ax 2j)xx4ja 1x 2jx2ja 1:
Thusx normaliseshax 2jiand so doesa. Thenhax 2ji/G andGis me- tacyclic, as required.
NOTE. In Case 4 we have shown that one of the following applies:
(i)X/G; (ii)A/G; (iii)axa 1x4j,x22Z(G) andx4j61.
Example 5 in Section 5 shows that (iii) does occur.
Case5.Suppose that X is finite.LetjXj n, so we may assume thatn is not a prime power. Ifnis odd, thenX/G, by Case 3. Therefore suppose that n is even, but not a power of 2.
We may assume thatA6/GandX6/G.
LetX2 hx2ibe the 2-component ofX. By the note above, (ii) or (iii) must apply toAX2. We distinguish these possibilities.
(a)Suppose that A/AX2. Then ax2 a 1: 10
LetX0 hx0ibe the 20-component ofX. ThenX0/GandAX0is nilpotent, by Case 3. AlsoAX0 is not abelian, sinceA6/G. Therefore
ax0axi0anduxi061.
As a conjugate ofA,hauiqn G. ConsiderhauiX2. Withhauithe infinite cyclic quasinormal subgroup, (i) above cannot apply, otherwiseX2/Gand thenX/G. If (iii) applies, then
(au)x2 (au) 1x4j2 a 1u,
by (10). But then we obtainx4j2 1, contradicting (iii). We are left with (ii).
Then (au)x2 (au) 1a 1u, again by (10). Souau 1 and this contra- dicts the nilpotency ofAX0.
(b)Suppose that(iii)applies to A as a quasinormal subgroup of AX2. Then for somej,
ax2a 1x4j2 andx4j2 61.
If [A;X0]1, thenG hax22ji jXandGis metacyclic. Thus suppose that [A;X0]61. By Case 3,
X0/AX0: 11
So again with u[a;x0], we consider hauiqnhauiX2 and the three possiblities in the note after Case 4. SinceX6/G, (i) cannot apply. If (ii) holds, then hauiis normalised by x2 and hence the action is by inver- sion. So
u 1a 1a 1x4j2u,
i.e.x4j2 1, a contradiction. Finally, suppose that (iii) applies. Then for some integerk,
(au)x2u 1a 1x4k2 a 1x4j2u.
But then by (11) we must havex4k2 x4j2 anduau 1, contradicting (7).
This completes the proof of Lemma 4.1. p
REMARK. Even in a finite group GAX, with A and X cyclic and A qn G, it does not follow in general thatGis metacyclic. (See Example 1 in Section 5.)
For the proof of Theorem D we need one more result.
LEMMA4.2. Let A haibe a cyclic quasinormal subgroup of a group G ha;x;yi. Then
[A;G](A\[A;G])h[a;x]ih[a;y]i:
12
Proof.WhenGis a finite 2-group, this is Lemma 3.3 in [3]. HereGmay be infinite, but we still need the case whereAis finite. This is a routine generalisation of Lemma 3.3 of [3], already used in the proof of Theorem 3.4 above. For convenience we include the argument here.
Thus suppose first thatAis finite. ThenAGis finite, by Lemma 2.7, and we can argue by induction onjAGj. We may assume that [A;G]61. So let N(61) be a normal subgroup ofGcontained inAor in [A;G]. By induction we have fromG=N
[A;G]N(A\[A;G])h[a;x]ih[a;y]iN:
13
Suppose thatAG61. Then there is a normal subgroupA1 ofGwithjA1j prime and A14A. TakeNA1. If N4[A;G], thenN4A\[A;G] and (13) becomes (12). On the other hand, ifN\[A;G]1, then intersecting both sides of (13) with [A;G] again gives (12).
Now suppose thatAG1. Then we may assume, without loss of gen- erality, thatL[A1;x]61, whereA1is again a subgroup of prime order in A. Since A1qn G, we have L4Z(G), by Lemma 2.4. So we can take NL. But thenN4h[a;x]iand again (13) becomes (12).
Finally suppose thatAis infinite. By Lemma 2.7,jAG:Ajis finite and soAG=AGis finite. ThereforeAG61. Suppose thatA\[A;G]1. By the case when Ais finite, withNAG we have (13). Then intersecting with [A;G] we get (12). But ifA\[A;G]61, then there is a normal subgroupN ofGwith 16N4A\[A;G]. Again (13) holds and this is (12). p PROOF OFTHEOREMD. Let A haibe an infinite cyclic quasinormal subgroup of a group G. Suppose that [A;G] is periodic. Then clearly A\[A;G]1. By Theorem 2.2, [A;G] is abelian andaacts on it as a power automorphism. Conversely, suppose that A\[A;G]1. Again [A;G] is abelian by Theorem 2.2. By Lemma 4.1,
[A;G] h[a;g]jg2Gi:
Let g2GandHAhgi. By Lemma 2.7,jAH:Ajis finite and therefore j[a;g]jis finite. It follows that [A;G] is periodic. This proves (ii).
