ON THE K-RING OF THE CLASSIFYING SPACE OF THE GENERALIZED QUATERNION GROUP MEHMETKIRDARANDSEVI˙LAYO¨ZDEMI˙R 3 Abstract. WedescribetheK-ringoftheclassifyingspaceofthegeneralizedquaterniongroupintermsofgenerators 1 andtheminimalsetofrelations. Wealsocomputetheorderofthemaingeneratorinthetruncated rings. 0 2 n a 1. Introduction J 3 The K-ringofthe classifyingspace BQ2n ofthe generalizedquaterniongroupQ2n, n≥3, is describedin [2]and [4]. In this note, we will make the description of these rings in a simpler way, by a minimal set of relations on a ] minimal set of generators. We will also make nice connections of these computations with the computations done T for the lens spaces. K In particular, we will compute the order of the main generator in truncated rings, in a much shorter way than . h that is done in [4]. at As usual, the description of these rings will be done from the representation rings of the groups via the Atiyah- m Segal Completion Theorem (ASCT) which roughly says that the K-ring of the classifying space of a group is the completion of the representation ring of this group at the augmentation ideal of the representation ring. [ We will also check the minimality of the relations we found through the Atiyah-Hirzebruch Spectral Sequence 3 (AHSS), as usual. v Inconnectionwithandparalleltothis problem,thereaderisadvisedtolookatthecomputationsdoneforcyclic 7 8 and dihedralgroups. A quick surveyfor the K-Rings of the classifying spacesof cyclic groups,i.e. lens spaces,and 0 of the classifying spaces of dihedral groups can be found in [3]. 3 . 2 2. Representations 1 2 The quaternion group Q2n, where n ≥ 3, is generated by two elements x and y with the relations x2n−1 = 1, 1 x2n−2 =y2 andxyx=y. Note thatx generatesacyclic groupoforder2n−1 andy generatesa cyclic groupoforder v: 4. Note also that x2n−2, which is equal to y2, generate a cyclic group of order 2. We keep in mind these natural Xi group inclusions of the cyclic groups in Q2n. There arefourone dimensionalirreducible complex representationsofQ2n. We willdenote them by1,η1,η2 and ar η3. They are explained by 1×1 matrices in [4]. Thereare2n−2−1twodimensionalirreduciblecomplexrepresentationsofQ2n.Wewilldenotethembydi where 1≤i≤2n−2−1. Actually, d make sense for any integer i and this will be clarified below. i Since all we need will be the relations that they can generate, we will not describe these representations by matrices here. The descriptions of these representations by 2×2 matrices are given in [4]. Before presenting the relations let us set: 2n =2m=4k. And we have this convention from now on. Note that k ≥2 and it is a power of 2 too. Now, we will list all possible relations in the representation ring R(Q ). First of all η = η η . And since 4k 3 1 2 η2 = η2 = 1, we also have the relation η2 = 1. For d ’s, we have the start d = 1+η and we have the end 1 2 3 i 0 1 d =η +η . k 2 3 The main relation, which is the most important of all, is didj =di+j +di−j. This relation make sense for any integer couple i,j because of the following fact: di =dm−i for all integers i. Date:13December,2012. 2000 Mathematics Subject Classification. Primary55R50;Secondary20C10. Key words and phrases. Topological K-Theory,Representation Theory,QuaternionGroups. 1 2 MEHMETKIRDARANDSEVI˙LAYO¨ZDEMI˙R And another set of relations are for the products of the one dimensional and two dimensional representations, and they are: η1di = di and η2di = dk−i for all i. Since, η3 = η1η2, it follows that η3di = dk−i for all i, the same as η . 2 We deduce from the relations above that, the representation ring of Q2n is just generated by η1,η2 and d1, by means of tensor products and direct sums. And the minimal polynomials on η ,η and d that define the ring can 1 2 1 be found from these relations. But, we will do that in our new variables. 