Professional Documents
Culture Documents
Controlled KK-theory II: A Mayer-Vietoris Principle and The UCT
Controlled KK-theory II: A Mayer-Vietoris Principle and The UCT
Abstract
Contents
1 Introduction 2
1
3 Controlled KK-groups in the odd case 17
1 Introduction
In this introduction, anything called A or B is a separable C ˚ -algebra. A
C ˚ -algebra A is said to satisfy the Universal Coefficient Theorem (UCT) of
Rosenberg and Schochet [31] if for any C ˚ -algebra B, there is a canonical
short exact sequence
Thus one can say informally that a C ˚ -algebra satisfies the UCT if all KK-
groups involving A are determined by the associated K-theory groups. The
UCT is known to be satisfied for a large class of separable C ˚ -algebras: for
example, it is satisfied for all (separable) inductive limits of type I C ˚ -algebras
[31, 1.17], and for the C ˚ -algebra of any (second countable) amenable groupoid
[36, Proposition 10.7]. However, Skandalis [33, page 571] has shown that
there are C ˚ -algebras that do not satisfy the UCT.
Now, the examples given by Skandalis are not nuclear. Whether or not
the UCT holds for all nuclear C ˚ -algebras is an important open problem.
2
One reason for this is the spectacular recent progress [15, 16, 13, 12, 35]
(see also [7]) in the Elliott program [11] to classify simple, separable, nuclear
C ˚ -algebras by K-theoretic invariants. Establishing the range of validity of
the UCT is now the only barrier to getting the “best possible” classification
result in this setting.
The goal of this paper is to introduce a new sufficient condition for the
UCT, and in particular to give a new reformulation of the UCT problem for
nuclear C ˚ -algebras in terms of a structural property of Kirchberg algebras.
Decomposable C ˚ -algebras
Our sufficient condition for the UCT is based on a new structural property
of C ˚ -algebras that we call decomposability. Roughly, a C ˚ -algebra is decom-
posable if there are almost central elements that “cut” the C ˚ -algebra into
two finite-dimensional C ˚ -algebras that have well-behaved (approximate) in-
tersection. The definition of decomposability is inspired by the properties of
having nuclear dimension one as in [40] and real rank zero as in [6], and also
by the authors’ earlier work on the Künneth formula (partly in collaboration
with Oyono-Oyono) [25, 37]. Indeed (we discuss this in more detail below),
if a C ˚ -algebra has real rank zero and nuclear dimension one, then there
are almost central elements that “cut” it into two finite-dimensional C ˚ -
subalgebras, but without necessarily giving any control on the intersection.
The condition on the intersection is important for our K-theoretic compu-
tations (similarly to the very closely related approximate ideal structures
introduced in [37]), so we make it part of our property.
If X and Y are subsets of a metric space and ą 0, we write X Ď Y if
for all x P X there is y P Y such that dpx, yq ď .
3
(iii) there is a finite-dimensional C ˚ -subalgebra F of A such that hp1 ´
hqX Ď F , and such that if B is the unit ball of F , then B Ď Fh and
B Ď F1´h .
If the same condition holds but with (iii) omitted, A is said to be weakly
decomposable.
Example 1.2. We will show in Appendix A below that the Cuntz algebras
On are decomposable for finite n, and that crossed products arising from
minimal actions of Z on the Cantor set are decomposable.
Example 1.3. In [19], Jaime and the first author show that all C ˚ -algebras
with nuclear dimension one and real rank zero are weakly decomposable.
Conversely, [19] also shows that a weakly decomposable C ˚ -algebra has nu-
clear dimension one (and in particular is nuclear). We do not know if (weak)
decomposability implies real rank zero in general.
In particular, these results imply that Kirchberg algebras are always
weakly decomposable, as [41, Theorem 1] shows they have always have real
rank zero and [5, Theorem G] shows they always have nuclear dimension one.
More generally, thanks to the recent successful resolution of (part of) the
Toms-Winter conjecture [8, 9], nuclear dimension one is now known to be a
very common property of simple nuclear C ˚ -algebras. On the other hand,
real rank zero has been known to be fairly widespread among simple nuclear
C ˚ -algebras for some time: see for example [4, 41, 29]. It thus seems fair
to say that weak decomposability is a common property of (simple) nuclear
C ˚ -algebras.
Non-example 1.4. In [19], Jaime and the first author show that any de-
composable C ˚ -algebra has torsion free K1 -group (following a suggestion of
Putnam). As there are Kirchberg algebras with any countable abelian group
as K1 -group [27, Theorem 3.6], there are Kirchberg algebras that are not
decomposable. In particular, weak decomposability is strictly weaker than
decomposability.
4
Main results
The main theorem of this paper is as follows.
We discuss the proof in more detail below, but let us first say a little
about the connection to Kirchberg algebras. Recall that a C ˚ -algebra is a
Kirchberg algebra if it is separable, simple, nuclear, and purely infinite. We
recall the following theorem of Kirchberg: see [28, Corollary 8.4.6].
Theorem 1.6 (Kirchberg). To prove the UCT for all separable, nuclear C ˚ -
algebras, it suffices to prove the UCT for any unital Kirchberg algebra with
zero K-theory.
Theorems 1.5 and 1.6 imply that if any unital Kirchberg algebra with zero
K-theory is decomposable, then the UCT holds for all separable nuclear C ˚ -
algebras. Conversely, if the UCT holds for all separable nuclear C ˚ -algebras,
then from the Kirchberg-Phillips classification theorem [21, 26] (see also [28,
Corollary 8.4.2] for the statement we want here), any unital Kirchberg alge-
bra with zero K-theory will be isomorphic to O2 , and so decomposable by
Example 1.2.
5
The first key ingredient in the proof of Theorem 1.5 comes from our ear-
lier work [38]. There we define controlled KK-theory groups in the following
way. Assume that A and B are separable C ˚ -algebras, and assume for sim-
plicity that A is unital and nuclear. Let HB :“ `2 b B be the standard
Hilbert B-module. Using an infinite multiplicity, faithful, unital representa-
tion A Ď Bp`2 q Ď LpHB q, identify A with a subset of the C ˚ -algebra LpHB q
of adjointable operators on HB . For a finite subset X Ď A and ą 0, ˜ define
¸
1 0
P pX, Bq to be the set of projections in M2 pLpHB qq such that p ´
0 0
is in the compact operators KpHB q, and such that }rp, xs} ď for all x P X.
The associated controlled KK-theory group is then defined to be
between the KL-group of A and B (see [27, Section 5] for the original def-
inition of KLpA, Bq in the UCT case, and [10, Section 5] for the general
definition) and the inverse limit of our controlled KK-groups defined by
letting X get larger, and smaller.
Now, assume that A is a decomposable C ˚ -algebra with K˚ pAq “ 0 as in
the statement of Theorem 1.5. We want to show that KKpA, Aq “ 0. Using
a result of Dadarlat [10, Corollary 5.3], it suffices to show that KLpA, Aq “ 0,
and therefore that lim KK0 pX, Bq by the result above. Let then prpX, sqX,
Ð
be an element of this group. To show that this element is zero, it suffices
to show that for any and X there exists δ ď and Y Ě X such that the
natural “forget control map”
6
paper. Let h and Fh , F1´h , and F be as in the definition of decomposability
for the given set X and some very small parameter δ. Let Y be the union
of X and finite spanning sets for each of Fh , F1´h , and F . Then one can
construct a diagram
KK 1 pF, Aq / KK0 pX, Aq
where the vertical arrow is the forget control map. This diagram has the
property that if rpY,δ s goes to zero under the map
then the image of rpY,δ s under the forget control map KKδ0 pY, Aq Ñ KK0 pX, Aq
is in the image of the map
In particular if the groups at the left and right are zero, then the group in
the middle is also zero.
Our analysis of the diagram in line (1) is based on a concrete construction
of this classical Mayer-Vietoris sequence that can be adapted to our controlled
1
It is not in the literature as far as we can tell. It can be derived from the usual long
exact sequence in KK-theory using, for example, the argument of [39, Proposition 2.7.15].
7
setting: the basic idea has its roots in algebraic K-theory, going back at least
as far as [23, Chapter 2]. Having said this, there is significant work to be done
adapting these classical ideas to the analytic superstructure that we built in
[38], and the resulting formulas and arguments end up being quite different.
There is also an additional step, in that the Mayer-Vietoris construction
naturally builds a diagram analogous to line (1), but with all groups being
controlled KK-groups, not the usual KK-groups as appear on the left and
right of line (1). One can, however, replace the controlled KK-groups with
usual KK-groups using an averaging technique based on compactness of the
unitary groups of F , Fh , and F1´h .
The details of our argument look a little different from the discussion
above, but we hope this gives a rough idea of why Theorem 1.5 is true.
8
matrix b, a ‘ b denotes the “block sum” pn ` mq ˆ pn ` mq matrix defined by
˜ ¸
a 0
a ‘ b :“ .
0 b
9
although concrete formulas for the Mayer-Vietoris boundary map unfortu-
nately seem to be missing from the C ˚ -algebra literature. The formulas we
use are instead inspired by classical formulas from algebraic K-theory [23,
Chapter 2], but heavily adapted to reflect our analytic setting.
The results of Section 5 essentially reduce showing our main vanishing
result for a decomposable C ˚ -algebra to showing a vanishing result for finite-
dimensional C ˚ -algebras. Section 6 then establishes the required vanishing
result for finite-dimensional C ˚ -algebras. This key point is to use an av-
eraging argument over finite-dimensional groups of unitaries to replace ap-
proximate commutativity with actual commutativity. Essentially, this iden-
tifies suitable controlled KK-groups, where the first variable spans a finite-
dimensional C ˚ -algebra, with honest KK-groups. Finally in the main body
of the paper, Section 7 puts everything together and gives the proof of The-
orem 1.5.
The paper concludes with Appendix A, which gives examples of decom-
posable C ˚ -algebras. We first use a technique of Winter and Zacharias [40,
Section 7] to show that the Cuntz alegbras On with n finite are decompos-
able. We then use our joint work with Guentner on dynamic asymptotic
dimension [17] to show that crossed products arising from Cantor minimal
systems are decomposable.
Acknowledgements
The authors were supported by the US NSF (DMS 1564281, DMS 1700021,
DMS 1901522, DMS 2000082) throughout the writing of this paper, and
gratefully acknowledge this support. We also thank Wilhelm Winter for
some helpful comments on a preliminary draft.
10
Hilbert B-module `2 b B. We identify LB with the “diagonal subalgebra”
1Mn b LB Ď Mn b LB “ Mn pLB q for each n.
Our goal in this section is to define a variant of the controlled KK-theory
groups of [38, Section 6]. We then show that taking this variant makes no
real difference, at least in the sense that an appropriate limit still gives KL-
groups.
(iii) the classes rσppqs, rσpqqs P K0 pCq defined by the images of p and q under
`
the canonical quotient map σ : Mn pKB q Ñ Mn pCq are the same.
