by Alan L. Carey
1. Preamble
It may seem hard to believe but I am not exactly sure when I first met Paul. However the occasion of a meeting organised by David Evans in 1987 in Warwick may well qualify. Our first substantial mathematical interaction, however, did not occur until an Oxford meeting organised by John Roe in the early 90s. There Paul, Varghese Mathai and I enjoyed a number of entertaining conversations and Paul made some very helpful comments which played a role in securing Mathai a prestigious Australian Research Council Fellowship soon after. Paul has been to Australia on several occasions visiting Sydney, Melbourne, Newcastle and finally Canberra. He achieved some notoriety for breaking into song at the Australian Mathematical Society conference dinner in La Trobe University in 2007. He was in very good spirits after giving the best plenary of the meeting a day or two beforehand. Paul was awarded an honorary doctorate from the Australian National University in 2013 in recognition of his lengthy and significant support of Australian mathematicians.
Our substantial research interaction began in Texas Christian University at a meeting in honour of Jonathan Rosenberg. Before launching into that it is useful to recall some history of twisted K-theory and this will occupy the next section. After that I will cover the basics on twisted K-theory. Then in the main section of the paper I will describe twisted geometric cycles from two viewpoints, BB (before Baum) and AB (after Baum). The story ends on a happy note.
2. History
The story brings together a number of mathematical topics: gerbes, continuous
trace
Continuous trace algebras have their roots in work of
Grothendieck
[e3]
and
Dixmier
and
Douady
[e1],
who studied
bundles of algebras over a paracompact space
The other line of development
is the theory of continuous trace
Twisted K-theory now comes into the picture. Given a paracompact space
These two disparate lines formed the motivation for the work of Murray [e15] on bundle gerbes. Bundle gerbes provide a differential geometric way to approach twisted K-theory, as explained in [e24], that is closely aligned with the geometry that is used in string theory.
The additional ingredient in this picture that we need from the string
theory side is the notion of
In
[e16]
Minasian
and
Moore
made the proposal that
2.1 Mathematical perspective
The contributions of physicists raise some immediate questions. When
there is no twist, the
relationship between K-theory and index theory of elliptic
operators is now well-established, including
the geometric cycle approach
to K-homology. In fact one version of the Atiyah–Singer index theorem due to
Baum,
Higson,
and
Schick
[3]
establishes a relationship between the analytic viewpoint provided
by elliptic differential operators and the geometric viewpoint provided by
the notion of geometric cycle introduced in the fundamental paper of Baum
and
Douglas
[1].
The viewpoint that geometric cycles in the sense of
[1]
are a model for
It is thus tempting to conjecture that there is an analogous picture of
In this enterprise, Paul entered the picture via his interaction with Bai-Ling Wang over the ideas introduced in [e33]. I want to explain this in detail in this essay. I will also fill in many of the details for the matters discussed previously in this introduction. I have borrowed very freely from a review article that I wrote with Bai-Ling Wang after our meeting with Paul in Texas Christian University [e37] (though whether it was an epiphany is hard to determine).
It is essential to remark that none of the existing arguments that are used in the geometric cycle approach to index theory (the general index problem of Paul) extend to cover the twisted version for CW-complexes. In the final section I am pleased to announce that there is now a very general argument that applies to all instances.
This article contains nothing original but does draw together the many threads that have led us to the present understanding of K-homology in the twisted case.
3. Topological and analytic twisted K-theory
Though our focus is twisted K-homology, in the literature on this topic, extensive use is made of Poincaré duality. For this reason it is important to review first some aspects of twisted K-theory.
We begin with the notion of a “twisting”.
Let
A twisting of complex K-theory on
The action
Definition:
The
(topological) twisted K-groups of
Due to Bott periodicity, we only have two different twisted K-groups

Take a pair of
twistings
Let
Let
It is important to recognise that these groups are only defined up to
isomorphism by the Dixmier–Douady class
3.1. Twisted K-theory for torsion twistings
There are some subtle issues in twisted K-theory and to handle these we
have chosen to use the language of bundle gerbes, connections and curvings
as explained in
[e15].
We explain first the so-called “lifting bundle
gerbe”

