On the Rarity of Quasinormal Subgroups
JOHNCOSSEY(*) - STEWARTSTONEHEWER(**)
ABSTRACT- For each primepand positive integern, Berger and Gross have defined a finitep-groupGHX, whereHis a core-free quasinormal subgroup of ex- ponentpn 1andXis a cyclic subgroup of orderpn. These groups are universal in the sense that any other finitep-group, with a similar factorisation into sub- groups with the same properties, embeds inG. In our search for quasinormal subgroups of finitep-groups, we have discovered that these groupsGhave re- markably few of them. Indeed whenpis odd, those lying inHcan have exponent onlyp,pn 2orpn 1. Those of exponentpare nested and they all lie in each of those of exponentpn 2andpn 1.
1. Introduction.
A subgroupQof a groupG, such thatQHHQfor all subgroupsHof G, is said to be quasinormal (sometimespermutable) inGand we write Q qn G. The concept was introduced by Ore in 1937 (see[6]) and in [7] he proved that, in finite groups, quasinormal subgroups are always subnormal.
He also proved that quasinormal subgroups are modular. Recall that a subgroupMof a groupGismodularif, for all subgroupsXandY ofG,
hX;Mi \Y hX;M\Yi ifX4Y and
hX;Mi \Y hX\Y;Mi ifM4Y:
Indeeda subgroup Q of a finite group G is quasinormal in G if and only if Q is modular and subnormal in G. Thus the concepts of modularity and
(*) Indirizzo dell'A.: Mathematics Department, School of Mathematical Sciences, Australian National University, Canberra, ACT 0200.
E-mail: John.Cossey@anu.edu.au
(**) Indirizzo dell'A.: Mathematics Institute, University of Warwick, Coventry CV4 7AL, England.
E-mail: S.E.Stonehewer@warwick.ac.uk
quasinormality coincide in finite p-groups. In 1973 Maier and Schmid proved in [5] that if Gis a finite group andQ qn G withQcore-free, i.e.
QG1, thenQ lies in the hypercentre of G. It follows fairly easily from this that the complexities of the embedding ofQinGreduce to the case whereG is a p-group, i.e. where Q is a modular subgroup. Bearing in mind that modularity is a property invariant under subgroup lattice isomorphisms, the quasinormal subgroups of a finitep-groupGare surely relevant when the structure ofGis approached via its lattice of subgroups.
It was shown in [10] that givenQ qn GandQ abelian, then Qnqn G, providedn is odd or divisible by 4 (even whenGis infinite). Apart from this, very little appears to be known about which subgroups ofQare also quasinormal inG.
ClearlyQ qn Gif and only ifQXis a subgroup for each cyclic subgroup X of G. Thus the situationGQX is an obvious starting point for in- vestigations. The case when G here is a finite p-group (for p an odd prime), andQis an abelian quasinormal subgroup ofG(withXcyclic), is studied in [2]. It is shown that there are two composition series of G passing throughQ, all the members of which are quasinormal in G. The attempt to remove the hypothesis that Q is abelian has produced the present work. We have discovered a situation very far removed from the abelian case. In fact there is a family of finitep-groupsG(for each odd primep) withQ qn Gsuch that
(i) there are subgroupsL3M4Q, withL,M qn G;
(ii) Ghas no quasinormal subgroup strictly betweenLandM;
(iii) Lhas exponentpandMhas exponentpm; and
(iv) mand the nilpotency class (even derived length) ofM=Lcan be greater than any given positive integer.
These groupsGwere constructed by Berger and Gross in [1] and they are defined as follows. Letnbe a positive integer andpbe an odd prime.
We defineGnto be the additive groupZof integers modulopnZ. Letxnbe the permutation ofGn given by
xn: pnZ `7!pnZ ` 1:
Then xn belongs to the symmetric group of degree pn. For 04m4n, define
xn;mxpnn m; of orderpm:
Let Xn hxni, a cyclic group of order pn; and let Xn;m hxn;mi, the subgroup of orderpm. DefineDn;mto be the set of elements in the additive
groupGn of orderpm. So
Dn;m flpn mj04l4pm 1; pj44lgfor 14m4n
andDn;0 f0g. ThenjDn;mj pm 1(p 1) for 14m4n. The permutation xn;m shifts bypn mand fixes the setDn;m1, i.e. the set of lpn m 1 (pj44l) between 0 andpn 1, for 04m4n 1. Thusxn;macting on the setDn;m1 (of cardinality pm(p 1)) is the product ofp 1 disjoint cycles pn;m;i of lengthpm, 14i4p 1, each cycle addingpn m. So
pn;m;i(ipn m 1; ipn m 1pn m;. . .; ipn m 1(pm 1)pn m):
Now for 04m4n 1, define An;mYp 1
i1
pcn;m;ii Xp 1
i1
ci0
:
Each permutation in this set is a product of disjoint cycles, each containing an integer of the formlpn m 1(pj44l) and shifting by a multiple ofpn m. In fact An;m is an abelian group isomorphic to the direct product of p 2 cyclic groups of order pm. (See [1], Lemma 3.1 (1).) Its elements fix every integer not inDn;m1.