If pp1 , q1 q is an element of Pn1 ,κ, pX, Bq and pp2 , q2 q is an element of Pn2 ,κ, pX, Bq,
define
˜˜ ¸ ˜ ¸¸
p1 0 q1 0
pp1 ‘ p2 , q1 ‘ q2 q :“ , P Pn1 `n2 ,κ, pX, Bq.
0 p2 0 q2
Define
8
ğ
P8,κ, pX, Bq :“ Pn,κ, pX, Bq,
n“1
(i) pp, qq „ pp ‘ r, q ‘ rq for any element pr, rq P P8,κ, pX, Bq with both
components the same;
11
(ii) pp1 , q1 q „ pp2 , q2 q whenever these elements are in the same path com-
ponent of P8,κ, pX, Bq.2
0
Finally, define KKκ, pX, Bq to be P8,κ, pX, Bq{ „.
Lemma 2.2. Let X be a subset of the unit ball of LB , and let ě 0 and
0
κ ě 1. Then the semigroup KKκ, pX, Bq constructed above is an abelian
group with neutral element the class r0, 0s of the zero idempotent.
0
Proof. It is clear from the definition that KKκ, pX, Bq is a monoid with
identity element the class r0, 0s. A standard rotation homotopy shows that
0
KKκ, pX, Bq is commutative. To complete the proof we need to show that
any element rp, qs has an inverse. We claim that this is given by rq, ps.
Indeed, applying a rotation homotopy to the second variable shows that
pp ‘ q, q ‘ pq „ pp ‘ q, p ‘ qq, and the element pp ‘ q, p ‘ qq is trivial by
definition of the equivalence relation.
12
Proof. Split the Hilbert space p is acting on as H0 ‘ H1 , where H0 is the
image of p and H1 is its orthogonal complement (note that neither H0 nor
H1 are trivial, as p is neither 0 nor 1). With respect to this decomposition,
˜ ¸
p has matrix 10 a0 for some a P BpH1 , H0 q, and we see that
Proof. The estimates for }1 ´ p} and }p ´ p˚ } are immediate from Lemma 2.3
(and direct checks for the degenerate cases p “ 0 and p “ 1). The estimate
for 2p ´ 1 follows as 2p ´ 1 “ p ´ p1 ´ pq.
13
(i) The element r is a projection that satisfies rp “ p and pr “ r.
(ii) Each rt is an idempotent such that }rt } ď κ for all t, and the map
t ÞÑ rt is κ-Lipschitz.
14
Moreover as z ě 1 we have }z ´1 } ď 1. Hence for any x P X,
}r ´ s} ď }pp ´ qqp˚ zp´1 } ` }qpp˚ ´ q ˚ qzp´1 } ` }qq ˚ pzp´1 ´ zq´1 q}. (4)
φ˚
(
0
KKλ, pX, Bq
15
commutes, where the horizontal and diagonal arrows are the forget control
maps of line (3).
Proof. Let pp, qq be an element of Pn,λ,δ pX, Bq. Let r and s be the projections
associated to p and q respectively as in Definition 2.5. Using Lemma 2.6 parts
(i) and (iii), and the fact that δ ď {p8κ3 q, we may define a map
Allowing n to vary, and noting that the process of taking associated projec-
tions takes homotopies to homotopies (by part (iv) of Lemma 2.6) and block
sums to block sums, we get a well-defined homomorphism
0 0
φ˚ : KKκ,δ pX, Bq Ñ KKλ, pX, Bq.
0
KKλ,δ pX, Bq / 0
KKκ,δ pX, Bq .
φ˚
(
0
KKλ, pX, Bq
For this, it suffices to show that if pr, sq P Pn,λ, pX, Bq is constructed from
pp, qq P Pn,λ,δ pX, Bq Ď Pn,κ,δ pX, Bq as above, then pr, sq and pp, qq are in the
same path component of Pn,λ, pX, Bq. This follows from parts (ii) and (iii)
of Lemma 2.6.
Definition 2.8. Fix κ ě 1. Let Xκ denote the set of all pairs pX, q where
X is a finite subset of A1 , and ą 0. Put a partial order on X by stipulating
that pX, q ď pY, δq if the relation in line (2) holds (with λ “ κ).
16
Analogously to [38, Remark 6.7], the set Xκ is directed: an upper bound
for pX1 , 1 q and pX2 , 2 q is given by pX1 YX2 , mint1 , 2 uq. The forgetful maps
0
in line (3) therefore turn the collection pKKκ, pX, BqqpX,qPXκ into an inverse
0
system, so we may form the inverse limit lim KKκ, pX, Bq. The next result
Ð
is our last goal in this section.
Proof. Using Lemma 2.7 we can build an increasing sequence pXn q of finite
subsets of A1 with dense union, and a decreasing sequence pn q of positive
numbers that tend to zero such that there is a commutative ladder diagram
/ 0
KKκ, pXn , Bq / KK 0 / / KK 0
¨¨¨ κ,n´1 pXn´1 , Bq ¨¨¨ κ,1 pX1 , Bq ,
n
O pnq O pn´1q
p1q
φ˚
O
φ˚ φ˚
) ' %
/ 0
KK1, pXn , Bq / 0
KK1, pXn´1 , Bq / / KK 0
¨¨¨ n n´1
¨¨¨ 1,1 pX1 , Bq
pnq
where the maps φ˚ are those from Lemma 2.7. It follows from this that
0 0
lim KKκ, pX, Bq – lim KK1, pX, Bq.
Ð Ð
and the last limit on the right is isomorphic to KLpA, Bq by [38, Lemma
A.25].
17
Hilbert B-module `2 b B. We identify LB with the “diagonal subalgebra”
1Mn b LB Ď Mn b LB “ Mn pLB q for each n.
Our goal in this section is to introduce an odd parity version of the
controlled KK-theory groups of the previous section.
Definition 3.1. Let X be a subset of the unit ball of LB , and let ě 0 and
κ ě 1. Let n P N, and define Un,κ, pX, Bq to be the subset of those invertible
`
elements u in Mn pKB q satisfying the following conditions:
If u1 is an element of Un1 ,κ, pX, Bq and u2 is an element of Un2 ,κ, pX, Bq,
define ˜ ¸
u1 0
u1 ‘ u2 :“ P Un1 `n2 ,κ, pX, Bq.
0 u2
Define
8
ğ
U8,κ, pX, Bq :“ Un,κ, pX, Bq,
n“1
(ii) u1 „ u2 if both are elements of the same path component of U8,2κ, pX, Bq.5
1
Finally, define KKκ, pX, Bq to be U8,κ, pX, Bq{ „.
5
Equivalently, both are in the same Un,2κ, pX, Bq, and are in the same path component
of this set. Notice also the switch from κ to 2κ, which is needed for our proof that
1
KKκ, pX, Bq is a group.
18
Lemma 3.2. Let X be a subset of LB , and let ě 0 and κ ě 1. Then the
1
semigroup KKκ, pX, Bq constructed above is an abelian group with neutral
element the class r1s of the unit.
1
Proof. It is clear that KKκ, pX, Bq is a monoid, and the class r1s is neutral
1
by definition. A standard rotation homotopy shows that KKκ, pX, Bq is
commutative. It remains to show that inverses exist. We claim that for
u P Un,κ, pX, Bq, the inverse of the class rus is given by ru´1 s. Indeed, consider
the homotopy
˜ ¸˜ ¸˜ ¸˜ ¸
u 0 cosptq ´ sinptq 1 0 cosptq sinptq
ut :“ ´1
, t P r0, π{2s.
0 1 sinptq cosptq 0 u ´ sinptq cosptq
This connects u ‘ u´1 and 12k , so it suffices to show that this passes through
U2n,2κ, pX, Bq. For the commutator condition, we compute that for x P X,
˜ ¸˜ ¸
rx, us 0 cos2 ptq cosptq sinptq
rx, ut s “ .
0 ru´1 , as cosptq sinptq ´ cos2 ptq
The scalar matrix on the right has norm | cosptq|, and the matrix on the left
has norm at most maxt}rx, us}, }rx, u´1 s}u ď , so }rx, ut s} ď . For the norm
condition, we compute that
˜ ¸˜ ¸ ˜ ¸
u 0 cos2 ptq cosptq sinptq sin2 ptq ´ cosptq sinptq
ut “ ` .
0 ´u´1 cosptq sinptq ´ cos2 ptq cosptq sinptq sin2 ptq
The first scalar matrix appearing above has norm | cosptq|, and the second
has norm | sinptq|. We thus have that }ut } ď κ| cosptq| ` | sinptq|, which is at
most6 2κ as required.
19
Definition 3.3. Let X be a subset of the unit ball of LB , and let ě 0 and
κ ě 1. Define
˚ 0 1
KKκ, pX, Bq :“ KKκ, pX, Bq ‘ KKκ, pX, Bq.
Analogously to the case of KK 0 discussed in line (3) above, if Y and X
are subsets of LB , ě 0, δ ě 0, κ ě 1, and λ ě 1 are such that
Un,λ,δ pY, Bq Ď Un,κ, pX, Bq and Un,2λ,δ pY, Bq Ď Un,2κ, pX, Bq (5)
for all n, then we have a canonical forget control map
1 1
KKλ,δ pY, Bq Ñ KKκ, pX, Bq. (6)
Similarly, for appropriate Y , X, ě 0, δ ě 0, κ ě 1, and λ ě 1 and with
notation as in Definition 3.3 we get a forget control map
˚ ˚
KKλ,δ pY, Bq Ñ KKκ, pX, Bq (7)
by combining the maps in lines (3) and (6).
20
Proposition 4.1. Let X be a subset of the unit ball of LB , κ ě 1, and ě 0.
Let pp0 , qq and pp1 , qq be elements of Pn,κ, pX, Bq, and let u P Un,κ, pX, Bq be
such that up0 u´1 “ p1 . Then the elements pp0 ‘0n , q‘0n q and pp1 ‘0n , q‘0n q
are in the same path component of P2n,κ3 ,3κ2 pX, Bq, and in particular, pp0 , qq
and pp1 , qq define the same class in KKκ03 ,3κ2 pX, Bq.
The analogous statement holds with the roles of the first and second com-
ponents reversed.
Proof. Define
˜ ¸˜ ¸˜ ¸
cosptq ´ sinptq 1 0 cosptq sinptq `
vt :“ P M2n pKB q.
sinptq cosptq 0 u ´ sinptq cosptq
4.2 Normalization
0
Our goal in this subsection is to show that we can assume cycles of KKκ, pX, Bq
1
and KKκ, pX, Bq have prescribed “scalar part”, at least up to some deteri-
oration of κ and . This will be important for our Mayer-Vietoris boundary
maps later in the paper.
