Abelian Groups that cannot be Factored Without Periodic Factor.
SAÂNDORSZABOÂ(*)
ABSTRACT- The list of the finite abelian groups that cannot be factored into two of its subsets without one factor being periodic is the same as the list of the finite abelian groups that cannot be factored into any number of its subsets without one factor being periodic.
1. Introduction.
Throughout this paper we will use multiplicative notation for abelian groups. LetGbe a finite abelian group. We denote the identity element of Gbye. LetB;A1;. . .;Anbe subsets ofG. We define the productA1 An
to be
fa1 an:a12A1;. . .;an2Ang:
SupposeBA1 An. We say that the productA1 Anisdirectif eachb inBis uniquely expressible in the form
ba1 an; a12A1;. . .;an2An:
IfBis a direct product ofA1;. . .;An, then the equationBA1 An is said to be afactorizationofB.
A subsetAof a finite abelian groupGwill be callednormalizedife2A.
The factorizationGA1 An is called normalized if each factor is nor- malized. We say thatAisperiodicif there is an elementa2Gsuch that a6eandaAA. The elementais called aperiodofA. Note that ifaand bare periods ofA, thenabis a period ofAunlessabe. It follows that the
(*) Indirizzo dell'A.: Institute of Mathematics and Informatics,University of PeÂcs, IfjuÂsaÂg u. 6 - 7624 PeÂcs, Hungary.
E-mail: sszabo7@hotmail.com
Mathematics Subject Classification (1991); Primary 20K01; Secondary 52C22
periods ofAaugmented with the identity element form a subgroupHofG.
Moreover there is a subsetBofGsuch thatABHis a factorization ofA.
If the groupGis a direct product of cyclic groups of orders t1;. . .;ts re- spectively, then we express this fact shortly saying that G i s of type
(t1;. . .;ts). If from the factorizationGA1 Anit follows that one of the
factors is periodic for any possible choice of the factorsA1;. . .;An, then we say that G has the Hn-property. (Hn is an abbreviation for the HajoÂs property withnfactors.) Here we do not allow factors with only one ele- ment and do not consider groups with only one element.
In 1949, G. HajoÂs [4] called for determining all finite abelian groups with theH2-property. The complete list of these groups first appeared in 1962 A. D. Sands [8].
(pa;q);
(p3;2;2);
(p;q;2;2);
(p2;q2);
(p2;2;2;2);
(p;3;3);
(22;22);
(p2;q;r);
(p;22;2);
(32;3);
(p;p):
(p;q;r;s) (p;2;2;2;2);
(2a;2);
(1)
Here p;q;r;s are distinct primes the p2 and p3 cases are not ex- cluded anda1 is an integer. Groups whose type is on list (1) and their subgroups have the H2-property and other groups do not have the H2- property.
This result has applications in various fields. In geometry [11], combi- natorics [5], coding theory [3], number theory [13], Fourier analysis [6].
If a group has theHn-property for each possible choice ofn, then we say thatGhas theH-property. It is plain that groups with theH-property have theH2-property. We will show that groups with theH2-property are in fact the same as groups with theH-property. We express this fact more formally as a theorem.
THEOREM1. Let G be a finite abelian group. If G has the H2-property, then G has the H-property.
2. Replacing factors.
We say that in the factorizationGABthe factorBcan bereplacedby B0 if GAB0 is also a factorization of G. If c is an element of G, then multiplying both sides of the factorization by c we get the factorization GGcA(Bc). In other wordsBcan be replaced byBcfor eachc2G.
Note thatBis periodic if and only ifBcis periodic. LetGA1 Anbe a factorization. Let a12A1;. . .;an2An multiplying the factorization by a11 an1 we get the normalized factorization G(a11A1) (an1An).
Thus when we study factorizations with periodic factors we may restrict our attention to normalized factorizations.
Note that GAB is a factorization if and only if jGj jAkBj and AA 1\BB 1 feg. From this it is plain thatBcan be replaced byB 1. Let GAB be a factorization of G. Then each g2G is uniquely ex- pressible in the formgab,a2A,b2B. We callatheA-coordinateofg and we denote it bya(g). Similarly, we callbtheB-coordinateofgand we denote it by b(g). The coordinates of g make sense only relative to the factorizationGAB. IfAis a subset of a finite abelian groupGandqi s an integer, thenAq will denote the setfaq:a2Ag.
LEMMA1. Let GAB be a factorization of G and let A fa1;. . .;ang.
For each g2G the elements a(ga1);. . .;a(gan) form a permutation of a1;. . .;an.
