Modular Permutations on Z.
FRANCESCODELCASTILLO(*)
ABSTRACT- In this paper the groupMof permutationssofZfor which an integer nn(s)>0exists such that (zn)szsnfor everyz2Zis studied.Mis countably infinite locally (abelian-by-finite) and contains all finitely generated (abelian-by-finite) groups as subgroups. The commutator subgroup M0is an infinite simple group and the quotient groupM=M0is isomorphic toZ. Finally, all abelian groups that can be represented as modular permutation groups are determined: these are countable abelian groups whose quotient over the torsion subgroup is free.
1. Definitions and first results.
Let s be a permutation of the set Z of the integers: we say that n2Nn0 is amoduleforsif
(zn)szsn;
for everyz2Zor, equivalently, ifscommutes with the translationtnand in this case we say thatsismodular. If itexists, such anis notunique, since any multiple of it is still a module. The set of all modular permuta- tions is a group, denoted by M. For s2 M, we denote by mod (s) t he smallestmodule fors. Thennis a module forsif and only if mod (s) divides n. Mn is the subgroup of M whose elements are the permutations for which n is a module. A modular permutation s2Mn preserves the n- congruence relation, i.e. z1z2(modn) if and only if z1sz2s(modn).
Thus, a modular permutation s2Mn induces a permutationsn of then cosets ofZ=nZ. This fact allow us to write, for everyz2Z,
zs(qz;na[z]n)n([z]n)sn (1:1)
(*) Indirizzo dell'A.: Dipartimento di Matematica ``Ulisse Dini'', UniversitaÁ degli Studi di Firenze, Viale Morgagni, 67a, 50134 Firenze.
E-mail: [email protected]
whereqz;n, [z]nare resp. the quotient and the remainder of the (euclidean) division ofzbynanda[z]nis an integer depending only on [z]n. With a slight abuse we adopt the same notation for both the remainder and the classes of n-congruence and will letsn act on both these objects; for this reason we will choose the setf0;. . .;n 1gas support for the symmetric group Snin place of the more usual f1;. . .;ng. Conversely we can associate to each r2Sn a permutation er2Mn justby setting a[z]n0 in the 1.1:
zerqz;nn([z]n)r. Thus, for everyn1, two homomorphism are defined ln:Sn !Mn
r7!er
fn:Mn!Sn
s7!sn
such that lnfn1. fn is therefore surjective and ln determines an iso- morphic copyfSnof SninM. LetDnbe the kernel offn: a permutationl2Dn can be written, according to 1.1, as
zlza[z]nn:
The features of l(which fixes each n-congruence class and acts ``inside'' each of them as a translation bya[z]n) suggest us to introduce a special class of modular permutations: forn1 andI f0;. . .;n 1gthe permutation ln;Idefined, forz2Z, by
zln;I zn if zi (modn) for somei2I z otherwise
is called modular translation. In caseI figwe justwriteln;i andln;i is called anelementary modular translation(shortenedEMT). Itis clear that ln;IQ
i2Iln;i and mod (ln;i)n. It is pure routine to check that
Dn hln;iji0;. . .;n 1iand thatDnis a free abelian group of ranknfor
which theln;iform a basis.
Furthermore, Mn splits as the semidirect productDnjSfn: if x2Mn, once takensxfnln2Sfnone has immediatelyxs 12Dnand thenxls, with l2Dn;s2Sfn uniquely determined. To this particular factorization we will refer (unless otherwise stated) in the following when we say
``xls2Mn''. We also observe that this factorization means that Mis locally (abelian-by-finite).
Sfnin its action by conjugation onDnpermutes the elements of the basis according to the natural action of Sn, namely one has (ln;i)s~ln;is; thus Mnis isomorphic to the permutational wreath productZwrnSn. This al- lows us to state the first of two theorems about groups that can be em- bedded inM, which is somehow a converse of the local (abelian-by-finite) structure ofM.
THEOREM1.1. Let G be a finitely generated (abelian-by-finite) group.
Then G can be embedded inM.
PROOF. LetA>!G!!Hbe exact, withAabelian andHfinite.Ais the directproductTF of its torsion subgroupTand a free abelian group of rank, say,k. PutNFGthe core ofFinG;Nis still free abelian of rankk.
The indexjG:Njis finite and thenUG=Nis a finite group of order, say,n.