Proof. Let r be the associated projection for p as in Definition 2.5. Using the
formulas in Lemma 2.6 part (i), we see that yp1 ´ p ` rq “ p1 ´ p ` rqy “ 1, so
y is invertible with inverse y ´1 “ 1 ´ p ` r. Using Lemma 2.6 part (i) again,
p1 ´ rqp “ 0, whence p˚ p1 ´ rq “ 0 also. Hence y ˚ y “ p1 ´ rq ` p˚ p. As p1 ´ rq
21
and p˚ p are orthogonal positive elements, }y ˚ y} “ maxt}1 ´ r}, }p˚ p}u ď κ2 ,
so }y} ď κ. We see similarly that y ´1 py ´1 q˚ “ p1 ´ pqp1 ´ pq˚ ` r. Corollary
2.4 gives that }1 ´ p} ď κ, so }y ´1 } ď κ similarly. Applying the same process
to q gives the associated projection s and an invertible x with s :“ xqx´1
a projection and such that }x} ď κ and }x´1 } ď κ. As p and q are similar
idempotents, r and s are similar projections, whence [3, Proposition 4.6.5]
gives a unitary w such that wrw˚ “ s. The invertible u :“ x´1 wy has the
claimed property.
22
As c has spectrum contained in rκ´1 , κs, and as γ stays at least 1{p2κq
away from that interval, the continuous functional calculus implies that
}pz ´ cq´1 } ď 2κ for z in the image of γ, and similarly for d. As the length of
γ is at most 6κ and as the largest value |f | takes on the image of γ is p2κq1{2 ,
we see from line (8) that
1
}c´1{2 ´ d´1{2 } ď 6κ ¨ p2κq1{2 ¨ 2κ ¨ 2κ}c ´ d},
2π
which completes the proof.
23
It remains to show that the path r0, t1 s Ñ C defined by t ÞÑ wt is p46 Lq-
Lipschitz.
We first note that for s, t P r0, t1 s, we have that
by assumption that ppt q is L-Lipschitz. Using that }xt } ď 2, this implies that
the map t ÞÑ x˚t xt is at most p8Lq-Lipschitz. As }xt } ď 2 and }x´1 t } ď 2, we
˚
have that 1{4 ď xt xt ď 4. Hence Lemma 4.4 (with κ “ 4) implies that the
path t ÞÑ px˚t xt q´1{2 is p8 ¨ 44 Lq-Lipschitz. Finally, this implies that the path
t ÞÑ wt is at most p8 ¨ 44 L ` 2Lq-Lipschitz, and we are done.
Recall that for l P t1, ..., nu, we let 1l P Mn pCq be the rank l projection
with l ones in the top-left part of the diagonal and zeros elsewhere.
1
Define P8,κ, pX, Bq to be the disjoint union of these sets as n ranges over N.
Here is the first of our main goals for this subsection: it allows control of
0
the “scalar part” of cycles for KKκ, pX, Bq.
(i) Any element v P Pn,κ, pX, Bq is in the same path component of Pn,κ5 , pX, Bq
1
as an element of Pn,κ 5 , pX, Bq.
1
(ii) If two elements pp0 , q0 q and pp1 , q1 q of Pn,κ, pX, Bq are connected by an
L-Lipschitz path in Pn,κ, pX, Bq, then they are connected by a p49 κ3 Lq-
1
Lipschitz path in Pn,κ,8κ 3 pX, Bq.
24
Lemma 4.7. For any κ ě 1, the set tu P GLn pCq | }u} ď κ and }u´1 } ď κu
is path connected.
2 ´1 2
by the C ˚ -identity and the functional calculus. We see }u´1 t } ď }u }
similarly. Using that the unitary group of Mn pCq is connected, connect u1{2
to the identity by a path of unitaries put qtPr1{2,1s . The concatenated path
put qtPr0,1s then connects u to the identity and passes through the set in the
statement.
`
Proof of Proposition 4.6. Let σ : Mn pKB q Ñ Mn pCq be the canonical quo-
tient map. For part (i), as pp, qq is in Pn,κ, pX, Bq, the classes rσppqs and
rσpqqs in K0 pCq are the same. Hence there is l P N such that σppq and σpqq
both have rank l, and so there exist invertible elements u, v P Mn pCq such
that uσppqu´1 “ 1l and such that vσpqqv ´1 “ 1l . Moreover, by Lemma 4.2,
we may assume that }u} ď κ2 and }v} ď κ2 , and similarly for u´1 and v ´1 .
Using Lemma 4.7, we get paths put qtPr0,1s and pvt qtPr0,1s of invertibles in
Mn pCq connecting u and v respectively to the identity, and such that all
elements of these paths and their inverses have norm at most κ2 . To com-
plete the proof, consider the path pput pu´1 ´1
t , vt qvt qqtPr0,1s . As scalar matrices
commute with X, this passes through Pn,κ5 , pX, Bq, and connects pp, qq to
1
an element of Pn,κ 5 , pX, Bq as claimed.
For part (ii), say pp0 , q0 q and pp1 , q1 q are as given. Then there exists l P N
such that σpp0 q “ σpq0 q “ 1l “ σpp1 q “ σpq1 q. Let r0 be the projection
associated to p0 as in Definition 2.5. As in Lemma 2.6, part (ii), the path
t ÞÑ p1 ´ tqp0 ` tr0 , t P r0, 1s is κ-Lipschitz and connects p0 and r0 through
idempotents of norm at most κ. Moreover, Lemma 2.6, part (iii) implies that
}rp1 ´ tqp0 ` tr0 , xs} ď 8κ3 for all x P X and all t P r0, 1s. Note also that
25
σpp1 ´ tqp0 ` tr0 q “ 1l for all t. Similarly, we get s0 which has the same
properties with respect to q0 . We have thus shown that pp0 , q0 q is connected
1
to the element pr0 , s0 q via a κ-Lipschitz path in Pn,κ,8κ 3 pX, Bq. Completely
we have that pr0 , s0 q and pr1 , s1 q are connected by a p6κ3 Lq-Lipschitz path
of projections in Pn,1,8κ3 pX, Bq, say pprt , st qqtPr0,1s .
Now, consider the path pσprt q, σpst qqtPr0,1s in Mn pCq ‘ Mn pCq, which is
p6κ Lq-Lipschitz. Lemma 4.3 gives p46 ¨ 6κ3 Lq-Lipschitz paths put qtPr0,1s and
3
pvt qtPr0,1s of unitaries in Mn pCq such that σprt q “ ut σpr0 qu˚t and σpst q “
vt σps0 qvt˚ for all t P r0, 1s. The path ppu˚t rt ut , vt˚ st vt qqtPr0,1s then passes
1 6 3
through Pn,1,8κ 3 pX, Bq, is p2 ¨ 4 ¨ 6κ Lq-Lipschitz, and connects pr0 , s0 q to
pu˚1 r1 u1 , v1˚ s1 v1 q.
Summarizing where we are, we have the following paths
1
(i) A κ-Lipschitz path through Pn,κ,8κ 3 pX, Bq, parametrized by r0, 1s, and
1
(iii) A κ-Lipschitz path through Pn,κ,8κ 3 pX, Bq, parametrized by r0, 1s, and
26
We now move on to controlling the “scalar part” of cycles for KK 1 .
1
Define U8,κ, pX, Bq to be the disjoint union of these sets as n ranges over N.
(i) Any element v P Un,κ, pX, Bq is in the same path component of Un,κ2 , pX, Bq
1
as an element of Un,κ 2 , pX, Bq.
1
(ii) If two elements v0 , v1 P Un,κ, pX, Bq are in the same path component of
1
Un,κ, pX, Bq, then they are in the same path component of Un,κ 2 , pX, Bq.
`
Proof. For part (i), let σ : Mn pKB q Ñ Mn pCq be the canonical quotient, and
´1
set w “ σpu q. Using Lemma 4.7, there is a path pwt qtPr0,1s of invertibles
connecting w “ w1 to the identity and all with norm at most κ. Then the
path pwt vqtPr0,1s is in Un,κ2 , pX, Bq and connects v to the element u :“ w1 v,
which satisfies σpuq “ 1, and so 1 ´ u P Mn pKB q.
For part (ii), let pvt qtPr0,1s be a path in Un,κ, pX, Bq connecting v0 and
v1 . Let wt “ σpvt´1 q, and note that w0 “ w1 “ 1. Moreover, }wt } ď κ for
1
all t. Then ut :“ wt vt is a path connecting v0 and v1 in Un,κ 2 , pX, Bq as
required.
27
Lemma 4.10. Let κ ě 1, and let p0 and p1 be idempotents in a C ˚ -algebra
C with norm at most κ, and such that }p0 ´ p1 } ď 1{p12κ2 q. Then there is
a path ppt qtPr0,1s of idempotents connecting p0 and p1 , and with the following
properties:
Proof. For each t P r0, 1s, define rt :“ p1 ´ tqp0 ` tp1 P C, and define ut :“
p1 ´ rt qp1 ´ p0 q ` rt p0 P C ` . We compute that 1 ´ ut “ p2p0 ´ 1qpp0 ´ rt q.
Hence by Corollary 2.4, }1 ´ ut } ď 2κ}p0 ´ p1 } ď 1{6. Hence ut is invertible,
}ut } ď 7{6, and (by the usual Neumann series representation of the inverse),
}u´1 ´1
t } ď 6{5. We thus see that each pt :“ ut p0 ut is a well-defined idempotent
in C. We claim that the path ppt qtPr0,1s has the desired properties. Note
first that r0 “ p0 , whence u0 “ 1, and so q0 “ p0 . On the other hand,
u1 p0 “ r1 p0 “ p1 p0 “ p1 u1 . Hence u1 p0 u´1 1 “ p1 . Thus the path ppt q really
does connect p0 and p1 .
For part (i), note that as ut p0 “ rt p0 , we get
1 6 6
}pt } “ }rt p0 u´1 ´1 ´1
t } ď }prt ´ p0 qp0 ut } ` }p0 ut } ď 2
κ ` κ ď 2κ.
12κ 5 5
For part (ii), we compute using the identity 1 ´ ut “ p2p0 ´ 1qpp0 ´ rt q that
as claimed. Finally, for part (iii), we compute using Corollary 2.4 that
1
}us ´ ut } “ }p2p0 ´ 1qprs ´ rt q} ď }2p0 ´ 1}|s ´ t|}p0 ´ p1 } ď 2κ|s ´ t|
12κ2
1
“ |s ´ t|
6κ
and so
36 1 6
}u´1 ´1 ´1 ´1
s ´ ut } “ }ut put ´ us qus } ď |s ´ t| “ |s ´ t|.
25 6κ 25κ
Hence
as claimed.
The next lemma allows us to control the “speed” of a homotopy (at the
price of moving to larger matrices). The basic idea is not new: see for
example [24, Proposition 1.31]. We give a complete proof, however, as we
need to incorporate commutator estimates.
29
Proof. Let ppt , qt q be an arbitrary homotopy in Pn,κ, pX, Bq connecting pp0 , q0 q
and pp1 , q1 q. Let δ ą 0 be such that if s, t P r0, 1s satisfy |s ´ t| ď δ, then
}ps ´ pt } ď 1{p12κ2 q and }qs ´ qt } ď 1{p12κ2 q. Let 1 “ t0 ă t1 ă .... ă tk “ 1
be a sequence of points in r0, 1s such that ti`1 ´ ti ď δ for all i. We claim
that this k works, and to show this we build an appropriate homotopy by
concatenating the various steps below.