A torsion twisting
There is an Azumaya bundle associated to
There is an equivalent definition of twisted K-theory using bundle
gerbe modules
(see
[e24]
and
[e32]).
A bundle gerbe module
for and is the target map of is compatible with the bundle gerbe multiplication which means
Note that the natural representation of
the Grothendieck group of the category of
bundle gerbe modules,the Grothendieck group of the category of
bundle modules.
One important example of torsion twistings comes from real oriented vector
bundles. Consider an oriented real vector bundle
Proof.
Denote by
4. Topological and analytic twisted K-homology
Let
The stable homotopy groups of
For a relative CW-complex
Analytic twisted K-homology, denoted
by
For a relative CW-complex
The proof of this theorem requires Poincaré duality between twisted K-theory and twisted K-homology (we describe this duality in the next theorem), and the isomorphism (Theorem 1) between topological twisted K-theory and analytic twisted K-theory.
Fix an isomorphism
There are natural isomorphisms
from twisted K-homology (topological, resp. analytic) to twisted K-theory
(topological, resp. analytic) of a smooth manifold
Analytic Poincaré duality was established in
[e31]
and
[e35],
and topological Poincaré duality was established in
[e33].
Theorem 4 and the exact sequences for a pair
Remark:
In fact,
Poincaré duality as in Theorem 5 holds for any compact Riemannian
manifold
5. Geometric cycles and geometric twisted K-homology
This is the main section of this essay. Here we outline both the ideas due to Bai-Ling Wang and the ideas due mostly to Paul. One may easily move between these points of view.
As usual

Definition:
Given
a smooth oriented manifold

As shown in
[e33],
this way of thinking about twisted
A morphism between
is homotopic to through a continuous map is homotopic to through a continuous mapthe composition of homotopies
is homotopic to
Two
Orientation reversal in the Grassmannian model defines an involution
5.1. Bai-Ling Wang’s approach to twisted geometric cycles or BB
I will first review the original approach to twisted geometric cycles and then review the ideas originating with Paul.
Definition:
A geometric
cycle for
Two geometric cycles
Let
Direct sum–disjoint union: If
and are two geometric cycles with the same -twisted structure, thenBordism: Given two geometric cycles
and we call them bordant if there exists an -twisted manifold and such that and Here denotes the manifold with the opposite -twisted structure. vector bundle modification: Suppose we are given a geometric cycle and a vector bundle over with even-dimensional fibres. Denote by the trivial rank-1 real vector bundle. Choose a Riemannian metric on and let be the sphere bundle of Then the vertical tangent bundle of admits a natural structure with an associated -graded spinor bundle . Denote by the projection which is K-oriented. Then
Definition:
Denote by
Remark:
If
in a geometric cycle for is a compact manifold with boundary, then has to be a class inIf
is a continuous map and is a twisting, then there is a natural homomorphism of abelian groups sending toLet
be a closed subspace of and be a twisting on A relative geometric cycle for is a quintuple such that is a smooth manifold (possibly with boundary), equipped with an -twisted structureif
has a nonempty boundary, then is a K-class in represented by a -graded vector bundle over or a continuous map
The relation
5.2. -cycles or AB
The
difficulty with the original approach to these twisted geometric cycles
outlined in the previous subsection is
that it is not clear how to construct them (that is, they
do not solve the
“general twisted index problem” that we will announce below). A new approach
was suggested by Paul and explained in our joint paper
[4].
The main difference between this new approach and the approach described
earlier in this article is that we make much heavier use of the theory of
continuous
trace
We begin by recalling a few concepts.
Given an oriented real Euclidean vector bundle
of rank K over a paracompact Hausdorff topological space a structure on is a lift of the oriented frame bundle to a principal -bundle where is the unique (for ) nontrivial central extension of by See Appendix D in [e11] for an equivalent definition of structures (and note that there is a well-known modification needed for ). A real vector bundle with a structure is called a vector bundle. structures are orientation conditions for complex K-theory in the sense that a vector bundle is a real vector vector bundle with a given complex spinor bundle or a K-theory Thom class. See Section 4 in [3] and Theorem C.12 in [e11] for more discussions of this. In particular, a spinor bundle for determines an orientation of(two-out-of-three principle) Let
be a short exact sequence of oriented real vector bundles on a paracompact Hausdorff topological space Then structures for any two of the vector bundles determine a structure for the third vector bundle.A
Riemannian manifold is a Riemannian manifold (perhaps with boundary) whose tangent bundle is a vector bundle. If has a boundary at each boundary point, the outward normal vector defines a trivial rank-1 real vector bundle over and is an exact sequence of real Euclidean vector bundles over Therefore, the two-out-of-three principle implies that if a manifold has a boundary, then this boundary, is a manifold in a canonical way.
5.2.1. Twisting data
If
Definition:
Let
If
For a
Let
On the other hand if
Definition:
A spinor bundle for
Lemma 8: Let
Proof. The set of Hilbert–Schmidt operators on
We import our previous notation: if
We now expound some elementary results with proofs to give the flavour of the Kasparov theory we used.
Proof. Let
5.2.2. K-cycles for twisted K-homology
As above,
Definition:
A twisted-by-
is a compact manifold without boundary, is a continuous map, (the -group of the -algebra
Remark:
The twisted K-cycles defined here are closely related to the cycles in the
original formulation of the Baum–Connes conjecture.
Later we will define the notion of
Keeping
bordism,
vector bundle modification.