According to the Kaluznin-Krasner theorem ([2], theorem 6.2.8.), Gis iso- morphic to a subgroup ofN wr U'Zkwr U. We now show how t he lat t er can be embedded inM. First we can embed the standard wreath product in the permutational wreath product (via the Cayley regular representation for finite groups)Zk wrU Sn; now,ZkwrU Sn(Zk)njSn9ZknjSkn, where, in the last embedding, Snis identified with the subgroup of Sknact ing on t hen blocks fik;. . .;ik(k 1)g of length k. The theorem is proved, since ZknjSkn'ZwrknSkn'Mkn. p
2. The commutator subgroup ofM.
In this section we show thatM0 is an infinite simple group (Theorem 2.8) and that the quotient groupM=M0is isomorphic toZ(theorem 2.5).
For this purpose it is useful to have another way of writing the elements of M in terms of certain basic elements. Let Gn be the subgroup of Mn
generated by the modular translations whose modules dividen, Gn hlm;ijmdividesn; i0;. . .;m 1i
and let Gn be the image ofGn through the homomorphismfn. Since the image of lm;i is easily checked to be the cycle 0 m . . . n m of lengthn=min Sn, we haveGn h 0 m . . . n m jmdividesni.
THEOREM2.1. For every n2, if Gn (Gn)fn, the following equality holds:
Gn Sn if n is even and not a prime power An if n is odd and not a prime power a p-Sylow subgroup of Sn if n is a prime power
8<
:
In order to prove the theorem we recall a result due to Jordan ([4], Theorem 13.9):
THEOREM(Jordan, 1873). Let p be a prime and G a primitive group of degree npk withk3. If G contains an element of degree and order p, then G is either alternating or symmetric.
PROOF OFTHEOREM2.1. The firsttwo cases (nnota prime power) can be proved via Jordan's theorem. Letnpqkwithp<qdifferentprimes andk2N. Since the cycle 0 qk. . . n qk 2Gnis an elementof order and degreep, one just has to check thatGnis primitive. Letf0g 6Dbe a block containing 0 and kl1;0fn 0 1 . . . n 1. Let u be the smallestnon-zero elementofD; since 0kuu,Dcontains the whole orbit of 0 under the action ofku and, since it is the smallest nonzero element, u mustbe a divisor ofn and we musthaveD f0;u;. . .;n ug. Ifu61, consider a divisorvofnsuch thatvis neither a multiple nor a divisor ofu.
If rlv;0fn, one has the contradiction 0rv62D, uru2D: then we musthaveu1 andD f0;1;. . .;n 1g, i.e.Gn is primitive. Now, ifnis odd the image of anylm;iis a cycle of odd length and therefore it lies in An; otherwise, ifnis even,Gncontains some odd permutation and must be the whole of Sn.
Proving the case ofna prime power is a matter of counting elements.
We recall that the order of a p-Sylow subgroup of Spk is pr, where r1p pk 1. We then proceed by induction on k2N. Let Pk Gpk. The case k1 is trivial. Suppose now that Pk is a p-Sylow subgroup of Spkand considerPk1Spk1. Denote again byPk the image ofPkvia the usual embedding of Spkin the pointwise stabilizer of the set fpk;pk1;. . .;pk1 1g in Spk1, let kl1;0fpk1 and g2Spk1 be a per- mutation such that, for 0zpk 1, zgpz. LetP0k(Pk)g; t hen
P0k h((lpj;0)fpk)g j0j<ki h(lpj;0)fpk1 j0<j<k1i andP0khas supportf0;p;2p;. . .;pk1 pg. For 0<ip 1, letPik(P0k)ki. ThePik's are subgroups ofPk1and have pairwise disjointsupports. Thus they generate their direct product P h(lpj;i)fpk1j0<j<k1; 0ip 1i, whose order isjP0kjpp(ppk). Furt hermore,k62Pand (Pik)kp Pik.
