FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY

WITH R/Z-COEFFICIENTS

PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

Institut de Mathématiques de Jussieu, (UMR 7586), Université Paris Diderot (Paris 7)

UFR de Mathématiques, CP 7012 - Bâtiment Sophie Germain
arXiv:1308.0218v1 [math.OA] 1 Aug 2013

5 rue Thomas Mann, 75205 Paris CEDEX 13, France

paolo.anton@gmail.com, azzali@math.jussieu.fr, skandalis@math.univ-paris-diderot.fr

Abstract. Let M be a closed manifold and α ∶ π1 (M ) → Un a representation. We give

a purely K-theoretic description of the associated element [α] in the K-theory of M with
R/Z-coefficients. To that end, it is convenient to describe the R/Z-K-theory as a relative
K-theory with respect to the inclusion of C in a finite von Neumann algebra B. We use
the following fact: there is, associated with α, a finite von Neumann algebra B together
with a flat bundle E → M with fibers B, such that Eα ⊗ E is canonically isomorphic
with Cn ⊗ E , where Eα denotes the flat bundle with fiber Cn associated with α. We also
discuss the spectral flow and rho type description of the pairing of the class [α] with the
K-homology class of an elliptic selfadjoint (pseudo)-differential operator D of order 1.

Contents
1. Introduction 2
2. Preliminary constructions 4
2.1. Full modules 4
2.2. Mapping cones 4
2.3. Relative K-theory 5
2.4. The bootstrap category 5
2.5. K-theory of von Neumann algebras 6
3. K-theory with coefficients 6
3.1. K-theory with rational and real coefficients 6
3.2. K-theory with Z/nZ-coefficients 7
3.3. K-theory with Q/Z-coefficients 7
3.4. K-theory with R/Z-coefficients 8
3.5. Chern–Weil theory 9
4. Atiyah’s theorem for covering spaces 11
5. [α]new in K 1 (M ; R/Z) associated with a flat bundle α∶ Γ → U (n) 13
6. The pairing with [D] ∈ K1 (M ) and the spectral flow description 15
6.1. Various notions of connections 15
6.2. The pairing as an element of K1 (Zi0 ) 15
6.3. The pairing as an element of R/Z 16
6.4. The pairing as a rho invariant 17

This research took place at the Institut de Mathématiques de Jussieu, CNRS - University Paris Diderot.
Paolo Antonini is funded by the Projet ANR KInd.
Sara Azzali is funded by an INdAM-Cofund Fellowship.
We thank these institutions for their support.
1
2 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

References 18

1. Introduction
Secondary invariants of geometric elliptic operators, such as the rho invariant of a
unitary representation α ∶ Γ → Un of the fundamental group, gain stability properties
only when reduced modulo Z. Indeed, Atiyah, Patodi and Singer in the seminal papers
[APS2, APS3] proved that the modulo Z class of the reduced rho invariant of an elliptic
selfadjoint operator D can be described as the pairing of the K-homology class [D] with a
K-theory class [α] with R/Z-coefficients associated with α. This result is called the index
theorem for flat bundles. The construction of [α] ∈ K 1 (M, R/Z) is using cohomology, and
is based on the fact that K-theory with real coefficients is isomorphic to H ∗ (M ; R).
The model of K ∗ (X, R/Z) in [APS3] is built of two addenda: a torsion part K ∗ (X, Q/Z) =
lim K ∗ (X, Z/n!Z), and a free part which is the image of K ∗ (X, R) ∶= K ∗ (X) ⊗ R; it relies
Ð→
on the functorial properties of ordinary K-theory.
Atiyah, Patodi and Singer then suggested that a direct description can be given in terms
of von Neumann algebras. This idea has been an inspiration for many authors. See in
particular [DHK, DHK2, Hu, KP, Ba].
Beyond the Atiyah–Patodi–Singer (APS) one, a number of models of the R/Z-K-theory
can be found in the literature, each one has its own features and flavor. Karoubi’s and
Lott’s models are based on Chern–Weil and Chern–Simons theory [Ka, Lo]. Basu imple-
mented the APS suggestion building a model with bundles of modules over von Neumann
algebras [Ba]. Apart from the model of Karoubi, the constructions in [Lo, Ba] are based
in a more or less explicit way on the notion of connection.
The purpose of this paper is to give a canonical construction of the R/Z-K-theory class
associated with a flat bundle using operator algebraic tools, and to compute the pairing
with K-homology as a Kasparov product. In particular, the models of K-theory with
coefficients used here are purely operator theoretic.
Note that for a C ∗ -algebra A in the bootstrap category, one can define the K-theory
of A with R-coefficients as K∗ (A; R) ∶= K∗ (A ⊗ B), where B is any II1 -factor: in fact,
by the Künneth property, the group K∗ (A ⊗ B) is independent of B up to a canonical
isomorphism - and coincides with the APS model for A commutative.
The model for the K-theory with R/Z-coefficients that we use is simply the ordinary
relative K-theory of the inclusion A ↪ A ⊗ B or, equivalently, the group K∗−1 (A ⊗ Ci0 )
where Ci0 is the mapping cone of the unital inclusion i0 ∶ C ↪ B for any II1 -factor B. This
is immediately shown to be well defined for any C ∗ -algebra in the bootstrap category. Let
us mention that this mapping cone has been used recently by Deeley to construct a model
of K-homology with R/Z-coefficients [De].
To show that all the models are equivalent under canonical isomorphisms, we use also
operator algebraic definitions of the K-theory with coefficients in Q, Z/nZ and Q/Z. Our
models are direct noncommutative generalizations of the APS ones: we have the inclusion
of C in n × n-matrices - and inductive limits (UHF algebras), whereas the APS definition
employs the K-theory of the mapping cones of maps of degree n on the circle - and
projective limits.
Let M be a closed manifold with fundamental group Γ. If Eα is a flat vector bundle over
M with holonomy α∶ Γ → U (n), Atiyah, Patodi and Singer’s class [α]AP S ∈ K 1 (M ; R/Z)
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 3

is based on the idea that Eα defines a torsion element in the reduced K-theory of M , in
fact there exists k ∈ N∗ such that the sum of k copies of Eα is trivial. The class of Eα in
K-theory with Z/kZ-coefficients is then added to the transgression form corresponding to
the two flat connections on the trivial bundle with fiber Cnk thus obtained.
Our main result is the direct description of the element [α] ∈ K 1 (M ; R/Z) associated
with α: this is given by the pair of bundles Eα , Cn with the (almost) canonical isomorphism
after tensoring with a II1 -factor B. Our construction consists of the following points:
(1) there is a canonical flat bundle E with fiber a II1 -factor B associated with a mor-
phism u ∶ Γ → U (B) (where U is the unitary group of B);
(2) there is a canonical isomorphism Eα ⊗ E Ð→ Cn ⊗ E;
(3) the bundle E is trivial in K-theory and may actually be assumed to be trivial.
The II1 -factor B can be taken to be L∞ (Un ) ⋊ Γ - i.e. the II1 von Neumann algebra
Morita equivalent to the foliated flat bundle M̃ ×α Un - and the bundle E encodes the
bundle of frames for Eα .
The canonical isomorphism in (2) derives from the fact that a vector bundle becomes
trivial when it is lifted to the bundle of frames.
The point (3) is based on the fact that a flat bundle with fibers a II1 -factor is trivial
in K-theory. This is easily seen applying Atiyah’s L2 -index theorem for covering spaces
[Ati] (and its generalizations [Lü, Sc] by Lück and Schick to every trace on C ∗ (Γ) (1)),
together with the property that K-homology separates points of K ∗ (M ; R). Conversely,
Chern–Weil theory can be used to prove the same result, and this in turn gives a different
proof of the L2 -index theorem and its generalizations.
We then establish the independence from all choices involved in the construction of [α],
and further show - using a Chern–Simons transgression argument with coefficients in von
Neumann algebras - that our element is the same as the one constructed by Atiyah, Patodi
and Singer.
We finally pair the K-theory class [α] with a K-homology class [D] ∈ K∗ (M ) repre-
sented by a first order elliptic operator. The pairing is realized as a Kasparov product
KK 0 (C, C(M ) ⊗ Ci0 ) × KK 1 (C(M ), C) Ð→ KK 1 (C, Ci0 ) = K1 (Ci0 ) = R/ Z .
To compute the intersection product [α] ⊗ [D], it is convenient to look at [α] in
KK(C, Zi0 ⊗ C(M )), the K-theory of the double cylinder of the inclusion C(M ) ↪
C(M ) ⊗ B. This provides a path which interpolates between the operator D α obtained
as a twisted tensor product of D by the flat bundle Eα , and the operator D n obtained by
tensoring D by the trivial bundle of rank n = rk(Eα ). The KK-product puts D α and D n
as affiliated to the same II∞ factor with (relatively) compact resolvent, that are bounded
perturbations of each other since they have the same principal symbol. Furthermore, they
have discrete spectrum and the corresponding spectral projections have integer von Neu-
mann dimension since they are obtained by tensoring with B the Dirac operators D α and
Dn .
The pairing, as an element of R/Z, is given by a type II spectral flow: for every a ∈ R,
there is a well defined index between the two projections χ[a,∞[ (D α ) and χ[a,∞[ (D n ) whose
difference is relatively compact; this index does not depend on a up to Z.

