EQUIVARIANT K-THEORY OF COMPACT LIE GROUPS WITH INVOLUTION 4 PO HU, IGOR KRIZ AND PETR SOMBERG 1 0 2 n a Abstract. For a compact simply connected simple Lie group G J withaninvolutionα,wecomputetheG⋊Z/2-equivariantK-theory 0 of G where G acts by conjugation and Z/2 acts either by α or by 3 −1 g 7→α(g) . Wealsogivearepresentation-theoreticinterpretation ] of those groups, as well as of KG(G). T K . 1. Introduction h t a Brylinski and Zhang [2] computed, for (say) a simple compact Lie m groupG,theG-equivariantK-theoryringK (G)withGactingonitself G [ by conjugation, as the ring Ω of Ka¨hler differentials of R(G) over R(G)/Z 1 Z (see also Adem, Gomez [1] for related work). Let α be an involutive v automorphism of G. Then we can consider actions of G ⋊ Z/2 on G 7 where G acts by conjugation and the generator of Z/2 acts either by 2 8 the automorphism α or by the map γ : g 7→ α(g)−1. The main result 7 of the present paper is a computation of K (G) in both cases as . G⋊Z/2 1 a module over K (∗) = R(G⋊Z/2). (Here R(H) is the complex 0 G⋊Z/2 4 representation ring of a compact Lie group H.) 1 The involutive automorphism α determines a compact symmetric : v space G/Gα, and we were originally interested in these computations i X as a kind of topological invariant of symmetric pairs of compact type. r (Recall, in effect, that G/Gα is a connected component of Gγ via the a embedding x 7→ α(x)x−1.) It turns out, however, that the groups K (G) are a rather crude invariant of symmetric pairs, since they G⋊Z/2 essentiallyonlydependonwhetherαisanouterorinnerautomorphism ofG; ifαisaninnerautomorphism, G⋊Z/2becomesacentralproduct, which behaves essentially thesameasthedirect product fromourpoint of view. Nevertheless, having a complete calculation is still interesting, as are some of the methods involved in it. The main ingredient of the method we present here is the construction of Brylinski-Zhang [2, 3] of The authors acknowledge the support of the Eduard Cˇech ECE in Prague GA CR P201/12/G028,and NSF grants DMS 1104348and DMS 1102614. 1 2 PO HU,IGOR KRIZAND PETR SOMBERG the element dv ∈ Ω = K (G) for a finite-dimensional complex R(G)/Z G representation v of G. That construction, unfortunately, was presented incorrectly in [2] (in fact, the elements written there are 0), and so we developed an alternate construction of those elements using induc- tion from the normalizer of a maximal torus. However, Brylinski [3] communicated the correct construction to us. The construction [3] is completely geometric, and supersedes our previous induction method (which, for that reason, we omit from this presentation). In fact, the construction [3] turns out to be equivariant with respect to both the α and γ actions. This allows an “obvious guess” of what K (G) G⋊Z/2 should be. We validate that guess following the methods of Green- lees and May [9], involving Tate cohomology. (Essentially, the main point is that under suitable finiteness hypothesis, a Z/2-equivariant map of Z/2-CW complexes which is an equivalence non-equivariantly is an equivalence equivariantly because the Tate cohomology can be computed as an algebraic functor of the “geometric fixed points”.) We realized, however, that the construction [3] can be generalized to give a representation-theoretical interpretation of the groups K (G), G K (G). Such an interpretation is strongly motivated by the work G⋊Z/2 of Freed, Hopkins and Teleman [6] who showed that if τ is a regular G-equivariant twisting of K-theory on G, then the twisted equivariant K-theoryK (G) isisomorphic tothefreeabeliangrouponirreducible G,τ projective representations of level τ−h∨ (where