Letnow a(lp;0)fpk1 and bak. We have 16b2Pk1nP,bp1 and b normalizes P. Thus, hP;bi Pjhbi and hP;bi has order p(1ppk), i.e. itis a p-Sylow subgroup of Spk1. Finally, since b acts on P essentially in the same way as k, we get Pjhbi hP0k;ki
h(lpj;0)fpk1j0j<k1i Pk1. p COROLLARY2.2. Ifnis even and not a prime power, thenGnMn. PROOF. Letx2Mn. Since the restriction offntoGn is surjective one can takeg2Gnsuch thatgfnxfn, and thenxgdfor somed2DnGn
and thereforex2Gn. p
COROLLARY2.3. Mis generated by the modular translations.
Thus, ifx2 M, besides the factorization xls in some Mn, we also havexQ
laii with theli's elementary modular translations and theai's integers. The latter expression is not unique but will be useful in this section as well as in the next one.
Consider the application Cn:M !Z defined, for x2 M, by xCnnP1
i0(zx z).Cn is the sum of the shiftings of the first nintegers under the action ofx; the restrictioncnofCnt o Mnis a homomorphism, as a directcheck shows: letx;y2Mn, then
(xy)cn Xn 1
z0
((zx)y z)Xn 1
z0
((zx)y zx)Xn 1
z0
(zx z)
Xn 1
z0
[(a[z]nnzx)y (a[z]nzx)]xcnXn 1
z0
(zy z)xcnycnxcn: Furthermore,Sfn ker(cn) andln;icn n. Thus ndividesxcn for every x2Mn.
PROPOSITION 2.4. The application C defined, for x2 M, by xC
lim
i!1
xCi
i is well defined and, ifx2Mn,xCxcn
n . In particularCis a surjective homomorphism ofMontoZ.
PROOF. Letxbe an elementof Mnand writeiqn[i] (division ofi byn). We have
xC lim
i!1
1 i
X
qn 1
z0
(zx z)Xi 1
zqn
(zx z)
!
lim
i!1
1
i q(xcn)[i] 1X
z0
(zx z)
!
lim
i!1
q(zcn) qn[i]lim
i!1
P
[i] 1
z0 (zx z)
i :
(2:1)
Ifzcnknthe first limit of the last member of the (2.1) isk, while the second one equals 0, since [i] 1P
z0 (zx z)
nP1
z0jzx zj (an i-independent
constant). p
What can be observed in the last proposition is that, ifxQn
i1laii is a factorization ofxin terms of EMT's, sinceliC1, thenxCPn
i1ai. Thus, although the EMT's that occur in the factorization ofxare notuniquely determined, the sum of theai's is indeed unique. In a certain sense the EMT's play a role inMsimilar to that of transpositions in finite symmetric groups and one can also regard the value ofCatxas a sortof ``signature''.
However the usefulness of EMT's is clear in the following results.
THEOREM2.5. M0kerCand thenM=M0'Z.
PROOF. The inclusionM0kerC is obvious sinceC has an abelian image. In order to prove the opposite inclusion, we note that, if xPk
i1liai2kerCwith theliEMT's, sinceP
ai0, we can suppose, up to multiplication by an element ofM0,ai( 1)iandkeven (sayk2r).
Thus it suffices to show that, for every n;m2N and every possible i;j, ln;i1lm;j2 M0.
Suppose then thatxln;i1lm;j. Ifnm1 there is nothing to prove.
If nm2, let s2Sn be such that isj. Then one has ln;i1ln;jln;i1(ln;i)s~[ln;i;es]2[Dn;Sfn]M0n. We also note that [Dn;Sfn]
kerC\Dn (since ford2Dn;es2Sfn, clearly (d 1es 1des)C0).
Finally, if n6m, let s2l:c:m:(n;m) and consider ln;i andlm;j as elements of Ms: bot hln;ifs andlm;jfs are odd permutation of Ss, and then ln;i1lm;j2AfsDs M0sDs. Thusln;i1lm;j2(M0sDs)\kerCM0s(Ds\kerC)
M0s[Ds;Ses]M0s. p In order to obtain information about the commutator subgroupM0we need to better understand the structure of the groups Mn. From now on we assume n5 and setKn kerC\Mn. Justby simple observations, as applying the isomorphism and the correspondence theorems, it is easy to verify the following facts:
Dn\Kn[Dn;Sfn][Dn;Mn] is a free abelian group of rankn 1 with basisfln;01ln;iji1;. . .;n 1g;
Kn[Dn;Mn]jSfn; ifUnDnjAfn, M0nUn; M0n[Dn;Mn]jAfn<4Kn. LEMMA2.6. For n5,M00nM0n.
