PERIODICITY OF HERMITIAN K-GROUPS A.J.BERRICK,M.KAROUBIANDP.A.ØSTVÆR 1 1 0 2 0. Introduction and statements of main results n a By the fundamental work of Bott [11] it is known that the homotopy groups of J classical Lie groups are periodic, of period 2 or 8. For instance, the general linear 1 and symplectic groups satisfy the isomorphisms: 1 πn(GL(R))∼=πn+8(GL(R)) ] T π (Sp(C))=π (Sp(C)) n ∼ n+8 K π (GL(C))=π (GL(C)) . n ∼ n+2 h These periodicity statements wereinterpretedby Atiyah,Hirzebruchandothers in t a theframeworkoftopologicalK-theoryofaBanachalgebraA: recallthatthereare m isomorphisms [ Ktop(A)=Ktop (A), n ∼ n+p 1 where Kntop(A) = πn−1(GL(A)) if n > 0 and K0top(A) = K(A) is the usual v Grothendieckgroup. Herepisthe periodwhichis 2or8accordingasAiscomplex 6 or real. We refer to [37] and [50] for an overview of the subject, both algebraically 5 and topologically. 0 2 Afew yearslater,afterhigheralgebraicK-theorywasintroducedbyQuillen,an . analogous periodicity statement was sought, of the form 1 0 K (A)=K (A), n ∼ n+p 1 1 where A is now a discrete ring. The first computations showed that a periodicity : isomorphismofthisformisfarfromtrueinbasicexamples. However,ifweconsider v K-theory with finite coefficients, and n is at least a certain bound d, then some i X periodicity conjectures appeared feasible, at least for certain rings of a geometric r nature. These conjectures were formulated for different prime power coefficient a groups, and are essentially of the following type (n d) ≥ K (A; Z/m)=K (A; Z/m). n ∼ n+p The relationshipbetweenthe prime powerm andthe associatedsmallestperiod p is given by the following convention, which we maintain throughout the paper. Convention 0.1. For Z/m coefficients, where m = ℓν with ℓ prime, the smallest period p is given by sup 8, ℓν−1 if ℓ=2, p= 2(ℓ 1)ℓν−1 otherwise. (cid:26) −(cid:0) (cid:1) Date:December10,2010. Firstand second authors partiallysupported by the National Universityof Singapore R-146- 000-097-112. Thirdauthor partiallysupportedbyRCN185335/V30. 1 2 BKO JANUARY12,2011 Using techniques of algebraic geometry and a comparison theorem with ´etale K-theory, numerous examples listed below showed that these conjectures hold. In thecaseofa2-power,thefirstthreeareparticularcasesofTheorem2in[52],based on the fundamental work of Voevodsky [62]. In the case of an odd prime power, the first four examples are consequences of the Bloch-Kato conjecture. Before giving these examples, we define the mod 2 virtual ´etale cohomological dimension vcd (A) of a commutative ring A as the mod 2 ´etale cohomological 2 dimension of A ZZ[µ4] obtained by adjoining a primitive fourth root of unity to ⊗ A. Forconvenience,ifℓisodd,thenvcd (A)denotesthemodℓ´etalecohomological ℓ dimension cd (A) of A. For ℓ fixed, here are the examples we consider. ℓ (1) Any field k of characteristic char(k) = ℓ for which vcd (k) < . In this ℓ 6 ∞ case, d=vcd (k) 1 if vcd (k)=0 and d=0 otherwise. ℓ ℓ − 6 (2) The ring [1/ℓ] of ℓ-integers in any number field F. In this case, d = F O vcd (F) 1=1 (cf. [43] when ℓ=2). ℓ − (3) Any finitely generated and regular Z[1/ℓ]-algebra A with finite mod ℓ vir- tual´etalecohomologicaldimension. Inthiscased=sup vcd (k(s)) 1,0 , ℓ { − } where k(s) is the residue field at any point s Spec(A). The same state- ∈ ment holds when replacing Z[1/ℓ] by Q or by any other field k of charac- teristic =ℓ. 6 The regularity assumptionon A can be dispensed with when workingwith negative K-theory [4], [30], [31], [61]. As shown in [53, Theorem 4.5], this does not change the bound d. (4) GroupringsR[G],whereGisfiniteandRisaringofℓ-integersinanumber