Next, we elaborate on these two steps for the case of twisted K-cycles.
Isomorphism: We say
Bordism: We say
is a compact manifold with boundary; is a continuous map from to
Vector bundle modification:
Let
Composition Lemma:
Let
Proof. We first construct a bordant manifold
Two twisted-by-
Definition:
Let
is a compact manifold with boundary, is a continuous map,
As in the untwisted case, the
additive inverse of
There is
a natural map
Paul has often enunciated the untwisted version of the following problem.
The General Twisted Index Problem:
When
The issue in this problem is constructibility of the K-cycle.
In
[4]
the following is proved using the notion of normal bordism
(introduced next) to establish the six-term
exact sequence in Kasparov
K-homology of
Theorem 10:
Let
5.2.3. Normal bordism
One of the main innovations in
[4]
was the elucidation of the
fundamental role of normal bordism.
Let
Definition:
Two twisted-by-
Lemma 11: Normal bordism is an equivalence relation.
Remark: The content of the next lemma is that for normal bordism there is no need to use elementary steps etc. The idea of normal bordism was first constructed by Jakob in [e22] for an alternative definition of generalised homology theory, and further applied in [e28] and [e38] in the study of various versions of geometric K-homology.
Lemma 12:
Two
twisted-by-
A twisted-by-
5.2.4. The group of -cycles
In this section, we introduce another notion of K-cycles for
Definition:
A
is a closed oriented Riemannian manifold; is a complex vector bundle on is a continuous map from to is a spinor bundle for
Remark:
If
is even-dimensional, then is the twisting datum on whose fibre at is the complexified Clifford algebra As usual, is the tangent space to at On the other hand, if is odd-dimensional, given choose a positively oriented orthonormal basis for Set and define by Then does not depend on the choice of positively oriented orthonormal basis for Also is in the centre of andNow set
Then is the twisting datum on whose fibre at isThe existence of a spinor bundle
for implies By standard algebraic topology, is the third (integral) Stiefel–Whitney class of so the existence of implies which is the Freed–Witten anomaly cancellation condition for Type IIB -branes as explained in [e33].
Definition:
Keeping
bordism,
direct sum - disjoint union,
vector bundle modification.

These three elementary moves can be precisely defined as follows.
Isomorphism: Two
Bordism: Two
Direct sum-disjoint union: Let
Vector bundle modification: Let
Let
5.2.5. The charge map
Definition:
Given
a
Denote by
the Gysin homomorphism associated to the canonical sectionObserve that, as twisting data on
a trivialisation of gives an equivalence a stable isomorphism of bundles of elementary -algebras. Hence the given spinor bundle for determines a spinor bundle for Then is a spinor bundle for and therefore yields an isomorphism of abelian groups
Question:
Is the
map
Given a positive answer to this question then we would have the following corollary:
Consequence:
Let
6. The latest chapter in the story
The authors of [4] have, both separately and in collaboration, come up with various proposals for answering the question posed above but a complete argument eluded us. Recently Paul and collaborators [5] answered the question in the affirmative, thus explaining how the original ideas of Bai-Ling Wang, in understanding the Freed–Witten anomaly cancellation condition, form the key assumption for creating twisted geometric cycles.
Moreover this very latest proof is the most general so far. It allows by one method to establish all previous versions of the general index problem in both the twisted and untwisted cases.
In this new argument Wang’s appeal to a twisted version of
the Conner–Floyd
splitting theorem is replaced by a twisted analogue of a theorem of
Hopkins
and
Hovey.
They proved that
Recall that given a CW-complex
This result is the key to proving that the
geometric model for twisted
K-homology in terms of
The upshot for the present essay is that all of the models for the group of twisted K-homology classes discussed previously are isomorphic. I will not discuss the details of [5] here. Paul is preparing an expository account which will provide a useful overview.