h∨ is the dual Coxeter number) of the loop group LG. This suggests that untwisted K-theory K (G) should correspond to G representations at the critical level of the Lie algebra Lg. We found that this is indeed true, but the representations one encounters are not lowest weight representations (which occur, for example, in the geometric Langlands program). Instead, the fixed point space of the infinite loop space K (G) turns out to be the group completion of the G 0 spaceoffiniterepresentationsoftheloopgroupLGwithanappropriate topology. Here by a finite representation we mean a finite-dimensional representation which factors through a projection LG → Gn given by evaluationatfinitelymanypoints(cf. [16]). (Itispossibletoconjecture that every finite-dimensional representation of LG is finite, although it may depend on the type of loops we consider; in this paper, we restrict our attention to continuous loops.) In fact, we also prove that this is true Z/2-equivariantly with respect to involutions, i.e. that the fixed point space of K (G) is the group completion of the space G⋊Z/2 0 of representations of LG ⋊ Z/2, where Z/2 acts on LG via its action on G in the case of Z/2 acting on G by α, and simultaneously on G EQUIVARIANT K-THEORY OF GROUPS WITH INVOLUTION 3 and on the loop parameter by reversal of direction in the case when Z/2 acts on G by γ. The present paper is organized as follows: In Section 2, we review the construction of Brylinski-Zhang [2, 3] and study its properties with respect to the involution on G. In Section 3, we compute the R(G ⋊ Z/2)-modules K∗ (G). In Section 4, we discuss the computation in G⋊Z/2 more concrete terms, and give some examples. In Section 5, we give an interpretation of K∗(G) in terms of representation of the loop group G LG, and in Section 6, we make that interpretation Z/2-equivariant, thus extending it to K∗ (G). G⋊Z/2 2. The Brylinski-Zhang construction Let G be a simply connected compact Lie group, T a maximal torus, N its normalizer, W the Weyl group. Let R(G) denote, as usual, the complex representation ring. Recall that if u ,...,u are 1 n the fundamental weights of G (n = rank(G)), then the weight lattice T∗ = Hom(T,S1) is freely generated by u ,...,u and we have 1 n R(T) = Z[u ,u−1,...,u ,u−1], 1 1 n n R(T) ⊃ R(G) = Z[u ,...,u ] 1 n where u is the sum of elements of the W-orbit of u . i i Let, for a map of commutative rings S → R, Ω denote the ring of R/S Ka¨hler differentials of R over S. Then one easily sees by the Ku¨nneth theorem that we have an isomorphism: K∗(T) = Ω (1) T R(T)/Z = Z[u ,u−1,...,u ,u−1]⊗Λ [du ,...,du ] 1 1 n n Z 1 n Theorem 1. (Brylinski-Zhang [2]) Suppose G acts on itself by conju- gation. Then there is a commutative diagram of rings (2) K (G) res. // K (T) G T ∼= ∼= (cid:15)(cid:15) (cid:15)(cid:15) Ω // Ω R(G)/Z R(T)/Z = = (cid:15)(cid:15) (cid:15)(cid:15) Z[u ,...,u ]⊗Λ[du ,...,du ] ⊂ // Z[u±1,...,u±1]⊗Λ[du ,...,du ]. 1 n 1 n 1 n 1 n 4 PO HU,IGOR KRIZAND PETR SOMBERG Moreover, the isomorphisms in (2) can be chosen in such a way that the generator dW ∈ K (G) for a complex (finite-dimensional) G- G representation W is represented in G-equivariant K-theory by a com- plex of G-bundles (3) G×R×W φ // G×R×W where φ is a homomorphism which is iso outside of G×{0}×W, and is given by φ(g,t,w) = (g,t,tw) for t < 0 (4) φ(g,t,w) = (g,t,−tg(w)) for t ≥ 0. Proof. The proof of the diagram (2) is given in [2]. Since there is a mistake in the formula (4) in [2], (corrected in [3]), we give a proof here. In view of the commutativity of diagram (2), and injectivity of the diagonal arrows, it suffices to prove the statement for T instead