field,asshownin[66]. Hered=1. Forsomeexplicitcomputationssee[42]. (5) The ringC(X)ofrealorcomplexcontinuousfunctionsonacompactspace X, as shown in [18], [46]. In this case d=1. Intheseexamples,theperiodicityisomorphismbetweenthegroupsK (A; Z/m) n andK (A; Z/m)isdefinedbytakingcup-productwitha“Bottelement”b . For n+p K p = 2ν−1 with ν 4, one can construct this element in the group K (Z; Z/2p), p ≥ such that its image in the topological K-group1 K (R; Z/2p) = Z/2p is the class p ∼ mod2pofageneratorinK (R)=Z. Thecup-productalludedtoaboveisapairing p ∼ ∪ :Kn(A; Z/m) Kp(Z; Z/2p) Kn+p(A; Z/m). × −→ WerefertoSection1forprecisedefinitionsandtheextensiontooddprimepowers. Aswecanseeintheseexamples,akeyroleisplayedbytheinfinitegenerallinear group GL(A). However, it was already shown in the works of Bott and Borel [10], andalsointopologicalapplications,thatotherinfinite seriesassociatedto classical Lie groups may be considered as well. More precisely, if we consider a ring with involutionAanda signofsymmetry ε= 1generalizingthe orthogonal(ε=1)or ± symplectic (ε = 1) case, one defines higher hermitian K-groups, denoted in this − paper KQ (A), in a parallelwayto algebraicK-groupsK (A). These groupsare ε n n associated to the infinite ε-orthogonal group O(A). A typical example is when ε A is commutative and ε = 1, in which case one recovers the infinite symplectic − groupon A. We refer to the survey paper [37] already mentioned above for precise definitions. 1WeshallwriteKn insteadofKntop whendealingwiththefieldRofrealnumbersorthefield Cofcomplexnumberswiththeirusualtopology,andlikewiseforspectra. Hermitianperiodicity January12,2011 3 The main purpose of this paper is to show that a periodicity statement in alge- braicK-theoryimpliesasimilaroneinKQ-theory,when1/2 A. SinceKQ-theory ∈ with coefficients Z/m, with m = 2ν, is the most important and difficult case, we state the main theorems in this context, leaving the case of odd prime power co- efficients to the end of this Introduction and to Section 5 of the main body of the paper. Forthe firststepintheargument,we introduceaparameterq thatisessentially p, apart from a slight modification in the case m = 16. Specifically, we make the following convention. Convention 0.2. 8 if m 8, ≤ q = 16 if m=16,  m/2 otherwise.  Inotherwords,q =pexceptwhenm=16,inwhichcaseq =2p. Itismeaningful  to speak of periodicity maps raising dimension by q, since q is a multiple of p. As a convenient notation, we write KQ (resp. K) for the KQ-theory (resp. K-theory) with coefficients in Z/m, the relationship between m and the period p being as in Convention 0.1. One of our main theorems is the following. Theorem 0.3. With the above definitions, assume that there exists an integer d such that the cup-product map ∪bK :Kn(A) Kn+p(A) −→ with the Bott element in K (Z; Z/2p) is an isomorphism whenever n d. Then, p ≥ for n d+q 1, there is also an isomorphism ≥ − KQ (A)= KQ (A). ε n ∼ε n+p Surprisingly, the isomorphism between the KQ-groups is in general not given by cup-product with a Bott element (see Remark 4.8 in Section 4). This relates to thefactthathermitianK-theorypossessesmorethanoneBottelement,aswenow describe. WhereasinalgebraicK-theoryuniversalBottelementsaretobefoundin the K-groupsofthe integersZ, here,becausewe areworkingwithringscontaining 1/2, our Bott elements are to be found in the hermitian K-groups of the ring of 2-integers Z′ =Z[1/2]. As in algebraic K-theory, using the methods of [5], in this paper we prove the existence of a “positive Bott element” b+ in KQ (Z′; Z/2p) whose image 1 p in K (Z′; Z/2p)=K (Z; Z/2p) is the Bott element in K-theory alluded to above. p ∼ p Onthe other hand, one of the maindifferences between