1As a side remark, we note that this result remains true for any trace on ℓ1 (Γ) by showing that Atiyah’s
method extends.
4 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

Finally, since the spectral flow of Dirac type operators is related through variational
formulas to the eta invariant (see [BF, BGV, MZ2, CP1]), we get the index theorem for
flat bundles [α] ⊗ [D] = ξ(D α ) − ξ(D n ) ∈ R/Z.

2. Preliminary constructions
We start with some well known constructions for C ∗-algebras that we shall use in the
following.
2.1. Full modules. Let A be a unital C ∗-algebra (or just a ring). A (right) A-module
F is said to be full if span{ℓ(x), x ∈ F, ℓ ∈ L(F, A)} = A (where L(F, A)} denotes
the space of A-linear maps from F to A). If A is a non unital C ∗ -algebra, F is full if
span{ℓ(x), x ∈ F, ℓ ∈ L(F, A)} is dense in A.
Lemma 2.1. Let A be unital and F be a finitely generated projective full module over A.
Let E, E ′ be finitely generated projective modules over A. Then [E] = [E ′ ] in K0 (A) if
and only if ∃k ∈ N ∶ E ⊕ F k ≃ E ′ ⊕ F k .
Proof. By hypothesis there exists a finitely generated projective G such that E⊕G ≃ E ′ ⊕G.
Write 1 = ∑ ℓi (xi ) with ℓ1 , . . . ℓn ∈ F ∗ and xi ∈ F ; thus we construct the module map
f ∶ Gn → A , (y1 , . . . , yn ) ↦ ∑i ℓi (yi ) which is onto. Since G is finitely generated, there is a
onto module map g ∶ Am → G; we deduce a surjective module map h∶ F nm → G.
Then it also holds F nm ≃ G ⊕ G′ (projectivity of G), and then E ⊕ F nm ≃ E ′ ⊕ F nm .
The converse is obvious. 
Lemma 2.2. Let A be a unital C ∗ -algebra, and E1 and E2 be finitely generated projective
and full A-modules. Suppose [E1 ] = [E2 ] in K0 (A). Then there exists n such that E1n ≃ E2n .
Proof. Since E1 and E2 are full, there exists k such that E1k ⊕ E1 ≃ E1k ⊕ E2 and E1 ⊕ E2k ≃
E2 ⊕ E2k .
Then E1a+1 ⊕ E2b ≃ E1a ⊕ E2b+1 , as soon as a ≥ k or b ≥ k. We obtain E12k ≃ E22k . 
2.2. Mapping cones. The mapping cone of a morphism ϕ∶ A → B of (unital) C ∗ -algebras,
is the algebra Cϕ = {(a, γ) ∈ A ⊕ C0 ((0, 1], B) ∣ γ(1) = ϕ(a)}.
Let now A0 , A1 , B be unital C ∗ -algebras.
Definition 2.3. Given two morphisms ϕi ∶ Ai → B, i = 0, 1, define the double cylinder
algebra as
Zϕ0 ,ϕ1 = {(a0 , σ, a1 ) with ai ∈ Ai , σ ∈ C([0, 1], B) s.t. σ(i) = ϕi (ai ), i = 0, 1} (2.1)
When ϕ0 = ϕ1 = ϕ we will denote the double cylinder Zϕ0 ,ϕ1 by Zϕ .
Note that for A0 = 0 this construction is the cone Cϕ1 . When A0 = B and ϕ0 is the
identity of B, the double cylinder is sometimes called the mapping cylinder of ϕ1 .
For the double cylinder Zϕ there is a split exact sequence
s
0 / Cϕ / Zϕ t /A /0 (2.2)
r π1

with r(a, σ) = (0, σ, a) and π1 (a0 , σ, a1 ) = a1 , and the splitting given by s(a) = (a, ϕ(a), a)
so that K∗ (Zϕ ) ≃ K∗ (Cϕ ) ⊕ K∗ (A). In particular, there is a map from K0 (Zϕ ) to the
summand K0 (Cϕ ) of the form x ↦ x − (s ○ π1 )∗ x.
Note that if ϕ is injective, then
Cϕ = {γ ∈ C0 ((0, 1], B) ∣ γ(1) ∈ ϕ(A)} and Zϕ = {γ ∈ C([0, 1], B) ∣ γ(0), γ(1) ∈ ϕ(A)}.
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 5

If B is a II1 -factor, the inclusion i0 ∶ C → B gives K0 (Ci0 ) = 0 and K1 (Zi0 ) = K1 (Ci0 ) =

R/Z.
2.3. Relative K-theory. Although relative K-theory coincides with the K-theory of a
mapping cone, it helps in giving a more direct description of some K-theory elements. For
instance, it has been recently used by Deeley to construct a model of K-homology with
R/Z-coefficients [De, Prop. 2.2, Ex. 5.3].
Let ϕ ∶ A → B be a homomorphism of unital C ∗ -algebras. The group K0 (ϕ) is given by
generators and relations:
● Its generators are triples (E+ , E− , u) where E+ , E− are finitely generated projective
A-modules, and u∶ E+ ⊗A B → E− ⊗A B is an isomorphism.
● Addition is given by direct sums.
● A homotopy is a triple associated with the map C([0, 1]; A) → C([0, 1]; B) induced
by ϕ.
● Trivial elements are those triples (E+ , E− , u) for which u = ϕ(v) for some isomor-
phism v∶ E+ → E− . The group K0 (ϕ) is formed as the set of those triples divided
homotopy and addition of trivial triples.
● For non unital algebras, we just put K0 (ϕ) = K0 (ϕ̃), where ϕ̃ ∶ Ã → B̃ is obtained
by adjoining units everywhere.
● The group K1 (ϕ) is also formed in this way. Its generators are pairs (u, f ) where
u is a unitary in Mn (A) and f is a continuous path in the unitaries of Mn (B)
joining ϕ(u) to 1n .
● There is a natural isomorphism K∗ (ϕ) ≃ K∗ (Cϕ ). This morphism is almost tauto-
logical for K1 . For K0 , it is described as follows. Assume that ϕ is unital and use
the isomorphism K∗ (Zϕ ) = K∗ (Cϕ ) ⊕ K∗ (A) associated the split exact sequence
(2.2). The class of a triple (E+ , E− , U ) ∈ K0 (ϕ) is then the image in the summand
K0 (Cϕ ) of the Zϕ -module
FU = {(x+ ; f ; x− ) ∈ E+ × C ([0, 1]; E+ ⊗A B) × E− ; f (0) = x+ ⊗ 1 and U f (1) = x− ⊗ 1}.
Let e+ , e− ∈ Mn (A) be projections such that E+ = e+ An , E− = e− An and u ∈
Mn (B) such that u∗ u = ϕ(e+ ), uu∗ = ϕ(e− ) and satisfies ux = U (x) for x ∈ e+ An =
E+ . Let f, g ∈ C([0, 1]) be given by f (t) = cos tπ
2 and g(t) = sin 2 . Then we can
tπ

write FU = eZϕ where

2n

e+ 0 f 2 ϕ(e+ ) f gu∗
e = (( ),( ),( )) ∈ M2n (Zϕ ).
0 0
0 0 f gu g ϕ(e− )
2 0 e−
2.4. The bootstrap category. The bootstrap category is the smallest class N of sepa-
rable nuclear C ∗-algebras containing commutative ones and closed under KK-equivalence
[Bl, 22.3.4].
Every C ∗-algebra A in the bootstrap category satisfies the Künneth formula for tensor
products [Bl, Theorem 23.1.3]; i.e. for every C ∗-algebra B
0 / K● (A) ⊗ K● (B) α / K● (A ⊗ B) σ / TorZ (K● (A), K● (B)) /0 (2.3)
1

where we denote K● = K0 ⊕ K1 as a graded group. Here α has degree 0, and σ has degree
1. If one of K● (A) and K● (B) is torsion free, then α is an isomorphism.
Remark 2.4. If A is in the bootstrap category and B is a von Neumann algebra, then
K1 (B) = 0 and K0 (B) is torsion free, see section 2.5. Therefore K∗ (A) ⊗ K0 (B) ≃ K∗ (A ⊗
B).
6 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

2.5. K-theory of von Neumann algebras. Any von Neumann algebra A is uniquely
decomposed in a product
A = AIf × AI∞ × AII1 × AII∞ × AIII ,
consequently the K-groups uniquely split
K∗ (A) = K∗ (AIf ) × K∗ (AI∞ ) × K∗ (AII1 ) × K∗ (AII∞ ) × K∗ (AIII ).
It is easy to see that if A is properly infinite then K0 (A) = 0 while it is always the case
that K1 (A) = 0. It follows that the K theory is reduced to that of the finite piece
K0 (A) = K0 (AIf ) ⊕ K0 (AII1 ).
If A is finite, there is a unique center valued trace truA ∶ A Ð→ Z(A) with the property
that two projections p, q ∈ Mn (A) are equivalent if and only if truA (p) = truA (q). Then it
follows from the universal property of K0 that truA defined on projections extends to a von
Neumann center valued dimension which is an injection
dimuA ∶ K0 (A) Ð→ Z(A)sa ∶= {a ∈ Z(A), a = a∗ }.
If A is type II1 , this is an isomorphism.