PROOF. [Dn;Mn]M00n: it suffices to show that every generator ln;01ln;i of [Dn;Mn] belongs to M00n. If 06j6i one can take s2An
such that jsj and 0si. With this choices one has at once ln;01ln;i[ln;0ln;1j;es]2[[Dm;Mn];[fSn;Sfn]]M00n. Even more trivially Afn[Afn;Afn]M00n and putting things together gives M0n
[Dn;Mn]jAfnM00n. p LEMMA2.7. For n5, let H be a normal proper subgroup ofM0n. Then H[Dn;Mn]. In particular H is abelian.
PROOF. SupposeH/M0nandH6[Dn;Mn]. Since M0n=[Dn;Mn] is sim- ple, we haveH[Dn;Mn]M0nand consequently
M0n
H H[Dn;Mn]
H ' [Dn;Mn] H\[Dn;Mn]:
Thus in particular M0n=His abelian andHM00nM0n. Thus,HM0n. p The above results are enough to prove thatM0is simple.
THEOREM2.8. M0is a simple group.
PROOF. Let16N be a normal subgroup of M0 and set, for every n5,NnN\M0n.Nnis normal in M0nand then must coincide with M0n or be abelian and contained in [Dn;Mn]. We now show that the latter cannot always be the case. Suppose that, for everyn5,Nn[Dn;Mn].
Let m5 be a fixed integer and take xmQ1
i0 lam;ii 2Nm. If k>0, xis also contained in Nkm and then one also has xkmQ1
i0 lbkm;ii . This means that, givenz2Z,x``fixes''zor ``moves'' itby a multiple ofkm, no matter how a large k>0 one chooses. This necessarily implies that x1 and then N1, which contradicts the initial assumption N61. Then for some m5 NmM0m and, for every k>0, one has NkmM0km, as NmNkmimplies thatNkmis notabelian. This conclude the proof since N S
k0Nkm S
k0M0km M0. 3. Abelian subgroups ofM.
Theorem 1.1 describes a large class of groups that can be embedded in M. In this section we focus our attention to abelian groups and will com-
pletely determine those abelian groups that can be represented as sub- groups ofM.
We denote by Fn the free abelian group of rankn( @0). Given a se- quence (possibly infinite)Pof prime numbers,CP is the group Dir
p2PCpand C1P is the group Dir
p2PCp1.
Forg2 Mwe consider the equationxkgand, if a solution exists, we call itak-rootofg. Itis clear thatmod (g) divides mod (x). Suppose now that xis ak-rootofgandgls;xdt2Mn; one hasgxkd0tk for some d02Dnand in particulartks, i.e.tis ak-rootofs: in a certain sense this allows us to subordinate the existence of roots inM to the existence of roots in finite symmetric groups, for which the following result holds ([3]):
PROPOSITION3.1. s2Snhas ak-root if and only if for every positive integerl, the number ofl-cycles in the canonical decomposition ofsis a multiple offkgl.
In the statement of the above proposition fkgl denotes the part of k which shares prime divisors withl, i.e. ifPis the set of prime divisors ofl, fkglis theP-partofk.
If an elementgls2Mnadmits ak-rootx, itis notnecessary thatx lies in Mn: anyway, bothgandxwill be in some Mm, withma multiple ofn.
For this reason it is useful to know how to factorizegin the group Mmfor multiplesmofn.
LEMMA 3.2. Let n2N, i2 f0;1. . .;n 1g and k0. Then the fol- lowing equalities hold:
ln;ilkn;i(k 1)n(i in . . . i(k 1)n);
ln;i1lkn;i1(i in . . . i(k 1)n) 1:
Thus, if a is a positive integer,
lan;ilkn;i(k 1)nlkn;i(k 2)n lkn;i(k a)n(i in . . . i(k 1)na ln;ialkn;in1 lkn;i2n1 lkn;ian1 (i in . . . i(k 1)n a:
For gls2Mn with lnQ1
i0lan;ii we put e(g)nP1
i0jaij. We have the following technical result:
LEMMA3.3. Let gls2Mn, m2N. Thene(gm)me(g).