algebraicand hermitian K-theory in our context is the existence of another element2 u in −1KQ−2(Z′), which plays an important role in the fundamental theorem in hermitian K-theory [33]. We now define the negative Bott element b− in hermitian K-theory to be the image of the element up/2 in the group 1KQ−p(Z′; Z/2p). To make the statement of Theorem 0.3 more precise, we note that the cup- product with the positive Bott element in KQ (Z′; Z/2p) determines a direct 1 p system of abelian groups KQ (A) KQ (A) KQ (A) . ε n −→ε n+p −→ε n+2p −→··· 2We recallthat the negative K-groups of aregular noetherian ring(for instance Z or Z′)are trivial. 4 BKO JANUARY12,2011 Symmetrically, cup-product with the negative Bott element in 1KQ−p(Z′; Z/2p) determines an inverse system of abelian groups KQ (A) KQ (A) KQ (A). ···−→ε n+2p −→ε n+p −→ε n Thetheoremabovecannowberestatedinamorepreciseform. (Recallthatthe overbar denotes Z/m coefficients.) Theorem 0.4. Let A be any ring (with 1/2 A), m, p and q be 2-powers as in ∈ Conventions 0.1 and 0.2, and let d Z, such that the cup-product with the Bott ∈ element b in K (Z; Z/2p) induces an isomorphism K p ∼ K (A) = K (A) n n+p −→ whenever n d. Then, for n d, there is an exact sequence ≥ ≥ θ− θ+ KQ (A) lim KQ (A) ···−→ε n+1 −→ ε n+1+ps −→ θ− θ+ lim KQ (A) KQ (A) lim KQ (A) → ε n+ps −→ε n −→ ε n+ps ←− −→ where θ+ (respectively θ−) is induced from the cup-product with the positive Bott element b+ (resp. the negative Bott element b−). Moreover, for n d+q 1, there ≥ − is a short split exact sequence θ− θ+ 0 lim KQ (A) KQ (A) lim KQ (A) 0. → ε n+ps −→ε n −→ ε n+ps → ←− −→ It turns out that the inverse limit is not always trivial. This point is discussed in Section 2 (where the inverse limit vanishes) and Section 4 (where it does not). However,forrings ofgeometricnatureandoffinite mod2 virtual´etalecohomo- logical dimension, we conjecture that the inverse limit is trivial. Definition 0.5. We say that a ring A is hermitian regular if lim KQ (A) and ε n+ps lim1 KQ (A) are trivial3. ←− ε n+ps ←− Remark 0.6. Itshouldbenotedthatsubsequenttotheoriginalsubmissionofthis paper at the beginning of February 2010, as a consequence of a more recent theo- remofHu,KrizandOrmsby[25]incharacteristic0,theauthorsandM.Schlichting proved independently that a field of characteristic 0 that is of finite mod 2 virtual ´etale cohomological definition is hermitian regular. Furthermore, Schlichting ex- tended this theorem for fields of characteristic p > 0 with the same cohomological properties. This affirms Conjecture 6.6 of the present paper, which implies in turn our Conjecture 0.14 and therefore considerably extends the number of examples of commutative rings (and schemes) that are hermitian regular. The details of the proofs will appear in a forthcoming joint paper of the authors and Schlichting [7]. A particular example quoted below is given by suitable rings of integers in a num- ber field. In Theorem 0.10, we state the periodicity theorem in this case with an independent proof which will be given in Section 2. A more general theorem is as follows. 3As a matter of fact, with our hypothesis about periodicity of the K-groups,we always have lim1=0,sincetheinversesystemsatisfiestheMittag-LefflerconditionasweshallseeinSections 3and4. Hermitianperiodicity January12,2011 5 Theorem 0.7. Let A be a ring which is hermitian regular and satisfies the hypoth- esis of the previous theorem for its K-groups. Then for n d, the cup-product with ≥ the positive Bott element induces an isomorphism ∼ KQ (A) = KQ (A). ε n −→ε n+p More generally, in order to fully exploit the spectrum approach and to improve the previous theorems, we