3. K-theory with coefficients

A model for the K-theory of a C ∗ -algebra A with coefficients in a countable abelian
group Λ is K∗ (A; Λ) ∶= K∗ (A ⊗ BΛ ) where BΛ is a C ∗ -algebra in the bootstrap category
with a specified isomorphism K0 (BΛ ) ≃ Λ and such that K1 (BΛ ) = 0. If Λ is uncountable,
such a model can be provided using (uncountable) inductive limits.
In this section, we describe K-theory with R/Z-coefficients in terms of von Neumann
algebras. The two models coincide for C ∗ -algebras in the bootstrap category. This von
Neumann description is suitable for the construction of the K-theory class of a flat bundle
given in section 5.
In order to relate our model with the one of Atiyah, Patodi and Singer [APS3, Sec.
5], we briefly discuss models for K-theory with coefficients in Q, R, Z/nZ and Q/Z and
compare them with the respective versions in [APS2].
3.1. K-theory with rational and real coefficients. For a (locally) compact space X,
S (X, Q) ∶= K (X) ⊗ Q and KAP S (X, R) ∶= K (X) ⊗
∗ ∗ ∗ ∗
Atiyah–Patodi–Singer’s model is KAP
R. More generally, we may set K∗AP S (A, Q) ∶= K∗ (A) ⊗ Q and K∗AP S (A, R) ∶= K∗ (A) ⊗ R
for any C ∗ -algebra A.
Our description of K-theory with rational coefficients is the following.
Definition 3.1. Let D be the universal UHF algebra, i.e. the one which satisfies K0 (D) =
Q. It is obtained as an inductive limit of matrices, D = ⊗ Mn (C). Define
n∈N
K∗ (A, Q) ∶= K∗ (A ⊗ D) (3.1)
Since D is in the bootstrap category, K1 (D) = 0 and K0 (D) = Q, the Künneth formula
gives a natural isomorphism K∗ (A) ⊗ Q → K∗ (A ⊗ D), and therefore our K-theory with
rational coefficients coincides with K∗AP S (A, Q).
In order to describe K-theory with real coefficients, we use the following result:
Lemma 3.2. Let B be a II1 -factor, and A be a C ∗-algebra in the bootstrap category. Then
the group K∗ (A ⊗ B) is canonically isomorphic to K∗ (A) ⊗ R, and therefore it does not
depend on B up to a canonical isomorphism.
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 7

Proof. The Künneth isomorphism

K0 (A) ⊗ R / K0 (A ⊗ B)
∼
(3.2)
is given by the canonical map [x] ⊗ t ↦ [x ⊗ pt ]. 
In particular, if B1 is a II1 -subfactor of B2 , the induced map K∗ (A⊗B1 ) Ð→ K∗ (A⊗B2 )
is a canonical isomorphism, independent of the inclusion B1 ⊂ B2 . In the following, B will
always denote a II1 -factor. Then for A in the bootstrap category we can define, indepen-
dently of B a model for K-theory with coefficients in R, which is naturally isomorphic
with the APS one.
Definition 3.3. Let A be in the bootstrap category.
K∗ (A, R) ∶= K∗ (A ⊗ B) (3.3)
3.2. K-theory with Z/nZ-coefficients. Let in ∶ C ↪ Mn (C) be defined by in (1) = Id.
Definition 3.4.
K∗ (A, Z/nZ) ∶= K∗−1 (A ⊗ Cin ) (3.4)
Let us relate it with the K-theory with Z/nZ-coefficients as defined by Atiyah–Patodi–
Singer in [APS2, Sec. 5].
Let fn ∶ S 1 → S 1 given by fn (z) = z n . The APS definition of the K-theory with Z/nZ-
coefficients for a closed (locally) compact space X is the relative K-theory of fn × idX ∶
S 1 × X → S 1 × X, [APS2, (5.2)].
Define fn∗ ∶ C(S 1 ) → C(S 1 ) by fn∗ (ϕ)(z) = (ϕ ○ fn )(z) = ϕ(z n ).
The K-theory of the cone algebras Cin and Cfn∗ are computed by means of the exact
sequences of the cones and give K1 (Cin ) ≃ K0 (Cfn∗ ) ≃ Z/nZ, and K0 (Cin ) ≃ K1 (Cfn∗ ) = 0.
Note that the inclusion SMn (C) → Cin induces the quotient map Z → Z/nZ in K1 ,
and therefore the isomorphism K1 (Cin ) ≃ Z/nZ is canonical; in the same way, the iso-
morphism K0 (Cfn∗ ) ≃ Z/nZ is canonical, and there is therefore a canonical isomorphism
ψ ∶ K1 (Cin ) → K0 (Cfn∗ ). By the UCT [Bl, Prop. 23.10.1], KK 1 (Cin , Cfn∗ ) is isomor-
phic to Hom(Z/nZ, Z/nZ). The isomorphism ψ determines therefore a generator of
KK 1 (Cin , Cfn∗ ).
It follows that Cin and Cfn∗ are naturally KK 1 -equivalent, or, equivalently, the algebras
SCin and Cfn∗ are canonically KK-equivalent.
It is of course possible to give an explicit construction of the canonical element of
KK 1 (Cin , Cfn∗ ).
As a consequence, let A be any separable C ∗ -algebra. Denote now with in,A ∶ A ↪
A⊗Mn (C), fn,A ∗
∶ C(S 1 , A) → C(S 1 , A) the analogous maps as above. Then Cin,A = A⊗Cin
and Cfn,A∗ = Cfn∗ ⊗ A are canonically KK 1 -equivalent.
S (X, Z/nZ) ≃ K∗+1 (C0 (X) ⊗ Cin ) .
∗
In particular, for a (locally) compact space X, KAP
In other words, the definition 3.4 of K-theory of Z/nZ coefficients coincides (for abelian
C ∗ -algebras) with the one given in [APS2].

3.3. K-theory with Q/Z-coefficients.

Definition 3.5. Let A be in the bootstrap category. Let D be the universal UHF algebra,
and A ↪ A ⊗ D the obvious inclusion, define
K∗ (A, Q/Z) ∶= K∗−1 (CA↪A⊗D ) (3.5)
8 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

S (X, Q/Z) is defined as the inductive

∗
Remark 3.6. For a (locally) compact space X, KAP
limit of KAP S (X, Z/n!Z) [APS2, (5.3)]. More generally, for every C ∗ -algebra A, we may
∗

set K∗AP S (A, Q/Z) ∶= lim K∗AP S (A, Z/n!Z). Observe that the construction in APS can be
Ð→n
made using the projective limit. If

T = lim .... S 1
×n ×(n+1)
/ S1 / S 1 ...
←Ð
then K∗AP S (A, Q/Z) is the relative K-theory of A ⊗ C(S 1 ) → A ⊗ C(T ). Note also that
C(T ) = C ∗ (Q).
We have CA↪A⊗D = lim CA↪A⊗Mn! (C) . As K∗ (A, Z/nZ) ≃ K∗AP S (A, Z/nZ), using con-
Ð→n
tinuity of K-theory, we find K∗ (A, Q/Z) ≃ K∗AP S (A, Q/Z).
3.4. K-theory with R/Z-coefficients.
Definition 3.7. We propose the following realization of R/Z-K-theory of a C ∗ -algebra A
in the bootstrap category.
K∗ (A, R/Z) ∶= K∗+1 (Cone(A ↪ A ⊗ B)),
where B is any II1 -factor. Note that Definitions 3.3 and 3.7 are well posed: first of all
since A is nuclear there is no ambiguity in the C ∗-tensor product A ⊗ B. Furthermore the
right hand side is in both definitions independent, up to natural canonical isomorphism,
of the factor B.
The latter definition is indeed a realization of R/Z-K-theory. The Bochstein change of
coefficients is the long exact sequence associated to the mapping cone C ∗-exact sequence.
Atiyah, Patodi and Singer’s description of R/Z-K-theory goes back to an idea of Segal
[APS3, Sec. 5]. It is made up of two addenda. The first component is the torsion part,
contained in K ∗ (X, Q/Z). The free part is in K ∗ (X, R) = K ∗ (X) ⊗ R. More precisely
S (X, R/Z) ∶= cokernel{(p, −j) ∶ K (X, Q) Ð→ K (X, Q/Z) ⊕ K (X, R)}
∗ ∗ ∗ ∗
KAP
where p is the natural projection and j is the natural injection.
We may of course put
K∗AP S (A, R/Z) ∶= cokernel{(p, −j) ∶ K∗ (A, Q) Ð→ K∗ (A, Q/Z) ⊕ K∗ (A, R)}
Finally we prove:
Proposition 3.8. For an algebra A in the bootstrap category, our realization of K∗ (A, R/Z)
coincides with K∗AP S (A, R/Z).
Proof. Let D be the universal UHF algebra, B a II1 -factor, and denote with iD ∶ C ↪ D
and iB ∶ C ↪ B the obvious inclusions. It is enough to prove that
K∗ (A ⊗ CiB ) = Coker {(p, −j)∶ K∗+1 (A ⊗ D) Ð→ K∗ (A ⊗ CiD ) ⊕ K∗+1 (A ⊗ B)} (3.6)
where p is induced from the natural map (C(M ) ⊗ D)(]0, 1[) Ð→ CiD and j is induced
from any unital inclusion D ↪ B. To verify (3.6), consider the diagram

