Antonini, Azzali and Skandalis proved that, for every countable discrete group Γ, the canonical map EΓ→EΓ induces a split injection in equivariant KK-theory with real coefficients. In this paper, we give a more elementary proof of the corresponding rational injectivity result in ordinary equivariant KK-theory without using II1 factors.
Introduction
Throughout, Γ is a countable discrete group. A Γ-C∗-algebra is a C∗-algebra equipped with an action of Γ by ∗-automorphisms. An extremely potent tool in noncommutative geometry, particularly in relation with the Novikov conjecture, is Kasparov’s equivariant KK-theory [5]. For separable Γ-C∗-algebras A and B, Kasparov’s equivariant KK-theory defines an abelian group KKΓ(A,B) consisting of homotopy classes of equivariant Kasparov modules. It is contravariant in A and covariant in B with respect to equivariant ∗-homomorphisms, and is homotopy invariant, stable and split exact. Equivariantly semisplit extensions give six-term exact sequences in both variables. The theory also satisfies Bott periodicity:
KKΓ(A,B)≅KKΓ(Σ2A,B)≅KKΓ(ΣA,ΣB)≅KKΓ(A,Σ2B).
where ΣkA=C0(Rk,A) and Γ acts trivially on Rk. For n∈Z, we write
KKnΓ(A,B):=KKΓ(A,ΣkB),k∈{0,1},k≡n(mod2).
The Kasparov product is an associative bilinear pairing
KKΓ(A,B)⊗ZKKΓ(B,C)⟶KKΓ(A,C),x⊗y⟼x⊗By.
An equivariant ∗-homomorphism φ:A→B determines a class [φ]∈KKΓ(A,B). When Γ is trivial, we write KK(A,B). For a general Γ, there is a natural forgetful homomorphism KKΓ(A,B)→KK(A,B).
A subset S of a Γ-space X is called Γ-compact if S⊆Γ⋅K for some compact subset K⊆X. A Γ-CW-complex is called finite if it has finitely many Γ-orbits of cells, and proper if all its stabilizer subgroups are finite; see [7 Section 1].
Definition 1.1. Let X be a Hausdorff Γ-space, let B be a Γ-C∗-algebra, and let n∈Z. Choose k∈{0,1} with k≡n(mod2), and define
KKnΓ(X,B):=⟶Y,AlimKKΓ(C0(Y),ΣkA),
Key words and phrases.K-homology, K-theory.
The firt author is supported in part by National Natural Science Foundation of China (No. 12471123). The second author was supported in part by Grant for Excellent Young Scholars in Tianyuan Mathematics. where Y ranges over the closed, Γ-invariant, Γ-compact subspaces of X that are locally compact and second countable, and A ranges over the separable Γ-invariant C∗-subalgebras of B, both ordered by inclusion.
Let EΓ be the universal space for proper and free actions of the group Γ, and let EΓ be the universal space for proper actions of the group Γ. Then there exists a natural Γ-equivariant continuous map π:EΓ→EΓ. Antonini, Azzali and Skandalis [1 Theorem 5.4] proved that the induced map
π∗,R:KK∗,RΓ(EΓ,A)⟶KK∗,RΓ(EΓ,A)
is split injective, where the groups are defined using equivariant KK-theory with real coefficients. In this paper, we give an elementary proof of rational injectivity in ordinary equivariant KK-theory, without using real coefficients. We state the main result as follows.
Theorem 1.2. Let Γ be a countable discrete group. For every Γ-C∗-algebra A, the homomorphism
π∗:KK∗Γ(EΓ,A)⊗ZQ⟶KK∗Γ(EΓ,A)⊗ZQ
is injective, where π∗ is induced by the natural Γ-equivariant continuous map π:EΓ→EΓ.
We choose countable Γ-CW-models for EΓ and EΓ. Since every Γ-compact subset of either model is contained in a finite Γ-subcomplex, the limit over Y in Definition 1.1 may be taken over the finite Γ-subcomplexes of the chosen model.
Acknowledgements. We thank Professor Guoliang Yu for very inspiring discussions.
AI disclosure. ChatGPT was used solely for language editing and grammatical refinement during the preparation of this manuscript. All mathematical proofs and core results were developed by the authors.