may consider a pointed CW-complex X and define the group K (A) as the group of homotopy classes of pointed maps from X to (A), X K where denotes the K-theory spectrum. If X is a pointed sphere Sn, we recover K Quillen’s K-group K (A). For brevity, we shall also write K (A) instead of n X+t KX∧St(A), and KX−t(A) instead of KX(StA), where StA denotes the t-iterated suspension of A (see for instance [37] for the definition of the suspension and the basicdefinitionsofvariousK-theories). Weadoptthesameconventionsforhermit- ian K-theory, and also for algebraic or hermitian K-theory with coefficients, and finally for spectra. The previous theorem can now be generalized as follows. Theorem 0.8. Let A be any ring (with 1/2 A), m, p and q be 2-powers as in ∈ Conventions 0.1 and 0.2, and let d Z, such that the cup-product with the Bott ∈ element b in K (Z; Z/2p) induces an isomorphism K p ∼ K (A) = K (A) n n+p −→ whenever n d. Then, if X is a (d 1)-connectedspace, there is an exact sequence ≥ − θ+ KQ (A) lim KQ (A) ε X+1 −→ ε X+1+ps −→ θ− θ+ lim KQ (A) KQ (A) lim KQ (A) . −→ ε X+ps −→ε X −→ ε X+ps →··· ←− −→ If X is (d+q 2)-connected, there is a split short exact sequence − θ− θ+ 0 lim KQ (A) KQ (A) lim KQ (A) 0. → ε X+ps −→ε X −→ ε X+ps → ←− −→ Finally, if A is hermitian regular and if X is (d 1)-connected, the cup-product − with the positive Bott element induces an isomorphism KQ (A)= KQ (A). ε X ∼ε X+p Corollary 0.9. For any (d+q 2)-connected space X and A as above (not nec- − essarily hermitian regular), there is a periodicity isomorphism KQ (A)= KQ (A). ε X ∼ε X+p ForsuitablesubringsA inanumberfieldF,thepreviousresultsmaybe stated S more precisely, by using the methods of [6]. The rings A , defined in Section 2 S below, generalize both the ring of S-integers (when S is finite) and the number field F itself. (More general examples are considered in Section 6 and in [7].) Theorem 0.10. Let F be a totally real 2-regular number field as considered in [6]; also, let m and p be 2-powers as in Convention 0.1. Then, for all integers n > 0, the inverse limit lim KQ (A ) is trivial (i.e. A is hermitian regular) and the ε n+ps S S “positive” Bott m←a−p βn = ∪b+ :εKQn(AS)−→εKQn+p(AS) 6 BKO JANUARY12,2011 is an isomorphism. More generally, if X is any connected CW-complex, the Bott map β : KQ (A ) KQ (A ) X ε X S −→ε X+p S is an isomorphism. For completeness we mention the odd-primary analog of Theorem 0.4, which is provedinSection5. ItsapplicationsarerelatedtotheBloch-Katoconjectureaswe mentionedatthe beginning. We notethatthehypothesis1/2 Amaybe dropped ∈ in this case. Theorem 0.11. Let p and m be odd prime powers as in Convention 0.1. Let b K be the associated Bott element in K (Z; Z/m) (see Section 1 for details). Now let p A be any ring and assume that, whenever n d, cup-product with b induces an K ≥ isomorphism K (A)=K (A). n ∼ n+p Then there exists a “mixed Bott element” b in KQ (Z′) such that for n d, the 1 p ≥ cup-product with b induces an isomorphism between the related KQ-groups ∼ β : KQ (A) = KQ (A). n ε n −→ε n+p More generally, if X is a (d 1)-connected CW complex, then the cup-product map − with b induces an isomorphism ∼ β : KQ (A) = KQ (A). X ε X −→ε X+p InSection6wenotethatworkinprogressbySchlichting[56]allowsustoextend ourresultsfromcommutativeringstoschemesS thatareseparated,noetherianand of finite Krull dimension. More precisely, following Jardine’s method for algebraic ´et K-theory [29] we define an “´etale” KQ-theory, denoted by KQ (S), where the ε n coefficient groups are prime powers. The ´etale KQ-theory shares many properties with the ´etale K-theory introduced by Dwyer and Friedlander [15]. For