return

Celebratio Mathematica

Paul Baum

Twisted K-homology and Paul Baum

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 C al­geb­ras, K-ho­mo­logy (both ana­lyt­ic and geo­met­ric) and D-branes from string the­ory. This is a broad set of top­ics but they all fit in­to the jig­saw puzzle.

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 X with fibre either a mat­rix al­gebra or the com­pact op­er­at­ors on an in­fin­ite-di­men­sion­al Hil­bert space. They are char­ac­ter­ised by an as­so­ci­ated in­vari­ant; a class in de­gree-3 Čech co­homo­logy of the base was iden­ti­fied (it is now known as the Dixmi­er–Douady class). From these pa­pers two in­de­pend­ent lines of de­vel­op­ment began. The first, the the­ory of gerbes, began with [e5] and was ex­ten­ded and ap­plied to prob­lems in string the­ory in [e13].

The oth­er line of de­vel­op­ment is the the­ory of con­tinu­ous trace C-al­geb­ras. These are al­geb­ras of sec­tions of the bundles in­tro­duced by Dixmi­er and Douady. Their his­tory can be found in the mono­graph of Rae­burn and Wil­li­ams [e17].

Twis­ted K-the­ory now comes in­to the pic­ture. Giv­en a para­com­pact space X with a “twist”, namely a de­gree-3 co­homo­logy class [α] on X, there is a prin­cip­al bundle with fibre the pro­ject­ive unit­ary group of a sep­ar­able Hil­bert space over X char­ac­ter­ised by [α]. Now re­gard the pro­ject­ive unit­ary group as the auto­morph­ism group of the com­pact op­er­at­ors on this Hil­bert space. Then, fol­low­ing Rosen­berg [e10], the twis­ted K-the­ory of X is the K-the­ory of the con­tinu­ous trace C-al­gebra of sec­tions of the bundle over X as­so­ci­ated to the giv­en prin­cip­al bundle with twist [α]. The twist only spe­cifies an equi­val­ence class of con­tinu­ous trace al­geb­ras, so this means that twis­ted K-the­ory with twist [α] is only defined by this meth­od up to an iso­morph­ism.

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 D-branes [e14]. They were pro­posed as a mech­an­ism for provid­ing bound­ary con­di­tions for the dy­nam­ics of open strings mov­ing in space-time. Ini­tially they were thought of as sub­man­i­folds. As D-branes them­selves can evolve over time, one needs to study equi­val­ence re­la­tions on the set of D-branes. An in­vari­ant of the equi­val­ence class is the to­po­lo­gic­al charge of the D-brane, which should be thought of as an ana­logue of the Dir­ac mono­pole charge as these D-brane charges are as­so­ci­ated with gauge fields (con­nec­tions) on vec­tor bundles over the D-brane. These vec­tor bundles are known as Chan–Paton bundles.

In [e16] Minas­i­an and Moore made the pro­pos­al that D-brane charges should take val­ues in K-groups and not in the co­homo­logy of the space-time or the D-brane. However, they pro­posed a co­homo­lo­gic­al for­mula for these charges which might be thought of as a kind of in­dex the­or­em in the sense that, in gen­er­al, in­dex the­ory as­so­ci­ates to a K-the­ory class a num­ber which is giv­en by an in­teg­ral of a closed dif­fer­en­tial form. In string the­ory there is an ad­di­tion­al field on space-time known as the H-flux which may be thought of as a glob­al closed 3-form. Loc­ally it is giv­en by a fam­ily of “2-form po­ten­tials” known as the B-field. Math­em­at­ic­ally these B-fields are in fact de­fin­ing a de­gree-3 in­teg­ral Čech class on the space-time, that is, the “twist”. Wit­ten [e18], ex­tend­ing [e16], gave a phys­ic­al ar­gu­ment for the idea that D-brane charges should be ele­ments of K-groups and, in ad­di­tion, pro­posed that the D-brane charges in the pres­ence of a twist should take val­ues in twis­ted K-the­ory (at least in the case where the twist is tor­sion). The math­em­at­ic­al ideas he re­lied on were due to Donovan and Ka­roubi [e4]. Sub­sequently Bouwknegt and [e20] ex­ten­ded Wit­ten’s pro­pos­al to the non­tor­sion case us­ing ideas from [e10]. A geo­met­ric mod­el (that is, a “string geo­metry” pic­ture) for some of these string the­ory con­struc­tions and for twis­ted K-the­ory was pro­posed in [e24] us­ing the no­tion of bundle gerbes and bundle gerbe mod­ules. Vari­ous re­fine­ments of twis­ted K-the­ory that are sug­ges­ted by these ap­plic­a­tions are also de­scribed in the art­icle of Atiyah and Segal [e27] and their ideas play a role in the next de­vel­op­ment.

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 D-branes in the un­twis­ted case is ex­pounded in [e29], [e36], [e34]. Note that in this view­point D-branes are no longer sub­man­i­folds (as en­vis­aged ori­gin­ally by phys­i­cists) but the im­ages of man­i­folds un­der a smooth map.

It is thus tempt­ing to con­jec­ture that there is an ana­log­ous pic­ture of D-branes as a type of geo­met­ric cycle in the twis­ted case as well. More pre­cisely the ques­tion is wheth­er there is a way to for­mu­late the no­tion of “twis­ted geo­met­ric cycle” (in ana­logy with [2] and [1]) and to prove an in­dex the­or­em in the spir­it of [3] for twis­ted K-ho­mo­logy. An ap­proach to an­swer­ing this ques­tion was pro­posed in [e33], though there is a step in the proof that is omit­ted. Wang un­der­stood that the miss­ing step re­quires a twis­ted ver­sion of the Con­ner–Floyd split­ting the­or­em [e7] for Spinc-bor­d­ism; however a com­plete ver­sion of Bai-Ling Wang’s ar­gu­ment has not ap­peared. The good news is that there is an ar­gu­ment that is sim­il­ar in spir­it to what Wang pro­posed due to Paul and col­lab­or­at­ors that over­comes not just this one dif­fi­culty but provides a con­struc­tion of twis­ted K-ho­mo­logy for CW-com­plexes [5]. In the last sec­tion we will de­scribe briefly this new ap­proach that is in pre­par­a­tion. It is im­port­ant to em­phas­ise that string geo­metry ideas from [e19] played a key role in find­ing the cor­rect way to gen­er­al­ise [2].