Berger and Gross define the group
Gn hxn;An;mj04m4n 1i;
andHn to be the stabiliser of 0 inGn. So Gn HnXn;
a finitep-group. AlsoHn has exponentpn 1. The main result of [1] is:- THEOREM. The subgroup Hn is core-free and quasinormal in Gn. Moreover, if H is a core-free quasinormal subgroup of a finite p-group GHhxi, wherehxiis a cyclic group of order pn, then there is a unique embeddingcof G in Gn such thatc(x)xn andc(H)4Hn.
Our main results concern the quasinormal subgroups ofGn that lie in Hn. In the Berger-Gross Theorem above, the primepis arbitrary. How- ever, the casep2 requires much additional analysis. Accordingly in all our work here we assume that
the prime p is odd:
We shall prove
THEOREM 1. Let n52 and let Q be a non-trivial quasinormal sub- group of Gnlying in Hnof exponent pk.Then one of the following holds:-
(i) k1;
(ii) n54and kn 2;
(iii) n53and kn 1.
Moreover for n54, the quasinormal subgroups of exponent p are nested and are contained in all those of exponentpn 2and exponentpn 1. Thus there is a maximal quasinormal subgroup L of exponent p and a quasinormal subgroupMof exponentpn 2such thatL4M and
there are no quasinormal subgroups of Gnin the interval[M=L]:
ClearlyL3M. AlsoM=Lhas exponentpn 3and there is no bound to its derived length asnincreases. (See [9].)
In Theorem 2, we find all the quasinormal subgroups ofGnlying inHn
and of exponent p. Theorem 3 gives first some necessary conditions for subgroups ofGn, lying inHnand of exponentpn 1, to be quasinormal inGn; then necessary and sufficient conditions for certain subgroups of exponent pn 1to be quasinormal inGn. Theorem 4 is similar, dealing with subgroups of exponentpn 2. Finally in Theorem 5 we show that there are no quasi- normal subgroups of Gn (n55), lying in Hn and of exponent pk, for 24k4n 3. Theorem 1 follows from these results.
We use standard notation for familiar concepts, together with that above introduced by Berger and Gross. Also we switch from multiplicative to additive notation for computation with modules over groups. In this connection there is a considerable amount of calculation involved and the following list should be helpful.
(a)b;g: the cyclic permutation containinga, shifting byb, of lengthg. Hereais a non-negative integer and b and g are positive integers such that bgpn.
fbi;kjrk5i51g: a basis (moduloVk 1(Hn)) of the elementary abelian group Vk(Hn)=Vk 1(Hn).
B`k;k: the subgrouphbi;kj`k5i51i.
Bk: the subgroupBrk;k.
Ck: the centraliser inGn of the element xn;k, i.e. CGn(xn;k), chiefly when k2.
rk: the rank of the elementary abelian group Vk(Hn)=Vk 1(Hn), for 14k4n 1.
Xn: the cyclic grouphxniof orderpn.
xn;m: the elementxpnn m, 04m4n.
Xn;m: the cyclic subgrouphxn;miof orderpm.
fwijr15i51g: a basis of the elementary abelian group V1(Hn), and wib1;i.
Wj: the subgrouphwijj5i51i(=Bj;1).
Zpm: a cyclic group of orderpm.
Vk(G): the subgroup of the finitep-groupGgenerated by the elements of order at mostpk.
2. Description of the groupsGn. It is shown in [1], Lemma 3.14, that
Vk(Gn) hxniAn;kxinji50iVk 1(Gn);
1
for 14k4n 1. Also
Vk(Gn)=Vk 1(Gn)is elementary abelian of rank pn k 1(p 1):
2
We see this as follows. By [1], Lemma 3.3, if n52, then there is an epi- morphismtn:Gn ! Gn 1, viz.g7!tn(g) where
tn(g): amodpn 1Z7!bmodpn 1Z 3
if g: amodpnZ7!bmodpnZ;
for all integersa,bandg2Gn. In particulartnmapsHnontoHn 1andxnto xn 1. The kernel oftnis elementary abelian, by [1], Lemma 3.3 (9), and it is precisely V1(Gn), by [1], Lemma 3.10. AlsojGnj ppn1, by [1], Corollary 3.15. Thus
V1(Gn)is elementary abelian of rank pn 1 pn 2pn 2(p 1):
Applyingtnrepeatedly for decreasing values ofn, we see that Gn=Vk 1(Gn)Gn k1;
4
for 24k4n; and
V1(Gn=Vk 1(Gn))Vk(Gn)=Vk 1(Gn)V1(Gn k1);
5
this last subgroup being elementary abelian of rankpn k 1(p 1), proving (2). We shall use (4) and (5) repeatedly throughout our work.