Using that the norm of the (second) scalar matrix appearing above is
| cosptq|, one checks that this connects the result of the previous stage
to
(iii) From Corollary 2.4, each idempotent 1 ´ pti has norm at most κ. For
each i P t1, ..., ku, using that }p1 ´ pti q ´ p1 ´ pti´1 q} ď 1{p12κ2 q, Lemma
30
4.10 gives a path of idempotents connecting 1´pti and 1´pti´1 and with
the following properties: it is 1-Lipschitz; it consists of idempotents of
norm at most 2κ; each idempotent r in the path satisfies }rr, xs} ď 21κ2
for all x P X. We get similar paths with respect to the elements 1 ´ qti ,
and use these paths to connect the result of the previous stage to
(iv) Use an analog of the homotopy in step (ii) in each block of the form
pti ‘ 1 ´ pti (and similarly for q) to connect this to
` ˘
p1n ‘ 0n q ‘ ¨ ¨ ¨ ‘ p1n ‘ 0n q ‘ptk , p1
loooooooooooooooomoooooooooooooooon n ‘ 0n q ‘ ¨ ¨ ¨ ‘ p1n ‘ 0n q ‘qtk .
loooooooooooooooomoooooooooooooooon
k times k times
(v) Finally, noting that ptk “ p1 and qtk “ q1 , use a rotation homotopy
parametrized by r0, π{2s to connect this to pp1 ‘1nk ‘0nk , q1 ‘1nk ‘0nk q.
This passes through Pp2k`1qn,κ, pX, Bq and is 2κ-Lipschitz.
Concatenating the five homotopies above gives a 2κ-Lipschitz homotopy,
parametrized by r0, 2π ` 1s, that passes through Pp2k`1qn,2κ, pX, Bq and con-
nects pp0 ‘ 1nk ‘ 0nk , q0 ‘ 1nk ‘ 0nk q and pp1 ‘ 1nk ‘ 0nk , q1 ‘ 1nk ‘ 0nk q.
Reparametrizing by r0, 1s, we get a p16κq-Lipschitz homotopy as required.
Before we get to the main result of this subsection, we give one more
elementary lemma; we record it as it will be used multiple times below.
Lemma 4.12. Say x and y1 , ..., yn are elements of a Banach algebra such
that }rx, yi s} ď δ and }yi } ď m for all i. Then if y :“ y1 y2 ¨ ¨ ¨ ym , we have
}rx, ys} ď nmn´1 δ.
Proof. This follows from the formula
n ´ ź
ÿ ¯ ´ ź ¯
rx, ys “ yj rx, yi s yj ,
i“1 1ďjăi iăjďn
31
which itself follows from induction on n and the usual Leibniz formula rx, y1 y2 s “
y1 rx, y2 s ` rx, y1 sy2 .
Here is the main result of this subsection, which says that homotopic
idempotents are similar, while keeping some (alarmingly bad!) control on
commutator and norm estimates. The basic idea of the proof is contained in
[24, Corollary 1.32], but as usual we need to do a bit more work in order to
get our estimates.
upp ‘ 1m ‘ 0m qu´1 “ q ‘ 1m ‘ 0m .
It follows that each vi is invertible, }vi } ď 2, and (by the Neumann series
formula for the inverse) }vi´1 } ď 2. Note also that as the homotopy pppt , qt qq
32
1
passes through Pp2k`1qn,2κ,200κ5 pX, Bq all the elements pti must have the same
`
scalar part (i.e. the same image under the canonical map Mn`2m pKB q Ñ
Mn`2m pCq), and so the elements vi must satisfy 1 ´ vi P Mn`2m pKB q. More-
over, for x P X, using line (9) again we see that
Hence moreover
1
At this point we have that each vi is an element of Un`2m,2,p10κq 6.
´1
Note also that vi pti “ pti´1 pti “ pti´1 vi , and so vi pi vi “ pti´1 for each
i. Define v to be the product v1 v2 ¨ ¨ ¨ vN , so v satisfies v ´1 p0 v “ p1 , or
in other words v ´1 pp ‘ 1m ‘ 0m qv “ 1l ‘ 1m ‘ 0m . Note that 1 ´ v P
Mn`2m pKB q. As }vi } ď 2, as }vi´1 } ď 2 for each i, we have that }v} ď 2N and
similarly }v ´1 } ď 2N . Moreover, for any x P X, Lemma 4.12 gives }rv, xs} ď
N 2N ´1 ¨ p10κq6 and similarly }rv ´1 , xs} ď N 2N ´1 ¨ p10κq6 . Applying the
same construction with pqt q in place of ppt q, we get an invertible element w
such that w´1 pq ‘ 1m ‘ 0m qw “ 1l ‘ 1m ‘ 0m , such that 1 ´ w P Mn`2m pKB q,
such that }w} ď 2N , }w´1 } ď 2N , and such that }rw, xs} ď N 2N ´1 ¨ p10κq6
and }rw´1 , xs} ď N 2N ´1 ¨ p10κq6 for all x P X. Define u “ wv ´1 . As
N “ r413 κ4 s, this has the claimed properties.
33
Our goal in this section is to construct and analyse a“Mayer-Vietoris
boundary map” in controlled KK-theory. The main results of the section
prove the existence of this boundary map (Proposition 5.1) and show it has
an exactness property (Proposition 5.5).
Here is the promised construction of a boundary map.
(i) With w P Un,κ, php1 ´ hqX Y thu, Bq, use Proposition 4.9 part (i) to
1
find an element u P Un,κ2 , php1 ´ hqX Y thu, Bq that is in the same path
(ii) Define
in Mn pLB q, and
˜ ¸˜ ¸˜ ¸˜ ¸
1 c 1 0 1 c 0 1
v “ vpu, hq :“ P M2n pLB q. (11)
0 1 ´d 1 0 1 ´1 0
(iii) Define « ˜ ¸ ˜ ¸ ff
1 0 ´1 1 0
Brws :“ v v , .
0 0 0 0
In order to make the proof more palatable, we split off some computa-
tions as two technical lemmas. The arguments given for these lemmas are
elementary, but quite lengthy. We record them for the sake of completeness,
but recommend the reader skips the proofs.
34
Lemma 5.2. Let u P Mn pLB q be an invertible element such that 1 ´ u P
Mn pKB q, and let h P LB be a positive contraction. Then the elements c “
cpu, hq and d “ dpu, hq from line (10) above have the following properties.
Proof. We just look at the case of cd ´ 1 for both parts; the case of dc ´ 1
is similar. Note first that because 1 ´ u is in Mn pKB q, we must have that
1 ´ u´1 is in Mn pKB q also. We compute that
Using that u and u´1 equal 1 modulo the ideal Mn pKB q, we compute that
this equals 0 modulo Mn pKB q. Hence cd ´ 1 is in Mn pKB q
Looking at part (ii), note that the terms hurh, u´1 s and rh, usp1 ´ hq in
line (12) above have norms at most κ and respectively. Hence cd ´ 1 is
within pκ ` 1q of h2 ` hp1 ´ hqu´1 ` hp1 ´ hqu ´ 2h ` h2 , which equals
hp1 ´ hqpu ` u´1 ´ 2q.
35
Proof. From the definition of v in line (11) above,
˜ ¸
cpdc ´ 2q 1 ´ cd
v“
dc ´ 1 ´d
and
˜ ¸˜ ¸˜ ¸˜ ¸ ˜ ¸
0 ´1 1 ´c 1 0 1 ´c ´d dc ´ 1
v ´1 “ “ .
1 0 0 1 d 1 0 1 1 ´ cd cpdc ´ 2q
Hence ˜ ¸ ˜ ¸
1 0 ´1 cdp2 ´ cdq cpdc ´ 2qpdc ´ 1q
v v “
0 0 p1 ´ dcqd pdc ´ 1q2
and so
˜ ¸ ˜ ¸ ˜ ¸
1 0 ´1 1 0 ´pcd ´ 1q2 pcd ´ 1qcpdc ´ 2q
v v ´ “ . (13)
0 0 0 0 p1 ´ dcqd pdc ´ 1q2
This formula, part (i) of Lemma 5.2, and the fact that Mn pKB q is an ideal
in Mn pLB q imply that
˜ ¸ ˜ ¸
1 0 ´1 1 0
v v ´ P M2n pKB q,
0 0 0 0
˜ ¸ ˜ ¸ ˜ ¸
1 0 ´1 ` 1 0 1 0
whence v v is in M2n pKB q, and v v ´1 and define
0 0 0 0 0 0
the same element (so in particular, the same K-theory class) under the image
`
of the canonical quotient map σ : M2n pKB q Ñ M2n pCq. Note moreover that
3 ´1 3
}v} ď pκ ` 2q and }v } ď pκ ` 2q . As κ ě 1, this implies that
› ˜ ¸ ›
› 1 0 ›
›v v › ď pκ ` 2q6 ď 36 κ6 .
´1 ›
›
› 0 0 ›
36
defines an element of P2n,36 κ6 ,216 κ5 pX, Bq it remains to show that
›« ˜ ¸ ˜ ¸ ff›
› 1 0 1 0 ›
› x, v v ´1 ´ › ď 216 κ5 . (14)
› ›
› 0 0 0 0 ›
for all x P X Y thu. We focus on the case when x is in X: the case when
x “ h follows from similar (and much simpler) estimates that we leave to the
reader.
Working towards the estimate in line (14), we compute that the element
in line (13) equals
˜ ¸˜ ¸
cd ´ 1 0 1 ´ cd cpdc ´ 2q
. (15)
0 dc ´ 1 ´d dc ´ 1
On the other hand, using part (ii) of Lemma 5.2, the first matrix in line (15)
above is within pκ ` 1q of hp1 ´ hqpu ` u´1 ´ 2q (we identify this as usual
with the diagonal matrix with both entries equal to hp1 ´ hqpu ` u´1 ´ 2q).
Hence the difference in line (13) is within 8pκ ` 1q5 of
˜ ¸
1 ´ cd cpdc ´ 2q
hp1 ´ hqpu ` u´1 ´ 2q .
´d dc ´ 1
37
Hence for x P X,
›« ˜ ¸ ˜ ¸ ff›
› 1 0 ´1 1 0 ›
› x, v v ´
› ›
›
› 0 0 0 0 ›
›« ˜ ¸ ff›
5
›
´1 1 ´ cd cpdc ´ 2q ›
ď 16pκ ` 1q }x} ` › x, hp1 ´ hqpu ` u ´ 2q ›.