PROOF. IflnQ1
i0lan;ii, one hasgmlls1lsm1sm andlsj nQ1
i0lan;iisj. Thus,
e(gm)Xn 1
i0
X
m 1
j0
aisj
X
i;j
jaisjj mXn 1
i0
jaij me(g): p
PROPOSITION 3.4. Let gls2Mn be an element of infinite order.
Then, for everyk >e(g), the equationxk gadmits no solution inM.
PROOF. As usual setlnQ1
i0lan;ii. We consider firstthe caseg2Dn, i.e.
s1, and proceed by way of contradiction. Put aP
i jaij e(g) and suppose that, for somek>a, an elementx2 Mexists such thatxkl. We can regardxas an elementof some Mtnwithtlarge enough so thatt>jaij for every i and, if ai60, kai divides t. For such a choice of t we have gdt2Mtn, witht nQ1
i0taii
and theti's are disjoint``t-cycles'' inSftn. Furthermore, when ai60, taii is the product of jaij nontrivial cycles of lengtht=jaij(and in particulart61). Letzi(zt;i) be the number of cycles of lengtht=jaijin the decomposition oft: we havezinP1
i0jaij afor every i, and this inequality holds for every choice oftwith the desired properties.
Now, ifxhy2Mtn, we haveykt, i.e. tadmits ak-rootinSftnand, according to Proposition 3.1, forai60,k fkgt=jaijdivideszia, which contradicts the initial assumptionk>a.
To conclude the proof, supposes61 and letmbe the order ofs. Thus gm2Dn. If x is a k-rootof g one also has xkmgm and then
kme(gm)me(g) and thus,ke(g). p
As a consequence of the previous proposition we can easily describe the structure of a torsion-free abelian subgroup ofM.
THEOREM3.5. Let G be a torsion-free abelian subgroup ofM. Then G is free.
PROOF. By Proposition 3.4 everyg2Gadmitsk-roots only for a finite number of indexesk. SinceGis torsion-free the equationxkgadmits at mostone solution inG. Then every elementofGadmits only a finite number of roots and this is enough to conclude thatGis free. p
Examples of free abelian subgroups ofMwe have found so far are the Dn's, of finite rankn. A natural question is to ask whetherMhas a free abelian subgroup of infinite rank and such a question can be answered positively. Such a subgroup can be defined as the union of an ascending chain of free abelian subgroups of M: letm2 be a fixed integer, set x00 andG0 hlm;0i; fori1 letxibe such that
0xi2im 1
xi6xj (mod 2jm); for everyj<i: (
If we defineGi hl2jm;xj jj0;. . .;ii, the groupG S
i2NGiis free abelian of infinite countable rank, i.e. G'F@0. Further, if one ``removes'' the generatorlm;0from the generating set ofG, a groupHis generated which is still isomorphic to F@0and whose supportis disjointfrom the setmZ.
We turn our attention to torsion subgroups ofM. We have already seen thatMcontains every finite group as a subgroup (Theorem 1.1) and a natural embedding one can think of is the regular representation in somefSn. If we adopt an argument similar to that we have used above for the free abelian groupGof infinite rank, we can easily embed inMdirectproducts of coun- tably many finite cyclic groups. LetP (p1;p2;. . .) be a sequence of prime numbers; for every primepiwe can find a modular permutation of orderpiin such a way that these permutations have pairwise disjoint supportsSi. Let n>p1and considerte12Sfnwheret1 0 1 . . . p 1 2Sn. Fori2 consider:
a1;. . .;api the smallest positive integers not contained iniS1
j1Si; kithe smallest positive integer such that2kin>api;
ti a1 a2 . . . api 2S2kin.
Thentei2Sg2kin is an elementofMof orderpiand thetei's generate a subgroup ofMisomorphic toCP. With small changes to the argument, as restricting the domain where to choose the integersai, one can constrain the support of the generated subgroup in any infinite subset of Zwith some modular feature, for example the set mZ mentioned above. It is therefore clear thatM contains some subgroup isomorphic to F@0CP, for every countable sequencePof prime numbers.