By [1], Lemma 3.10,
V1(Gn)V1(Hn)Xn;1: Also, by (1),
V1(Gn) hxniAn;1xinji50i;
an elementary abelian group of rankpn 2(p 1), by (2). By [1], Lemma 3.2 (4), xn;1lies in the centre of Gn:
Let
r1pn 2(p 1) 1rank ofV1(Hn);
6
for n52. Recall that Hn is core-free in Gn and so V1(Gn) is an in- decomposableXn-module. We find a canonical basis forV1(Gn) as follows.
Define
wr1xpnn2 1(pn;1;11 pn;1;2)xn(pn2 1)2V1(Gn):
7
We introduce the following shorthand notation for cyclic permutations of integers modulopnZwhich just add a fixed integer:-
(a)b;g(a; ab; a2b;. . .; a(g 1)b);
8
wherebgpn. Then
pn;1;11 pn;1;2(pn 2)pn11;p(2pn 2)pn 1;p2An;1:
Note that the two cycles here are disjoint, coming from different orbits of xn;1. Also
wr1(1)pn11;p(1pn 2)pn1;p;
where again the two cycles are disjoint. Therefore wr1 fixes 0 and so wr1 2V1(Hn).
Forr1 15i50, define
wi[wi1;xn] 9
inductively. Then dropping the suffixes in (8) for the moment and writing rr1andxxn, we have
wr 1(1)(1pn 2) 1(2) 1(2pn 2)(1)(2) 1(1pn 2) 1(2pn 2), wr 2(1) 1(2)2(3) 1(1pn 2)(2pn 2) 2(3pn 2),
wr j(1)( 1)j1(2)?. . .(j1pn 2),
w1(1) 1(2)?. . .(pn 1 1), w0(1)(2)?. . .(pn 1).
Any 2 cycles above are either the same or disjoint. Hence they commute.
Also wi, for r1(r)5i51, fixes 0 and so belongs to V1(Hn). Clearly
wr1; wr1 1;. . .; w0 are linearly independent as elements of the module
V1(Gn), since each contains a cycle that does not occur in its predecessors.
The rank ofV1(Gn) isr11, by (2). Therefore
fwijr15i51gis a basis for V1(Hn) 10
and
fwijr15i50gis a basis for V1(Gn):
11
Clearly V1(Hn) is an indecomposable Xn-module. For if not, then the centraliser of Xn in Gn is not cyclic and therefore intersects V1(Hn) non-trivially. Then the normal closure of this intersection in Gn would lie in Hn, giving a contradiction. Also we must have w02 hxn;1i and xn;1(1)(2). . .; (pn 1), with cycles of length p, shifting by pn 1. Thus
w0xn;1xpnn1: 12
NOTE. For anyi,j, using additive module notation, we have [wi;xpnj]wi(xpnj 1)wi(xn 1)pj wi pj: 13
Herewi, fori50, is taken to be 1.
Next, for 24k4n 1, we find a basis for Vk(Gn)=Vk 1(Gn) as in- decomposableXn-module of rank
pn k 1(p 1)rk1;
14
say (see (2)). Notation will necessarily becomes more complicated. By [1], Corollary 3.11,
Vk(Gn)Vk(Hn)Xn;k: 15
Thus by analogy with (7), define
brk;kxpnn k 1(pn;k;11 pn;k;2)xn(pn k 1 1)
for 14k4n 1. Then using the notation (8), we have pn;k;11 pn;k;2(pn k 1)pn k1 ;pk(2pn k 1)pn k;pk2An;k and
brk;k(1)pn k1 ;pk(1pn k 1)pn k;pk: 16
Sincebrk;k fixes 02Gn, it belongs toVk(Hn). Whenk1, we have br1;1wr1:
Under the isomorphism (5),brk;k moduloVk 1(Gn) corresponds to (1)pn k1 ;p(1pn k 1)pn k;p
17
inV1(Gn k1), since in the setZ=pn k1Zon whichGn k1acts,pn k10.