of G. By Ku¨nneth’s theorem, it suffices to further consider T = S1. In the case T = S1, let z be the tautological 1-dimensional complex representation of S1 (considered as the unit circle in C). Then the element of K−1(S1) given by (4) for W = zn is equal to the element S1 e of K0 (S2) given by Hn − 1 where H is the tautological bundle on S1 S2 =e CP1 (with trivial action of S1). But it is well known that H = u+1 where u ∈ K0 (S2) is the Bott periodicity element, and thus, (recalling S1 that u2 = 0e), Hn −1 = (u+1)n −1 = nu. Thus, choosing the Bott element as u gives the required isomorphism in the right hand column of (2) for T = S1, and thus the statement (cid:3) follows. Proposition 2. Let G be as above, let α be an involutive automorphism of G, and let W be a finite-dimensional complex representation such that (5) α∗(W) ∼= W (where α∗(W) is the representation of G on W composed with the au- tomorphism α). Then, given the choices described in Theorem 1, if the generator a of Z/2 acts on G by α, then dW is in the image of the restriction (forgetful map) (6) K1 (G) → K1(G). G⋊Z/2 G EQUIVARIANT K-THEORY OF GROUPS WITH INVOLUTION 5 When the generator a of Z/2 acts on G by γ, dW is in the image of the restriction (forgetful map) (7) KA (G) → K1(G) G⋊Z/2 G where A is the 1-dimensional real representation of G ⋊Z/2 given by the sign representation of the quotient Z/2. Proof. Recall that when (5) holds, then a choice of the isomorphism (5) canbe madeto give a representation ofG⋊Z/2 on W. Moreover, there are precisely two such choices, differing by tensoring by the complex 1-dimensional sign representation of Z/2. Consider first the case when the generator a of Z/2 acts on G by α. Then consider the Z/2-action on the Brylinski-Zhang construction G×R×W φ // G×R×W (8) (cid:15)(cid:15) (cid:15)(cid:15) G×R×W φ // G×R×W where the generator of Z/2 acts by (g,t,w) (g,t,w) ❴ ❴ (cid:15)(cid:15) (cid:15)(cid:15) (α(g),t,α(w)) (α(g),t,α(w)). When the generator a of Z/2 acts on G by γ, consider the Z/2-action (8) where the generator of Z/2 acts by (g,t,w) (g,t,w) ❴ ❴ (cid:15)(cid:15) (cid:15)(cid:15) (α(g)−1,−t,α(w)) (α(g)−1,−t,α(g−1w)) (Note that α(α(g)α(g−1w)) = w, so the action of a on the right hand side is involutive. One readily sees that it also intertwines the action of G via the automorphism α.) To verify that the homomorphism φ commutes with the involution in the case of a acting on G via γ, since we already know the action is involutive, it suffices to consider t < 0. In this case, we have (g,t,w) ✤ // (g,t,tw) ❴ ❴ (cid:15)(cid:15) (cid:15)(cid:15) (α(g)−1,−t,α(w)) ✤ // (α(g)−1,−t,tα(g)−1α(w)). 6 PO HU,IGOR KRIZAND PETR SOMBERG (cid:3) 3. The computation of equivariant K-theory In this section, we will compute K (G) where the generator a G⋊Z/2 of Z/2 acts by α or γ. First observe that in both cases, the gener- ator a of Z/2 acts on K∗(G) ∼= Ω by automorphisms of rings. G R(G)/Z The action on R(G) is given by a permutation representation given by the permutation of irreducible representations by the automorphism α. Alternately, one may think in terms of the action of α on Weyl group orbits of weights. Let u ,...,u be the fundamental weights of the 1 n simply connected group G determined by the Lie algebra g. Let σ be the involution on {1,...,n} given by α∗u = u . i σ(i) Consider now the short exact sequence (9) 1 → G → G⋊Z/2 → Z/2 → 1. By Z/2 , we shall mean the suspension spectrum of the G⋊Z/2-space + Z/2 by the action (9). We define SA by the cofibration sequence + Z/2 ι // S0 → SA + where S0 is the G⋊Z/2-sphere spectrum and ι is the collapse map (for terminology, see [13]). We have, of