One can go further and consider a single g image (under the homo- morphismln) of a cycle of prime lengthpin the symmetric group Sn. As an elementingSpn,gis the image of a permutationx2Spnwhich is the product of pcycles of lengthp:xclearly admits ap-rootx1in Spnandx1 is a cycle of
lengthp2. Thus,gadmits ap-rootg1xe1ingSpn. One can iterate the argu- mentand find, for everyi2N, an elementgi2Sgpin such thatgpi gi 1:g and thegi's generate a subgroup ofMisomorphic to the PruÈferp-groupCp1. By applying this argument to each of thetei's above, one can build a subgroup ofMisomorphic toCP1and, again, one can confine the support of the group into the setmZ. Since any abelian torsion group can be embedded in a direct productof PruÈferp-groups, one has immediately the following results.
PROPOSITION3.6. Mcontains every countable abelian torsion group as a subgroup.
PROPOSITION 3.7. Let P be a countable sequence of prime numbers andAPF@0CP1. ThenAP is isomorphic to a subgroup ofM.
Proposition 3.7 individuates a large class of abelian groups that can be embedded in M: t he AP's and their subgroups. We now show that all abelian subgroups of M belong to that class. Suppose A is an abelian subgroup ofM, letTbe its (maximal) torsion subgroup and consider the quotient groupA=T. Let16aT2A=T(in particularais of infinite order inA) and supposexTis ak-rootofaTinA=T: for somet2T,xkta. If jtj n, xknan and, according to Theorem 3.4, kne(an)ne(a) and thenke(a). As a consequence, every elementofA=Tadmits only a finite number of roots and thenA=Tis free abelian.Tis therefore a direct factor of AandAis isomorphic to a subgroup of someAP as defined in Propo- sition 3.7. What we have seen so far proves the following theorem.
THEOREM3.8. Let A be a countable abelian group and T be its torsion subgroup: A can be represented as a subgroup ofMif and only if A=T is free abelian.
Since we have showed (theorem 2.8) that the commutator subgroupM0 of M is simple, a natural question could arise, whether it is possible to restate the last and theorem 1.1 withM0in place ofM.
For theorem 3.8 it is easily seen that it is just a matter of embedding F@0
inM0, since any embedding ofCP1already lies there. Such an embedding can be simply achieved by modifying the group G S
i2NGi constructed before: one considers fGi h(l2m;x1) 1l2jm;xj jj2;. . .;ii M0n and Ge S
i2
fGi M0 which is still isomorphic to F@0 with support disjoint from mZ.
For theorem 1.1 the same is not as immediate as for theorem 3.8, but still possible. The homomorphism f :Dn !D2n, defined by (ln;i)f
l2n;i(l2n;in) 1, is injective and commutes with every automorphism of Sfn; furthermore, (Dn)f [D2n;Sf2n]D2n\ker (C)D2n\ M0. Thus, itis possible to embed MnDnjSfn into [D2n;Sf2n]jSf2n M0 and conclude that every finitely generated abelian-by-finite group can be embedded in M0(this also shows that every finitely generated (abelian-by-finite) group can be embedded into the commutator subgroup of a finitely generated (abelian-by-finite) group).
FINAL REMARK. Just before this work was submitted, a paper by M. R.
Dixon, M. J. Evans and H. Smith appeared ([1]) in which the existence of a locally (abelian-by-finite) simple group that is not locally finite is estab- lished; the authors construct such a group by means of a direct limit in- volving wreath products of direct products of infinite cyclic groups by finite alternating groups, which are in some way basic ``bricks'' not too dissimilar to the ones of the groupMof modular permutations. However, our work was originally intended with the main purpose of obtaining a better un- derstanding of the group M, whose definition sounds quite natural and reasonable, even though almost no traces were found in the literature.
REFERENCES
[1] M. R. DIXON- M. J. EVANS- H. SMITH,Embedding groups in locally (soluble- by-finite) simple groups, Journal of Group Theory9(2006), pp. 383-395.
[2] M. I. KARGAPOLOV - JU. I. MERZLJAKOV, Fundamentals of the Theory of Groups, Graduate Texts in Mathematics, vol.62, Springer (1979).
[3] N. POUYANNE,On the number of permutations admitting an m-th root, The Electronic Journal of Combinatorics9(2002).
[4] H. WIELANDT, Permutationsgruppe, Math. Inst. Univ. TuÈbingen (1955).
[Translated in english: Finite Permutation Groups, Academic Press, New York (1964). Reprinted inMathematische Werke, Walter de Gruyter, Berlin, Vol.1(1994), pp. 119-198.]
Manoscritto pervenuto in redazione il 10 luglio 2006