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 H be an in­fin­ite-di­men­sion­al, com­plex and sep­ar­able Hil­bert space. We shall con­sider loc­ally trivi­al prin­cip­al PU(H)-bundles over a para­com­pact Haus­dorff to­po­lo­gic­al space X; the struc­ture group PU(H) is equipped with the norm to­po­logy. The pro­ject­ive unit­ary group PU(H) with the to­po­logy in­duced by the norm to­po­logy on U(H) (see [e2]) has the ho­mo­topy type of an Ei­len­berg–MacLane space K(Z,2). The clas­si­fy­ing space of PU(H), de­noted by BPU(H), is a K(Z,3). The set of iso­morph­ism classes of prin­cip­al PU(H)-bundles over X is giv­en by (Pro­pos­i­tion 2.1 in [e27]) ho­mo­topy classes of maps from X to any K(Z,3) and there is a ca­non­ic­al iden­ti­fic­a­tion [X,BPU(H)]H3(X,Z).

A twist­ing of com­plex K-the­ory on X is giv­en by a con­tinu­ous map α:XK(Z,3). For such a twist­ing, we can as­so­ci­ate a ca­non­ic­al prin­cip­al PU(H)-bundle Pα through the usu­al pull-back con­struc­tion from the uni­ver­sal PU(H) bundle de­noted by EK(Z,2), as sum­mar­ised by the dia­gram (1)PαEK(Z,2)XαK(Z,3) We will use PU(H) as a group mod­el for a K(Z,2). We write Fred(H) for the con­nec­ted com­pon­ent of the iden­tity of the space of Fred­holm op­er­at­ors on H equipped with the norm to­po­logy. There is a base-point-pre­serving ac­tion of PU(H) giv­en by the con­jug­a­tion ac­tion of U(H) on Fred(H): (2)PU(H)×Fred(H)Fred(H).

The ac­tion (2) defines an as­so­ci­ated bundle over X which we de­note by Pα(Fred)=Pα×PU(H)Fred(H). We write {ΩXnPα(Fred)=Pα×PU(H)ΩnFred} for the fibre-wise it­er­ated loop spaces.

Defin­i­tion: The (to­po­lo­gic­al) twis­ted K-groups of (X,α) are defined to be Kn(X,α):=π0(Cc(X,ΩXnPα(Fred))), the set of ho­mo­topy classes of com­pactly sup­por­ted sec­tions (mean­ing they are the iden­tity op­er­at­or in Fred off a com­pact set) of the bundle of Pα(Fred).

Due to Bott peri­od­icity, we only have two dif­fer­ent twis­ted K-groups K0(X,α) and K1(X,α). Giv­en a closed sub­space A of X, (X,A) is a pair of to­po­lo­gic­al spaces, and we define re­l­at­ive twis­ted K-groups to be Kev/odd(X,A;α):=Kev/odd(XA,α).

Take a pair of twist­ings α0,α1:XK(Z,3), and a map η:X×[1,0]K(Z,3) which is a ho­mo­topy between α0 and α1, rep­res­en­ted dia­gram­mat­ic­ally by the fig­ure shown to the right. Then there is a ca­non­ic­al iso­morph­ism Pα0Pα1 in­duced by η. This ca­non­ic­al iso­morph­ism de­term­ines a ca­non­ic­al iso­morph­ism on twis­ted K-groups (3)η:Kev/odd(X,α0) Kev/odd(X,α1). This iso­morph­ism η de­pends only on the ho­mo­topy class of η. The set of ho­mo­topy classes of maps between α0 and α1 is la­belled by [X,K(Z,2)]. Re­call the first Chern class iso­morph­ism Vect1(X)[X,K(Z,2)]H2(X,Z), where Vect1(X) is the set of equi­val­ence classes of com­plex line bundles on X. We re­mark that the iso­morph­isms in­duced by two dif­fer­ent ho­mo­top­ies between α0 and α1 are re­lated through an ac­tion of com­plex line bundles.

Let K be the C-al­gebra of com­pact op­er­at­ors on H. The iso­morph­ism PU(H)Aut(K) via the con­jug­a­tion ac­tion of the unit­ary group U(H) provides an ac­tion of a K(Z,2) on the C-al­gebra K. Hence, any K(Z,2)-prin­cip­al bundle Pα defines a loc­ally trivi­al bundle of com­pact op­er­at­ors, de­noted by Pα(K)=Pα×PU(H)K.

Let Γ0(X,Pα(K)) be the C-al­gebra of sec­tions of Pα(K) van­ish­ing at in­fin­ity. Then Γ0(X,Pα(K) is the (unique up to iso­morph­ism) stable sep­ar­able com­plex con­tinu­ous-trace C-al­gebra over X with Dixmi­er–Douady class [α]H3(X,Z) (here we identi­fy the Čech co­homo­logy of X with its sin­gu­lar co­homo­logy; see [e10] and [e27]).

The­or­em 1 ([e27],[e10]): The to­po­lo­gic­al twis­ted K-groups Kev/odd(X,α) are ca­non­ic­ally iso­morph­ic to ana­lyt­ic K-the­ory of the C-al­gebra Γ0(X,Pα(K)) Kev/odd(X,α)Kev/odd(Γ0(X,Pα(K))), where the lat­ter group is the K-the­ory of Γ0(X,Pα(K)), defined to be limkπ1(GLk(Γ0(X,Pα(K)))). Note that the K-the­ory of Γ0(X,Pα(K)) is iso­morph­ic to Kas­parov’s KK-the­ory [e6], [e9], [e8] KKev/odd(C,Γ0(X,Pα(K)).

It is im­port­ant to re­cog­nise that these groups are only defined up to iso­morph­ism by the Dixmi­er–Douady class [α]H3(X,Z). To dis­tin­guish these two defin­i­tions of twis­ted K-the­ory we will write Ktopev/odd(X,α) and Kanev/odd(X,α) for the to­po­lo­gic­al and ana­lyt­ic twis­ted K-the­or­ies of (X,α) re­spect­ively. Twis­ted K-the­ory is a 2-peri­od­ic gen­er­al­ised co­homo­logy the­ory: a con­trav­ari­ant func­tor on the cat­egory con­sist­ing of pairs (X,α), with the twist­ing α:XK(Z,3), to the cat­egory of Z2-graded abeli­an groups. Note that a morph­ism between two pairs (X,α) and (Y,β) is a con­tinu­ous map f:XY such that βf=α.