course, K∗ S0 = R(G⋊Z/2) , G⋊Z/2 even K∗ Z/2 = R(G) . G⋊Z/2 + even Here, the subscript ? means that the given R(G⋊Z/2)-module is even locatedintheeven dimension ofthe Z/2-gradedring K∗. Furthermore, we have an exact sequence 0 → K0 SA → R(G⋊Z/2) → R(G) → K1 SA → 0 G⋊Z/2 G⋊Z/2 where the middle arrow is restriction. Therefore, K1 SA is the free G⋊Z/2 abelian group on irreducible G-representations which do not extend to G ⋊ Z/2. Recall that Z/2 acts on the set of isomorphism classes of irreducible representations of G; R(G⋊Z/2) is the free abelian group on the regular orbits, and on two copies of each fixed orbit. Therefore, K0 SA can be thought of as the free abelian group on irreducible G⋊Z/2 G-representations which do extend to G⋊Z/2-representations. Equiv- alently, K0 SA = Z{u ∈ T∗ dominant |α∗u = u}, G⋊Z/2 EQUIVARIANT K-THEORY OF GROUPS WITH INVOLUTION 7 K1 SA = Z{regular α∗-orbits of dominant weights}. G⋊Z/2 LetS(ǫ) forǫ ∈ ZdenoteSA−1 = Σ−1SA resp. S0 depending onwhether ǫ is odd or even. Let ≻ denote any chosen linear ordering of the set of subsets of {1,...,n}. Let I be the set of subsets σ (10) {i < ··· < i } ⊆ {1,...,n} 1 k such that {σ(i ),...,σ(i )} ≻ {i ,...,i } 1 k 1 k and let J be the set of subsets (10) such that σ {σ(i ),...,σ(i )} = {i ,...,i }. 1 k 1 k Let orb(S) for a σ-invariant set S denote the number of regular (=Z/2- free) σ-orbits of S when a acts on G by α, and of all σ-orbits of S when a acts on G by γ. Theorem 3. There exists an isomorphism of R(G⋊Z/2)-modules K∗ (G) ∼= G⋊Z/2 (11) KG∗⋊Z/2( _ ΣkZ/2+ ∨ _ ΣkS(orb{i1,...,ik})) {i1<···<ik}∈Iσ {i1<···<ik}∈Jσ G ⋊ Z/2 acts on the wedge summands on the right hand side of (11) through the projection to Z/2. Proof. We first construct a G⋊Z/2-equivariant stable map u of each S wedge summand of (11) into the E -algebra ∞ (12) F(Λ ,K ) + G⋊Z/2 where Λ is G on which g ∈ G ⊂ G ⋊ Z/2 acts by conjugation and α acts by γ such that the wedge of all the maps (12) induces an isomor- phism on G-equivariant coefficients. Here F(?,?), as usual, denotes the (equivariant) function spectrum, see [13]. First, recall the isomorphism [2] π (F(Λ ,K )G) = K∗(G) ∗ + G⋊Z/2 G (13) ∼= Ω = Z[u ,...,u ]⊗Λ[du ,...,du ] R(G)/Z 1 n 1 n of Theorem 1, induced by (1). Forregular(=σ-free)orbits, themapweneedfollowsfromG-equivariant considerations: Send, G-equivariantly, Sk → F(Λ ,K ) + G⋊Z/2 8 PO HU,IGOR KRIZAND PETR SOMBERG by the generator du ∧···∧du i1 ik of (13) and then use the fact that (G ⋊ Z/2)⋊? is the left adjoint to the forgetful functor from G⋊Z/2-spectra to G-spectra (cf. [13]). Next, for σ-invariant sets 1 ≤ i < ··· < i ≤ n which consist of a 1 k single orbit, we have k ≤ 2. If k = 1, the map follows from Proposition 2. If k = 2, we have a G-equivariant map u : S1 → F(Λ ,K ) + G⋊Z/2 given as the generator du ∧ du of π (F(Λ ,K )G) = K∗(G). i1 i2 ∗ + G⋊Z/2 G The G⋊Z/2-equivariant map u : S1+A → F(Λ ,K ) {i1,i2} + G⋊Z/2 we seek may then be defined as G⋊Z/2 N u G where N is the multiplicative norm (see [8, 11]). Finally, we may define u := u ∧···∧u , S1∐···∐Sℓ S1 Sℓ using Bott periodicity to identify S2 with S2A. Thus, taking a wedge sum of these maps, we have a map X := F( ΣkZ/2 ∨ ΣkS(orb{i1,...,ik}),K ) W + W G⋊Z/2 (14) ↓ f Y := F(Λ ,K ) + G⋊Z/2 of K -modules, inducing an isomorphism of G-equivariant coeffi- G⋊Z/2 cients (using the Wirthmu¨ller isomorphism [13] and, again, Bott peri- odicity). This implies that (14) induces an equivalence on Borel coho- mology: (15) F(EZ/2 ,fG) : F(EZ/2 ,XG)Z/2 ∼ // F(EZ/2 ,YG)Z/2. + + + We need to conclude that (14) induces an equivalence on G⋊Z/2-fixed points, i.e. that we have an equivalence (16) (fG)Z/2 : (XG)Z/2 ∼ // (YG)Z/2. To this end, consider XG, YG as Z/2-equivariant spectra. EQUIVARIANT K-THEORY OF GROUPS WITH INVOLUTION 9 Lemma 4. Denote R = R := (R(G⋊Z/2)/indG⋊Z/2R(G))⊗Q, G G R = R := (R(G⋊Z/2)/indG⋊Z/2R(G))∧ ⊗Q. G G 2 b c Then for the Z/2-spectra Z = XG,YG, the spectra ΦZ/2Z, Z (see [13, b 9]) are rational, and we have an isomorphism (17) Z ∼= (ΦZ/2Z) ⊗ R ∗ ∗ R b b natural with respect to the map (14). (Here indG⋊Z/2 : R(G) → R(G⋊ G Z/2) denotes the induction, and (?)