Now suppose that [A;G] is not periodic. By what we have already proved, this is equivalent to A\[A;G]61. By Lemma 2.5,AG is gener- ated by ascendant cyclic subgroups and so it is locally nilpotent, by [6],
Theorem 2. Thus [A;G] is locally nilpotent and so there is an elementg2G such that [a;g] has infinite order. Butanormalisesh[a;g]i, by Lemma 4.1, and soacentralises [a;g]. This applies to any commutator [a;g] of infinite order. Suppose that [a;h] has finite order. Then
[a;hg][a;g][a;h]g
has infinite order and soacentralises [a;hg]. PutH ha;g;hi ha;g;hgi.
By Lemma 4.2,
[A;H](A\[A;H])h[a;g]ih[a;hg]i and soacentralises [A;H]. Thereforeacentralises [a;h].
It follows that A centralises [A;G]. Thus AG centralises [A;G] and [A;G] is abelian. Hence AG is abelian. This completes the proof of
Theorem D. p
The power automorphisms in (i) are not always universal. For example, letHbe an abelian group of typep1and letabe ap-adic integer such that a1mod p(a1mod4 ifp2). Then the map
h7!ha; 14
for allh2H, defines an automorphism ofH. LetGbe the split extension of H by the cyclic group A generated by a. Then every subgroup of G is quasinormal (see [11], Theorem 2.4.11). Ifa61, then [A;G]H. Clearly the power automorphism (14) is not universal in general.
We end this Section, as we did in Section 3, with a brief discussion of how our results extend tolocally cyclicsubgroups. LetAbe a torsion-free locally cyclic group. Then the automorphism group Aut(A) of A is iso- morphic to a subgroup of the multiplicative group of the rationals. Also Aut(A) contains an element a of infinite order, if and only if we have ApA, for at least one primep. In this case, no non-trivial cyclic subgroup ofAis left invariant bya. (See [5].)
THEOREM 4.3. Let A be a torsion-free locally cyclic quasinormal subgroup of a group G. Then the following conditions are equivalent.
(i) There is a non-trivial cyclic subgroup of A which is quasi- normal in G.
(ii) Every subgroup of A is quasinormal in G.
(iii) For all g2G,jhgi:Chgi(A)jis finite.
PROOF. Assume that (iii) holds. Let a2A,a61. Let g2Gand put T ha;gi, L hA;gi and A1A\T. If jL:Aj 1, then A/L, by
Lemma 2.6, and soA1 haihgi/T. SinceA1is finitely generated, we have hai4A1 ha1i/T and sohai/T. If jL:Aj is finite, thenhai4A1qn T and jT:A1j is finite. ThusA1 is finitely generated, hence cyclic, and so hai4A1 ha1iqn T. Then haiqn T, by Lemma 2.3. It follows that haiqn G. Therefore (iii) implies (ii).
Now suppose that (iii) fails, i.e.jhgi:Chgi(A)j 1, for some elementg.
ThenAgA, by Lemma 2.6, and ginduces an automorphism ofAof in- finite order. But thenhaig6 haifor anya2A,a61, as stated above. Thus
haiis not quasinormal inG, and (i) fails. p
THEOREM 4.4. Let A be a torsion-free locally cyclic quasinormal subgroup of a group G. Then
(i) [A;G]is abelian, A acts on it as a group of power automorph- isms and AGis a modular group;
(ii) [A;G]is periodic if and only if A\[A;G]1;
(iii) if A\[A;G]61, then AGis abelian; and
(iv) for every natural number n, Anis quasinormal in G.
The proofs of the above statements can safely be left as exercises, following the arguments of Lemma 4.2, Theorem D and Theorem 4.3. Of course (iv) holds for any abelian quasinormal subgroupAof any groupG, providednis odd or divisible by 4. (See [15], Theorem 1.)
5. Examples.
All the results that have been proved in the previous Sections have depended on showing that a groupGAX, withAandXcyclic subgroups and A quasinormal, is metacyclic, provided certain hypotheses are sa- tisfied. We begin this final Section by showing that hypotheses are re- quired here.
EXAMPLE1. LetBandX be cyclic groups of orders 7 and 9, respec- tively, and letB hbi,X hxi. Form the split extension
HBjX;
where bxb2. Now letC be a cyclic group of order 3 generated by the elementcand form the split extension
GHjC;
wherebcbandxcx4. The action ofconHhas order 3 and the relations ofHare preserved. SoGexists and has order 33:7.