Also (17) is br1;1in Gn k1; and under the homomorphism (3), xn7!xn 1. Therefore defining
bi;k[bi1;k;xn] 18
forrk 15i50, we see from (10) and (11) that
fbi;kjrk5i51gis a basis forVk(Hn) moduloVk 1(Gn);
even moduloVk 1(Hn), since (as fork1)bi;k 2Hnifk60. Also fbi;kjrk5i50gis a basis forVk(Gn) moduloVk 1(Gn):
19
Note thatbi;1wi, for alli. Also
tnmapsbi;2inGn tobi;1(wi) inGn 1; 20
for alli. We shall use the `w' -notation when we wish to emphasise the fact that we are consideringV1(Gn) asXn- (orGn-)module.
By [1], Lemma 3.1 (4),
bi;kcommutes withxn;k; 21
for alln 15k51 andrk5i50. Let
Bk hbi;kjrk5i51i;
14k4n 1. ThenBkhas exponentpkand Yn 1
k1
Bk
Xn Gn: 22
LEMMA1. Let14k4n 1.Then
(i) hBk;Xn;ki Zpk Zpk (with rk1factors); and (ii) [wi;BkXn;k]1, for pn k 15i51,24k4n 1.
PROOF. (i) By (21), [Bk;Xn;k]1. Let CkCGn(xn;k). So, writing XXn,hBk;Xi4CkandCk(Hn\Ck)X. Then by (15),
Vk(Ck)Vk(Hn\Ck)Xn;kDk;
say. SinceDk is a characteristic subgroup ofCk, it follows that the derived subgroupD0k is normalised byX. ButD0k4Hnand so (D0k)Gn4Hn. There- fore sinceHnis core-free inGn, we must haveD0k1, i.e.
Dk is abelian:
23
However,hBk;Xn;ki4Dkand sohBk;Xn;kiis abelian of exponentpk. Thus (i) follows from (19). Observe that
Bk hbrk;ki hbrk 1;ki hb1;ki;
which is homogeneous of exponentpk. (ii) Letk52. By (13),
[wi;xn;k]wi pn k 1;
for 14i4pn k 1. Thereforewi2Dk and (ii) follows from (23). p We have bases (19) for the V-layers of Gn (as Xn-modules) and the homomorphism tn: Gn ! Gn 1 maps these bases to those of Gn 1 (see (20)). In fact the bases that we have chosen have another very useful property withinGn itself. We claim first that
bpr2;2br2;1(wr2):
24
For, from (16) withk2, where the two cycles are disjoint, bpr2;2Yp 1
i0
(1ipn 2)pn11;p(1pn 3ipn 2)pn 1;p: 25
The cycles in (25) are disjoint and hence commute. By (11) and (19) (with k2),
V2(Gn)B1B2Xn;2:
Therefore by Lemma 1 (i), and using the notation of that Lemma, D2(D2\B1)B2Xn;2
andD2is abelian, by (23), and is normalised byXn. Thus again by Lemma 1 (i), Dp2 Bp2Xn;1is elementary abelian
of rank r21 and is an Xn-submodule of V1(Gn). Hence, setting Wj hwijj5i51iforr15j50, we must have
bpr2;22Dp2\HnWr2: 26
Now from the equations following (9) and a simple calculation, we have br2;1(1)pn11;p(2)?pn1;p. . .(1pn 3(p 1)pn 2)pn1;p:
27
Observe that the last cycle here is the same as the last cycle in (25). But from (26),
bpr2;2rY2 1
i0
bar2i i;1;
for integersai. Therefore, again by the equations following (9), we must have ar2 1 ar2 2 a10
and so
bpr2;22 hbr2;1i:
Thus (24) follows by comparing the exponents of (1)pn 1;pin (25) and (27).
From (18) we deduce that
bpr2 1;2[br2;2;xn]p[bpr2;2;xn];
sinceBX2n4D2, which is abelian. Therefore, by (24) and (9), bpr2 1;2[br2;1;xn]br2 1;1:
Continuing in this way, we see that
bpi;2bi;1(wi);
28
r25i51. Also, by (18),
b0;2[b1;2;xn]xn;2 mod V1(Gn);
29
using (12) and (20). Moreoverhb0;2;xn;2i4C2 and V1(Gn)\C2Wpn2 1Xn;1 (by (13)). Therefore
bp0;2xn;1(b0;1):
30
Combining the above results, we can now obtain useful information about adjacentV-layers inGn. For 24k4n 1, we have from (4)
V2(Gn k1)V2(Gn=Vk 1(Gn))Vk1(Gn)=Vk 1(Gn):
31
Thus we deduce from (28) and (30) that
bpi;k1bi;k mod Vk 1(Hn);
32
forrk15i51; and
bp0;k1xn;kb0;k mod Vk 1(Gn):
More information about adjacent V-layers of Gn will follow from our next key result.