Preliminaries
We recall exact sequences in the first variable of equivariant KK-theory and the action of equivariant vector bundles, including its compatibility with boundary maps. We then construct rational idempotents using the equivariant Chern character. For an abelian group M, we write MQ=M⊗ZQ. If an idempotent e acts on M, we write eM=Im(e:M→M).
Exact sequences of KK-theory in the first variable
For nuclear proper Γ-C∗-algebras in the first variable, [6 Corollary A.4] identifies equivariant KK-theory with equivariant E-theory. Under this identification, [3 Proposition 5.8 and Theorem 6.18] give the following statement.
Proposition 2.1 ([3]). Let Γ be a countable discrete group, and let
0⟶JjDqQ⟶0(1)
be an equivariant extension of separable nuclear proper Γ-C∗-algebras. There is a natural boundary class δq∈KK1Γ(Q,J) such that, for every separable Γ-C∗-algebra B, the sequence
is exact, where the boundary map ∂ is determined by
∂(x)=δq⊗Jx.(2)
For a morphism between two such extensions, with induced maps ϕ:J→J′ and ψ:Q→Q′, naturality means that
δQ⊗J[ϕ]=[ψ]⊗Q′δQ′.(3)
Vector bundles and KKΓ
Throughout this subsection, all Γ-spaces are assumed to be second countable, locally compact and Hausdorff. We recall some standard facts from [10 Section 3.4]; see also Kasparov’s original paper [5].
For a proper Γ-space X, let KΓ0(X) be the Grothendieck group associated to the monoid of isomorphism classes of finite-dimensional complex Γ-vector bundles over X. Tensor product of Γ-vector bundles gives KΓ0(X) the structure of a commutative ring.
Let V→X be a Γ-equivariant complex vector bundle, equipped with a Γ-invariant Hermitian metric. The space Γ0(V) of continuous sections vanishing at infinity is a Hilbert C0(X)-module, and multiplication defines a Γ-equivariant homomorphism
ρ:C0(X)⟶K(Γ0(V)),
where K(Γ0(V)) denotes the C∗-algebra of all compact operators on the Hilbert module Γ0(V). With the trivial grading and F=0, this gives an equivariant Kasparov module in the sense of [5 Definition 2.2], and hence a class
[V]:=[(Γ0(V),ρ,0)]∈KK0Γ(C0(X),C0(X)),
where [V] denotes the class of the vector bundle V in KΓ0(X). We now have a unital ring homomorphism
where multiplication on the target is the Kasparov product.
For every Γ-C∗-algebra B, we define
ξ⋅x:=ξ⊗C0(X)x,ξ∈KΓ0(X),x∈KKnΓ(C0(X),B).(4)
By the bilinearity and associativity of the Kasparov product, this makes KKnΓ(C0(X),B) a unital KΓ0(X)-module. Extending scalars gives an action of KΓ0(X)Q on KKnΓ(C0(X),B)Q.
Proposition 2.2 ([10 Lemma 3.4.4(a) and Proposition 3.4.5]). Let Γ be a countable discrete group, X a proper Γ-space, B a separable Γ-C∗-algebra, and n∈Z. Then the following statements hold.
(1) Let Y be another proper Γ-space, and let f:Y→X be a proper Γ-equivariant continuous map. For every ξ∈KΓ0(X) and x∈KKnΓ(C0(Y),B), one has
f∗(f∗ξ⋅x)=ξ⋅f∗x,
where f∗ξ is defined by pullback of vector bundles, and
f∗:KKnΓ(C0(Y),B)⟶KKnΓ(C0(X),B)
is induced by the pullback homomorphism C0(X)→C0(Y), h↦h∘f.
(2) Let Z⊆X be a closed Γ-invariant subspace, and put U=X∖Z. Let
∂:KKnΓ(C0(U),B)⟶KKn−1Γ(C0(Z),B)
be the boundary map given by (2). For every ξ∈KΓ0(X) and x∈KKnΓ(C0(U),B), one has
∂(ξ∣U⋅x)=ξ∣Z⋅∂(x).
where ξ∣U and ξ∣Z denote the restrictions of ξ to U and Z, respectively.
Both identities extend Q-bilinearly to ξ∈KΓ0(X)Q and the corresponding rationalized KK-groups.