example, there exists a comparison map ´et σ : KQ (S) KQ (S). ε n −→ε n Foroddprimepowers,thereisaninvolutionontheoddtorsiongroup KQ (S). ε n ´et ´et Let εKQn(S)+ and εKQn(S)− denote the corresponding eigenspaces. On the otherhand,foranyprimepower(oddoreven),thecup-productmapwiththeBott elementb+ defined aboveinduces a directsystemofgroups,whose colimitwe shall denote by KQ (S) β−1 , in the notation of [60]. Next, we state two theorems ε n and a conjecture in this context. (cid:2) (cid:3) Theorem 0.12. With the coefficient group Z/ℓν, where ℓ is an odd prime, there is an isomorphism KQ (S) β−1 = KQ´et(S) ε n ∼ε n for all n if cdℓ(S)< . Moreover, th(cid:2)e com(cid:3)parison map σ induces an isomorphism ∞ ´et KQ (S) = KQ (S) ε n + ∼ε n for n sup cdℓ(k(s)) 1 s∈S. ≥ { − } Hermitianperiodicity January12,2011 7 Recall that S is uniformly ℓ-bounded with bound d if for all residue fields k(s) we have cd (k(s)) d. In the event that S is uniformly ℓ-bounded with bound d, ℓ ≤ then cd (S) n+d where n denotes the Krull dimension of S; an elegant proof ℓ ≤ for this inequality is given in [39, Theorem 2.8]. At the prime 2 we prove the following theorem, reminiscent of the main results in [16] and in [60]. Theorem 0.13. With the coefficient group Z/2ν, there is an isomorphism KQ (S)[β−1]= KQ´et(S) ε n ∼ε n for all n if vcd (S)< . Moreover, the comparison map 2 ∞ ´et σ : KQ (S) KQ (S) ε n −→ε n is a split surjection for n sup vcd2(k(s)) 1 s∈S +q 1. ≥ { − } − More generally, we make the following conjecture. Conjecture 0.14. Withthe coefficient groupZ/2ν the map σ is bijective whenever n sup vcd2(k(s)) 1 s∈S. ≥ { − } Using algebro-geometricmethods, in Theorem6.5below we showhow to reduce this conjecture to the case of fields. As mentioned above, the characteristic 0 case was solved independently by the authors and M.Schlichting, while the positive characteristic case was solved by Schlichting. A proof of this conjecture in general will appear in a joint paper with Schlichting [7]. Let us now briefly discuss the contents of the paper. In Section 1, for 2-power coefficients we carefully construct the Bott elements that play an important role in this work, as referred to above. Section2issomewhatindependentofthe othersections. Inparticular,weprove a refined version of our theorems in the case A is the ring of integers in a totally real 2-regular number field. (This version is a particular case of the considera- tions in Section 6 for schemes. Assuming Conjecture 0.14, which will be proven in [7], Theorem 2.1 may be given an independent proof in a much more general framework.) InSection3,weintroducewhatwecall“higherKSC-theories”. Thesetheoriesin some sense measure the deviation of “negative” periodicity of the KQ-groups. On theotherhand,they arebuiltbysuccessiveextensionsoftheK-groups. Therefore, they are periodic if the K-groups are periodic. Section4isdevotedtotheproofofourmainTheorems0.4and4.5(forarbitrary rings with 2 invertible and mod 2ν coefficients). The proof is roughly divided into two steps as follows. In the first one, we prove a cruder periodicity statement for n d + q 1. In the second, we use the KQ-spectrum and an argument ≥ − aboutcohomologytheories to provethe periodicity theorems in full generality. We conclude this section with an upper bound of the KQ-groups in terms of the K- groups. InSection5,westudy the caseofoddprime powers,whichisparadoxicallysim- pler in our framework. The main observation is that the KQ-ring spectrum splits naturally as the product of two ring spectra, the first one being the “symmetric” part of the K-theory spectrum. Section6ismoregeometricinnatureandgeneralizesthepreviousconsiderations (whenAiscommutative)tonoetherianseparatedschemesoffiniteKrulldimension. 