A ⊗ VD / A ⊗ Ci
p
D

A ⊗ VB / A ⊗ Ci
 
B
h
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 9

where VB = {(a, s) ∈ C0 (] − 1, 0]) × B([0, 1[) s.t. iB (a(0)) = r(0)} and VD is defined in
analogous way. Note that the inclusion A ⊗ SB → A ⊗ VB induces an isomorphism in
K-theory as follows from the exact sequence 0 → (A ⊗ B)(]0, 1[) → VB → A(]−1, 0]) → 0.
Moreover h is surjective. Therefore, we have an associated Mayer-Vietoris exact sequence.
Now j∗ ∶ K∗ (A ⊗ VD ) ≃ K∗+1 (A ⊗ D) Ð→ K∗+1 (A ⊗ B) ≃ K∗ (A ⊗ VD ) is injective.
Finally the Mayer–Vietoris sequence gives
0 Ð→ K∗ (A ⊗ D) Ð→ K∗+1 (A ⊗ CiD ) ⊕ K∗ (A ⊗ B) Ð→ K∗+1 (A ⊗ CiB ) Ð→ 0. 
Remark 3.9. There are in the literature several other realizations of R/Z-K-theory:
Karoubi’s desription: is a cohomological interpretation based on Chern–Weil and
Chern–Simons theory [Ka2] - this will be explained below. At first sight, it works
only for manifolds but as pointed out by Atiyah Patodi and Singer it can be
generalized to compact spaces by an embedding trick. Cycles for K 1 (X, R/Z) are
triples ((E, ∇E ), (F, ∇F ), ω) where E, F are Hermitian bundles with Hermitian
connections ∇E , ∇F and ω is a transgression odd degree differential form such that
dω = Ch(∇E ) − Ch(∇F ).
A suitable notion of sum and equivalence of cycles is defined leading to an abelian
group which is a realization of K 1 (X, R/Z).
Basu’s description: is a bundle theoretic relative description [Ba]. It can be thought
of as a realization of the suggestion of Atiyah, Patodi and Singer to describe
K 1 (X, R/Z) with bundles of von Neumann algebras. Cycles are couples of finite
rank vector bundles E, F such that there exists a von Neumann bundle V (relative
to a semi-finite or II∞ von Neumann algebra) such that E ⊗ V ≃ F ⊗ V . Addition
of cycles is defined by direct sum and there is a natural equivalence relation. The
equivalence classes form a group which is a realization of K 1 (X, R/Z).
3.5. Chern–Weil theory. Let M be a closed manifold. Then the group K∗ (C(M ), R)
S (M, R) ∶= K (M ) ⊗ R as defined in [APS2]. This
∗ ∗
as defined above is isomorphic to KAP
is in turn isomorphic to H (M ; R) using Chern–Weil theory.
∗

Let A be a unital C ∗ -algebra. Let W Ð→ M be a smooth bundle of finitely generated

Hilbert A-modules over the manifold M : using the Serre–Swann theorem, this is a finitely
generated projective module over C ∞ (M ; A), endowed with a nondegenerate C ∞ (M )
valued scalar product. We refer to [MF] and [Sc, Sections 2,3] for the general theory of
such bundles. A connection ∇ is extended as usual to forms
∇(ω ⊗ s) = dω ⊗ s + (−1)deg ω ω∇s, ω ∈ Ω∗ (M ), s ∈ Γ(W ).
The curvature ∇2 is a 2-form with values endomorphisms. If the connection is metric, the
curvature is skew–adjoint.
3.5.1. Chern–Weil theory for K0 . If Ω = ∇2 ∈ Ω2 (M ; EndA (W )) is the curvature, the
exponential
Ω∧⋅⋅⋅∧Ω
exp ( ) ∶= ∑ ∈ Ω2∗ (M ; EndA (W )) ≃ Ω2∗ (M ) ⊗C ∞ (M ) EndA (W )
Ω
2iπ k (2iπ)k k!

is well defined as a finite sum.

A trace τ ∶ A Ð→ C extends to Mn (A) and therefore to EndA (E) ≃ pMn (A)p for
every finitely generated projective A-module E ≃ pAn (where p ∈ Mn (A) is a projection).
If W Ð→ M is a finitely generated projective A-module bundle, we thus obtain a well
10 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

defined trace, which is a C(M )-linear map still denoted τ ∶ EndA (W ) → C(M ). The
Chern character of W associated with τ is defined as follows. Put

Ch τ (Ω) ∶= (id ⊗τ )( exp ( )) ∈ Ω2∗ (M ; C).

Ω
2iπ
From the trace property it follows that the Chern form is closed. Its cohomology class
does not depend on the connection. If τ is self adjoint, this class is seen to be real by
taking ∇ metric.
Hilbert A-module bundles define the K-theory of a compact space with coefficients
in A. Indeed K(M ; A) is the K-group of the category of finitely generated projective
Hilbert A-module bundles. The Serre–Swann functor sending a bundle to the module of
its continuous functions establishes a canonical isomorphism
K 0 (M ; A) ≃ K0 (C(M ; A)) ≃ K0 (C(M ) ⊗ A).
If A is a II1 -factor, from the commutativity of the diagram

K 0 (M ) ⊗ K0 (A) / K 0 (M ; A) ≃ K0 (C(M ) ⊗ A)
≃
(3.7)
✐✐✐✐
Ch ⊗τ ✐✐ ✐✐✐✐✐
✐✐ Chτ
 t ✐✐✐
H 2∗ (M ; R)
✐

one gets that Chτ is an isomorphism.

3.5.2. Chern–Weil theory for K1 . The odd K-theory with coefficients in A is defined by
K 1 (M ; A) ∶= Kc0 (M × R; A) ≃ K0 (C(M ) ⊗ C0 (R) ⊗ A) = K0 (C(M ) ⊗ SA)
where SA = C0 (R; A) is the suspension of A.
We then write K 1 (M ; A) × K 0 (M ; A) = K 0 (M × S 1 ; A). Also H 2∗ (M × S 1 , C) ≃
H (M, C). The map H 2k (M × S 1 , C) → H 2k−1 (M, C) is obtained by integration along
∗

the S 1 fibers. Using these identifications, if τ is a trace on A, we have a commuting

diagram:
K 0 (M × S 1 ; A) / H 2∗ (M × S 1 , C)
Chτ

 
K 1 (M ; A) / H 2∗+1 (M ; C)
Chod
τ

which defines the odd Chod τ in the bottom line. In particular, if A is a II1 -factor, then Chτ
is an isomorphism.
Let W Ð→ S 1 × M be a smooth bundle of finitely generated Hilbert A-modules over
S 1 × M and ∇ a connection on W . Write S 1 = R/Z and use ∇ in order to trivialize
the bundle W along [0, 1]. We thus obtain a constant bundle W0 over M and a family
(∇u )u∈[0,1] of connections on W0 . In other words, we may write (over M × [0, 1]),

∇ = (∇u ) + du
∂
⋅
∂u
Then ∇2 = ∇
˙ u ∧ du + ∇2u , whence:
∇2u
τ (W, ∇) = (2πi) ∫ τ (∇
˙ u exp ( )) du ∈ Ω2∗+1 (M ; C).
1
−1
Chod
0 2πi
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 11

Alternatively, an element of K 1 (M ; A) is represented by a continuous (smooth) map

α ∶ M Ð→ Un (A). The corresponding bundle over M × S 1 is obtained by gluing the trivial
bundle M × [0, 1] × An by means of α.
One then may compute Chτ for this bundle and get ([G, Prop. 1.2])
∞
Ch τ (α) = ∑ (−1)k (id ⊗τ )(α−1 dα)2k+1 ∈ Ω2∗+1 (M ; C) .
k!
(2k + 1)!(2πi)k+1
(3.8)
k=0

3.5.3. Chern–Simons class. Let A a C ∗ -algebra and τ ∶ A → C a continuous trace. For a

bundle E of finitely generated projective Hilbert A-modules equipped with two connections
∇0 , ∇1 , the odd Chern–Simons differential form is
∇2u
CSτ (∇0 , ∇1 ) ∶= (2πi)−1 ∫ τ (∇
˙ u exp ( )) du ∈ Ω2∗+1 (M ; C)
1
(3.9)
0 2πi
where ∇u ∶= (1 − u)∇0 + u∇1 . Recall that d(CSτ (∇0 , ∇1 )) = Chτ (∇0 ) − Chτ (∇1 ).
Definition 3.10. A flat Hilbert A-module bundle is a pair (W, ∇) where W is a Hilbert
A-module bundle, and ∇ is a flat connection.
A flat bundle isomorphism between two flat Hilbert A-module bundles (W, ∇W ) and
(V, ∇V ) is a bundle isomorphism ψ∶ W Ð→ V which preserves the connection.