PROOF. Clearly, a(gai)2A. So we will show that a(gai)a(gaj) i m- pliesaiaj. From the equations
gaia(gai)b(gai); gaja(gaj)b(gaj) we get the equations
ga(gai)b(gai)ai1; ga(gaj)b(gaj)aj1:
Then b(gai)ai1b(gaj)aj 1 and ajb(gai)aib(gaj). Now as ai;aj2A, b(gai),b(gaj)2B, from the factorizationGABit follows thataiaj.
This completes the proof.
LEMMA2. Let GAB be a factorization of G and let q be a prime such that q6 j jAj. Then GAqB is a factorization of G.
PROOF. Choose ana2A,g2Gand defineTto be the set of allqtuples (x1;x2;. . .;xq); x1;x2;. . .;xq2A
for which a(gx1x2 xq)a. First note thatjTj jAjq 1. Indeed, choose x1;x2,. . .;xq 12Aarbitrarily, then by Lemma 1,a[(gx1x2 xq 1)xq]a has a unique solution for xq. Next note that i f (x1;x2;. . .;xq)2T, then
(x2;. . .;xq;x1)2T. We defi ne a graphG. The verti ces ofGare the elements
of T and we draw an arrow from the node (x1;x2;. . .;xq) to the node
(x2;. . .;xq;x1). The graphG is a union of disjoint cycles. The cycles are of length 1 or of lengthq. Whenx1x2 xq, then the node (x1;x2;. . .;xq) is on a cycle of length 1. Whenx1;x2;. . .;xqare not all equal, then the node (x1;x2;. . .;xq) is on a cycle of lengthq. Asq6 j jAjthere must be a cycle of length 1 inG. In other words there is anx12Asuch thata(gxq1)a.
Consider the factorization GAB. From gxq1a(gxq1)b(gxq1) using a(gxq1)a we getgxq1ab(gxq1), thengax1qb(gxq1). Hereax1q2aA q, b(gxq1)2B and the equation holds for each g2G. It follows that G(aA q)B. Then GA qB. Note that jGj jAkBj, jA qj jAj imply jGj jA qkBjand soGA qBis a factorization. NowA qcan be replaced byAqand so it follows thatGAqBis a factorization.
This completes the proof.
LEMMA3. Let GAB be a factorization such that e2A,jAj p is a prime. Then GA0B is a factorization of G, where A0 fe;a;a2;. . .;ap 1g, a2An feg.
PROOF. Note thatGAtBis a factorization ofGwhenevert2,p6t.
Indeed, astis a product of primes we can apply Lemma 2 several times starting with the factorizationGAB. LetA fe;a1;a2;. . .;ap 1g. The fact thatGAtBis a factorization is equivalent to that the sets
eB;at1B;at2B;. . .;atp 1B
form a partition ofG. SetA0 fe;ak;a2k;. . .;apk 1g. The fact thatGA0Bis a factorization is equivalent to that the sets
eB;akB;a2kB;. . .;apk 1B
form a partition of G. Si nce G is finite it is enough to show that aikB\ajkB ; for each i;j, 0i<jp 1. Assume the contrary that aikB\ajkB6 ;. Multiplying by aki we get eB\aj ik B6 ;. Set tj i.
Clearly, 1tp 1 and sotis prime top. NoweB\atkB6 ;contradicts the fact thatGAtBis a factorization ofG.
This completes the proof.
3. Periodicity forcing factorization types.
IfGA1 Anis a factorization ofGandjA1j q1;. . .;jAnj qn, then we call then tuple (q1;. . .;qn) the typeof the factorization. If from each GA1 Anfactorization of type (q1;. . .;qn) it follows that at least one of
the factors is always periodic we call (q1;. . .;qn) a periodicity forcing factorization type for G. In 1965 L. ReÂdei [7] proved that a factorization type (q1;. . .;qn) is always periodicity forcing ifq1;. . .;qnare primes.
LEMMA4. Let q1p1 ps, where p1;. . .;psare primes. If(q1;. . .;qn) is a periodicity forcing factorization type for an abelian group, then so is (p1;. . .;ps;q2;. . .;qn).
PROOF. Suppose that (q1;. . .;qn) is a periodicity forcing factorization type for the finite abelian groupGand consider a normalized factorization
GB1 BsA2 An; (2)
where
jB1j p1;. . .;jBsj ps;jA2j q2;. . .;jAnj qn:
We would like to show that at least one of the factorsB1;. . .;Bs,A2;. . .;An is periodic. If one of the factors is periodic then there is nothing to prove so we assume that none of the factors is periodic.