› ›
› ´d dc ´ 1 ›
(17)
As }rx, hs} ď , we have }rx, hp1 ´ hqs} ď 2; combining this with line (16)
gives
›« ˜ ¸ ff›
›
´1 1 ´ cd cpdc ´ 2q ›
x, hp1 hqpu u 2q
› ›
› ´ ` ´ ›
› ´d dc ´ 1 ›
› « ˜ ¸ ff›
› 1 ´ cd cpdc ´ 2q ›
ď 2 ¨ 8pκ ` 1q5 ` ›hp1 ´ hq x, pu ` u´1 ´ 2q ›.
› ›
› ´d dc ´ 1 ›
Combining this with line (17) gives
›« ˜ ¸ ˜ ¸ ff›
› 1 0 ´1 1 0 ›
› x, v v ´
› ›
›
› 0 0 0 0 ›
› « ˜ ¸ ff›
5
›
´1 1 ´ cd cpdc ´ 2q ›
ď 32pκ ` 1q ` ›hp1 ´ hq x, pu ` u ´ 2q ›.
› ›
› ´d dc ´ 1 ›
(18)
˜ ¸
1 ´ cd cpdc ´ 2q
Every entry of the matrix pu ` u´1 ´ 2q can be written
´d dc ´ 1
as a sum of at most 30 terms, each of which is a product of at most 5
elements from the set tu, u´1 , h, 1u, each of which has norm at most κ. As
}rhp1 ´ hqx, ys} ď for all y P tu, u´1 , h, 1u, Lemma 4.12 gives
›« ˜ ¸ ff›
› 1 ´ cd cpdc ´ 2q ›
› hp1 ´ hqx, pu ` u´1 ´ 2q › ď 4 ¨ 30 ¨ 5 ¨ κ4 . (19)
› ›
› ´d dc ´ 1 ›
On the other hand, }rhp1 ´ hq, ys} ď 2 for all y P tu, u´1 , h, 1u, whence
›« ˜ ¸ ff ›
› 1 ´ cd cpdc ´ 2q ›
´1
› hp1 ´ hq, pu ` u ´ 2q x› ď 4 ¨ 30 ¨ 5 ¨ κ4 . (20)
› ›
› ´d dc ´ 1 ›
38
Finally, note that
« ˜ ¸ ff
1 ´ cd cpdc ´ 2q
hp1 ´ hq x, pu ` u´1 ´ 2q
´d dc ´ 1
« ˜ ¸ ff
1 ´ cd cpdc ´ 2q
“ hp1 ´ hqx, pu ` u´1 ´ 2q
´d dc ´ 1
« ˜ ¸ ff
1 ´ cd cpdc ´ 2q
` hp1 ´ hq, pu ` u´1 ´ 2q x,
´d dc ´ 1
Proof of Proposition 5.1. Assume that w P Un,κ, php1 ´ hqX Y thu, Bq, and
1
let u P Un,κ 2 , php1 ´ hqX Y thu, Bq be in the same path component as w in
Un,κ2 , php1 ´ hqX Y thu, Bq; u is guaranteed to exist Proposition 4.9 part (i).
Define v :“ vpu, hq as in line (11), so Lemma 5.3 gives an element
˜ ˜ ¸ ˜ ¸¸
1 0 ´1 1 0
v v , P P2n,36 κ12 ,216 κ10 pX Y thu, Bq.
0 0 0 0
1
Moreover, if u0 :“ u, and u1 is another choice of element in Un,κ 2 , php1´hqX Y
39
in P2n,36 κ24 ,216 κ20 pXYthu, Bq. In particular, the class Brws P KK306 κ24 ,216 κ20 pXY
thu, Bq does not depend on the choice of u, so at this point we have a well-
defined set map
Un,κ, php1 ´ hqX Y thu, Bq Ñ KK306 κ24 ,216 κ20 pX Y thu, Bq.
We next claim that this map sends block sums on the left to sums on the
right.
For this, assume that w1 and w2 are elements of Un,κ, php1´hqX Ythu, Bq.
1
Let u1 and u2 be elements of Un,κ 2 , php1 ´ hqX Y thu, Bq that are connected
1
to w1 and w2 respectively in Un,κ2 , php1 ´ hqX Y thu, Bq. For i P t1, 2u let
vi “ vpui , hq be as in line (11), and let v :“ vpu1 ‘ u2 , hq P M4n pLB q. Then
the pairs
˜ ˜ ¸ ˜ ¸ ˜ ¸ ˜ ¸¸
1n 0 ´1 1n 0 ´1 1n 0 1n 0
v1 v ‘ v2 v , ‘
0 0 1 0 0 2 0 0 0 0
and ˜ ˜ ¸ ˜ ¸¸
12n 0 ´1 12n 0
v v ,
0 0 0 0
in M4n pLB q‘M4n pLB q differ by conjugation by the same (scalar) permutation
matrix in each component, and so define the same class in KK306 κ24 ,216 κ20 pX Y
thu, Bq.
At this point, we have a monoid homomorphism
Un,κ, php1 ´ hqX Y thu, Bq Ñ KK306 κ24 ,216 κ20 pX Y thu, Bq.
1
We claim that it respects the equivalence relation defining KKκ, php1´hqX Y
thu, Bq. First, we check that w‘1k goes to the same class as w. As we already
know we have a monoid homomorphism, it suffices to show that 1k goes to
zero in KK306 κ24 ,216 κ20 pX Y thu, Bq. For this, note that if v :“ vp1k , hq is as in
line (11), then v “ 12k , whence the image of 1k in KK306 κ24 ,216 κ20 pX Y thu, Bq
is the class r1k ‘ 0k , 1k ‘ 0k s, which is zero by definition.
Let us now show that elements of Un,κ, php1 ´ hqX Y thu, Bq that are
homotopic through Un,2κ, php1 ´ hqX Y thu, Bq go to the same class. For
40
this, say that w0 and w1 are homotopic through Un,2κ, php1 ´ hqX Y thu, Bq.
1
Choose u0 and u1 in Un,κ 2 , php1 ´ hqX Y thu, Bq that are connected to w0
and w1 in Un,κ2 , php1 ´ hqX Y thu, Bq as in Proposition 4.9 part (i). Using
Proposition 4.9 part (ii), u0 and u1 are connected by a homotopy put qtPr0,1s
1
in Un,4κ4 , php1 ´ hqX Y thu, Bq. Let vt :“ vput , hq be as in line (11). Then
41
exists, and moreover has the following “exactness” property.
Let P r0, δ{2s, and let κ P r1, λs. Assume that Yh , Y1´h , and Y are self-
adjoint subsets of the unit ball of LB such that Y Ď Yh and Y Ď Y1´h . With
1
notation as in Definition 4.5, assume that pp, qq P Pn,κ, pX Y Yh Y Y1´h , Bq.
Assume that h P LB is a positive contraction such that }rh, xs} ď for all
0
x P X Y tp, qu. Let rp, qsh and rp, qs1´h be the classes in KKκ,2 phX Y Yh , Bq
0
and KKκ,2 pp1 ´ hqX Y Y1´h , Bq respectively given by Lemma 5.4. Assume
that the images of rp, qsh and rp, qs1´h under the respective forget control maps
0 0
KKκ,2 phX Y Yh , Bq Ñ KKλ,δ phX Y Yh , Bq
and
0 0
KKκ,2 pp1 ´ hqX Y Y1´h , Bq Ñ KKλ,δ pp1 ´ hqX Y Y1´h , Bq
of line (3) are zero. Then with notation as in Definition 4.8, there exists
1
an element u P U8,N 1 ,N1 δ
php1 ´ hqX Y thu Y Y q such that Brus “ rp, qs in
0
KKN2 ,N2 δ pX Y thu, Bq.
Just as for the proof of Proposition 5.1, to make the proof more palatable,
we split off some computations as two technical lemmas. As in the previ-
ous case, the arguments we give for these lemmas are elementary, but quite
lengthy (in fact, much longer than the earlier ones). We record them for the
sake of completeness, but again recommend that the reader skips the proofs.
Proof. We split the computations into the points labeled (i), (ii), (iii), (iv),
and (v) below.
42
(i) As uh ´1 P Mn pKB q and u1´h ´1 P Mn pKB q, we see that u´1 P Mn pKB q.
Completely analogously,
}ry, ps} ď }ry ´ yh , ps} ` }ryh , ps} ď }p} ` 2 ď pκ ` 2q ď 3κ. (23)
Putting together the estimates in lines (21), (22), (23), and (24), we see
that
}ru, hs} ď }rh, u1´h s}}1´p}`}u1´h }}rh, 1´ps}`}rh, uh s}}p}`}uh }}rh, ps}.
43
As }ru1´h , hs} ď γ, as }ruh , hs} ď γ, and as }rp, hs} ď , this implies
that
}ru, hs} ď }1 ´ p}γ ` }u1´h } ` κγ ` }uh }.
As }p} ď κ (whence }1 ´ p} ď κ by Corollary 2.4), as }uh } ď ν, as
}u1´h } ď ν this implies that
}rhp1 ´ hqx, ps} ď }rx, ps} ` }r1 ´ h, ps} ` }rh, ps} ď 3.
}rhp1 ´ hqx, us} ď }rhp1 ´ hqx, u1´h s}}1 ´ p} ` }u1´h }}rhp1 ´ hqx, 1 ´ p}
` }rhp1 ´ hqx, uh s}}p} ` }uh }}rhp1 ´ hqx, p}
ď 2}1 ´ p}γ ` }u1´h }3 ` 2}p}γ ` }uh }3.
44
Putting the points (i), (ii), (iii), (iv), and (v) above together we see indeed
1
that (with notation as in Definition 4.8), u is an element of Un,2ν 2 ,12ν 2 γ php1 ´
This satisfies the same norm estimates as w, and so we get the norm estimates
Lemma 5.3 and the fact that }u} ď 2ν 2 implies that }v} ď p2ν 2 ` 2q3 and
}v ´1 } ď p2ν 2 ` 2q3 . As ν ě 1, we thus see that
45
From lines (28) and (29), we thus see that
S :“ th, 1 ´ h, p, q, 1 ´ p, 1 ´ q, uh , u´1 ´1 ´1
h , u1´h , u1´h , u, u , c, du, (31)
Note moreover that for every element s P S we have that oat least ne of the
following holds: (a) }rs, xs ď 16ν 2 γ; or (b) }rs, p1 ´ hqxs} ď 16ν 2 γ; or (c)
}rs, p1 ´ hqxs} ď 16ν 2 γ; or (d) }rs, hp1 ´ hqxs} ď 16ν 2 γ. In any of these cases,
using that }rs, hs} ď 12ν 2 γ, we get that for any s P S, }rs, hp1 ´ hqxs} ď
40ν 2 γ. Applying Lemma 4.12 again, we therefore see that
whence ˜ ¸
rcdc, hs ´ 2rc, hs rcd, hs
rh, vs “ .
rh, dcs rd, hs
and so
}rh, vs} ď }rcdc, hs} ` 2}rc, hs} ` }rcd, hs} ` }rh, dcs} ` }rd, hs}.