8 BKO JANUARY12,2011 Here we rely heavily on the fundamental theorem in hermitian K-theory provedin the scheme framework by Schlichting [56]. Finally,Sections7and8aredevotedtoselectedapplications: ringsofintegersin numberfields,smoothcomplexalgebraicvarieties,andringsofcontinuousfunctions on compact spaces. Another application, to hermitian KQ-theory of group rings, is a consequence of an appendix to this paper by C.Weibel [66]. Acknowledgements. Wewarmlythanktherefereeforveryrelevantcomments onapreviousversionofthis paper. We extendourthanksto MarcoSchlichtingfor discussions resulting in our joint work [7]. 1. Bott elements in K- and KQ-theories Let ℓ be a prime number and 0/ℓν the mod ℓν Moore spectrum. In [1, 12], S § Adams constructed ∗-equivalences KO Aℓν :Σp 0/ℓν 0/ℓν. S −→S Thedimensionshiftpissup 8,2ν−1 ifℓ=2and2(ℓ 1)ℓν−1 ifℓisodd. Asshown { } − by Bousfield in [12, 4], work of Mahowald and Miller implies that a spectrum § E is -local if and only if its mod ℓ homotopy groups are periodic via A for every ℓ KO prime ℓ. We shall refer to the periodicity manifested in -local spectra as Bott KO periodicity. Note that -localizations are the same as -localizations [12, 4]. KO KU § IngeneralthereareseveralchoicesofanelementAℓν asaboveiftheonlycriterion is that it induces a -isomorphism. We are interested in particular choices of KO elements pertaining to classical Bott periodicity. Let u denote a generator of the infinite cyclic group π (BU). Then for r 1 the Bott element u2r in π (BU) is 2 4r ≥ independentofthechoiceofu. Wedenotebyv theelementofπ (BO)mappingto 8r theBottelementinπ (BU)underthemapinducedbycomplexificationc:BO 8r → BU. The mod 2ν Bott element in degree 8r>0 is the generator v =id v KO ( 0/2ν; Z/2ν)=[ 0/2ν, 0/2ν] . S0/2ν 8r 8r ∧ ∈ S S KO∧S The element A2ν is called an Adams periodicity operator if it maps to the mod 2ν Bott element in degree p under the naturally induced -Hurewicz map KO π∗( 0/2ν; Z/2ν) ∗( 0/2ν; Z/2ν) S →KO S for 0/2ν. When ℓ = 2, the definition of a mod ℓν Bott element is the same as S 6 above,except that is replacedby . Crabb and Knapp[14] have shownthat KO KU there exist Adams periodicity operators for all ℓ and ν 1. ≥ Bysmashingtheunitmap 0 ofaringspectrum with 0/ℓν andpushing S →E E S forward the class in πp( 0/ℓν; Z/ℓν) represented by the map Aℓν, one obtains a S class in the group π ( ; Z/ℓν) that we call a Bott element. p Next, for m = 2p,Ewhere p = 2υ−1 is a 2-power 8, we study mod m Bott elements in more detail for K- and KQ-theory in the≥example of Z′. The case of an odd prime is dealt with in Section 5. Tobegin,weshallconsider“Bottelements”inK (Z′; Z/m)and KQ (Z′; Z/m), p 1 p whose images in K (R; Z/m) and KQ (R; Z/m), respectively, are generators de- p 1 p duced from classical Bott periodicity for the real numbers (as KQ-modules). This is well-known for the algebraic K-groups; it is included here for the sake of com- pleteness. Hermitianperiodicity January12,2011 9 B¨okstedt’s square of algebraic K-theory spectra introduced in [9] (Z′) (R)c K # −→ K # ↓ ↓ (F ) (C)c K 3 # −→ K # was verified to be homotopy cartesian by Rognes-Weibel in [49], [65], as a conse- quence of Voevodsky’s proof of the Milnor conjecture. Here means 2-adic com- # pletions and c means connective cover. Smashing with 0/2ν yields a homotopy S cartesian square (an overbar indicates reduction mod m): (Z′) (R)c K −→ K ↓ ↓ (F ) (C)c 3 K −→ K DenotebyK the correspondingmodm homotopygroups. ByBottperiodicityand the isomorphism Kp−1(Z′) Kp−1(F3), there is a split short exact sequence → 0 K (Z′) K (R) K (F ) K (C) 0. p p