If the order of each element inBiis a power ofpi, then we defineCito beBi. Assume that there are elements inBiwhose order is not a power of pi. In factorization (2), by Lemma 2,Bican be replaced by a normalized subsetCisuch that the orders of each element inCiis a product of at most two distinct primes andCi is not a subgroup ofG. Note that asjCij i s a prime ande2Ci,Ciis periodic if and only ifCiis a subgroup ofG. Setting CC1 Cs we get a factorizationGCA2 An. The type of this fac- torization is (q1;q2;. . .;qn). So by our assumption one of the factors
C;A2;. . .;Anis periodic. If one ofA2. . .;Anis periodic, then we are done
so we assume thatCis periodic. Now a periodic subsetCofGis factored into subsets C1;. . .;Cs. In addition, there is a subset A of G such that GCA is a factorization. Simply set AA2 An. The hypotheses of Theorem 2 of [2] are satisfied therefore this theorem gives that at least one of the factorsC1;. . .;Cs must be periodic.
This contradiction completes the proof.
IfAis a subset andxis a character ofG, then we will use the notation x(A) to denote the sum X
a2A
x(a):
In this paper character ofGalways means irreducible character ofG. The set of all charactersxofGwithx(A)0 is called theannihilatorset ofA and will be denoted by Ann(A).
LEMMA5. Let G be a finite abelian group with the H2-property and let q1p be a prime. In the p3case we assume that the p-component of G is cyclic. (In the p2 case nothing is assumed about the p-com- ponent of G.) Then(q1;q2;q3)is a periodicity forcing factorization type for G.
PROOF. LetGA1A2A3be a normalized factorization ofGsuch that jA1j q1,jA2j q2,jA3j q3. We would like to show that one of the factors A1,A2,A3is periodic. If one ofA1,A2,A3is periodic, then there is nothing to prove. So we assume that none of the factors is periodic. Assume thatp3.
In this case thep-component ofGis cyclic. By Lemma 3,A1can be replaced byA01 fe;a;a2;. . .;ap 1gfor eacha2A1n feg. IfA01is a subgroup ofG for eacha2A1n feg, then as thep-component ofGis cyclic,A1is equal to the unique subgroup ofGthat haspelements. ButA1 is not a subgroup.
This contradiction gives that for somea2A1n fegthe subsetA01 is not a subgroup. This is equivalent toap6e. For the remaining part of the proof we choose an A01 which is not a subgroup. In the p2 case we set A01A1 fe;ag. Herea26easA1is not a subgroup. Thusap6eholds in both of the casesp2 orp3.
From the factorizationG(A01A2)A3, by theH2-property ofG, it fol- lows thatA01A2orA3is periodic. SinceA3is not periodic,A01A2is periodic, say with periodg, that is,A01A2gA01A2. We claim thatx(A2)0 holds for all charactersxofGwithx(g)61 andx(ap)61. In order to prove the claim assume that x(g)61 and x(ap)61. As x(g)61 from x(A01A2)x(g)
x(A01A2) we get 0x(A01A2)x(A01)x(A2). Fromx(ap)61 it follows that x(A01)60. Thereforex(A2)0. The fact thatx(ap)61 andx(g)61 imply x(A2)0 is equivalent to
Ann hapi
\Ann hgi
Ann(A2):
By Theorem 2 of [10], there are subsetsX,YofGsuch that A2Xhapi [Yhgi;
where the products are direct and the union is disjoint. Similarly, from the factorizationGA2(A01A3) it follows thatA01A3is periodic, say with period h. Then there are subsetsU,V ofGsuch that
A3Uhapi [Vhhi;
where the products are direct and the union is disjoint.
IfX ;, thenA2is periodic with periodg. IfU ;, thenA3is periodic with periodh. So we may assume that X6 ; andU6 ;. Choose x2X,
u2U. Multiply the factorization GA1A2A3 by gx 1u 1 to get the factorization
GGgA1(A2x 1)(A3u 1):
Now hapi A2x 1, hapi A3u 1 contradict the definition of the factor- ization.
This completes the proof.
4. Proof of Theorem 1.
We consider a finite abelian groupGwith theH2-property. ThusGi s a subgroup of a group whose type is on list (1) and we try to establish thatG has theHn-property for each possible choice ofn. Then1 case is trivial so we assume thatn2.
LetGbe a subgroup of a group of type (pa;q) and letGA1 Anbe a factorization ofG. IfGi s ap-group, thenjA1j;. . .;jAnjare powers ofp.