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” Gα [e15] as­so­ci­ated to the prin­cip­al PU(H)-bundle π:PαX and the cent­ral ex­ten­sion (4)1U(1)U(H)PU(H)1. This is con­struc­ted by start­ing with π:PαX, form­ing the fibre product Pα[2], which is a group­oid Pα[2]=Pα×XPα π1  π2 Pα with source and range maps π1:(y1,y2)y1andπ2:(y1,y2)y2. There is an ob­vi­ous map from each fibre of Pα[2] to PU(H) and so we can define the fibre of Gα over a point in Pα[2] by pulling back the fibra­tion (4) us­ing this map. This en­dows Gα with a group­oid struc­ture (from the mul­ti­plic­a­tion in U(H)) and in fact it is a U(1)-group­oid ex­ten­sion of Pα[2].

A tor­sion twist­ing α is a map α:XK(Z,3) rep­res­ent­ing a tor­sion class in H3(X,Z). Every tor­sion twist­ing arises from a prin­cip­al PU(n)-bundle Pα(n) with its clas­si­fy­ing map XBPU(n), or a prin­cip­al PU(H)-bundle with a re­duc­tion to PU(n)PU(H). For a tor­sion twist­ing α:XBPU(n)BPU(H), the cor­res­pond­ing lift­ing bundle gerbe Ga shown to the right is defined by Pα(n)[2]Pα(n)PU(n)Pα(n) (as a group­oid) and the cent­ral ex­ten­sion 1U(1)U(n)PU(n)1.

There is an Azu­maya bundle as­so­ci­ated to Pα(n) arising nat­ur­ally from the PU(n) ac­tion on the n×n matrices. We de­note this as­so­ci­ated Azu­maya bundle by Aα. An Aα-mod­ule is a com­plex vec­tor bundle E over M with a fibre-wise Aα ac­tion Aσ×MEE. The C-al­gebra of con­tinu­ous sec­tions of Aα, van­ish­ing at in­fin­ity if X is non­com­pact, is Mor­ita equi­val­ent to a con­tinu­ous trace C-al­gebra Γ0(X,Pα(K)). Hence there is an iso­morph­ism between K0(X,α) and the K-the­ory of the bundle mod­ules of Aa.

There is an equi­val­ent defin­i­tion of twis­ted K-the­ory us­ing bundle gerbe mod­ules (see [e24] and [e32]). A bundle gerbe mod­ule E of Gα is a com­plex vec­tor bundle E over Pα(n) with a group­oid ac­tion of Gα, i.e., an iso­morph­ism φ:Gα×(π2,p)EE, where Gα×(π2,π)E is the fibre product of the source π2:GαPα(n)andp:EPα(n) such that

  1. pφ(g,v)=π1(g) for (g,v)Gα×(π2,p)E, and π1 is the tar­get map of Gα;

  2. φ is com­pat­ible with the bundle gerbe mul­ti­plic­a­tion m:Ga×(π2,π1)GαGα, which means φ(id×φ)=φ(m×id).

Note that the nat­ur­al rep­res­ent­a­tion of U(n) on Cn in­duces a Gα bundle gerbe mod­ule Sn=Pα(n)×Cn. Here we use the fact that Gα=Pα(n)U(n)Pα(n) (as a group­oid). Sim­il­arly, the dual rep­res­ent­a­tion of U(n) on Cn in­duces a Gα bundle gerbe mod­ule Sn=Pα(n)×Cn. Note that SnSnπAα des­cends to the Azu­maya bundle Aα. Giv­en a Gα bundle gerbe mod­ule E of rank K, as a PU(n)-equivari­ant vec­tor bundle, SnE des­cends to an Aα-bundle over M. Con­versely, giv­en an Aα-bundle E over M, SnπAαπE defines a Gα bundle gerbe mod­ule. These two con­struc­tions are in­verse to each oth­er due to the fact that Sn(SnπAαπE)(SnSn)πAαπEπAαπAαπEπE. There­fore, there is a nat­ur­al equi­val­ence between the cat­egory of Gα bundle gerbe mod­ules and the cat­egory of Aα bundle mod­ules, as dis­cussed in [e32]. In sum­mary, we have the fol­low­ing pro­pos­i­tion.

Pro­pos­i­tion 2 ([e24],[e32]): For a tor­sion twist­ing α:XBPU(n)BPU(H), twis­ted K-the­ory K0(X,α) has an­oth­er two equi­val­ent de­scrip­tions:
  1. the Grothen­dieck group of the cat­egory of Gα bundle gerbe mod­ules,

  2. the Grothen­dieck group of the cat­egory of Aσ bundle mod­ules.

One im­port­ant ex­ample of tor­sion twist­ings comes from real ori­ented vec­tor bundles. Con­sider an ori­ented real vec­tor bundle E of even rank over X with a fixed fibre-wise in­ner product. De­note by νE:XBSO(2k) the clas­si­fy­ing map of E. The twist­ing o(E):=W3νE:XBSO(2k)K(Z,3) will be called the ori­ent­a­tion twist­ing as­so­ci­ated to E. Here W3 is the clas­si­fy­ing map of the prin­cip­al BU(1)-bundle BSpinc(2k)BSO(2k). Note that the ori­ent­a­tion twist­ing o(E) is null-ho­mo­top­ic if and only if E is K-ori­ented.

Pro­pos­i­tion 3: Giv­en an ori­ented real vec­tor bundle E of even rank over X with an ori­ent­a­tion twist­ing o(E), there is a ca­non­ic­al iso­morph­ism K0(X,o(E))K0(X,W3(E)), where K0(X,W3(E)) is the K-the­ory of the Clif­ford mod­ules as­so­ci­ated to the bundle Cliff(E) of Clif­ford al­geb­ras.