Proof. For (1), let [f#]∈KK0Γ(C0(X),C0(Y)) be the class of the pullback homomorphism
Finally, both identities extend Q-bilinearly to the corresponding rationalized groups. □
The idempotents supplied by the Chern character
Lück and Oliver constructed, for every proper Γ-CW-complex X, a natural equivariant Chern character
chX∗:KΓ∗(X)⟶HΓ∗(X;Q⊗R(−));
see [8 Section 5]. Here the target is Bredon cohomology, and R(−) is the coefficient system assigning the complex representation ring R(H) to Γ/H for each finite subgroup H≤Γ, with structure maps induced by restriction and conjugation. The Chern character sends KΓ0(X) to even-degree cohomology and KΓ1(X) to odd-degree cohomology. It preserves products and commutes with boundary maps in Mayer–Vietoris sequences. For finite proper Γ-CW-complexes, it becomes an isomorphism of rings after rationalization.
Proposition 2.3 ([8 Theorem 5.5]). For any finite proper Γ-CW-complex X, the equivariant Chern character extends to an isomorphism of rings
chX∗⊗Q:KΓ∗(X)Q≅HΓ∗(X;Q⊗R(−)).
This isomorphism is natural in X.
For each finite subgroup H≤Γ, set
εH:=∣H∣[CH]∈R(H)Q,
where CH is the complex left regular representation of H. It is an idempotent, i.e., εH2=εH. The following lemma is an immediate consequence of [8 Theorem 5.5].
Lemma 2.4.For each finite proper Γ-CW-complex X, there is an idempotent eX∈KΓ0(X)Q with the following properties:
(1) Under KΓ0(Γ/H)Q≅R(H)Q,
eΓ/H=εH.
(2) For every equivariant map f:X→X′ between finite proper Γ-CW-complexes,
f∗eX′=eX.
(3) If the Γ-action on X is free, then
eX=1,
where 1 is the multiplicative unit of KΓ0(X)Q.
Proof. The idempotents εH are compatible with restriction and conjugation, and hence define a natural idempotent
cX∈HΓ0(X;Q⊗R(−)).
By Proposition 2.3, there is a unique eX∈KΓ0(X)Q such that
(chX0⊗Q)(eX)=(cX,0,0,…),
where all positive-degree components are zero. Since the rational Chern character is a ring isomorphism, cX2=cX implies eX2=eX. Its naturality, together with f∗cX′=cX, gives f∗eX′=eX.
For Γ/H, the rational Chern character is the identity under the canonical identifications with R(H)Q. Thus eΓ/H=εH. If the Γ-action on X is free, all stabilizers are trivial and ε{1}=1, so cX=1 and hence eX=1. □
Proof of Theorem 1.2
Let f:X→X′ be an equivariant continuous map between proper Γ-CW-complexes. For each finite Γ-subcomplex K⊆X, choose a finite Γ-subcomplex K′⊆X′ containing f(K). The restriction f∣K:K→K′ is proper and therefore induces
(f∣K)∗:KKnΓ(K,B)⟶KKnΓ(K′,B).
These homomorphisms are compatible with support inclusions and coefficient inclusions, and hence define
f∗:KKnΓ(X,B)⟶KKnΓ(X′,B)
on the inductive limits. This construction is functorial.
We apply this construction to the maps
j:EΓ⟶EΓ×EΓ,j(z)=(z,π(z)),q:EΓ×EΓ⟶EΓ,q(z,x)=z,
p:EΓ×EΓ⟶EΓ,p(z,x)=x.
Since q∘j=id, functoriality gives q∗∘j∗=id. Thus, for every Γ-C∗-algebra A, the homomorphism
j∗:KK∗Γ(EΓ,A)⟶KK∗Γ(EΓ×EΓ,A)
is split injective, with left inverse q∗. Since π=p∘j, we also have π∗=p∗∘j∗. It therefore suffices to prove that
p∗:KK∗Γ(EΓ×EΓ,A)Q⟶KK∗Γ(EΓ,A)Q
is injective. Passing to inductive limits over finite Γ-subcomplexes of EΓ and separable Γ-invariant coefficient subalgebras, we reduce to proving that
(pX)∗:KK∗Γ(EΓ×X,B)Q⟶KK∗Γ(X,B)Q
is injective for every separable Γ-C∗-algebra B and every finite proper Γ-CW-complex X, where
pX:EΓ×X⟶X,pX(z,x)=x.