IfGi s not ap-group, thenqdivides one ofjA1j;. . .;jAnjandqcan divide only one ofjA1j;. . .;jAnj. SayqjA1jand plainly each ofjA2j;. . .;jAnjmust be a power of p. Now the p-component of G i s cycli c and each of
jA2j;. . .;jAnji s a power ofp. The conditions of Theorem 2 of [9] are met.
By this theorem, one of the factorsA1;. . .;Ani s peri odi c and soGhas the Hn-property. Sincen2 was arbitraryGhas theH-property.
By Theorem 1 of [12] (or by Theorem 10 of [1]), a group of type (2a;2) has theH-property.
We inspect the remaining groupsGin a case by case manner and sort out the arising cases using ReÂdei's theorem or Lemma 4 or Lemma 5.
(1) Suppose the type ofGis one of the following (p3;2;2); (p2;2;2;2); (p;2;2;2;2):
(3)
Note thatjGjis a product of 5 primes. This is why we treat these cases together. Let GA1 An be a factorization of G. If the type of the factorization is (q1;q2;q3;q4;q5), then eachqiis a prime and ReÂdei's the- orem gives that one of the factors is periodic. If the factorization type is (q1;q2;q3;q4), then only oneqican be composite, sayq4. TheH2-property ofGgives that (q1q2q3;q4) is a periodicity forcing factorization type forG.
Now by Lemma 4, (q1;q2;q3;q4) is a periodicity forcing factorization type forG. If the factorization type is (q1;q2;q3), then at least one of theqi's is a
prime, sayq1i s a pri me, and Lemma 5 appli es. IfGis a proper subgroup of a group whose type i s on li st (3), then we can use a si mi lar but to some extent simpler argument.
(2) Let us assume that the type ofGis one of the next (p2;q2)
(p;4;2)
(p2;q;r) (p;q;2;2)
(p;q;r;s) (22;22):
(4)
Note thatjGjis a product of 4 primes. LetGA1 Anbe a factorization of G. If the factorization type is (q1;q2;q3;q4), then ReÂdei's theorem implies that this factorization type is periodicity forcing. If the factorization type is (q1;q2;q3), then Lemma 4 is applicable. IfGis a proper subgroup of a group whose type is on list (4), then an analogous but slightly simpler reasoning can be used.
(3) The case whenGis of type (p;3;3), (32;3), (p;p), that is, whenjGji s a product of at most 3 primes is rather trivial.
This inspection completes the proof.
REFERENCES
[1] K. AMIN- K. CORRAÂDI- A. D. SANDS,The HajoÂs property for2-groups, Acta Math. Hungar.89(2000), pp. 189±198.
[2] K. CORRAÂDI- S. SZABOÂ,Factorization of periodic subsets II, Math. Japonica.
36(1991), pp. 165±172.
[3] C. DEFELICE,An application of HajoÂs factorization to variable length codes, Theoret. Comp. Sci.164(1996), pp. 223±252.
[4] G. HAJOÂS,Sur la factorisation des groupes abeÁliens, CÏasopis PeÂs. Mat. Fys.74 (1949), pp. 157±162.
[5] R. HILL- R. W. IRVING,On group partitions associated with lower bounds for symmetric Ramsey numbers, European Journal of Combinatorics 3(1982), pp. 35±50.
[6] J. C. LAGARIAS- Y. WANG,Spectral sets and factorizations of finite abelian groups, J. Funct. Anal.145(1997), pp. 73±98.
[7] L. REÂDEI,Die neue Theorie der endlichen abelschen Gruppen und Verallge- meinerung des Hauptsatzes von HajoÂs, Acta Math. Acad. Sci. Hungar.16 (1965), pp. 329±373.
[8] A. D. SANDS,On the factorisation of finite abelian groups II, Acta Math. Acad.
Sci. Hungar.13(1962), pp. 153±169.
[9] A. D. SANDS,Factoring finite abelian groups, Journal of Alg.275(2004), pp.
540±549.
[10] A. D. SANDS- S. SZABOÂ,Factorization of periodic subsets, Acta Math. Acad.
Sci. Hungar.57(1991), pp. 159±167.
[11] S. K. STEIN,Algebraic tiling, Amer. Math. Monthly,81(1974), pp. 445±462.
[12] S. SzaboÂ, Factoring a certain type of 2-groups by subsets, Rivista di Matematica,6(1997), pp. 25±29.
[13] R. TIJDEMAN,Decomposition of the integers as a direct sum of two subsets, Number Theory (Paris 1992-1993) London Math. Soc. Lecture Note Ser. 215 Cambridge Univ. Press, Cambridge 1995, pp. 261±276.
Manoscritto pervenuto in redazione il 17 aprile 2004 e in forma finale il 19 ottobre 2005