Proof.   De­note by Fr the frame bundle of V, the prin­cip­al SO(2k)-bundle of pos­it­ively ori­ented or­thonor­mal frames, i.e., E=Fr×ρ2nR2k, where ρn is the stand­ard rep­res­ent­a­tion of SO(2k) on Rn. The lift­ing bundle gerbe as­so­ci­ated to the frame bundle and the cent­ral ex­ten­sion 1U(1)Spinc(2k)SO(2k)1 is called the Spinc bundle gerbe GW3(E) of E, whose Dixmi­er–Douady in­vari­ant is giv­en by the in­teg­ral third Stiefel–Whit­ney class W3(E)H3(X,Z). The ca­non­ic­al rep­res­ent­a­tion of Spinc(2k) gives a nat­ur­al in­clu­sion Spinc(2k)U(2k) which in­duces a com­mut­at­ive dia­gram U(1)Spinc(2k)SO(2k)=U(1)U(2k)PU(2k)=U(1)U(H)PU(H) This provides a re­duc­tion of the prin­cip­al PU(H)-bundle Po(E). The as­so­ci­ated bundle of Azu­maya al­geb­ras is in fact the bundle of Clif­ford al­geb­ras, whose bundle mod­ules are called Clif­ford mod­ules [e12]. Hence, there ex­ists a ca­non­ic­al iso­morph­ism between K0(X,o(E)) and the K-the­ory of the Clif­ford mod­ules as­so­ci­ated to the bundle Cliff(E).

4. Topological and analytic twisted K-homology

Let X be a CW-com­plex (or para­com­pact Haus­dorff space) with a twist­ing α:XK(Z,3) and Pα be the cor­res­pond­ing prin­cip­al K(Z,2)-bundle. Any base-point-pre­serving ac­tion of a K(Z,2) on a space defines an as­so­ci­ated bundle by the stand­ard con­struc­tion. In par­tic­u­lar, as a clas­si­fy­ing space of com­plex line bundles, K(Z,2) acts on the com­plex K-the­ory spec­trum K rep­res­ent­ing the tensor product by com­plex line bundles, where Kev=Z×BU(),Kodd=U(). De­note by Pα(K)=Pα×K(Z,2)K the bundle of based K-the­ory spec­tra over X. There is a sec­tion of Pα(K)=Pα×K(Z,2)K defined by tak­ing the base points of each fibre. The im­age of this sec­tion can be iden­ti­fied with X and we de­note by Pα(K)/X the quo­tient space of Pα(K) ob­tained by col­lapsing the im­age of this sec­tion.

The stable ho­mo­topy groups of Pα(K)/X by defin­i­tion give the to­po­lo­gic­al twis­ted K-ho­mo­logy groups Kev/oddtop(X,α). (There are only two due to Bott peri­od­icity of K.) Thus we have Kevtop(X,α)=limkπ2k(Pα(BU())/X) and Koddtop(X,α)=limkπ2k+1(Pα(BU())/X). Here the dir­ect lim­its are taken by the double sus­pen­sion πn+2k(Pα(BU())/X)πn+2k+2(Pα(S2BU())/X) and then fol­lowed by the stand­ard map πn+2k+2(Pα(S2BU())/X)b1πn+2k+2(Pα(BU()BU())/X) m πn+2k+2(Pα(BU())/X), where b:R2BU() rep­res­ents the Bott gen­er­at­or in K0(R2)Z, m is the base-point-pre­serving map in­du­cing the ring struc­ture on K-the­ory.

For a re­l­at­ive CW-com­plex (X,A) with a twist­ing α:XK(Z,3), the re­l­at­ive ver­sion of to­po­lo­gic­al twis­ted K-ho­mo­logy, de­noted by Kev/oddtop(X,A,α), is defined to be Kev/oddtop(X/A,α), where X/A is the quo­tient space of X ob­tained by col­lapsing A to a point. Then we have the ex­act se­quence Koddtop(X,A;α)Kevtop(A,α|A)Kevtop(X,α)Koddtop(X,α)Koddtop(A,α|A)Kevtop(X,A;α) and the ex­cision prop­er­ties Kev/oddtop(X,B;α)Kev/oddtop(A,AB;α|A) for any CW-tri­ad (X;A,B) with a twist­ing α:XK(Z,3). A triple (X;A,B) is a CW-tri­ad if X is a CW-com­plex, and A, B are two sub­com­plexes of X such that AB=X.

Ana­lyt­ic twis­ted K-ho­mo­logy, de­noted by Kev/oddan(X,α), is defined to be Kev/oddan(X,α):=KKev/odd(Γ0(X,Pα(K)),C), Kas­parov’s Z2-graded K-ho­mo­logy of the C-al­gebra Γ0(X,Pα(K)).

For a re­l­at­ive CW-com­plex (X,A) with a twist­ing α:XK(Z,3), the re­l­at­ive ver­sion of ana­lyt­ic twis­ted K-ho­mo­logy Kev/oddan(X,A,α) is defined to be Kev/oddan(XA,α). Then we have the ex­act se­quence Koddan(X,A;α)Kevan(A,α|A)Kevan(X,α)Koddan(X,α)Koddan(A,α|A)Kevan(X,A;α) and the ex­cision prop­er­ties Kev/oddan(X,B;α)Kev/oddan(A,AB;α|A) for any CW-tri­ad (X;A,B) with a twist­ing α:XK(Z,3).

The­or­em 4 (The­or­em 5.1 in [e33]): There is a nat­ur­al iso­morph­ism Φ:Kev/oddtop(X,α)Kev/oddan(X,α) for any smooth man­i­fold X with a twist­ing α:XK(Z,3).

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 HHH which in­duces a group ho­mo­morph­ism U(H)×U(H)U(H) whose re­stric­tion to the centre is the group mul­ti­plic­a­tion on U(1). So we have a group ho­mo­morph­ism PU(H)×PU(H)PU(H) which defines a con­tinu­ous map, de­noted by m, of CW-com­plexes BPU(H)×BPU(H)BPU(H). As BPU(H) is iden­ti­fied as K(Z,3), we may think of this as a con­tinu­ous map tak­ing K(Z,3)×K(Z,3)toK(Z,3), which can be used to define α+oX.