We fix the following notation, used repeatedly in the following. If (W, ∇W ) and (V, ∇V )
are flat Hilbert A-module bundles and α∶ W Ð→ V is a isomorphism which does not
necessarely preserve the connection, we use (3.9) to define the closed form

CSτ (α∗ ∇V , ∇W ) ∈ Ωodd (M ; C) . (3.10)

From the above discussion we immediately get
Proposition 3.11. If α ∶ V → V is an automorphism of a flat Hilbert A-module bundle,
then α defines a class in K1 (C(M ) ⊗ A) and the classes Chτ (α) and CSτ (α∗ ∇V , ∇V ) in
H ∗ (M ; C) coincide. ◻

4. Atiyah’s theorem for covering spaces

A key ingredient in our construction of the element [α] ∈ K 1 (M ; R/Z) is the fact that
every flat bundle on M with fibers a (finite) von Neumann algebra is trivial (in K-theory).
In fact this statement is equivalent to Atiyah’s theorem on covering spaces [Ati] (more
precisely, to the versions of Lück and Schick in [Lü, Sc]). See also remark 4.3 below for a
further generalization.
Let M be a closed manifold with π1 (M ) = Γ. For an elliptic operator D acting on the
sections of a bundle S Ð→ M , let D̃ be the lift on the universal covering. Recall that the
index class
Ind D̃ ∈ K∗ (C ∗ Γ)
is the KK-product [V]⊗C(M ) [D], where V = M̃ ×Γ C ∗ Γ is the so called Mishchenko bundle.
Indeed, under the identification of the sections of V ⊗ S with the Γ-invariant sections of
the lifted bundle π ∗ (V ⊗ S) = (C ∗ Γ ⊗ π ∗ (S)) × M̃ , one easily sees that D̃ is a connection
12 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

and a C ∗ Γ-Fredholm operator. Its index is the Connes–Moscovici index class given by the
idempotent ( ) + R for
0 0
0 I
S02 S0 (I + S0 )Q̃
R ∶= ( )
S1 D̃ −S12
where Q̃ is any almost local parametrix of D̃ and S0 = I − Q̃D̃, S1 = I − D̃Q̃ are smoothing.
Let now B a II1 -factor. A morphism u ∶ Γ → U(B) is a representation of Γ and therefore
extends to a ∗-morphism ū ∶ C ∗ Γ → B. Denote by Eu the bundle Eu ∶= M̃ ×u B over M with
fiber B, let Eu denote the corresponding (finitely generated projective right) (C(M ) ⊗ B)-
module of sections of Eu , and [Eu ] its class in K0 (C(M ) ⊗ B).
The von Neumann index of D (in K0 (B)) is
Ind D̃B = ū(Ind D̃) = [Eu ] ⊗C(M ) [D] (4.1)
Lemma 4.1. Eu is the trivial element in K0 (C(M ) ⊗ B).
Proof. The Chern character ch ∶ K0 (C(M ) ⊗ B) Ð→ H ev (M, R) is an isomorphism, as
recalled in Section 3.5. Then the statement follows since Eu is a flat bundle. 
Remark 4.2. ● Atiyah’s L2 -index theorem (Lück’s and Schick version [Lü, Sc]) fol-
lows immediately from Lemma 4.1. In fact, given any trace τ on C ∗ (Γ), we may
construct a II1 -factor B with a trace-preserving embedding C ∗ (Γ) → B. It imme-
diately follows from Lemma 4.1 and formula (4.1) that Ind D̃B = Ind D.
● Conversely, Lemma 4.1 follows from Atiyah’s L2 -index theorem (Lück’s and Schick
version [Lü, Sc]). Indeed, for any trace τ on B, τ (Ind B D̃) = τ (Ind D ⋅ [1]),
and the result follows from the fact that the pairing map K 0 (C(M )) ⊗ R Ð→
Hom(K 0 (M ), R), D z→ ⟨⋅, D⟩ is surjective since K 0 (M ) is finitely generated.
● It would be interesting to find a K-theoretic proof of Lemma 4.1 with the idea:
find a bigger II1 -factor B1 , i.e. a II1 -factor containing B, such that the bundle
Eu ⊗C(M )⊗B C(M ) ⊗ B1 becomes explicitly trivial.
Remark 4.3. (Atiyah’s theorem for continuous traces on ℓ1 (Γ)). The symmetric index of
an elliptic (pseudo)-differential operator D is well defined as an element indℓ1 (Γ) (D̃) of
K0 (ℓ1 (Γ)) (see [CM]).
A continuous linear form on ℓ1 (Γ) is of the form a ↦ ∑g∈Γ ag f (g) for some f ∈ ℓ∞ (Γ).
The trace property holds if and only if f is constant on conjugacy classes. Let τ be a trace
associated with f ∈ ℓ∞ (Γ).
Now τ (indℓ1 (Γ) (D̃)) = Tτ (R) where Tτ is defined as follows: let first p ∶ M̃ → M be the
covering map. Note that Γ acts freely and properly on M̃ and M = Γ/M̃ i.e. the fibers of
p are the orbits of Γ.
Let G = Γ/M̃ 2 - where Γ acts diagonally on M̃ ×M̃ . Then G is a Lie groupoid and Cc∞ (G)
is the algebra of Γ-invariant smooth kernels on M̃ with a bounded support condition. Note
that R ∈ Mn (Cc∞ (G)).
Let ϕ ∈ Cc∞ (G). Let x ∈ M̃ . Let ϕx ∶ Γ → C be defined by ϕx (g) = ϕ(gx, x). For h ∈ Γ
and g ∈ Γ, we have ϕ(ghx, hx) = ϕ(h−1 ghx, x) and therefore τ (ϕx ) = τ (ϕhx ). We then put
Tτ (ϕ) = ∫M τ (ϕx ) dx.
Now, if D̃ and Q̃ are taken local enough, we have R(x, gx) = 0 for all x ∈ M̃ and g ≠ 1,
and therefore
τ (indℓ1 (Γ) (D̃)) = τ (1) ε(indℓ1 (Γ) (D̃)) = τ (1) ind(D)
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 13

where ε ∶ Γ → C is the trivial representation of Γ.

5. [α]new in K 1 (M ; R/Z) associated with a flat bundle α∶ Γ → U (n)

The triviality property of Lemma 4.1 is crucial in what will be our definition of the
element in K 1 (M ; R/Z) associated to flat a vector bundle.
Let M = Γ/M̃ be a closed manifold with π1 (M ) = Γ and universal cover M̃ . A flat
bundle over M is just given by a homomorphism from Γ to the associated structure group.
Let B be a (finite) von Neumann algebra. A flat C(M ) ⊗ B module with fiber B n
is given by a homomorphism u from Γ to the group Un (B) of unitaries of Mn (B). The
associated bundle is M̃ ×Γ B n ; the corresponding finitely generated projective module over
C(M ) ⊗ B is the set of continuous maps f ∶ M̃ → B n such that f (gx) = u(g)f (x) for all
g ∈ Γ and x ∈ M̃ . A flat isomorphism of flat bundles associated with u+ ∶ Γ → Un (B) and
u− ∶ Γ → Un (B) is given by a unitary v ∈ Un (B) intertwining the morphisms u+ and u− .
We first observe the following
Proposition 5.1. Let M be a compact manifold and B be a finite factor. Let E be a flat
C(M ) ⊗ B module with fiber B (given by a homomorphism u ∶ π1 (M ) → U (B)). Then
there exists ℓ ∈ N∗ such that E ℓ ≃ (C(M ) ⊗ B)ℓ .
Proof. Follows immediately from Lemma 4.1 using Lemma 2.2. 
A flat Hermitian vector bundle Eα → M is given by a homomorphism α∶ Γ → Un .
In the following we denote with E + = C(M, Eα ) the corresponding (flat) C(M )-module
of sections and E − = C(M )n the module of sections of the trivial bundle of rank n.
Proposition 5.2. Let Eα be a flat unitary vector bundle as above.
a) There exists a finite factor B, a flat C(M ) ⊗ B module E with fiber B and an
isomorphism of flat bundles v∶ E − ⊗C(M ) E Ð→ E + ⊗C(M ) E.
By Proposition 5.1, up to tensoring B by Mn , we have an isomorphism ϕ∶ C(M ) ⊗ B → E.
Let wv be the isomorphism
wv ∶= (1E − ⊗ ϕ)−1 ○ v −1 ○ (1E + ⊗ ϕ)∶ E + ⊗ B → E − ⊗ B . (5.1)
b) The class of (E , E , wv ) in K (M, R/Z) is independent of all choices B, E, v, ϕ.
+ − 1