We proceed in two steps.
The orbit case. Put E=EΓ, and let X=Γ/H, where H≤Γ is a finite subgroup.
It follows from [3 Lemma 12.11] that the standard compression maps induce natural isomorphisms
CompT:KKnΓ(T×X,B)Q≅KKnH(T,B)Q,T=E,pt.
Here pt×X=X, and B on the right carries the restricted H-action.
Let
rT:KKnH(T,B)⟶KKn(T,B),T=E,pt
denote the forgetful maps.
We next define the transfer map using the induction construction in [10 Remark 3.6.8, Lemma 3.6.9 and Proposition 3.6.10]. For a finite H-subcomplex Δ⊆T, Proposition 3.6.10 of [10] gives an isomorphism
i{1}H:KKn(C(Δ),B)≅KKnH(ind{1}HC(Δ),B).
Here ind{1}HC(Δ)=C(H,C(Δ)), with H acting by left translation:
(γ⋅f)(γ′)=f(γ−1γ′),γ,γ′∈H.
Define the H-equivariant homomorphism
jΔ:C(Δ)⟶ind{1}HC(Δ),
by
jΔ(f)(γ)=γ−1(f),γ∈H.
Define
trΔ:=jΔ∗∘i{1}H:KKn(Δ,B)⟶KKnH(Δ,B).
The naturality of i{1}H and jΔ makes these maps natural for proper equivariant maps and compatible with inclusions of finite supports. Passing to the inductive limits therefore gives the transfer maps
We first compute the composites JT∘RT. By definition,
JT∘RT=CompT−1∘(trT∘rT)∘CompT.
Thus we begin with the composite trΔ∘rΔ for any finite H-subcomplex Δ of T, where rΔ is the corresponding forgetful map on Δ.
In degree zero, by [2 Proposition 20.2.4], we may represent x∈KK0H(Δ,B) by an H-equivariant Kasparov module
(E,φ,F),
where E is a countably generated graded Hilbert B-module equipped with a continuous, grading-preserving H-action, φ:C(Δ)→B(E) is an H-equivariant graded ∗-homomorphism, and F∈B(E) is a strictly H-invariant operator of degree one. For each γ∈H, let Eγ be a copy of the underlying graded vector space of E, equipped with the Hilbert B-module structure
ξ⋅γb=ξγ−1(b),⟨ξ,η⟩γ=γ(⟨ξ,η⟩).
Define the representation φγ on this module by
φγ(f)=φ(γ−1(f)),f∈C(Δ).
Equip ⨁γ∈HEγ with the H-action
(γ⋅ξ)γ′=ξγ−1γ′,γ,γ′∈H,ξ∈γ∈H⨁Eγ.
By the cycle description in [10 Lemma 3.6.9], followed by pullback along jΔ, the class (trΔ∘rΔ)(x) is represented by
γ∈H⨁Eγ,γ∈H⨁φγ,γ∈H⨁F(5)
Let {δγ:γ∈H} be the standard basis of CH. The map
U:γ∈H⨁Eγ⟶E⊗CH,U((ξγ))=γ∈H∑(γ⋅ξγ)⊗δγ,
is a B-linear unitary intertwining the representations and operators. It is also H-equivariant for the diagonal action on E⊗CH. Here H acts on CH by the left regular representation. Consequently, the cycle (5) is unitarily equivalent to
(E⊗CH,φ⊗1,F⊗1).
with the diagonal H-action. Hence
(trΔ∘rΔ)(x)=[Δ×CH]⋅x,(6)
where [Δ×CH]∈KH0(Δ) is the class of the trivial vector bundle Δ×CH equipped with the diagonal H-action. Suspension gives the same identity in every degree.
Let
V=Γ×HCH
be the Γ-equivariant vector bundle over X=Γ/H, with projection [γ,v]↦γH. Under the canonical identification KΓ0(X)≅R(H), the class [V] therefore corresponds to [CH]. By Lemma 2.4[1],
[V]=∣H∣eXin KΓ0(X)Q.
For T=pt, (6) gives
trpt∘rpt=[pt×CH]⋅−.
Since the fiber of V over H∈X is CH, the compression map is compatible with the corresponding vector-bundle actions: for every y∈KKnΓ(X,B)Q,
Let θΔ:ZΔ→X be the restriction of the second-coordinate projection E×X→X, given by
θΔ(γz,γH)=γH,γ∈Γ,z∈Δ.