LEMMA2. Let24k4n 1and for04i4pk 2(p 1), let ViWipn k 1:
Then Bkcentralises the series
V1(Hn)WrVpk2(p 1)> >Vi>Vi 1> >V01 33
of length pk 2(p 1).
PROOF. By Lemma 1 (ii),BkcentralisesV1. We argue by induction oni increasing and suppose that [Vi1;Bk]4Vifor somei50. Then by (13)
[Vi2;Xn;k]Vi1Xn;1; and so
[Vi2;Xn;k;Bk]4Vi:
Since [Xn;k;Bk]1, by Lemma 1 (i), the Three Subgroup Lemma (see for example [8], Lemma 2.13) implies
[Bk;Vi2;Xn;k]4ViXn;1(3Gn):
Therefore [Bk;Vi2]4Vi1Xn;1, again by (13). Thus [Bk;Vi2]4Vi1Xn;1\HnVi1
and the Lemma follows. p
COROLLARY1. For 04i4n 2, the factor groupVi2(Gn)=Vi(Gn) is nilpotent of classp 1and hence regular.
PROOF. By (31) we may assume thati0. Take k2 in Lemma 2.
ThenB2centralises the series (33) of lengthp 1. Also, moduloXn;1,Xn;2
centralises this series (by (13)). Therefore, sinceB2Xn;2 is abelian (by Lemma 1 (i)), the subgroup
V2(Gn)B1B2Xn;2
is nilpotent of class at mostp 1. In fact the class is exactlyp 1, because of the action ofXn;2 onV1(Gn)WrXn;1. The regularity follows, for ex-
ample, from [3], Corollary 12.3.1. p
3. Subgroups ofGn.
We can now describe all the quasinormal subgroups ofGn lying inHn
and of exponentp.
THEOREM 2. Let n52.Then a quasinormal subgroup of Gn of ex- ponent p, lying in Hn, has the form Wi, for some i.Moreover Wiqn Gnif and only if14i4p2 1.
PROOF. Observe from (6) that whenn2,i4p 2; and whenn3, i4p(p 1) 1. Therefore in both these cases we are saying that Wiqn GGnfor allifor whichWiis defined.
Let Q qn G with Q of exponent p and lying in HHn. Then Q4Wr1 V1(H). Also QXn is a subgroup and V1(QXn)QXn;1. Thus QXn;1is anXn-submodule ofV1(G) and so, since indecomposable modules are uniserial (see [4], Theorem VII, 5.3),QXn;1WiXn;1, for somei. Hence QWi.
We establish first the sufficiency of the condition oni. Thereforesup- pose that i4p2 1 and letghy, whereh2Handy2XXn, and put K hgi. We show thatWiKis a subgroup by considering 3 possibilities.
(i)Suppose that y2Xp2Xn;n 2. ThenycentralisesWi, by (13). Since Wi3H, it follows thatgnormalisesQ.
(ii)Suppose thathyi X. Then by [1], Lemma 3.12 (ii),V1(K)Xn;1. ThereforeWiK WiXn;1K is a subgroup, sinceWiXn;1/G.
(iii) Finallysuppose thathyi Xp. Clearly we may assume thatn53.
By Corollary 1, Vn 1(G)=Vn 3(G) is regular with elementary abelian de- rived subgroup. Therefore
gphpyp mod Vn 3(G):
Also hp2 hbj;n 2j14j4p 2i mod Vn 3(H);
by (32) and (14). Continuing takingp-th powers in this way, we obtain gpn2 2Wp 2Xn;1:
34
MoreoverjHK:Hj jHhyi:Hj jhyij pn 1. Thereforegpn22=H.
Now considerWiK. Ifi4p 1, thenycentralisesWi(by (13)) and so again g normalises Wi, i.e. WiK is a subgroup. On the other hand if p4i(4p2 1), then by (34)
WiK WiXn;1K and this is a subgroup as in (ii).
Conversely, suppose, for a contradiction, that Wiqn G where i5p2. Then by (6), n54. Let gb1;n 1xn;n 2 and K hgi. We have [b1;n 1;xn;n 1]1, by (21). So
gpn2bp1;nn21b1;1 w1; 35
using the argument of (iii) above. Consider the subgroupWiK. SinceV1(G) is elementary abelian, an element of orderpinWiK must have the form wg1, where w2Wi and g12V1(K)W1 (by (35). Thus V1(WiK) Wi3WiK. SinceWi3H, it follows that xn;n 2 also normalises Wi. But [wp2;xn;n 2]xn;1 (by (13)), a contradiction. This completes the proof of
Theorem 2. p
For the next theorems we need to be able to identify certain subgroups ofGn. Letn52,n 15k51 andrk5`k51. Define
B`k;k hbj;kj14j4`ki;
which is isomorphic toZpk Zpk (`k copies), by Lemma 1 (i).