Proof. a) Given a finite factor B and a flat C(M ) ⊗ B module E with fiber B asso-
ciated with a morphism u ∶ Γ → U (B), the flat bundle E − ⊗C(M ) E corresponds
to the morphism g ↦ 1 ⊗ u(g) ∈ Un (B) ⊂ Mn ⊗ B and the flat bundle E + ⊗C(M ) E
corresponds to the morphism g ↦ α(g) ⊗ u(g).
We therefore just have to construct a finite factor B with a morphism u ∶ Γ →
U (B) and an element v ∈ Un (B), such that, for all g ∈ Γ, we have v(1 ⊗ u(g)) =
(α(g) ⊗ u(g))v.
Let Γ1 = α(Γ) be the image of Γ in Un (with the discrete topology) and K = Γ1
its closure in Un . Set B1 = L∞ (K) ⋊ Γ1 , where Γ1 acts on K by left translation.
Denote by u ∶ Γ1 → L∞ (K) ⋊ Γ1 the canonical inclusion; the inclusion map v ∶ K →
Un ⊂ Mn is an element of Mn (L∞ (K)) ⊂ Mn (B1 ); by definition (1 ⊗ ug )v(1 ⊗ ug )−1
is the function x ↦ v(g −1 x) = α(g)−1 v(x), i.e. the element (α(g)−1 ⊗ 1)v in
Mn ⊗ L∞ (K) ⊂ Mn (B1 ).
The desired equality v(1 ⊗ u(g)) = (α(g) ⊗ u(g))v follows.
By density of Γ1 (discrete) in K (compact), we have that B1 is a finite factor.
14 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

b) The choice of E is equivalent to the choice of a morphism u ∶ Γ → U (B); the choice of

ϕ is equivalent to a continuous map ψ ∶ M̃ → U (B) such that ψ(gx) = u(g)ψ(x) for
all g ∈ G and x ∈ M̃ . The element wv−1 is then the map x ↦ (1Mn ⊗ ψ(x))−1 v(1Mn ⊗
ψ(x)) from M̃ to Un (B) which satisfies ψ(gx) = (α(g) ⊗ 1)ψ(x) and thus defines
an isomorphism of bundles E − ⊗ B → E + ⊗ B.
Assume that we are given (Bj , uj , vj , ψj ) with j = 1, 2. Put B = B1 ⊗B2 (von
Neumann tensor product of course), and define u = u1 ⊗ u2 ∶ g ↦ u1 (g) ⊗ u2 (g)
and ψ = ψ1 ⊗ ψ2 ∶ x ↦ ψ1 (x) ⊗ ψ2 (x); put also V1 = v1 ⊗ 1B2 and V2 = β(v2 ), where
β ∶ Mn ⊗ B2 → Mn ⊗ B1 ⊗ B2 is the obvious inclusion.
We find in this way two new families (B, u, V1 , ψ) and (B, u, V2 , ψ) with the
desired properties.
For x ∈ M̃ , we have wV1 (x) = (1Mn ⊗ ψ(x))−1 V1−1 (1Mn ⊗ ψ(x)) = wv1 (x) ⊗
B2 ; it follows that the elements of K1 (M, R/Z) corresponding to (B1 , u1 , v1 , ψ1 )
and (B, u, V1 , ψ) coincide. Also wV2 (x) = β(wv2 (x), and thus the elements of
K1 (M, R/Z) corresponding to (B2 , u2 , v2 , ψ2 ) and (B, u, V2 , ψ) coincide.
Finally, write V1 = V2 W where W is a unitary in Mn (Q) where Q = B∩{u(g); g ∈
Γ}′ . By connectedness of the unitary group of a von Neumann algebra, it follows
that the elements of K1 (M, R/Z) corresponding to (B, u, V1 , ψ) and (B, u, V2 , ψ)
coincide. 

Definition 5.3. Given α∶ Γ → Un , E + and E − as above. For any E, ϕ, v as in Proposition

5.2. We define
[α]new ∶= [(E + , E − , wv )] ∈ K1 (C(M ), R/Z) .

Theorem 5.4. Let Eα be a flat unitary vector bundle as above. Consider any other given
data of: a finite factor B, a flat B-bundle F → M along with an isomorphism θ∶ E − ⊗ F →
E + ⊗ F (not required to preserve connections), and a trivialization ψ∶ F → C(M ) ⊗ B, and
denote with CS(θ ∗ ∇E + ⊗F ; ∇E − ⊗F ) the Chern–Simons transgression form (formula 3.9).
Let wθ = (1E + ⊗ ψ)−1 ○ θ ○ (1E − ⊗ ψ) ∶ E − ⊗ B → E + ⊗ B. We have

[(E + , E − , wθ−1 )] + j∗ (Ch −1

τ (CS(θ ∇E − ⊗F ; ∇E + ⊗F ))) = [α]new ∈ K (M, R/Z).
∗ 1

Proof. Up to tensoring F by E given by proposition 5.2, we may assume that there is a

flat connexion preserving isomorphism v∶ E + ⊗ F → E − ⊗ F.
Let now B and F → M be fixed, and θ∶ E − ⊗ F → E + ⊗ F and v∶ E + ⊗ F → E − ⊗ F two
isomorphisms, such that v is connexion preserving.
Put ℓ = v ○ θ; it is an automorphism of E − ⊗ F. As v ∗ (∇E − ⊗F ) = ∇E + ⊗F , we find (using
prop. 3.11)
CS(θ ∗ ∇E + ⊗F ; ∇E − ⊗F ) = CS(ℓ∗ ∇E − ⊗F ; ∇E − ⊗F ) = Ch(ℓ).
Moreover [(E + , E − , wv )] − [(E + , E − , wθ−1 )] = [(E − , E − , wv ○ wθ ] is the image by j ∶
K1 (C(M ); R) → K1 (C(M ); R/Z) of the class of the automorphism wv ○ wθ = (1− ⊗ ϕ)−1 ○
ℓ ○ (1− ⊗ ϕ) of E − ⊗ B. This is the same class as the automorphism ℓ of E − ⊗ F. 

The element [α]AP S is constructed in [APS3, 5] by means of an isomorphism θ ∶ E − ⊗B →

E + ⊗ B, where B = Mn (C) (for a suitable n): it is defined by [α]AP S = [(E + , E − , wθ−1 )] +
j∗ (Ch −1
τ (CS(θ ∇E − ⊗B ; ∇E + ⊗B ))). We immediately find:
∗

Corollary 5.5. Let α∶ Γ → U (n). The element [α]new coincides with [α]AP S . ◻
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 15

6. The pairing with [D] ∈ K1 (M ) and the spectral flow description

Let [α] = [(E + , E − , wv )] ∈ K0 (Ci0 ⊗ C(M )) of Section 5, where Ci0 is the mapping cone
of the unital inclusion i0 ∶ C → B - here B is a II1 factor. In order to stay within unital
algebras, we represent [α] in the K-theory of the double cylinder Zi0 ⊗ C(M ), as we saw
in (2.2). It is the image in the summand K0 (Ci0 ⊗ C(M )) of the Zi0 ⊗ C(M )-module
Fwv = {f ∈ C ([0, 1], E + ⊗ B) ; f (0) ∈ E + ⊗ 1 and wv f (1) ∈ E − ⊗ 1} .
The Kasparov product gives the natural pairing between the class [α] ∈ K0 (Ci0 ⊗C(M ))
and a K-homology class [D] ∈ K1 (M ). In this section we describe and interpret it. In
other words, we interpret the map K1 (M ) = KK 1 (C(M ), C) → R/Z induced by [α] via
the pairing
KK 0 (C, C(M ) ⊗ Zi0 ) × KK 1 (C(M ), C) Ð→ KK 1 (C, Zi0 ) = K1 (Zi0 ) = R/ Z . (6.1)