46
Each summand on the right hand side above is of the form }rh, ss} where
s P S 3 for S as in line (31). Hence line (32) implies that
whence
Combining this with lines (28), (29), (34), and (35), as well as that ν ě 1,
we see that
Now let us look at rx, vw´1 s for x P X. The definition of v from line (11)
gives
˜ ¸ ˜ ¸
cpdc ´ 1q 1 ´ cd c 0
vw´1 “ w´1 ´ w´1
dc ´ 1 0 0 d
˜ ¸˜ ¸ ˜ ¸
cd ´ 1 0 c ´1 c 0
“ w´1 ´ w´1 .
0 dc ´ 1 1 0 0 d
47
Hence the formula for w´1 from line (27) gives
˜ ¸˜ ¸
´1
cd ´ 1 0 cp1 ´ pqu 1´h cp ´ u 1´h p1 ´ pq
vw´1 “
0 dc ´ 1 p1 ´ pqu´11´h p
looooooooooooooooooooooooooooooooooomooooooooooooooooooooooooooooooooooon
y1
˜ ¸
1 ´ q uh p
´h
´u´1h q 1´p
looooooooooomooooooooooon
y2
˜ ¸
p1 ´ pqu´1 1´h p
´ p1 ´ hq .
´q u1´h p1 ´ pq
looooooooooooooooooooooomooooooooooooooooooooooon
y3
We now estimate }rvw´1 , xs} for each x P X by looking at each of the terms
y1 , y2 , and y3 separately.
(i) First, we look at y1 . Let x P X. Lemma 5.2 implies that the matrix
›˜ ¸ ›
› cd ´ 1 0 ›
´ hp1 ´ hqpu ` u´1 ´ 2q› ď pκ ` 1q (37)
› ›
›
› 0 dc ´ 1 ›
´1
(where, as usual, we identify hp1´hqpu`u
˜ ´2q with the corresponding
¸
´1
cp1 ´ pqu1´h cp ´ u1´h p1 ´ pq
diagonal matrix). Let z1 :“ . Corol-
p1 ´ pqu´1
1´h p
lary 2.4 gives that }1 ´ p} ď κ, whence
}z1 } ď 9ν 4 . (38)
48
Combining this with line (37), we see that
}rx, y1 s} ď }rx, y1 ´ hp1 ´ hqpu ` u´1 ´ 2qz1 s} ` }rx, hp1 ´ hqpu ` u´1 ´ 2qz1 s}
ď p2νq5 γ ` }rx, hp1 ´ hqpu ` u´1 ´ 2qz1 s}.
whence
}rrx, hp1 ´ hqs, pu ` u´1 ´ 2qz1 s} ď 28 ν 6 γ. (41)
Combining lines (39), (40), and (41) thus implies that
49
Now, as we have observed above already, every entry in the matrix
pu ` u´1 ´ 2qz1 is a sum of at most 8 elements from the set S 4 , where
S is as in line (31). Hence from line (33),
}rx, y1 s} ď 220 ν 8 γ.
We have that
« ˜ ¸ ff ˜ ¸
1 ´ q uh p rq, hs rh, uh ps
h, “ .
´u´1
h q 1´p ru´1
h q, hs rp, hs
50
As }rh, xs} ď ď γ, we have
›« ˜ ¸ ff› ›˜ ¸›
› 1 ´ q uh p › › 1´q u p ›
h
› rx, hs, › ď 2γ › ›.
› › › ›
´1 ´1
› ´uh q 1 ´ p › › ´uh q 1 ´ p ›
Combining this with lines (44) and (46) therefore implies that
}rx, y2 s} ď 210 ν 4 γ.
(iii) This can be handled very similarly to the case of y2 , giving the estimate
}rx, y3 s} ď 210 ν 4 γ for all x P X.
Putting together the concluding estimates of points (i), (ii), and (ii) above,
we see that }rx, vw´1 s} ď 221 ν 8 γ for all x P X. Combining this with line
(36), we see that
}rx, vw´1 s} ď 221 ν 10 γ (47)
51
for all x P X Y thu.
To complete the proof, let us estimate }rx, wv ´1 s} for x P X Y thu. Using
the formula rx, wv ´1 s “ wv ´1 rvw´1 , xswv ´1 , we see that
}rx, wv ´1 s} ď 237 ν 26 γ
and
1
u1´h P Um`2j,M,M δ pp1 ´ hqX Y thu Y Y1´h , Bq
such that
uh pp ‘ 1k ‘ 1j ‘ 0j qu´1
h “ q ‘ 1k ‘ 1j ‘ 0j
and
u1´h pp ‘ 1k ‘ 1j ‘ 0j qu´1
1´h “ q ‘ 1k ‘ 1j ‘ 0j .
52
thanks to the definition of the controlled KK 0 groups, so this makes no real
difference. Define now
u :“ u1´h p1 ´ pq ` uh p. (48)
of Proposition 5.1 exists and such that Brus “ rp, qs in KKN0 2 ,N2 δ pX Ythu, Bq.
We claim that N2 :“ 9p237 M 26 q9 works.
Increasing N2 if necessary7 , we may assume it is large enough compared
to N1 “ 12M 3 that Proposition 5.1 applies. Hence the homomorphism in
line (49) exists and if v :“ vpu, hq is as in line (11), we have
« ˜ ¸ ˜ ¸ ff
1 0 ´1 1 0
Brus “ v v , .
0 0 0 0
Define now
˜ ¸
u1´h p1 ´ pq ´q
w :“ P M2n pLB q. (50)
p p1 ´ pqu´1
1´h
53
0
Proposition 4.1 implies that in KKM 3 3 pX Y thu, Bq
1 ,3M1 δ
« ˜ ¸ ˜ ¸ ff
1 0 ´1 1 0
Brus “ v v ,
0 0 0 0
« ˜ ¸ ˜ ¸ ff
1 0 1 0
“ pvw´1 q´1 v v ´1 pvw´1 q ,
0 0 0 0
« ˜ ¸ ˜ ¸ ff
1 0 1 0
“ w w´1 , .
0 0 0 0
Computing, we see that
˜ ¸ ˜ ¸
1 0 1 ´ q 0
w w´1 “ ,
0 0 0 p
whence «˜ ¸ ˜ ¸ ff
1´q 0 1 0
Brus “ , (51)
0 p 0 0
0
in the group KKM 3 3 pX Y thu, Bq.
1 ,3M1 δ ˜ ¸
1´q q `
Note now that the matrix P M2n pKB q has norm at most
q 1´q
2κ (as }q} ď κ, and so }1 ´ q} ď κ by Corollary 2.4), and that it satisfies
›« ˜ ¸ ff›
› 1´q q ›
› x, ›ď
› ›
› q 1´q ›
˜ ¸
1´q q
for all x P X Y thu. Hence P U2n,2κ, pX Y thu, Bq. Applying
q 1´q
Proposition 4.1 again and using that κ ď M1 and ď γ and
˜ ¸˜ ¸˜ ¸
1´q q 1 0 1´q q
q 1´q 0 0 q 1´q
shows that this class is the same as
«˜ ¸ ˜ ¸ ff
1´q 0 1´q 0
,
0 p 0 q
0
in KKM 6 9 pX Y thu, Bq. Using a rotation homotopy, this is the same as
1 ,9M1 δ
0
rp, qs by definition of KKM 6 ,9M 9 δ pX Y thu, Bq and we are done.
1 1
54
6 Finite dimensional algebras
In this section (as throughout), B is a separable C ˚ -algebra, and LB and
KB are respectively the adjointable and compact operators on the standard
Hilbert B-module `2 b B. We identify LB with the “diagonal subalgebra”
1Mn b LB Ď Mn b LB “ Mn pLB q for each n.
We need an ad-hoc definition.
Definition 6.1. A C ˚ -subalgebra F Ď LB is large if it satisfies the following
conditions:
(i) F is contained in the C ˚ -subalgebra Bp`2 q of LB arising from the ten-
sorial decomposition HB “ `2 b B;
55
Lemma 6.3. Let κ ě 1. In addition to our usual assumptions on B, as-
sume that K˚ pBq “ 0. Let F Ď Bp`2 q Ď LB be a finite-dimensional large
subalgebra, and let Z be a subset of the unit ball of F that spans F . Then
˚
KKκ,0 pZ, Bq “ 0.
Proof. Let F 1 Ď LB be the commutant of F in LB , which is a C ˚ -algebra.
Then the maps
0
KKκ,0 pZ, Bq Ñ K0 pF 1 X KB q, rp, qs ÞÑ rps ´ rqs
and
1
KKκ,0 pZ, Bq Ñ K0 pF 1 X KB q, rus ÞÑ rus
are isomorphisms (we leave this to the reader to check). it thus suffices to
show that K˚ pKB X F 1 q “ 0.
Write F “ Mk1 ‘ ¨ ¨ ¨ ‘ MkN as a direct sum of matrix algebras, and for
each n P t1, ..., N u, let πn : F Ñ BpCkn q be the irreducible representation of
F corresponding to Mkn . As F is a large C ˚ -subalgebra of LB , the represen-
tation theory of finite-dimensional C ˚ -algebras gives us a decomposition
N ´
à ¯
kn
HB “ pH0 b Bq ‘ C b Hn b B ,
n“1
Hence
´ ¯ àN ´ ¯
F 1 X KB “ LpH0 b Bq X KB ‘ 1Ckn b LpHn b Bq X KB .
n“1
Each summand is just a copy of Kp`2 q b B, however (or possibly zero for
LpH0 b Bq X KB ). Hence K˚ pF 1 X KB q is isomorphic to a direct sum of either
N , or N ` 1, copies of K˚ pBq. Either way, it is zero.
56
We need some more preliminaries for before we can get to the proof of
Proposition 6.2.
Lemma 6.4. Let x be an element in a C ˚ -algebra C such that }x2 ´x} ď 1{16
and }x} ď κ for some κ ě 1. Then the spectrum of x is contained in
tz P C | |z| ď 1{4 or |z ´ 1| ď 1{4u, and for z P C we have
Proof. Using Lemma 6.5, we have }x2 ´x} ď p2κ`1q ă 1{16. Hence Lemma
6.4 and the holomorphic functional calculus imply that χpxq is a well-defined
idempotent. Concretely, we have
ż
1
χpxq “ pz ´ xq´1 dz,
2πi γ
Applying the estimate in line (52) to the terms }pz ´ pq´1 } and }pz ´ xq´1 }
we see that
1 2π
}χpxq ´ p} ď 81κ2 }x ´ p} ď 27κ2
2π 3
as required.
58
Lemma 6.7. For any κ ě 1 and P p0, 1{p1000κ5 qq, if F Ď Bp`2 q Ď LB is a
large finite-dimensional subalgebra, if Y is a finite subset of the unit ball of
LB at Hausdorff distance less than from the unit ball of F , and if Z is a
subset of the unit ball of F that spans F then there are homomorphisms
˚ ˚
φ : KKκ,0 pZ, Bq Ñ KKκ,κ pY, Bq
and
˚ ˚
ψ : KKκ,κ pY, Bq Ñ KK2κ,0 pZ, Bq
such that ψ ˝ φ is the canonical forget control map
˚ ˚
KKκ,0 pZ, Bq Ñ KK2κ,0 pZ, Bq
(where the last map is a forget control map as in line (7)) is the canonical
forget control map
˚ ˚
KKκ, pY, Bq Ñ KK4κ,84κ2 pY, Bq.