Then the pullback bundle θΔ∗V has the form
θΔ∗V≅Γ×H(Δ×CH),
where H acts diagonally on Δ×CH. In particular, its restriction to Δ×{H} is the trivial bundle Δ×CH with the diagonal H-action. As in the point case, the compression map is compatible with these bundle actions:
As Δ ranges over the finite H-subcomplexes of E, the subcomplexes ZΔ are cofinal among the finite Γ-subcomplexes of E×X. The compression, forgetful and transfer maps are compatible with the homomorphisms induced by support inclusions. Consequently,
JE∘RE=∣H∣⋅id.(8)
We next show that
(pX)∗:KKnΓ(E×X,B)Q≅eXKKnΓ(X,B)Q.(9)
Since E is contractible, homotopy invariance gives an isomorphism
(ppt)∗:KKn(E,B)Q≅KKn(pt,B)Q,
where ppt:E→pt is the collapse map. The naturality of compression, forgetful, and transfer maps give
so the image of (pX)∗ lies in the eX-summand. Conversely, if y∈eXKKnΓ(X,B)Q, then
((pX)∗∘σ)(y)=eX⋅y=y.
We have therefore proved (9). The inverse of (9) is the restriction of σ to the summand on the right. Finite proper Γ-CW-complex case. We also set E=EΓ. For any finite proper Γ-CW-complex X, we claim that
(pX)∗:KKnΓ(E×X,B)Q≅eXKKnΓ(X,B)Q.(11)
For brevity, write
Fn(W):=KKnΓ(E×W,B)Q,Gn(W):=eWKKnΓ(W,B)Q
for any finite proper Γ-CW-complex W. Let x∈Fn(W) be represented by xL∈KKnΓ(L,B)Q on a finite Γ-subcomplex L⊆E×W. The restricted projection pW∣L:L→W is proper. Since L is Γ-free, Lemma 2.4(2)–(3) gives
This shows that (pW)∗ takes values in Gn(W) for any finite proper Γ-CW-complex W. We proceed by induction on dimX. If dimX=0, then
X=α=1∐rΓ/Hα.
The groups Fn(X) and Gn(X) decompose as direct sums over these orbits. Under these decompositions,
(pX)∗=α=1⨁r(pΓ/Hα)∗.
Each summand is an isomorphism by (9), so (pX)∗ is an isomorphism as well.
Suppose that m=dimX≥1, and put Z=X(m−1). Write the m-cells as Γ/Hα×Dm, and set
C:=α∐Γ/Hα.
Excision and suspension identify the relative terms for (E×X,E×Z) and (X,Z) with the corresponding groups of C, shifted by m. Thus the projections give a commutative diagram of exact sequences
After excision and suspension, compatibility of the bundle actions with the long exact sequence for (X,Z) gives the commutative diagram of exact sequences
[1]P. Antonini, S. Azzali and G. Skandalis. The Baum–Connes conjecture localised at the unit element of a discrete group, Compos. Math. 156 (2020), no. 12, 2536–2559.arxiv.org/abs/1807.05892
[2]B. Blackadar. K-Theory for Operator Algebras, second edition, Mathematical Sciences Research Institute Publications 5, Cambridge University Press, Cambridge, 1998.
[3]E. Guentner, N. Higson and J. Trout. Equivariant E-theory for C*-algebras. Mem. Amer. Math. Soc. 148 (2000), no. 703.
[5]G. G. Kasparov. Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91 (1988), no. 1, 147–201.
[6]G. G. Kasparov and G. Skandalis. Groups acting properly on “bolic” spaces and the Novikov conjecture, Ann. Math. 158 (2003), 165–206.
[7]W. Lück and B. Oliver. The completion theorem in K-theory for proper actions of a discrete group. Topology 40 (2001), no. 3, 585–616.DOI
[8]W. Lück and B. Oliver. Chern characters for the equivariant K-theory of proper G-CW-complexes, in Cohomological Methods in Homotopy Theory (Bellaterra, 1998), Progr. Math. 196, Birkhäuser, Basel, 2001, 217–247.DOI
[10]J. Raven. An equivariant bivariant Chern character. Ph.D. thesis, Pennsylvania State University, 2004.