There are nat­ur­al iso­morph­isms from twis­ted K-ho­mo­logy (to­po­lo­gic­al, resp. ana­lyt­ic) to twis­ted K-the­ory (to­po­lo­gic­al, resp. ana­lyt­ic) of a smooth man­i­fold X where the twist­ing is shif­ted by αα+oX, where τ:XBSO is the clas­si­fy­ing map of the stable tan­gent space and α+oX de­notes the map XK(Z,3), rep­res­ent­ing the class [α]+W3(X)inH3(X,Z).

The­or­em 5: Let X be a smooth man­i­fold with a twist­ing α:XK(Z,3). There ex­ist iso­morph­isms Kev/oddtop(X,α)Ktopev/odd(X,α+oX) and Kev/oddan(X,α)Kanev/odd(X,α+oX), with the de­gree shif­ted by dimX(mod2).

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 (X,A) im­ply the fol­low­ing co­rol­lary.

Co­rol­lary 6: There is a nat­ur­al iso­morph­ism Φ:Kev/oddtop(X,A,α)Kev/oddan(X,A,α) for any smooth man­i­fold X with a twist­ing α:XK(Z,3) and a closed sub­man­i­fold AX.

Re­mark: In fact, Poin­caré du­al­ity as in The­or­em 5 holds for any com­pact Rieman­ni­an man­i­fold W with bound­ary W and a twist­ing α:WK(Z,3). This du­al­ity takes the form Kev/oddtop(W,α)Ktopev/odd(W,W,α+oW) and Kev/oddan(W,α)Kanev/odd(X,X,α+oW), with the de­gree shif­ted by dimW(mod2). From this, we have a nat­ur­al iso­morph­ism Φ:Kev/oddtop(X,A,α)Kev/oddan(X,A,α) for any CW pair (X,A) with a twist­ing α:XK(Z,3) us­ing the five lemma.

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 X is a para­com­pact Haus­dorff space and α:XK(Z,3) is a twist­ing over X.

Defin­i­tion: Giv­en a smooth ori­ented man­i­fold M with a clas­si­fy­ing map ν of its stable nor­mal bundle, we say that M is an α-twis­ted Spinc man­i­fold over X if M is equipped with an α-twis­ted Spinc struc­ture, that means, a con­tinu­ous map ι:MX such that the dia­gram shown to the right com­mutes up to a fixed ho­mo­topy η from W3ν and αι. Such an α-twis­ted Spinc man­i­fold over X will be de­noted by (M,ν,ι,η).

Pro­pos­i­tion 7: M ad­mits an α-twis­ted Spinc struc­ture if and only if there is a con­tinu­ous map ι:MX such that ι([α])+W3(M)=0. If ι is an em­bed­ding, this is the an­om­aly can­cel­la­tion con­di­tion ob­tained by Freed and Wit­ten in [e19].

As shown in [e33], this way of think­ing about twis­ted Spinc struc­tures gen­er­al­ises to cov­er high­er-de­gree twists. The lat­ter are im­port­ant in string to­po­logy.

A morph­ism between α-twis­ted Spinc man­i­folds (M1,ν1,ι1,η1)and(M2,ν2,ι2,η2) is a con­tinu­ous map f:M1M2 where the dia­gram shown to the right is a ho­mo­topy com­mut­at­ive dia­gram such that

  1. ν1 is ho­mo­top­ic to ν2f through a con­tinu­ous map ν:M1×[0,1]BSO;

  2. ι2f is ho­mo­top­ic to ι1 through a con­tinu­ous map ι:M1×[0,1]X;

  3. the com­pos­i­tion of ho­mo­top­ies (αι)(η2(f×id))(W3ν) is ho­mo­top­ic to η1.

Two α-twis­ted Spinc man­i­folds (M1,ν1,ι1,η1) and (M2,ν2,ι2,η2) are called iso­morph­ic if there ex­ists a dif­feo­morph­ism f:M1M2 such that the above holds. If the iden­tity map on M in­duces an iso­morph­ism between (M,ν1, ι1,η1) and (M,ν2, ι2,η2), then these two α-twis­ted Spinc struc­tures are called equi­val­ent.

Ori­ent­a­tion re­versal in the Grass­man­ni­an mod­el defines an in­vol­u­tion r:BSOBSO. Us­ing this, one may de­term­ine a unique equi­val­ence class of α-twis­ted Spinc struc­tures on M, called the op­pos­ite α-twis­ted Spinc struc­ture, simply de­noted by (M,ν,ι,η).

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 (X,α) is a quin­tuple (M,ι,ν,η,[E]), where [E] is a K-class in K0(M) and M is a smooth closed man­i­fold equipped with an α-twis­ted Spinc struc­ture (M,ι,ν,η).

Two geo­met­ric cycles (M1,ι1,ν1,η1,[E1]) and (M2,ι,2ν2,η2,[E2]) are iso­morph­ic if there is an iso­morph­ism f:(M1,ι1,ν1,η1)(M2,ι2,ν2,η2), as α-twis­ted Spinc man­i­folds over X, such that f!([E1])=[E2].

Let Γ(X,α) be the col­lec­tion of all geo­met­ric cycles for (X,α). We now im­pose an equi­val­ence re­la­tion on Γ(X,α) gen­er­ated by the fol­low­ing three ele­ment­ary re­la­tions:

  1. Dir­ect sum–dis­joint uni­on: If (M,ι,ν,η,[E1]) and (M,ι,ν,η,[E2]) are two geo­met­ric cycles with the same α-twis­ted Spinc struc­ture, then (M,ι,ν,η,[E1])(M,ι,ν,η,[E2])(M,ι,ν,η,[E1]+[E2]).

  2. Bor­d­ism: Giv­en two geo­met­ric cycles (M1,ι1,ν1,η1,[E1]) and (M2,ι2, ν2, η2,[E2]), we call them bord­ant if there ex­ists an α-twis­ted Spinc man­i­fold (W,ι,ν,η) and [E]K0(W) such that (W,ι,ν,η)=(M1,ι1,ν1,η1)(M2,ι2,ν2,η2) and ([E])=[E1][E2]. Here (M1,ι1,ν1,η1) de­notes the man­i­fold M1 with the op­pos­ite α-twis­ted Spinc struc­ture.