3. Cohomology Integral cohomology of Q , k≥2, is the following and can be found in [1]: 4k Z p=0 Hp(BQ ;Z)= Z2⊕Z2 if p=4s+2 4k  Z4k if p=4s, s≥1 0 if p is odd. Note that the odd dimensional cohomology vanishes. Because of that, the AHSS which converges to K(BQ ), 4k collapsesonpagetwosothattheK-ringiscompletelydeterminedbythe integralcohomologyandviseversa. Here, we also notice that the cohomology is periodic. The relations of the cohomology ring can also be found in [1]. Note that these relations are quite different than the relations in the K-ring. We don’t try to find connections between these relations. We will just compare the ordersoftheelementsofK(BQ )inthefiltrationsofthespectralsequencewiththesizesofthecohomologygroups 4k to prove that the relations are minimal. 4. K-Rings Corresponding to the representations η ,η and d , there are induced vector bundles over the classifying space 1 2 1 BQ and we denote them by the same letters. We will set the reduced vector bundles as v =η −1, v =η −1 4k 1 1 2 2 and, most importantly, the main element φ = d − 2. Due to the ASCT, the elements v ,v and φ generate 1 1 2 K(BQ ). All we need is to find the minimal relations on these generators so that the ring is well described. 4k First off all, since η2 =η2 =1, we have the following relations 1 2 v2 =−2v and v2 =−2v (Relations 1 & 2). 1 2 2 We note that the above relations are the standard relations for real line bundles over the classifying spaces. These small and simple relations on v ’s explain the cohomology groups H4s+2(BQ ;Z)=Z ⊕Z which are the i 4k 2 2 (4s+2)-th filtrations E∞4s+2,−4s−2 on the main diagonal of the AHSS. The first Z2 in the direct sum is generated by vs and the second Z is generated by vs where s≥1. 1 2 2 NextweshouldexplainZ ’swhichoccurinthecohomologyring,inotherwords,weshouldexplainthefiltrations 4k E∞4s,−4s, s≥1, of the AHSS. This will not be easy. Recall the natural inclusion of Z in Q defined by the element x ∈ Q . This inclusion rises a natural ring 2k 4k 4k homomorphism K(BQ )→K(BZ ). 4k 2k Under this homomorphism, the image of the virtual bundle d −2 in K(BZ ) is ηi +η−i −2 where η is the i 2k one and only generator of K(BZ ) and the one and only relation it satisfies is η2k = 1. We set w = η+η−1−2 2k in K(BZ ). The element w generates a subring of K(BZ ) which is isomorphic to the subring of KO(BZ ), 2k 2k 2k solely generated by w. Actually it is almost isomorphic, except a Z direct summand which is generated by the 2 tautological one dimensional reduced real bundle, traditionally denoted by λ. Note that KO(BZ ) is the real 2k topological K-theory of the space BZ . 2k Hence, under the natural homomorphism mentioned above, the image of d −2 in K(BZ ) is ψi(w) where ψi i 2k is the Adams operation of degree i. Lets recall the effect of (the real) Adams operation of degree i, on the main generator w of KO(BZ ): 2k i i i+j−1 ψi(w)= j j wj (cid:0) (cid:1)(cid:0)2j−1 (cid:1) Xj=1 j (cid:0) (cid:1) We name the above polynomial ”quadratic binomial of degree i” because of its connection to the real part of a rootofunity andbecause its coefficients arepolynomialsof i2. Inparticular,under the abovering homomorphism, φ maps exactly on w. ON THE K-RING OF THE CLASSIFYING SPACE OF THE GENERALIZED QUATERNION GROUP 3 On the other hand, in the ring K(BQ4k), we have dk+1 −dk−1 = 0. Practically, we observe that this gives a polynomial in φ of degree k+1, no one dimensional bundles involved. This is true for any d where i is odd. They i can be written as a polynomial of φ and only φ. In the ring KO(BZ ), we know that the main relation is ψk+1(w)−ψk−1(w) = 0, [3]. And, we deduce that 2k ψk+1(φ)−ψk−1(φ)=0 in the ring K(BQ ) too. We also conclude that 4k d −2=ψi(φ) when i is odd. i We can also prove this very important fact from the relations of the representation ring without referring to lens spaces. But, this would take longer. Lens spaces make this tricky. The polynomials g (φ)=ψk+1(φ)− ψk−1(φ) are given by the following series: 2k k 2k2+j−1 k+j−2 g (φ)=4kφ+ φj +φk+1 2k (j−1)(2j−1)(cid:18) 2j−3 (cid:19) Xj=2 Hence, the following relation is satisfied in K(BQ ): 4k g (φ)=0 (Relation 3). 2k From this relation, we deduce that φ satisfies a relation in the form 4kφ=f(φ)φ2 where f(φ)∈K(BQ4k) is a virtual bundle generated by φ. This explains the fact that the 4-th filtration E∞4,−4 on the lastpage of the main diagonalof AHSS is generatedby maybe φ (or maybe φ−v −v etc.) and is isomorphic 1 2 to H4(BQ ;Z)=Z . We don’t want to speculate much about the spectral sequence. 