6.1. Various notions of connections. Let us start with a class [D] ∈ K1 (M ) repre-
sented by the unbounded Kasparov module (H, D), where D is a first order self-adjoint
elliptic (classical) pseudodifferential operator acting on the sections of a Hermitian vector
bundle S → M , and H = L2 (M, S).
A particularly important case for D in our discussion is the case of a Dirac type operator,
i.e. an operator constructed in the following way:
● We assume that M is endowed with a Riemannian metric,
● The bundle S is a Clifford bundle i.e. it is endowed with a linear bundle map
cliff ∶ T ∗ M → End(S) such that for every ξ ∈ Tx∗ M then cliff(ξ) ∈ End(Sx ) is a
skew-adjoint and satisfies cliff(ξ)2 = −∥ξ∥2 1Sx ; denote by c ∶ S ⊗ T ∗ M → S the
corresponding map.
● The bundle S is further endowed with a metric Clifford connection ∇S in the sense
of [BGV, Def. 3.39].
● The corresponding Dirac operator is given by D(s) = c(∇S (s)) for any smooth
section s of S.
Let us distinguish different types of connections that arise in our discussion:
Definition 6.1. Let A be a unital C ∗ -algebra E a finitely generated projective module
endowed with a compatible Hilbert A-module structure. Let V → M be a smooth bundle
with fibers E.
● If D is a first order self-adjoint elliptic (classical) pseudodifferential operator acting
on the sections of a hermitian vector bundle S → M , a pseudodifferential connection
for D is a selfadjoint pseudodifferential A operator DV (in the sense of [MF]) acting
on the sections of the bundle V ⊗ S whose principal symbol is σDV = 1V ⊗ σD .
● If D is Dirac type, then a Dirac type connection for D is an operator of the form
(1V ⊗c)○∇V ⊗S where ∇V ⊗S is a metric Clifford connection on the Clifford A-bundle
V ⊗ S.
Of course, if D is Dirac type, then a Dirac type connection for D is a particular case of
a pseudodifferential connection.

6.2. The pairing as an element of K1 (Zi0 ). The product (6.1) is computed by any
choice of a D-connection G in the sense of [CS, Ku] on the Hilbert Zi0 -module V =
Fwv ⊗C(M )⊗Zi0 (L2 (M, S) ⊗ Zi0 ). Note that a pseudodifferential connection for D is a
D-connection in the sense of [CS, Ku] (the converse is of course not true).
16 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

We have
[α] ⊗C(M ) [D] = [(V, G)] . (6.2)
We can take any G as follows. It will be of the form G = (Gt ), i.e. G(f )(t) = Gt (ft ).
Since the bundle E − = C n ⊗M is trivial, we may set G1 = wv (1Cn ⊗D⊗1B )wv−1 . To construct
G0 , we may just take any selfadjoint pseudodifferential operator DE + on E + ⊗ S whose
principal symbol is 1E + ⊗ σD ; we may also assume that D is local enough (in particular
if D is differential), so that there is a canonical choice for DE + , called the twisted tensor
product of D with the flat bundle E + .
Put then G0 = DE + ⊗ 1B . The operators G0 and G1 have the same principal symbol
and therefore G1 − G0 is bounded. We may then choose any strongly continuous bounded
path qt of selfadjoint operators acting on E + ⊗ S ⊗ B such that q0 = 0, q1 = G1 − G0 and
set Gt = G0 + qt . In particular, we may define Gt = (1 − t)G0 + tG1 .

6.3. The pairing as an element of R/Z. Let us first recall a few facts about index in
the type II setting:
Let B be a factor of type II1 . Denote by τ the normalized trace on B. Let N = B⊗L(H)
the associated factor of type II∞ endowed with the trace TrN = τ ⊗ Tr where Tr is the
canonical trace on L(H). Let p = 1 ⊗ p0 ∈ N , where p0 is a minimal projection in L(H)
and identify B with pN p.
Let P± ∈ N be two projections such that P+ − P− is in the ideal KN of generalized
compact operators in N . Then ind(P+ , P− ) is the index of q ∶ E + → E − where E± = P± N p
considered a a Hilbert B-module and qx = P− x for x ∈ E + . Note that q is B-Fredholm with
quasi-inverse x ↦ P+ x. If P+ −P− is in the domain of TrN , then ind(P+ , P− ) = TrN (P+ −P− ).
Let D± be selfadjoint operators affiliated with N such that D+ − D− ∈ N (which means
that D+ and D− have same domain and there exists b ∈ N , with b = b∗ and D+ = D− + b).
Note that for λ± not in the spectrum of D± , we have
(D− − λ− )−1 − (D+ − λ+ )−1 = (D+ − λ+ )−1 (b + λ− − λ+ )(D− − λ− )−1 ,
therefore the resolvent of D+ is in KN if and only if the resolvent of D− is. Assume this is
the case. For a ∈ R, let χa be defined by
χa (t) = 1 if t ≥ a and χa (t) = 0 if t < a.
Then for all a ∈ R, we have χa (D+ ) − χa (D− ) ∈ KN and we put
sf a (D+ , D− ) = ind(χa (D+ ), χa (D+ )).
Note that for all a, b ∈ R, we have (by additivity of the index)
sf a (D+ , D− ) − sf b (D+ , D− ) = TrN ((χa − χb )(D+ )) − TrN ((χa − χb )(D− )).
Let us now come to (V, G) as defined above. We easily find:
Proposition 6.2. The class j(sf a (G0 , G1 )) ∈ R/Z is independent of a ∈ R. The class
[α] ⊗C(M ) [D] in R/Z is j(sf a (G0 , G1 )).

Proof. Since G0 = DE + ⊗ 1B and G1 = wv (1Cn ⊗ D ⊗ 1B )wv−1 , the spectra of G0 and

G1 are discrete and the trace of all the spectral projections is integer. In other words,
TrN ((χa − χb )(G0 )) ∈ Z and TrN ((χa − χb )(G1 )) ∈ Z, and therefore j(sf a (G0 , G1 )) ∈ R/Z
is independent of a ∈ R.
To see that the class [(V, G)] in R/Z is equal to j(sf a (G0 , G1 )), note that:
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 17

(1) there is a smooth function ϕ ∶ R → [−1 + 1] such that ϕ = −1 near −∞, ϕ = +1

near +∞, and ϕ = χa in Sp(G0 ) ∪ Sp(G1 ). The class of the unbounded Kasparov
module (V, G) is that of (V, ϕ(G)).
(2) If sf a (G0 , G1 ) = 0, one can find a path P = (Pt ) of projections in L(V ) such that
t ↦ Pt − P0 is a norm continuous path with values in K((L2 (M ; S ⊗ E + ) ⊗ B), and
such that P0 = χa (G0 ) and P1 = χa (G1 ): the class of (V, ϕ(G)) coincides with the
class of (V, 2P − 1) which is degenerate.
(3) For λ ∈ R∗+ , let Eλ be a finitely generated projective B-module with trace λ and
denote Vλ = C0 (]0, 1[, Eλ ). Let Fλ ∈ L(Vλ ) be given by (Fλ ξ)(t) = (2t − 1)ξ(t).
Then (Vλ , Fλ ) is the product of the (opposite of the) Bott element in K1 (C0 (]0, 1[)
by Eλ , and its class in K1 (C0 (]0, 1[) ⊗ B) = R is therefore equal to −λ.
(4) If sf a (G0 , G1 ) = λ > 0, then the class of (V, ϕ(G)) ⊕ (Vλ , Fλ ) is 0 by (2), hence
the class of (V, ϕ(G)) is j(λ). This is also true if sf a (G0 , G1 ) < 0 (for instance by
replacing G by −G). 

6.4. The pairing as a rho invariant. We will now explain how our approach can be
used to establish the formula of Atiyah–Patodi–Singer in [APS3, Sec.5].
We first recall the facts about the η invariant that will be used here. A very nice survey
can be found in the first chapter and in appendix D of [Bo].
(1) If P is an elliptic self-adjoint pseudodifferential operator of positive order, the eta
function η(P, .) = Tr(P ∣P ∣−1−s ) has a meromorphic continuation which is regular
at 0 (cf. [APS3, Thm. 4.5], [Gi2]). Put ξ(P ) = 21 (η(P ) + dim ker P )

(2) If P is differential of Dirac type, then η(P ) = √ ∫ Tr(P e−tP ) √ (cf [BF,
1 +∞ 2 dt

π 0 t
Thm 2.6]).
(3) The above remain true for P with coefficients in a finite von Neumann algebra B
([CG, Ram, CP1, LP, An]).
(4) For P with coefficients in B, put ξε (P ) = √ ∫ Tr(P e−tP ) √ + dim B ker P.
1 +∞ 2 dt 1
2 π ε t 2
If P1 , P0 are bounded perturbations of each other, then for any ε > 0, and any
smooth path Pt joining P0 with P1 (i.e. a path of the form t ↦ P0 + Qt where
t ↦ Qt is smooth) we have ([CP1, Cor. 8.10])
√
sf 0 (P0 , P1 ) = ξε (P0 ) − ξε (P1 ) + ∫ TrN (Ṗt e t ) dt.
ε 1
−εP 2
π 0
(where N is the II∞ factor associated with B).
(5) If V → M is a√ von Neumann bundle and Pt is a smooth family of Dirac type
∫ TrN (Ṗt e t ) dt converges, when ε → 0, to a local term,
ε 1
−εP 2
operators then
π 0

∫ Â(M ) ch S/Ŝ ⋅ CS(∇0 , ∇1 )

M

see [APS2, Sec. 4], [Bo, 1.5.1].