Proof. Let us look first at the case of KK 0 . Let first pp, qq P Pn,κ,0 pZ, Bq
0
define a class in KKκ,0 pZ, Bq. As p and q commute with F , and as every
y P Y is at distance at most from some f in the unit ball of F , we have
Hence pp, qq defines an element of Pn,κ,κ pY, Bq. It is not difficult to see from
here that we get a well-defined homomorphism
0 0
φ : KKκ,0 pZ, Bq Ñ KKκ,κ pY, Bq, rp, qs ÞÑ rp, qs. (54)
Our goal in the rest of the proof is to construct a map going in the opposite
direction, and prove that the two are mutually inverse up to the relevant
forget control map.
59
Let G be the (unital) C ˚ -subalgebra of LB generated by F and 1LB ´ 1F .
Note that G “ F or G – F ‘ C, and in particular is finite dimensional. Let
U be the (compact) unitary group of G, and let µ be the Haar measure on
U , normalized to have total mass one.
Let pp, qq be an element of Pn,κ,κ pY, Bq. Any u P U is within of some
element y P Y , whence
Define ż
x :“ upu˚ dµpuq P Mn pLB q. (56)
U
Note that each u P U commutes with the scalar matrices Mn pCq Ď Mn pLB q.
` `
As p is in Mn pKB q, this implies that upu˚ is actually in Mn pKB q for all u P U .
As the integral in line (56) above converges in norm, it follows that x is in
`
Mn pKB q also. Note also that x precisely commutes with G (equivalently,
with F ) and that }x} ď κ. Moreover, line (55) gives
›ż › ż
› ›
˚
}x ´ p} “ › urp, u sdµpuq› ď 2κdµpuq “ 2κ.
› ›
› U › U
}p ´ p1 } ď 54κ3 . (57)
`
Analogously we find an idempotent q1 in the commutant of F in Mn pKB q
3 5
such that }q ´ q1 } ď 54κ . As ă 1{p1000κ q, we see in particular that
}p1 } ď κ ` κ “ 2κ, and similarly }q1 } ď 2κ. Note moreover that if σ :
Mn pKB `
q Ñ Mn pCq is the canonical quotient map then }p´p1 } ď 1{p12κ2 q by
line (57) and choice of . Hence in K0 pCq, Lemma 4.10 gives rσppqs “ rσpp1 qs.
Similarly rσpqqs “ rσpq1 qs, and so rσpp1 qs “ rσpq1 qs. At this point we have
that pp1 , q1 q is a well-defined element of Pn,2κ,0 pZ, Bq.
60
The above process taking pp, qq to pp1 , q1 q respects homotopies and block
sums, and thus we get a well-defined homomorphism
0 0
ψ : KKκ,κ pY, Bq Ñ KK2κ,0 pZ, Bq, rp, qs ÞÑ rp1 , q1 s. (58)
We claim that φ and ψ as in lines (54) and (58) are mutually inverse up to
the appropriate forget control maps, which will complete the proof. Indeed,
ψ ˝ φ is the natural forget control map
0 0
KKκ,0 pZ, Bq Ñ KK2κ,0 pZ, Bq
even on the level of cycles pp, qq, as if pp, qq is in the image of φ, then the
process above replacing pp, qq with pp1 , q1 q does not do anything. To see
that φ ˝ ψ is the claimed forget control map, we have to show that rp, qs is
0
equivalent to rp1 , q1 s in KK4κ,84κ 2 pY, Bq. Indeed, Lemma 4.10 implies that
there is a path pppt , qt qqtPr0,1s connecting pp, qq and pp1 , q1 q in Pn,4κ,84κ2 pY, Bq,
so we are done.
The case of KK 1 is similar (and simpler). Precisely analogously to the
case of KK 0 , we get a map
1 1
φ : KKκ,0 pZ, Bq Ñ KKκ,κ pY, Bq, rus ÞÑ rus.
3 5
and }u´1
1 } ă κ ` 4κ , which is smaller than 2κ as ă 1{p1000κ q. Hence we
may define
1 1
ψ : KKκ,κ pY, Bq Ñ KK2κ,0 pZ, Bq, rus ÞÑ ru1 s.
61
To complete the proof, we note just as in the KK 0 case that ψ ˝ φ is the
natural forget control map
1 1
KKκ,0 pZ, Bq Ñ KK2κ,0 pZ, Bq.
˚
In particular, the forget control map in the line above factors through KK2κ,0 pZ, Bq.
This latter group is zero by Lemma 6.3, completing the proof.
62
Definition 7.1. Let A be a (separable) C ˚ -algebra. A representation π :
A Ñ LB is scalar if it factors through the natural inclusion Bp`2 q Ď LB
arising from the decomposition HB “ `2 b B.
Lemma 7.2. If A is nuclear and unital, then there exists a scalar, unitally
absorbing representation π : A Ñ LB .
For the rest of this section, we assume that A has been identified with a
subset of LB by a scalar unitally absorbing representation as in Lemma 7.2
above.
We recall a result of Dadarlat [10, Corollary 5.3].
63
and if X is a finite subset of A1 and ą 0, then there is a finite subset Z of
A1 that contains X and γ ď such that the forget control map
0 0
KKκ,γ pZ, Aq Ñ KKκ, pX, Aq
takes rpZ,γ , qZ,γ s to zero (and therefore rpX, , qX, s “ 0 as this forget control
map takes rpZ,γ , qZ,γ s to rpX, , qX, s by definition of an element of an inverse
limit). Replacing X with X Y X ˚ , we may assume that X is self-adjoint.
Let us then prove this. Let us iteratively put conditions on γ as follows.
0
KKκ,γ pX, Aq / KKκ05 ,γ5 pX, Aq (59)
5
φ˚
(
0
KKκ, pX, Aq
(in which the horizontal and diagonal maps are forget control maps as
in line (3)) commute.
64
Working through the above conditions, one sees that for any κ ě 1, there is
some γ ą 0 that makes the above all work.
Let h P A1 , Fh , F1´h , and F be as in the definition of decomposability
with respect to the parameter γ and the set X. Let Yh , Y1´h and Y be self-
adjoint, γ-dense finite subsets of the unit balls of Fh , F1´h and F respectively.
Set Z :“ X Y thu Y Y1´h Y Yh . We claim that the pair pZ, γq has the desired
property.
Let n be such that ppZ,γ , qZ,γ q is in Pn,κ,γ pZ, Aq. Using Proposition 4.6
1
part (i) there is l P N and pp, qq P Pn,κ 1 ,γ
pZ, Aq in the same path component
`
of Pn,κ1 ,γ pZ, Aq as ppZ,γ , qZ,γ q. Let l P N be such that if σ : Mn pKB q Ñ Mn pCq
is the canonical quotient map, then σppq “ σpqq “ 1l . Lemma 5.4 implies
that pp, qq defines classes
rp, qsh P KKκ01 ,2γ phX YYh , Aq and rp, qs1´h P KKκ01 ,2γ pp1´hqX YY1´h , Aq.
and
KKκ01 ,2γ pp1 ´ hqX Y Y1´h , Aq Ñ KKκ02 ,γ2 pp1 ´ hqX Y Y1´h , Aq
1
of line (3) are zero. We may now apply Proposition 5.5 to get u P U8,κ 3 ,γ3
php1´
hqX Y thu Y Y, Aq such that if N2 is as in Proposition 5.5 with respect to
λ :“ κ2 , then Brus “ rp, qs in KKN0 2 ,N2 γ2 pX Y thu, Aq.
We have by assumption that the Hausdorff distance between hp1´hqXYY
and the unit ball of F is at most γ, which is less than γ3 . As γ is small enough
so that γ3 ă {p1000κ53 q, Proposition 6.2 again gives that the canonical forget
control map
65
of line (6) is zero. Hence in particular, rus “ 0 in KKκ14 ,γ4 php1 ´ hqX, Aq.
Proposition 5.1 and the facts that κ5 ě N0 and γ5 ě N0 γ4 implies that that
the boundary map
0
KKκ,γ pX, Aq / KKκ05 ,γ5 pX, Aq .
5
φ˚
(
0
KKκ, pX, Aq
0
We have that rp, qs “ 0 in KKκ5 ,γ5 pX, Aq. Hence φ˚ rp, qs “ 0 in KKκ, pX, Aq.
0
By commutativity of the diagram, this also says that rp, qs “ 0 in KKκ, pX, Aq
and we are done.
66
Proposition A.1. For any n P N with n ě 2, the Cuntz algebra On is
decomposable.
(with t1, ..., nu0 by definition consisting only of the empty word). For each
µ “ pi1 , ...., ik q P Wn , define eµ :“ ei1 b ¨ ¨ ¨ b eik , and define e∅ to be any unit-
length element of H b0 “ C. Then the set teµ | µ P Wn u is an orthonormal
basis of Γpnq. For µ P Wn , write |µ| for the length of µ, i.e. |µ| “ k means
that µ “ pi1 , ..., ik q for some i1 , .., ik P t1, ..., nu. Then the canonical copy of
H bk inside Γpnq from line (60) has orthonormal basis teµ | |µ| “ ku.
For each i P t1, ..., nu let Ti be the bounded operator on Γpnq that acts
on basis elements via the formula
Ti : eµ ÞÑ ei b eµ .
algebra On .
67
Now, for x P R` , define rxs :“ mintn P N | n ě xu, and define9
2k´1
à 2k`rk{2s
à
Γ0,k :“ H bl and Γ1,k :“ H bl . (61)
l“k l“k`rk{2s
p0q
For i P t0, 1u, define Bi,k :“ BpΓi,k q. For each l, m P N, we identify H bl b
H bm with H bpl`mq via the bijection of orthonormal bases
` ˘ ` ˘
ei1 b ¨ ¨ ¨ b eil b ej1 b ¨ ¨ ¨ b ejm Ø ei1 b ¨ ¨ ¨ b eil b ej1 b ¨ ¨ ¨ b ejm .
Fix for the moment k P N (it will stay fixed until Lemma A.2 below). Then
for each j P N we get a canonical identification
2k´1
à pj`1qk´1
à
Γ0,k b H bjk “ H bl b H bjk “ H bl .
l“k l“jk
“:H0
p0q
Let id be the identity representation of B0,k on Γ0,k and write B0,k for the
p0q
image of B0,k in the representation on Γpnq that is given by
8
´à ¯
0H0 ‘ id b 1H bjk
k“0
“:H1
9
In [40, Section 7], Γ0,k is written Γk,2k and Γ1,k is written Γk`rk{2s,2k`rk{2s .