∧ denotes completion at 2.) 2 Proof of (16) using Lemma 4: Note that (14) also implies an equiva- lence on Borel homology: (18) (EZ/2 ∧fG) : (EZ/2 ∧XG)Z/2 ∼ // (EZ/2 ∧YG)Z/2 + + + and Tate cohomology (19) fG : XG ∼ // YG. c d c By (17), however, the map ΦZ/2fG : ΦZ/2(XG) → ΦZ/2(YG) (cid:3) is also an equivalence, and together with (18), this implies (16). Proof of Lemma 4: The spectra ΦZ/2(M), M are rational for any cell c module M over the E -ring spectrum K by a theorem of Green- ∞ Z/2 less and May [9] which asserts this for M = K . Additionally, the Z/2 methods of [9] (or a direct calculation) readily imply that ΦZ/2(KG ) = R, G⋊Z/2 ∗ (20) \ (KG ) = R. G⋊Z/2 ∗ b Now ?⊗ R is clearly an exact functor on R-modules, so by using the R long exactbsequence in cohomology, it suffices to filter XG, YG both into finite sequences of cofibrations such that the quotients Z satisfy (17). InthecaseofXG,thequotientsareeitheroftheformF(Z/2 ,KG ) + G⋊Z/2 for which the statement is trivial (both the geometric and Tate theory are 0) or KG , which is covered by (20), or ΣA(KG ), which is a G⋊Z/2 G⋊Z/2 cofiber of modules of the first two types. In the case of YG, use the decomposition of Λ into G-orbits with respect to conjugation of skeleta of the fundamental alcove, applying 10 PO HU,IGOR KRIZAND PETR SOMBERG also the fact that γ acts trivially on T. This is, in fact, a G ⋊ Z/2- CW decomposition, where the cells are of type H ⋊Z/2 where H is a compact Lie subgroup of G associated to a sub-diagram of the affine Dynkin diagram. Applying the computation of geometric and Tate Z/2-fixed points of K , we are done if we can prove H (21) R = R ⊗ R . H H RG G c c To this end, put R0 := R(H ⋊Z/2)/indH⋊Z/2R(H). H H Recall from [17] that R(G⋊Z/2) is a Noetherian ring and R(H⋊Z/2) is a finite module. Therefore, R0 is a finite R0-module. H G Now for any Noetherian ring P and a finite P-module M, we have (22) M∧ = M ⊗ P∧. 2 P 2 (Consider the presentation MP → MP → M → 0 n m and right exactness of (?)∧ in this case.) Rationalizing, (22) implies 2 (cid:3) (21). 4. Concrete computations and examples Let, again, G be a simply connected simple compact Lie group and the generator a of Z/2 act on the target G by α or by γ. To calculate K (G) as an R(G⋊Z/2)-module, in view of Theorem 3, it suffices G⋊Z/2 tocalculate theaction ofthe automorphism α onthe Weyl grouporbits of the fundamental weights of the group G. The key observation is that if α is an inner automorphism, then the actionis trivial simply because aninner automorphism doesnotchange the isomorphism class of a representation. Outer automorphisms of simply connected simple Lie groups cor- respond to automorphisms of the Dynkin diagram, and therefore are necessarily trivial for all types except A,D and E . Furthermore, the 6 permutation representation of Z/2 on orbits of fundamental weights is isomorphic to the permutationrepresentation on theset of simple roots (using the bijection between fundamental weights and simple roots: A fundamental weight is a point of the weight lattice with which one sim- ple root has minimal positive inner product, and the other simple roots have inner product 0).