SetABC, a cyclic group of order 21. We claim that A is quasinormal in G:
15
To see this, sinceB/G, we may factor byB, i.e. we may assume thatB1.
ThenGXjC, which is a modular group with all subgroups quasinormal.
Therefore (15) is true. However, G is not metacyclic. For, G0 hb;x3i C21. If GNK, withN/GandN andK both cyclic, then G04N. SinceG=G0C3C3, we must haveG0<N. ThusjNj 32:7 and so N3C9. B ut N3 must centralise B, while CG(B)B hx3i C con- taining no cyclic subgroup of order 9.
Another key result was Lemma 2.9 showing that when A hai is a cyclic quasinormal subgroup of odd prime power order in G and A\[A;G]1, then
[A;G] f[a;g]jg2Gg:
16
This result is also true whenAhas 2-power order, provided condition () holds. (It follows easily from [3], Lemma 2.7.) However, if () is not sa- tisfied, then (16) can fail to hold.
EXAMPLE2. LetHandXbe cyclic groups of orders 8 and 16, respec- tively, and letH hhiandX hxi. We form the split extension
GHjX;
with hxh 1. Letahx2 andA hai, a cyclic group of order 8. Then CG(A)HX2. AlsoA\HA\X1, soGAX and
[A;G] h[h;x]i H2C4:
Thus A\[A;G]1. Also the only elements of form [a;g] are 1 and h 2([a;x]). Therefore
[A;G]6 f[a;g]jg2Gg:
However,
A is quasinormal in G:
17
For, any cyclic subgroup of G of form hhix2ji commutes with A. On the other hand, hhixji, with j odd, has order 16 (because (hixj)2
hixjhix jx2jx2j) and intersects A trivially. So jAhhixjij 27 and hence Ahhixji G. Thus (17) is true.
REMARK. Clearly the subgroupXisnotquasinormal inGand so () is not satisfied here. Also each element of [A;G] has the form [ai;g], for some integeriand elementg. For many purposes this is as useful as (16) and it would be interesting to know if it isalwaysthe case.
We know (see Lemma 2.6) that ifAisanyquasinormal subgroup ofany groupGand ifXis an infinite cyclic subgroup ofGsuch thatA\X1, then X normalises A. It is natural to ask if this then implies that A is normal inG. In fact this is not the case, as is shown by the next two ex- amples.
EXAMPLE3. LetX hxibe an infinite cyclic group and letY hyibe a cyclic group of order 8. We form the split extension
KXjY;
wherexyx 1. Then letA haibe a cyclic group of order 2 and form the split extension
GKjA;
wherexax,yay5. This extension exists, because the relationsy81 andxyx 1are both preserved by thea-action (of order 2). Clearly
A is not normal in G.
However,Ais quasinormal inG. For, consider an arbitrary cyclic subgroup H hxiyjaki, 04j47, 04k41. We have
(xiyjak)axiy5jak:
Case(i).Suppose that j is even. Theny5jyj andAcentralisesH.
Case(ii).Suppose that j is odd. Then [xiyj;a][yj;a]y42Z(G) and hxiyj;aihas nilpotency class 2. Thus (xiyjak)5(xiyj)5ak. Also (xiyj)2y2j and (xiyj)5xiy5j. So (xiyjak)5xiy5jak(xiyjak)aandAnormalisesH.
ThereforeA4norm(G) and soA qn G, as claimed.
The previous group Gis isomorphic to C1j(C8jC2) and involves a dihedral action. We show that the same result can be achieved using odd primes instead of the prime 2.
EXAMPLE4. Letpbe an odd prime and letYbe a cyclic group of order p2 generated by y. Put Y1Y=Yp and y1yYp. Let RZY1, the in- tegral group ring ofY1and letIR(y1 1), the augmentation ideal ofR.
ThenIbecomes aY-module via the natural actionuyuy1, allu2I. We form the split extension K IjY, a free abelian group of rank p 1 extended byCp2.
LetAbe a cyclic group of orderpgenerated byaand define an action of AonKby
uau, allu2I,yay1p.
Then form the split extensionGKjA. ThusAis not normal inG. B ut, as before, we claim that
A4norm(G):
18
For,AcentralisesIYpAL, say. A typical cyclic subgroup ofG, not con- tained inL, is generated by an element of the formgyiuaj, wherei1 modpandu2I. Thengayi(1p)uaj. But we have
g1pyi(1p)uaj: 19
For,hyiu;aiis nilpotent of class 2, since its derived subgroup isYp, which is central inG. Therefore
g1p(yiu)1paj:
Also (yu)1pyu(yu)pyu(ypu(1y1y21 yp1 1))y1pu. Thus (yiu)1pyi(1p)uand (19) is true. Therefore (18) holds andA is quasi- normal in G.