68
p0q
and define B1,k to be the image of B1,k under the representation
8
´à ¯
0H1 ‘ id b 1H bjk .
k“0
Now, let f : r0, 1s Ñ r0, 1s be the function with graph pictured, where
the non-differentiable points occur at the x values 1{6, 2{6, 4{6, and 5{6.
p0q p0q
Let h0,k P B0,k be the operator on Γ0,k that acts on the summand H bl from
line (61) by multiplication by the scalar f ppl ´ kq{pk ´ 1qq. Similarly, let
p0q p1q
h1,k P B1,k be the operator on Γ1,k that acts on the summand H bl from line
(61) by multiplication by the scalar 1 ´ f ppl ´ k ´ rk{2sq{pk ´ 1qq. Let h0,k
p0q p0q
and h1,k be the images of h0,k and h1,k in B0,k and B1,k respectively. Note
that the operator on h0,k ` h1,k on Γpnq acts on the summand on H bl from
line (60) by multiplication by 1 as long as l ě k ` rk{2s. In particular,
Lemma A.2. For any T in the Cuntz-Toeplitz algebra Tn and i P t0, 1u, we
have that }rhi,k , T s} Ñ 0 as k Ñ 8.
Proof. We will focus on h0,k : the case of h1,k is essentially the same. It
suffices to consider the case where T is one of the canonical generators Ti of
the Cuntz-Toeplitz algebra. Let eµ be a basis element with |µ| “ jk ` l for
69
some j, l P N with l P t0, ..., k ´ 1u. Then we compute that rh0,k , Ti seµ “ 0 if
j “ 0, and that otherwise
` ˘
rh0,k , Ti seµ “ f ppl ` 1q{pk ´ 1qq ´ f pl{pk ´ 1qq ei b eµ .
The choice of function f implies that the right hand side above is approxi-
mately 6{k, so we are done.
Proof. We will focus on the case of h0,k : the other cases are similar. It
suffices to consider T a finite product S1 ...Sm , where each Sj is either one of
1{l
the generators Ti or its adjoint. Using Lemma A.2, we see that rh0,k , Sj s Ñ 0
as k Ñ 8 for any j, and any l P N with l ě 1. Hence the difference
` 1{p2mq 1{p2mq ˘` 1{p2mq 1{p2mq ˘ ` 1{p2mq 1{p2mq ˘
h0,k S1 ...Sm ´ h0,k S1 h0,k h0,k S2 h0,k ¨ ¨ ¨ h0,k Sm h0,k
70
For a word µ P Wn in t1, ..., nu, we may uniquely write µ “ µ0 µ1 , where the
lengths |µ0 | and |µ1 | satisfy |µ0 | P t0, ..., k ´ 1u, and |µ1 | P kN. Then the
bijective correspondence of orthonormal bases
eµ Ø eµ0 b eµ1
Identify the C ˚ -algebra BpΓk pnqq b Tnk with its image in the representation
on Γpnq arising from the above decomposition. The following is essentially
part of [40, Lemma 7.1].
Lemma A.4. With notation as above, BpΓk pnqq b Tnk contains the finite-
dimensional C ˚ -algebras we have called B0,k and B1,k , and in particular also
contains h0,k and h1,k .
Proof. In the notation of [40, Lemma 7.1], B0,k “ Λk pBpΓk,2k qq, and B1,k “
Λk pBpΓk`rk{2s,2k`rk{2s qq. Part (i) of [40, Lemma 7.1] says exactly that the
image of Λk is contained in BpΓk pnqq b Tnk , however, so we are done.
71
Corollary A.5. For any a P On , we have that the following all tend to zero as
k Ñ 8: }rqph0,k q, ιpaqs}, }rqph1,k q, ιpaqs}, dpqph0,k qιpaq, qpB0,k qq, dpqph1,k qιpaq, qpB1,k qq,
and dpqph0,k h1,k qιpaq, qpB0,k X B1,k qq.
Proof of Proposition A.1. Let ą 0, and let X be a finite subset of the unit
ball of On . Corollary A.5 implies that for any large k we have that for all
a P X and i P t0, 1u, the quantities }rqphi,k q, ιpaqs}, dpqphi,k qιpaq, qpBi,k qq,
and dpqph0,k h1,k qιpaq, qpB0,k X B1,k qq are smaller than {2. We may assume
moreover that k ” 1 modulo n ´ 1. Fix this k for the remainder of the proof.
As discussed on [40, page 488], we have a canonical unital inclusion Onk Ñ
On by treating suitable products of the generators of On as generators of
Onk . Moreover, dk is equal to k modulo n ´ 1. It follows that the K-theory
of Mdk pOn q is given by Z{pn ´ 1qZ in dimension zero and zero in dimension
one, with the class r1s of the unit in K0 represented by the residue of k in
Z{pn ´ 1qZ. Hence the K-theory invariants of Mdk pOn q and On agree, as we
are assuming that k ” 1 modulo n ´ 1. In particular, the Kirchberg-Phillips
classification theorem (see for example [28, Corollary 8.4.8]) gives a unital
isomorphism Mdk pOn q – On . Combining this with the inclusion Onk Ñ On
mentioned above gives a unital inclusion
72
h1,k q “ 1 (see line (62)), imply that this data satisfies the definition of de-
composability (Definition 1.1), so we are done.
Corollary A.7. Let Γ be a virtually cyclic group acting freely on the Cantor
set X. Then the crossed product CpXq ¸r Γ is decomposable.
In particular, if Z acts minimally on the Cantor set, then CpXq ¸r Z is
decomposable.
For the proof of Proposition A.6, it is convenient to split off two lemmas.
73
Lemma A.8. Let G be an étale groupoid. Let K be a subset of G, and let
H be the subgroupoid of G that K generates. Then if K is open, H is open
(and in particular is an étale groupoid in its own right). Moreover, if H is
relatively compact and K is closed then H is compact.
For the next lemma, recall that if X and Y are subsets of a metric space
and ě 0 then we write “X Ď Y ” to mean that for every x P X there exists
y P Y such that dpx, yq ď .
Proof. We first appeal to [14, Lemma 3.4]: in the language of that lemma,
each Hi is a CEER, and therefore for each i there is an isomorphism of
topological groupoids
ğmi
piq piq
Hi – Y j ˆ Pj
j“0
74
piq piq
where each Yj is a compact open subset of Gp0q and each Pj is a pair
groupoid on a finite set. We may assume further that this isomorphism
piq
is compatible in the obvious sense with the inclusion of each Yj into Gp0q .
piq
Further subdividing the sets Yj , we may assume that there are finitely many
piq piq
compact open subsets Y0 , ..., YN of Gp0q and pair groupoids P0 , ..., PN for
i P t0, 1, 2u as well as integers m1 ď m2 in t0, ..., N u such that
m2
ğ N
ğ
p1q p2q
H1 – Yj ˆ Pj , H2 – Yj ˆ Pj
j“0 j“m1
and m2
ğ p0q
H0 – Y j ˆ Pj ,
j“m1
p0q p1q p2q
and such that Pj Ď Pj X Pj for all j. Write I1 :“ t0, ..., m2 u, I2 :“
tm1 , ..., N u, and I0 :“ tm1 , ..., m2 u. Using that each Gp0q is totally discon-
nected and that for each i P t0, 1, 2u, each x P Xi is a uniformly contin-
uous function x : Hi Ñ C, we may further subdivide each Yj into com-
pact open subsets to assume that for each x P Xi there is an element z of
piq
spantχYj ˆtgu | j P Ii , g P Pj u such that }z ´ x}Cr˚ pGq ď (here χYj ˆtgu P
Cc pGq is the characteristic function of the compact open set Yj ˆ tgu). Set
piq
Fi :“ spantχYj ˆtgu | j P Ii , g P Pj u. Then F0 , F1 , F2 have the right proper-
ties.
Proof of Proposition A.6. Assume a finite subset X of the unit ball of Cr˚ pGq
and ą 0 are given. Up to an approximation, we may assume that X is a
subset of Cc pGq, whence there is a compact open set K Ď G such that X is
a subset of Cc pKq. Using [17, Lemma 8.20] there exists δ ą 0 such that if
φ : Gp0q Ñ C satisfies supgPK |φprpgqq ´ φpspgqq| ă δ, then for any f P Cc pGq
supported in K, we have }rf, φs} ď p{3q}f }}φ}. Using dynamic asymptotic
dimension one and [17, Proposition 7.1] there exists an open cover U1 , U2 of
Gp0q and continuous functions φ1 , φ2 : Gp0q Ñ r0, 1s such that: each φi is
supported in Ui ; φ1 pxq2 ` φ2 pxq2 “ 1 for all x P X; |φi pspgqq ´ φi prpgqq| ă δ
for all g P K and i P t1, 2u; and tg P K | spgq, rpgq P Ui u is contained in a
relatively compact open subgroupoid Hi of G for i P t1, 2u. As Gp0q is totally
75
disconnected, we can assume on shrinking U1 and U2 slightly that these sets
are closed and open. Hence in particular, each set tg P K | rpgq, spgq P Ui u
is compact and open. Hence the subgroupoid this generates is compact and
open by Lemma A.8, so we may also assume that H0 and H1 are compact
and open.
Now, define h :“ φ21 , which is a positive contraction such that 1 ´ h “ φ22 .
For any x P X, note first that
}hp1 ´ hqx ´ φ0 xφ0 } ď }φ1 φ22 }}rx, φ1 s} ` }φ2 }φ21 }}rx, φ2 s} ď 2{3.
References
[1] M. Amini, K. Li, D. Sawicki, and A. Shakibazadeh. Dynamic asymptotic
dimension for actions of virtually cyclic groups. ArXiv:2007.00960, 2020.
73
76
[3] B. Blackadar. Operator Algebras: Theory of C ˚ -Algebras and Von Neu-
mann Algebras. Springer, 2006. 22
[10] M. Dadarlat. On the topology of the Kasparov groups and its applica-
tions. J. Funct. Anal., 228(2):394–418, 2005. 6, 9, 63
77
[14] T. Giordano, I. Putnam, and C. Skau. Affable equivalence relations and
orbit structure of Cantor dynamical systems. Ergodic Theory Dynam.
Systems, 24(2):441–475, 2004. 73, 74
78
[24] H. Oyono-Oyono and G. Yu. On quantitative operator K-theory. Ann.
Inst. Fourier (Grenoble), 65(2):605–674, 2015. 9, 20, 29, 32
[29] M. Rørdam. The stable and the real rank of Z-absorbing C ˚ -algebras.
Internat. J. Math., 15(10):1065–1084, 2004. 4
[31] J. Rosenberg and C. Schochet. The Künneth theorem and the universal
coefficient theorem for Kasparov’s generalized K-functor. Duke Math.
J., 55(2):431–474, 1987. 2, 72
79
[35] A. Tikuisis, S. White, and W. Winter. Quasidiagonality of nuclear C ˚ -
algebras. Ann. of Math., 185(1):229–284, 2017. 3
80