Note that B`1;1W`1.
LEMMA 3. Let n52, rk5`k5pn k 1 1 for 14k4n 3, rn 25
`n 25`n 1and rn 15`n 151.Then
LB`1;1B`2;2. . .B`n1;n 1Xn;n 1 36
is a subgroup of Gn; and
B`1;1B`2;2. . .B`n1;n 1L\Hn 37
is a subgroup of Hn.
PROOF. Ifn2, thenLW`1Xn;13G2. Ifn3, thenLW`1B`2;2Xn;2
with r1p(p 1) 15`15`2 and r2p 25`251. By Lemma 1 (i), B`2;2Xn;2is a subgroup. AlsoW`1Xn;13G3. SinceLis the product of these two subgroups, againLis a subgroup.
Now suppose that the Lemma is true forn 1, somen54, and argue by induction onn. Then from (4) we obtain
LV1(Gn)Wr1L
is a subgroup. Recall that C2CGn(xn;2). It follows from (13), (22) and Lemma 1 (i), that
C2Wpn2 1Br2;2. . .Brn1;n 1Xn:
Then Wr1L\C2Wr1B`2;2. . .B`n 1;n 1Xn;n 1\C2
Wpn2 1B`2;2. . .B`n1;n 1Xn;n 1
is a subgroup. Since `15pn 2 1, forming the product of this subgroup with the normal subgroupW`1Xn;1, we obtainL. SoLis a subgroup.
Equation (37) follows from Dedekind's Intersection Lemma. p The quasinormal subgroups of exponent pn 1 that we shall exhibit in Theorem 3 all have the formL\Hngiven by (37); and those of exponent pn 2 in Theorem 4 have the formVn 2(L\Hn). As a consequence of the next result,we see that these subgroups are actually normal inHn.
LEMMA4. Let n53, rk5`k5rk p2for14k4n 3, rn 25`n 25`n 1 and rn 15`n 151.Define L as in(36).Then
(i) L is a subgroup of Gn; and (ii) for14k4n 1,
Vk(L)B`1;1B`2;2. . .B`k;kXn;k Lk; 38
say; and Lk3Gnfor14k4n 2.
Moreover, if rk5`k5rk p for14k4n 2and rn 15`n 151, then (iii) again L is a subgroup of Gnand L3Gn.
PROOF. (i) The hypotheses here imply those of Lemma 3. For, with n53 and 14k4n 3,
rk p2 (pn k 1 1)pn k 1(p 1) p2 pn k 1pn k 1(p 2) p2 5p2(p 2) p250:
ThereforeLis a subgroup ofGn.
(ii) It is easy to check that, for 14k4n 2,`k5`k1. We claim that, for 24k4n 1,
(B`k;kXn;k)pB`1;1B`2;2. . .B`k 1;k 1Xn;k 1: 39
We do not know yet that the right side of (39) is a subgroup. In order to prove (39), suppose that k2. ThenB`k;kXn;k Zp2 Zp2 (`k1 copies), by Lemma 1 (i). Thus (39) follows from (28), since`15`2. Therefore (39) is true for n3. We suppose that (39) is true forGn 1 (n54) and 24k4n 2 and argue by induction onn. Then for 34k4n 1, we have from (20)
(B`k;kXn;k)pWr1B`2;2. . .B`k 1;k 1Xn;k 1: 40
However, the left side of (40) and all but the first factor on the right belong toC2(CGn(xn;2)). Thus intersecting (40) withC2gives
(B`k;kXn;k)pWpn2 1;1B`2;2. . .B`k 1;k 1Xn;k 1
B`1;1B`2;2. . .B`k 1;k 1Xn;k 1; since one checks easily that
`15pn 2 1:
41
Therefore (39) is true.
Now we can prove (38). Clearly Vn 1(L)LLn 1. We proceed by induction onkdecreasing and suppose that (38) is true for somek4n 1.
Certainly
Vk 1(L)B`1;1. . .B`k 1;k 1Xn;k 1: 42
If the inclusion (42) is strict, then there is an element g2B`k;kXn;k with jgj pk 1andg2=Lk 1. But by Lemma 1 (i),g2(B`k;kXn;k)p, contradicting (39). Therefore (42) is an equality. Thus our induction argument goes through and we have proved (38).
The second part of (ii) will follow from Ln 23Gn: 43
When n3,LW`1B`2;2X3;2 andL1W`1X3;13G3. Therefore we sup- pose that n54 and that (43) is true for Gn 1 and proceed by induction.