  3. Spinc vec­tor bundle modi­fic­a­tion: Sup­pose we are giv­en a geo­met­ric cycle (M,ι,ν,η,[E]) and a Spinc vec­tor bundle V over M with even-di­men­sion­al fibres. De­note by R the trivi­al rank-1 real vec­tor bundle. Choose a Rieman­ni­an met­ric on VR, and let M^=S(VR) be the sphere bundle of VR. Then the ver­tic­al tan­gent bundle Tv(M^) of M^ ad­mits a nat­ur­al Spinc struc­ture with an as­so­ci­ated Z2-graded spinor bundle SV+SV . De­note by ρ:M^M the pro­jec­tion which is K-ori­ented. Then (M,ι,ν,η,[E])(M^,ιρ,νρ,ηρ,[ρESV+]).

Defin­i­tion: De­note by Kgeo(X,α)=Γ(X,α)/ the geo­met­ric twis­ted K-ho­mo­logy. Ad­di­tion is giv­en by the dir­ect sum - dis­joint uni­on re­la­tion. Note that the equi­val­ence re­la­tion pre­serves the par­ity of the di­men­sion of the un­der­ly­ing α-twis­ted Spinc man­i­fold. Now let K0geo(X,α) (resp. K1geo(X,α)) be the sub­group of Kgeo(X,α) de­term­ined by all geo­met­ric cycles with even-di­men­sion­al (resp. odd-di­men­sion­al) α-twis­ted Spinc man­i­folds.

Re­mark:

  1. If M, in a geo­met­ric cycle (M,ι,ν,η,[E]) for (X,α), is a com­pact man­i­fold with bound­ary, then [E] has to be a class in K0(M,M).

  2. If f:XY is a con­tinu­ous map and α:YK(Z,3) is a twist­ing, then there is a nat­ur­al ho­mo­morph­ism of abeli­an groups f:Kev/oddgeo(X,αf)Kev/oddgeo(Y,α) send­ing [M,ι,ν,η,E] to [M,fι,ν,η,E].

  3. Let A be a closed sub­space of X and α be a twist­ing on X. A re­l­at­ive geo­met­ric cycle for (X,A;α) is a quin­tuple (M,ι,ν,η,[E]) such that

    1. M is a smooth man­i­fold (pos­sibly with bound­ary), equipped with an α-twis­ted Spinc struc­ture (M,ι,ν,η);

    2. if M has a nonempty bound­ary, then ι(M)A;

    3. [E] is a K-class in K0(M) rep­res­en­ted by a Z2-graded vec­tor bundle E over M, or a con­tinu­ous map MBU().

The re­la­tion gen­er­ated by dir­ect sum - dis­joint uni­on, bor­d­ism and Spinc vec­tor bundle modi­fic­a­tion is an equi­val­ence re­la­tion. The col­lec­tion of re­l­at­ive geo­met­ric cycles, mod­ulo the equi­val­ence re­la­tion is de­noted by Kev/oddgeo(X,A;α).

5.2. D-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 C-al­geb­ras. We em­phas­ise that in this new ap­proach the pro­ject­ive unit­ary group is equipped with the strong op­er­at­or to­po­logy, not the norm to­po­logy as was used in earli­er sec­tions.

We be­gin by re­call­ing a few con­cepts.

  1. Giv­en an ori­ented real Eu­c­lidean vec­tor bundle F of rank K over a para­com­pact Haus­dorff to­po­lo­gic­al space Y, a Spinc struc­ture on F is a lift of the ori­ented frame bundle PSO(F) to a prin­cip­al Spinc(k)-bundle PSpinc(F), where 1U(1)Spinc(k)SO(k)1 is the unique (for k>2) non­trivi­al cent­ral ex­ten­sion of SO(k) by U(1). See Ap­pendix D in [e11] for an equi­val­ent defin­i­tion of Spinc struc­tures (and note that there is a well-known modi­fic­a­tion needed for k2). A real vec­tor bundle with a Spinc struc­ture is called a Spinc vec­tor bundle. Spinc struc­tures are ori­ent­a­tion con­di­tions for com­plex K-the­ory in the sense that a Spinc 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 F de­term­ines an ori­ent­a­tion of F.

  2. (two-out-of-three prin­ciple) Let 0FFF0 be a short ex­act se­quence of ori­ented real vec­tor bundles on a para­com­pact Haus­dorff to­po­lo­gic­al space Y. Then Spinc struc­tures for any two of the vec­tor bundles de­term­ine a Spinc struc­ture for the third vec­tor bundle.

  3. A Spinc Rieman­ni­an man­i­fold is a Rieman­ni­an man­i­fold W (per­haps with bound­ary) whose tan­gent bundle TW is a Spinc vec­tor bundle. If W has a bound­ary W, at each bound­ary point, the out­ward nor­mal vec­tor defines a trivi­al rank-1 real vec­tor bundle NW over W and 0T(W)TW|WNW0 is an ex­act se­quence of real Eu­c­lidean vec­tor bundles over W. There­fore, the two-out-of-three prin­ciple im­plies that if a Spinc man­i­fold W has a bound­ary, then this bound­ary, W, is a Spinc man­i­fold in a ca­non­ic­al way.

5.2.1. Twisting data

If H is a com­plex Hil­bert space, K(H) will de­note the C-al­gebra of all com­pact op­er­at­ors on H. Hil­bert spaces will be as­sumed to be sep­ar­able. Re­call that a C-al­gebra A is an ele­ment­ary C-al­gebra if there ex­ists a com­plex Hil­bert space H and an iso­morph­ism of C-al­geb­ras AK(H).

Defin­i­tion: Let X be a second-count­able loc­ally com­pact Haus­dorff to­po­lo­gic­al space. A twist­ing datum on X is a loc­ally trivi­al bundle A of ele­ment­ary C-al­geb­ras on X, that is, each fibre of A is an ele­ment­ary C-al­gebra with struc­ture group the auto­morph­ism group of K(H) for some com­plex Hil­bert space.

If E is a (loc­ally trivi­al) bundle of Hil­bert spaces on X, then K(E) is the twist­ing datum defined by K(E)x=K(Ex),xX. Any twist­ing datum A is loc­ally of the form K(E); i.e., for any xX there ex­ists an open set U in X with xU and a (loc­ally trivi­al) Hil­bert space vec­tor bundle E on U with A|UK(E).