p 3 p −→ −→ ⊕ −→ −→ On the other hand, Quillen’s homotopy fibration Ω (C)Ψ3−1Ω (C) (F ) (C)Ψ3−1 (C) 3 K −→ K −→K −→K −→ K yields an exact sequence K (C) ·m K (C) K (F ) K (C) ·m K (C), p+1 p+1 p 3 p p −→ −→ −→ −→ and hence the isomorphisms K (F )= K (C)=Z/m. p 3 ∼ m p ∼ Here A denotes the kernel of the multiplication by n map on an abelian groupA. n Hence, diagram chasing shows there are isomorphisms K (Z′)=K (R) and K (Z′)=K (F ). p ∼ p p ∼ p 3 Moreprecisely,thereexistsaBottelementb inK (Z′)mappingatthesametime K p to a generator of K (R) and to a generator of K (F ). p p 3 WeproceedinthesamemannerinordertoexplicateBottelementsinhermitian K-theory,having almostthe exactsame properties as their namesakesin algebraic K-theory. More precisely, we shall prove the following theorem: Theorem 1.1. Let p 8 a 2-power and m=2p. Then the group KQ (Z′; Z/m) 1 p is isomorphic to Z/m ≥Z/m Z/2. There is a Bott element b+ in KQ (Z′; Z/m) 1 p ⊕ ⊕ thatmaps at thesametimetoagenerator ofZ/min KQ (F ; Z/m)=Z/m Z/2 1 p 3 ∼ ⊕ and to a generator of KQ (R; Z/m), viewed as a module4 over KQ (R; Z/m). 1 p 1 0 Proof. In the following proof, we are going to use the results of [5, Theorem 6.1] and[6,Theorems1.2,1.5]. In[5],itisshownthatthesquareofhermitianK-theory completed connective spectra (Z′)c (R)c 1KQ # −→ 1KQ # ↓ ↓ (F )c (C)c 1KQ 3 # −→ 1KQ # 4TheringstructureforKQ-theorywithmod2ν coefficients iswell-definedifν≥4. 10 BKO JANUARY12,2011 is homotopy cartesian (Recall that (C) is just .) Reducing mod m yields 1 KQ KO another homotopy cartesian square: (Z′)c (R)c 1 1 KQ −→ KQ ↓ ↓ (F )c (C)c 1 3 1 KQ −→ KQ This in turn gives rise to a short exact sequence (1:1) 0 KQ (Z′) KQ (R) KQ (F ) KQ (C) 0, −→1 p −→1 p ⊕1 p 3 −→1 p −→ whichsplitssince KQ (R)isadirectsumoftwocopiesof KQ (C),sayG G. The 1 p 1 p ⊕ firstcopyofG, say G , is generatedby the imageof 1 under the Bott isomorphism 1 KQ (R) = KQ (R) (see Appendix B in [6]). The splitting is given by the 1 0 ∼ 1 p isomorphism between KQ (C) and the second copy of G, say G . Therefore, we 1 p 2 get an isomorphism KQ (Z′)=G KQ (F ) 1 p ∼ 1⊕1 p 3 In order to finish the proof of the theorem, we need to compute KQ (F ). By 1 p 3 [19], there is a Bockstein exact sequence (1:2) 0 Z/2 KQ (F ) Z/m 0. −→ −→1 p 3 −→ −→ In order to resolve this extension problem, consider the map KQ (F )/m=Z/m Z/2 KQ (F ) 1 0 3 ⊕ −→1 p 3 given by cup-product with any element that maps to the generator of KQ (C)= 1 p ∼ Z/m under the Brauer lift KQ (F ) KQ (C). This gives a splitting of the 1 p 3 → 1 p exact sequence (1:2), and therefore KQ (F )=Z/m Z/2. 2 1 p 3 ∼ ⊕ Remarks 1.2. By considering the forgetful map from the hermitian K sequence (1:1)toitsalgebraicKcounterpart,oneseesthattheBottelementof KQ (Z′; Z/m) 1 p mapstotheBottelementinthecorrespondingalgebraicK-theorygroupunderthe map induced by the forgetful functor. Moreover,all the results for F in the above 3 also hold for any finite field F with t elements, provided t 3(mod 8). t ≡± 2. Proof of the periodicity theorem for totally real 2-regular number fields Let A be the ring of 2-integers in a totally real 2-regular number field F with r real embeddings. In [6], we proved that the square of hermitian K-theory 2- completed connective spectra (A)c r (R)c εKQ # −→ εKQ # ↓(F )c Wr ↓ (C)c εKQ t # −→ εKQ # is homotopy cartesian (with t a carefully chWosen odd prime and where # denotes 2-adic completion). Therefore, the mod 2ν reduction of this square, namely (A)c r (R)c ε ε KQ −→ KQ ↓(F )c Wr ↓ (C)c ε t ε KQ −→ KQ is also homotopy cartesian, since εKQ−1(A)W= 0 by Lemmas 3.11 and 3.12 in [6]. Using this square, we deduce an enhanced version of our periodicity theorem.