Again from (20)
Wr1B`2;2. . .B`n2;n 2Xn;n 23Gn; and intersecting withC2gives
Wpn2 1B`2;2. . .B`n 2;n 2Xn;n 23C2:
Thus from (41) andW`1Xn;13Gn, we see thatLn 2 is normalised by C2. However, [Wr1;Ln 2]4Wr1 p2Xn;1, by Lemma 2, and so Wr1 normalises Ln 2. SinceWr1C2Gn, (43) follows.
(iii) It is easy to check that the hypotheses here imply those of parts (i) and (ii). SoLis a subgroup. Now however
[Wr1;L]4Wr1 pXn;14L;
by Lemma 2, and soWr1normalisesL. Also whenn3, by consideringL modulo V1(G3) and intersecting with C2, we see that L3Gn. Then the induction argument used to prove (43) can be applied here to giveL3Gn
for alln. p
4. Quasinormal subgroups of large exponent.
We shall see that the quasinormal subgroupsQofGnlying inHn, other than those of Theorem 2, all have exponentpmformn 2 orn 1. Also Vk(Q) modulo Vk 1(Hn) has `large' rank, for all 14k4m 1. We begin with those of exponentpn 1.
THEOREM3. (i)Let n53 and Q be a quasinormal subgroup of Gn of exponent pn 1and lying in Hn.Then for14k4n 1,
Vk(Q)B`k;k modVk 1(Hn);
44
and`k5rk p, for14k4n 2, and`n 151.
(ii)Again let n53and
QYn 1
k1
B`k;k
with`k5rk p for14k4n 2and`n 151.Then Q is a subgroup of Hnof exponent pn 1.Moreover Q qn Gnif and only if
`k5rk p1;
45
for14k4n 2.
PROOF. (i) Since QXn is a subgroup, we see that V1(QXn) is an Xn- module contained inV1(Gn) and therefore has the formWiXn;1. Also there is an elementhb1;n 12Qwithh2Vn 2(Hn), by Theorem 2. Thus sinceWr1 must normaliseQ,
[Wr1;hb1;n 1]4Q\Wr1 p;
by Lemma 2. However
this commutator subgroup does not belong to Wr1 p 1:
For, otherwise we would have [Wr1;b1;n 1]4Wr1 p 1. Then conjugating by xn, we would obtain (using (18))
[Wr1;b1;n 1b0;n 1]4Wr1 p 1Xn;1: 46
But b0;n 1xn;n 1 mod Vn 2(Gn), using (29) and the homomorphismtn. Thusxn;n 1would centraliseWr1Xn;1=Wr1 p 1Xn;1, a contradiction.
Thus (44) is true fork1. The rest of (i) follows by induction onnfrom Gn=V1(Gn)Gn 1.
(ii) The hypotheses here imply those of Lemma 4. ThusQis a subgroup ofHnof exponentpn 1(of the form (37)).
Now suppose that Q qn Gn. Assume for the moment that
`15r1 p1:
47
Then we obtain (45), again by induction onn. So it suffices to prove (47).
Letgwr1xnandK hgi. Then
V1(K) hxn;1i;
48
by [1], Lemma 3.12 (2). In fact this follows easily from a simple module calculation. Thus using additive notation,
gp(wr1xn)pwr1(1xn1xn2 xn(p 1))xn;n 1
wr1(xn1 1)p 1xn;n 1 wr1(xn 1)p 1xn;n 1 mod Wr1 pXn;1
wr1 p1xn;n 1 mod Wr1 pXn;1: 49
(The appearance ofwr1 p1here is the clue to establishing (47).) Similarly gp2wr1 p21xn;n 2 mod Wr1 p2Xn;1:
(We can obtain (48) by computinggpn1in this way.) Since`1>r1 p21, it follows from (48) that
xn;n 22QK:
50
Alsob1;n 12Q, by hypothesis, and soQKcontains [b1;n 1;g][b1;n 1;xn][b1;n 1;wr1]xn: However, by Lemma 2,
[b1;n 1;wr1]xn2W`x1nW`g14QK:
Hence
QKcontains [b1;n 1;xn]g1; 51
say. Nowg12QGn4QXn;n 1, sinceQXn;n 13Gn, by Lemma 4 (iii). From (4) and (12) we have
g1xn;n 1g2;
where g22Vn 2(Gn)Vn 2(Hn)Xn;n 2. Therefore g2hy, with h2 Vn 2(Hn) andy2Xn;n 2. Thusy2QK, by (50). Then
g12QXn;n 1)g22QXn;n 1)h2QXn;n 1\Hn Q)g22QK) )xn;n 12QK;
by (51).