When proving Lemma 4.1, we considered in Case 4 a groupGAX, whereA haiis an infinite cyclic quasinormal subgroup ofGandX hxi is cyclic of order 2n. We showed that if neitherXnorAis normal inG, then
axa 1x4j;
withx22Z(G) andx4j61. In fact this situation does occur.
EXAMPLE5. LetA haibe an infinite cyclic group and letY hyibe a cyclic group of order 2n 1(n53). LetHAYand letjbe an integer.
ThenHhas an automorphism of order 2 defined by a7!a 1y2j, y7!y.
So there is an extensionGofHby a group of order 2 defined by G ha;xjaxa 1x4j;x2n1i:
(See [12], Theorem 9.7.1 (ii).) ThenGAX, whereX hxiandax2x2a.
We claim that
A is quasinormal in G:
20
For, a typical cyclic subgroupHofGis generated by an elementaixk. Ifkis even, thenAcentralisesH. Ifkis odd, then (aixk)2x2k4ij, generatingX2 andX24Z(G). SoH2X2andAHG. Thus (20) is true. Provided 2n 2 does not divide j, we have a group of type described in (iii) of Case 4 of Lemma 4.1.
Our final example also concerns the case when the quasinormal sub- groupAis infinite cyclic. Situations where [A;G] are periodic or torsion- free are familiar. However, themixedcase can also occur.
EXAMPLE 6. Let H ha;yjy81;yay5i. Then H has an auto- morphism of order 2 defined by
a7!a 1; y7!y5: 21
So we can form a split extensionGHjX, whereX hxiis a cyclic group of order 2 with action onHdefined by (21). We claim thatA haiis qua- sinormal in G. To see this, sinceA2/G, we may factor byA2, i.e. assume that a21. Then easy calculations show thataconjugates each element ofGto its 5-th power. ThereforeA qn G. However,
[A;G] ha2i hy4i C1C2:
Acknowledgments. The second author is grateful to the Mathematics Institute, University of Warwick, for hospitality during the preparation of this paper.
REFERENCES
[1] G. BUSETTO,Proprieta di immersione dei sottogruppi modulari localmente ciclici nei gruppi, Rend. Sem. Mat. Univ. Padova,63(1980), pp. 269±284.
[2] J. COSSEY- S.E. STONEHEWER,Cyclic permutable subgroups of finite groups, J. Austral. Math. Soc.,71(2001), pp. 169±176.
[3] J. COSSEY- S.E. STONEHEWER,The embeddingof a cyclic permutable sub- group in a finite group, Illinois J. of Math.,47(2003), pp. 89±111.
[4] J. COSSEY- S.E. STONEHEWER,The embeddingof a cyclic permutable subgroup in a finite groupII, Proc. Edinburgh Math. Soc.,47(2004), pp. 101±109.
[5] L. FUCHS,Abelian Groups, Pergamon Press (New York, 1960).
[6] K.W. GRUENBERG,The Engel elements of a soluble group, Illinois J. of Math.,3 (1959), pp. 151±168.
[7] K. IWASAWA,On the structure of infinite M-groups, Jap. J. Math.,18(1943), pp. 709±728.
[8] J.C. LENNOX- S.E. STONEHEWER,Subnormal Subgroups of Groups, Oxford Mathematical Monographs (Clarendon Press, Oxford, 1987).
[9] O. ORE,On the application of structure theory to groups, Bull. Amer. Math.
Soc.,44(1938), pp. 801±806.
[10] D.J.S. ROBINSON,A Course in the Theory of Groups, Graduate Texts in Math.
80, 2nd edition (Springer, New York, 1996).
[11] R. SCHMIDT,Subgroup Lattices of Groups(Walter de Gruyter, Berlin, 1994).
[12] W.R. SCOTT,Group Theory, Prentice-Hall, Inc. (Englewood Cliffs, N.J., 1964).
[13] S.E. STONEHEWER, Permutable subgroups of infinite groups, Math. Z.,125 (1972), pp. 1±16.
[14] S.E. STONEHEWER, Modular subgroup structure in infinite groups, Proc.
London Math. Soc., (3),32(1976), pp. 63±100.
[15] S.E. STONEHEWER- G. ZACHER,Abelian quasinormal subgroups of groups, Rend. Mat. Acc. Lincei, (9),15(2004), pp. 69±79.
Manoscritto pervenuto in redazione il 15 dicembre 2005.