Let D be an elliptic differential operator. Recall that the rho invariant of α is ρ(α, D) =
ξ(D α ) − nξ(D).
Proposition 6.3. [APS3, Sec.5]: If D is a Dirac type operator, then [α] ⊗C(M ) [D] is the
class in modulo Z of ρ(α, D).
18 PAOLO ANTONINI, SARA AZZALI AND GEORGES SKANDALIS

Proof. If P and P ′ are two Dirac type operators acting √ on a1 von Neumann Clifford bundle,
we put θ(P, P ′ ) = sf 0 (P, P ′ )−ξ(P )+ξ(P ′ ) = limε→0 Tr(Ṗt e−εPt ) dt for any smooth
ε
∫
2

π 0
path Pt joining P with P ′ .
On the bundle E + ⊗ B ≃1E + ⊗ϕ E + ⊗ E ≃v E + ⊗ E ≃1E − ⊗ϕ−1 E − ⊗ B over M , we have
three flat bundle structures (since the two middle ones coincide). We thus get three Dirac
type operators as twisted tensor products of D with this flat bundle G0 = D α ⊗ 1B , G̃ and
G1 = wv (1Cn ⊗ D ⊗ 1B )wv−1 acting on E + ⊗ S ⊗ B.
Consider the Dirac operators P = D ⊗ 1B and P ′ = (1S ⊗ ϕ)D̃(1S ⊗ ϕ−1 ) acting on S ⊗ B.
We have G1 = wv (1Cn ⊗P )wv−1 and G̃ = wv (1Cn ⊗P ′ )wv−1 , therefore θ(G1 , G̃) = nθ(P, P ′ ).
By the locality property (5) of θ, and since G0 is the twisted tensor product of P with
the flat bundle E + and G̃ is the twisted tensor product of P ′ with the flat bundle E + , we
find θ(G0 , G̃) = nθ(P, P ′ ). And finally θ(G0 , G1 ) = 0. 

References
[An] P. Antonini, The Atiyah Patodi Singer signature formula for measured foliations, Ph.D Thesis, La
Sapienza, Rome http://arxiv.org/abs/0901.0143.
[Ati] M.F. Atiyah, Elliptic operators discrete groups and von Neumann algebras. Astérisque 32–33
(1976), 43–72.
[APS1] M.F. Atiyah, V.K. Patodi, I.M. Singer. Spectral asymmetry and Riemannian geometry. I. Math.
Proc. Cambridge Philos. Soc. 77 (1975), 43–49.
[APS2] , Spectral asymmetry and Riemannian geometry. II. Math. Proc. Cambridge Philos. Soc.
78 (1975), 405–432.
[APS3] , Spectral asymmetry and Riemannian geometry. III. Math. Proc. Cambridge Philos. Soc.
79 (1976), 71–99.
[Ba] D. Basu, K-theory with R/Z coefficients and von Neumann algebras, K-Theory 36 (2005), no. 3-4,
327–343 (2006).
[BGV] N. Berline, E. Getzler, M. Vergne, Heat kernels and Dirac operators, Springer-Verlag, New York
1992.
[BF] J.M. Bismut and D.S. Freed, The analysis of elliptic families II: Dirac operators, eta invariants,
and the holonomy theorem of Witten, Commun. Math. Phys. 107(1986), 103–163.
[Bl] B. Blackadar, K-theory for operator algebras, second edition, MSRI publications, Cambridge Uni-
versity Press.
[Bl1] , Operator algebras: theory of C*-Algebras and non Neumann algebras, Vol. 13, Encyclopae-
dia on operator algebras and non-commutative geometry, Springer 2006.
[Bo] M. Bohn, On Rho invariants of fibre bundles, http://arxiv.org/abs/0907.3530 [math.GT].
[CP1] A. Carey, J. Phillips, Spectral flow in Fredholm modules, eta invariants and the JLO cocycle,
K-Theory 31 (2) (2004), 135–194.
[CG] J. Cheeger, M. Gromov, On the characteristic numbers of complete manifolds of bounded curvature
and finite volume, in Differential geometry and complex analysis, Springer, Berlin 1985, 115–154.
[CM] A. Connes, H. Moscovici, Cyclic cohomology, the Novikov conjecture and hyperbolic groups, Topol-
ogy 29 (1990), no. 3, 345–388.
[CS] A. Connes, G. Skandalis, The longitudinal index theorem for foliations, Publ. Res. Inst. Math. Sci.
20 (1984), no. 6.
[De] R.J. Deeley, R/Z-valued index theory via geometric K-homology,
http://arxiv.org/abs/1206.5662 [math.KT].
[DHK] R.G. Douglas, S. Hurder, J. Kaminker, Eta invariants and von Neumann algebras, Bull. Amer.
Math. Soc 21 no.1 (1989).
[DHK2] R.G. Douglas, S. Hurder, J. Kaminker, Cyclic cocycles, renormalization and eta-invariants, Inv.
Math. 103, no.1 (1991), 101–180.
[G] E. Getzler, The odd Chern character in cyclic homology and spectral flow, Topology 32 (1993) (3),
489–507.
[Gi2] P. B. Gilkey, The residue of the global eta function at the origin, Adv. Math. 40 (1981), 290–307.
 FLAT BUNDLES, VON NEUMANN ALGEBRAS AND K-THEORY WITH R/Z-COEFFICIENTS 19

[Gi] P. B. Gilkey, The eta invariant and secondary characteristic classes of locally flat bundles, in
Algebraic and differential topology – global differential geometry, Teubner–Texte Math. 70, Leipzig
(1984) 49–87.
[Gr] M. Gromov, Kähler hyperbolicity and L2 -Hodge theory, J. Diff. Geom. 33 (1991), 263–292.
[Hu] S. Hurder, Eta invariants and the odd index theorem for coverings. Geometric and topological
invariants of elliptic operators, Contemp. Math., 105, Amer. Math. Soc., Providence, RI, (1990),
47–82.
[Ka] M. Karoubi, Homologie Cyclique et K-Théorie, Astérisque 149 (1987).
[Ka2] M. Karoubi, K-theory. An introduction. Grundlehren der Mathematischen Wissenschaften, Band
226. Springer-Verlag, Berlin–New York, 1978.
[KP] J. Kaminker, V. Perera, Type II Spectral Flow and the Eta Invariant, Canad. Math. Bull. 43 (1)
(2000), 69–73.
[Ku] D. Kucerovsky, The KK-product of unbounded modules, K-Theory 11 (1997), no. 1, 17–34.
[LP] E. Leichtnam, P. Piazza, Cut-and-Paste on foliated bundles, Contemp. Mathem. 366 (2005), 351–
392 ”Spectral Geometry of Manifolds with Boundary”, Amer. Math. Soc., Providence, RI.
[Lo] J. Lott, R/Z-index theory, Comm. Anal. Geom. 2 (1994), no. 2, 279–311.
[Lü] W. Lück, The relation between the Baum–Connes Conjecture and the Trace Conjecture, Invent.
math. 149 (2002), 123–152
[MZ1] X. Ma, W. Zhang, η-Invariant and flat vector bundles, Chin. Ann. Math 27B(1), (2006), 57–72.
[MZ2] , η-Invariant and flat vector bundles II, in “Inspired by S. S. Chern. A memorial volume in
honor of a great mathematician.” Hackensack, NJ: World Scientific. Nankai Tracts in Mathematics
11 (2006), 335–350.
[MF] A. S. Miščenko, A. T. Fomenko, The index of elliptic operators over C*-algebras. (Russian) Izv.
Akad. Nauk SSSR Ser. Mat. 43 (1979), no. 4, 831–859, 967.
[MN] I. Moriyoshi, T. Natsume, The Godbillon-Vey cyclic cocycle and longitudinal Dirac operators,
Pacific J. of Math. 172, no.2 (1996), 483–539.
[PS] P. Piazza, T. Schick, Bordism, rho-invariants and the Baum-Connes conjecture. J. of Noncommu-
tative Geometry 1 (2007), 27–111.
[Pu] I. F. Putnam, An excision theorem for the K–theory of C ∗ algebras, J. Operator Theory 38 (1997),
151–171.
[Ra] I. Raeburn, K-theory and K-homology relative to a II∞ -factor. Proc. Amer. Math. Soc. 71 (1978),
no. 2, 294–298.
[Ram] M. Ramachandran, Von Neumann index theorems for manifolds with boundary, J. Diff. Geom. 38
(1993), 315–349.
[Sc] T. Schick, L2 -index theorems, KK-theory, and connections, New York J. Math. 11 (2005), 387–443.
[Va] B. Vaillant, Indextheorie fur Überlagerungen. Diplomarbeit, Universität Bonn 1997,
http://styx.math.uni-bonn.de/boris/diplom.html.
[Wo] K. Wojciechowski, A note on the space of pseudodifferential projections with the same principal
symbol, J. Operator Theory 15 (1986), no. 2, 207–216.