For a C-al­gebra A, Aop de­notes the op­pos­ite C-al­gebra. As Banach spaces A=Aop, and re­mains un­changed. Thus ab in Aop is ba in A. If A is a twist­ing datum on X, then Aop is the twist­ing datum ob­tained by re­pla­cing each fibre Ax by Axop. If A and B are twist­ing data on X, then AB is the twist­ing datum on X whose fibre at xX is the C-al­gebra AxBx.

Let A be a twist­ing datum on X, and as­sume that the fibre of A is in­fin­ite-di­men­sion­al. As be­fore P de­notes the prin­cip­al PU(H) bundle on X whose fibre at xX is Px={C-algebra isomorphisms:K(H)Ax}. There is then the ca­non­ic­al iso­morph­ism of twist­ing data on X AP×U(H)K(H). Note that the prin­cip­al PU(H) bundle P is clas­si­fied by a con­tinu­ous map XBPU(H). Let DD(A) de­note the Dixmi­er–Douady in­vari­ant of A (in H3(X,Z)).

On the oth­er hand if A has fi­nite-di­men­sion­al fibres, we let B be any twist­ing datum on X with every fibre of B in­fin­ite-di­men­sion­al and then DD(A) is defined by DD(A)=:DD(AB)DD(B). For any two twist­ing data A,B on X we have DD(AB)=DD(A)+DD(B) and DD(Aop)=DD(A).

Defin­i­tion: A spinor bundle for A is a vec­tor bundle S of Hil­bert spaces on X to­geth­er with a giv­en iso­morph­ism of twist­ing data AK(S). A spinor bundle for A ex­ists if and only if DD(A)=0.

Lemma 8: Let A be any twist­ing datum on X. Then there is a ca­non­ic­al spinor bundle for AAop.

Proof.  The set of Hil­bert–Schmidt op­er­at­ors on H, de­noted by LH-S(H), is an ideal in K(H). The C-val­ued in­ner product T1,T2=Trace(T1T2) makes LH-S(H) in­to a Hil­bert space. Now let A be an ele­ment­ary C-al­gebra. Choose an iso­morph­ism of C-al­geb­ras ψ:AK(H). Now, ψ1(LH-S(H)) is an ideal in A and is in­de­pend­ent of the choice of ψ be­cause the Hil­bert–Schmidt op­er­at­ors are in­vari­ant un­der Aut(K(H))=PU(H). De­note this ideal by AH-S. The left-mul­ti­plic­a­tion and right-mul­ti­plic­a­tion ac­tions of A on AH-S com­bine to give a left ac­tion of AAop on AH-S which iden­ti­fies AAop with the com­pact op­er­at­ors on the Hil­bert space AH-S: AAopK(AH-S). If A is a twist­ing datum on X, let S be the vec­tor bundle of Hil­bert spaces on X whose fibre at xX is (Ax)H-S. Then S is a spinor bundle for AAop and is well-defined as our con­struc­tion is in­de­pend­ent of ψ.

We im­port our pre­vi­ous nota­tion: if A is a twist­ing datum on X, then Γ0(X,A) de­notes the C-al­gebra of all con­tinu­ous van­ish­ing-at-in­fin­ity sec­tions of A. Re­call that the com­pactly sup­por­ted Kas­parov group KKc(Γ0(X,A),C) is KKcj(Γ0(X,A),C):=limΔXΔ compactKKj(Γ(Δ,A),C),j=0,1, where Γ(Δ,A) is the C-al­gebra of all con­tinu­ous sec­tions of A re­stric­ted to Δ. Since X is a CW-com­plex, this is equal to the dir­ect lim­it over the fi­nite sub-CW-com­plexes of X. We will refer to KKc(Γ0(X,A),C) as the twis­ted Kas­parov K-ho­mo­logy of X.

We now ex­pound some ele­ment­ary res­ults with proofs to give the fla­vour of the Kas­parov the­ory we used.

Lemma 9: A spinor bundle S for ABop de­term­ines a stable iso­morph­ism between Γ0(X,A) and Γ0(X,B), and thus gives an iso­morph­ism ΨS:KKcj(Γ0(X,A),C)KKcj(Γ0(X,B),C),j=0,1.

Proof.  Let BH-S be the spinor bundle for BopB. Lemma 8 im­plies ABopBAK(BH-S). If S is a spinor bundle for ABop, then ABopBK(S)B. There­fore AK(BH-S)K(S)B. Note that the Dixmi­er–Douady in­vari­ants of K(BH-S) and K(S) are zero. For any com­pact sub­space ΔX, we have KKj(Γ(Δ,A),C)KKj(Γ(Δ,AK(BH-S)),C)KKj(Γ(Δ,K(S)B),C)KKj(Γ(Δ,B),C). Here the first and the third iso­morph­isms are provided by Mor­ita equi­val­ence bimod­ules BH-S and S re­spect­ively. Passing to the dir­ect lim­it, we get the de­sired iso­morph­ism ΨS.

5.2.2. K-cycles for twisted K-homology

As above, X is a loc­ally fi­nite CW-com­plex and A is a twist­ing datum on X.

Defin­i­tion: A twis­ted-by-A K-cycle on X is a triple (M,σ,φ) where

  • M is a com­pact Spinc man­i­fold without bound­ary,

  • φ:MX is a con­tinu­ous map,

  • σK0(Γ(M,φAop)) (the K0-group of the C-al­gebra Γ(M,φAop)).

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 D-cycles, mo­tiv­ated by D-branes in string the­ory, which may be re­garded as rep­res­ent­ing geo­met­ric cycles for (X,A). In fact the two res­ult­ing twis­ted K-ho­mo­lo­gies are iso­morph­ic.

Keep­ing X,A fixed, de­note by {(M,φ,σ)} the col­lec­tion of all twis­ted-by-A K-cycles on X. On this col­lec­tion im­pose the equi­val­ence re­la­tion gen­er­ated by the two ele­ment­ary steps

  • bor­d­ism,

  • vec­tor bundle modi­fic­a­tion.

Next, we elab­or­ate on these two steps for the case of twis­ted K-cycles.

Iso­morph­ism: We say (M,φ,σ) is iso­morph­ic to (M,φ,σ) if and only if there ex­ists a dif­feo­morph­ism ψ:MM pre­serving the Spinc-struc­tures and with com­mut­ativ­ity in the dia­gram shown to the right and in ad­di­tion ψ(σ)=σ, where ψ:K0Γ(M,φA)K0Γ(M,φA) is the iso­morph­ism of K-the­ory de­term­ined by the iso­morph­ism of C-al­geb­ras Γ(M,φAop)Γ(M,φAop).