4k 4k And by multiplying this relation by powers of φ, all 4s-th filtrations on the main diagonal of AHSS, i.e. all groups H4s(BQ ;Z)=Z in the cohomology,are similarly explained. 4k 4k But, we are still not done! It turns out that the relation 3 is not minimal. We will prove it when we talk about the minimal relation for the cross product v v . We also didn’t still explain the products v φ and v φ. Without 1 2 1 1 these relations, the ring can not be completely described, although more or less the filtrations of the diagonal of the AHSS are explained. Lets first find the minimal relations for the products v φ where i=1 or 2, and explain why they are not needed i to occupy any place on the AHSS. From the relations η d =d , it immediately follows that 1 1 1 v φ=−2v (Relation 4). 1 1 This takescareofthe productv1φ.Nextwewilltakecareoftheproductv2φ. Fromtherelation,η2d1 =dk−1, since dk−1 =ψk−1(φ)+2, we obtain v φ=ψk−1(φ)−φ−2v (Relation 5). 2 2 So v φ where i=1 or 2 are dependent variables, and we don’t have to search a place on the AHSS for them. i Finally lets explain what remained, in other words, lets find the minimal relation for the cross product v v . 1 2 It turns out that the main relation is not the Relation 3, but that one. In fact, we will throw away our favorite relation, the Relation 3, from the minimal set of relations. We will separate the cases n = 3 and n ≥ 4, since it turns out that Q is a little different than the bigger 8 generalized quaternion groups. For k =2, from the relation d2 =d +d =1+η +η +η , we have 1 2 0 1 2 3 v v =4φ+φ2−2v −2v (Relation 6, for n=3). 1 2 1 2 This is the main relation for K(BQ ). If we multiply this equation by φ+2, an amazing thing happens and we 8 find the Relation 3. In other words,we can kick off the Relation 3 from the minimal list of relations that describes the ring. For n≥4, starting from the relation d2 =d +d , by using the relation d2 =d +d repeatedly, one inside the 1 2 0 i 2i 0 other, we can obtain the relationd −d =ψk(φ). And along the way,amusingly,we obtain the polynomials ψi(φ) k 0 i where i is a power of 2, in terms of ψ2(φ). And thus, we have v v =ψk(φ)−2v (Relation 6, for n≥4). 1 2 2 We will throw away the Relation 3 for n ≥ 4 too. We multiply the Relation 6 above by φ+2 and then we use the Relations 4 and 5 properly in the equation we obtained, and again amazingly find the Relation 3. This also 4 MEHMETKIRDARANDSEVI˙LAYO¨ZDEMI˙R removes doubts from the obscure explanations we did above when we derived this relation trickily from the lens spaces. We sum up everything in: Theorem 1: K(BQ2n) is generated by v1,v2 and φ with the minimal set of relations (1),(2),(4),(5) and (6) above. 5. Orders In [4], Proposition 5.1, the order of the element φ in the truncated ring R(Q )(cid:30)φ2R(Q ), after a lot of work 4k 4k byhugematrices,isfoundas4k.Heusedtheseorders,toanswersomegeometricproblems,namelyproblemsabout immersion of spaces like S4N+3(cid:30)Q in real Euclidean spaces. But, from the relation 4kφ = f(φ)φ2 we found 4k above, this is evident. Similarly, we can find the order of φ in the truncated ring R(Q )(cid:30)φN+1R(Q ) by a careful counting. In the 4k 4k Relation3,weobservethatthe coefficientofφ2 is k(2k2+1) whoseprimary2factorisk =2n−2. Sincethe coefficient 3 of φ is 4k=2n, the jump betweenthem is 2n =4. So the totalcountupto φN+1 must be 4k.4N. Hence, we have 2n−2 Corollary 2: The order of φ in K(S4N+3(cid:30)Q ) is 2n+2N. 4k References [1] HayamiT.andSanadaK.: CohomologyRingoftheGeneralizedQuaternionGroupwithCoefficientsinanOrder,Communications inAlgebraVol.30,No.8,pp.3611–3628, 2002 [2] KensoF.: J-groups of the Quaternionic Spherical Space Forms,HiroshimaMath.J.24(1994), 365-399. [3] KırdarM.: Topological K-Theory of the Classifying Spaces of Cyclicand Dihedral Groups, Preprint. [4] PittD.: Free Actions of the Generalized Quaternion Groups on Spheres, Proc.LondonMath.Soc.(3)26(1973)1-18. Departmentof Mathematics,Faculty of ArtsandScience,NamıkKemalUniversity,Tekirdag˘,Turkey E-mail address: [email protected], [email protected]