Finally from (49) we now obtain
wr1 p12QK\HnQ:
Thus`15r1 p1, as required.
Conversely, suppose that `k5rk p1 for 14k4n 2 and`n 151.
We show thatQ qn Gn. Letghy, whereh2Hn,y2Xn, and letK hgi.
It suffices to show that QK is a subgroup. Let y2Xn;knXn;k 1, where 04k4n. (Assume thatXn; 11.) We claim that
gp2QXn;k 1: 52
Note thatQXn;n 13Gn, by Lemma 4 (iii). ThusQXn;iis a subgroup for alli;
and
Q3Hn: 53
Let hwb2b3. . .bn 1, where w2Wr1 and bi2Bi, 24i4n 1. Con-
sideringV1(Gn) asGn-module, we have
gp(wb2. . .bn 1y)pw((b2. . .bn 1y) 1 1)p 1(b2. . .bn 1y)p: Thus
gp2Wr1 p1C2; 54
by Lemma 1. We prove (52) by induction onn.
Supposen3. If k42, then gp2Wp 2X3;k 1, by Corollary 1, and so gp2QX3;k 1. Ifk3, then by (54)
gp2Wr1X3;2\Wr1 p1C2(Wr1\Wr1 p1C2)X3;24QX3;2:
So (52) holds forn3. We proceed by induction onn, assuming thatn54 and that (52) holds inGn 1. Thus by induction (usingGn=V1(Gn)Gn 1) and (54), we have
gp2Wr1Xn;1QXn;k 1\Wr1 p1C24(Wr1Q\Wr1 p1C2)Xn;1Xn;k 1
Wr1 p1(Wr1Q\C2)Xn;1Xn;k 14Wr1 p1(Wpn2 1Q)Xn;1Xn;k 1; by the argument used in Lemma 3. Therefore gp2QXn;1Xn;k 1. So (52) follows ifk52. But ifk41, thengp2Hnand sogp2Q(QXn;k 1). Thus we have established (52).
Now from (52) we obtainQhgpi4QXn;k 1. But
jQhgpij jQkhgpi:Q\ hgpij jQjpk 1 jQXn;k 1j:
Therefore
Qhgpi QXn;k 1; 55
a subgroup. By Lemma 4
Vk 1(QXn;n 1)Vk 1(Q)Xn;k 1N;
say, and N3Gn. If k4n 1, then by Lemma 1 (i), Xn;k centralises Q modulo N and therefore g normalises Q(by (53)). Otherwise kn and thenNQXn;n 1Qhgpi, by (55); i.e.QKis a subgroup.
This completes the proof of Theorem 3. p
Next we deal with the quasinormal subgroups of exponent pn 2. The argument follows that of Theorem 3, though there added complications.
Establishing the necessity of our conditions for quasinormality involves considering the product QK forK hwr1b1;n 1xn;2irather than hwr1xni;
and the prime 3 causes problems.
THEOREM4. (i)Let n54and Q qn Gn, Q4Hn and Q have exponent pn 2.Then for14k4n 2,
Vk(Q)B`k;k mod Vk 1(Hn);
56
and`k5rk p2, for14k4n 3, and`n 251.
(ii)Again let n54and
QYn 2
k1
B`k;k;
with`k5rk p2for14k4n 3and`n 251.Then Q is a subgroup of Hn
of exponent pn 2.Moreover Q qn Gnif and only if
`k5rk p(p 1)
for14k4n 3and`n 25p 2.
PROOF. (i) We argue exactly as in Theorem 3 (i) to obtain the form (56).
Then for the restrictions on `k we use the fact that there is an element hb1;n 22Q, with h2Vn 3(Hn). Thus [Wr1;hb1;n 2] lies inQ\Wr1 p2 (by Lemma 2), but not inWr1 p2 1.
(ii) WithLdefined as in Lemma 4, we have QVn 2(L)\Hn: HenceQis a subgroup.
Now suppose that Q qn Gn.We shall show that
`15r1 p(p 1) 57
and
`25p 2 whenn4:
58
Then the necessity of our conditions will follow by induction on n, as in Theorem 3. We shall prove that
wpn2 2Q;
59
i.e.`15pn 2. Assume this for the moment and let gwr1b1;n 1xn;2; K hgi:
Then sinceQKis a subgroup, (59) implies [wpn2;g]2QK:
Therefore substituting forgand expanding (using (13)), we get [wpn 2;b1;n 1]xn;12QK:
Thusxn;1 2QK. Considering actions on V1(Gn) and using Lemma 2, we obtain
gpwbp1;n 1xen;1; where
w2Wr1 p(p 1)nWr1 p(p 1) 1: 60
The appearance of the suffixr1 p(p 1) in (60) is the clue to establishing