Bor­d­ism: We say (M0,φ0,σ0) is bord­ant to (M1,φ1,σ1) if and only if there ex­ists (W,φ,σ) such that

  1. W is a com­pact Spinc man­i­fold with bound­ary;

  2. φ is a con­tinu­ous map from W to X;

  3. σK0(Γ(W,φAop));

  4. (W,φ|W,σ|W)(M0,φ0,σ0)(M1,φ1,σ1).

Vec­tor bundle modi­fic­a­tion: Let (M,φ,σ) be a twis­ted-by-A K-cycle on X, and let F be a Spinc vec­tor bundle on M of even rank. As in the un­twis­ted case, 1R de­notes the trivi­al real line bundle on M, S(F1R) is the unit sphere bundle of F1R and π:S(F1R)M is a fibra­tion. Let s:MS(F1R) be the ca­non­ic­al unit sec­tion of 1R. Then the giv­en Spinc struc­ture for F de­term­ines a Gys­in ho­mo­morph­ism (see [e32]) s:K0(Γ(M,φAop))K0(Γ(S(F1R),(φπ)Aop)). Here we use the fact that s(φπ)Aop=φAop. Then (M,φ,σ)(S(F1R),φπ,sσ). In the fol­low­ing (S(F1R),φπ,sσ) will be de­noted by F#(M,φ,σ) and will be re­ferred to as the modi­fic­a­tion of (M,φ,σ) by F.

Com­pos­i­tion Lemma: Let (M,φ,σ) be a twis­ted-by-A K-cycle on X, and let F be an even-rank Spinc vec­tor bundle on M. Let F1 be an even-rank Spinc vec­tor bundle on S(F1R). Then F1#(F#(M,φ,σ)) is (in a ca­non­ic­al way) bord­ant to (sF1F)#(M,φ,σ).

Proof.  We first con­struct a bord­ant man­i­fold W between S(F11R)andS(sF1F1R). Let D(F11R) be the unit ball bundle of F11R; then S(F11R)=(D(F11R)) is the un­der­ly­ing man­i­fold for F1#(F#(M,φ,σ)). Note that sFF1R is iso­morph­ic to the nor­mal bundle νι for the in­clu­sion map ι:MD(F11R), where ι is defined by the com­pos­i­tion of s:MS(F1R) and the zero sec­tion of the bundle F11R over S(F1R). This en­sures that we can identi­fy the sphere bundle of sF1F1R with the bound­ary of the ball bundle of ra­di­us 14 in the nor­mal bundle νι. Define W~=D(F11R)D1/4(νι). Then W~S(F11R)S(sF1F1R). Let φ~:W~D(F11R) π S(F1R)πFM be the ob­vi­ous pro­jec­tion and s~ be the com­pos­i­tion of the ca­non­ic­al unit sec­tions s and s1 for F1R and F11R re­spect­ively. Then (W~,φ~,s~σ) provides the bor­d­ism between F1#(F#(M,φ,σ)) and (sF1F)#(M,φ,σ). Here we ap­plied the facts that the push­for­ward map s~ is func­tori­al and is also a ho­mo­topy in­vari­ant. ☐

Two twis­ted-by-A K-cycles (M,φ,σ) and (M,φ,σ) on X are equi­val­ent if and only if it is pos­sible to pass from (M,φ,σ) to (M,φ,σ) by a fi­nite se­quence of the two ele­ment­ary steps. The K-cycle (or to­po­lo­gic­al) twis­ted-by-A K-ho­mo­logy of X, de­noted by Ktop(X,A), is the set of equi­val­ence classes of twis­ted-by-A K-cycles: Ktop(X,A):={(M,φ,σ)}/. Ad­di­tion in Ktop(X,A) is defined by dis­joint uni­on of twis­ted-by-A K-cycles: (M,φ,σ)+(M,φ,σ)=(MM,φφ,σσ).

Defin­i­tion: Let (M,φ,σ) be a twis­ted-by-A K-cycle on X. We say (M,φ,σ) bounds if and only if there ex­ists (W,φ~,σ~), where

  1. W is a com­pact Spinc man­i­fold with bound­ary,

  2. φ~:WX is a con­tinu­ous map,

  3. σ~K0(Γ(W,φ~Aop)),

  4. (W,φ~|W,σ~|W)(M,φ,σ).

As in the un­twis­ted case, the ad­dit­ive in­verse of (M,φ,σ) is (M,φ,σ). The equi­val­ence re­la­tion on twis­ted-by-A K-cycles (M,φ,σ) pre­serves the di­men­sion of M mod­ulo 2. There­fore, as an abeli­an group, Ktop(X,A) is the dir­ect sum Ktop(X,A)=K0top(X,A)K1top(X,A), where Kjtop(X,A) is the sub­group of Ktop(X,A) gen­er­ated by those twis­ted-by-A K-cycles (M,φ,σ) such that every con­nec­ted com­pon­ent of M has di­men­sion jmodulo2, j=0,1.

There is a nat­ur­al map ηX:Kjtop(X,A)KKcj(Γ0(X,A),C) defined as fol­lows. Giv­en a twis­ted-by-A K-cycle (M,φ,σ) on X, by ap­ply­ing Poin­caré du­al­ity in twis­ted K-the­ory (see [e31] and [e35]) KK0(C,Γ(M,φAop))KKj(Γ(M,φA),C), we have PD(σ)KKj(Γ(M,φA),C). De­note by φ:KK(Γ(M,φA),C)KKc(Γ0(X,A),C) the map of abeli­an groups in­duced by φ:MX. Then the nat­ur­al map ηX is giv­en by ηX(M,φ,σ)=φ(PD(σ)). It is routine to check that ηX is well-defined on Kjtop(X,A) and is func­tori­al in the fol­low­ing sense. Let f:YX be a con­tinu­ous map and X be equipped with a twist­ing datum A; then the fol­low­ing dia­gram com­mutes: Kjtop(Y,fA)fKjtop(X,A)ηYηXKKcj(Γ0(Y,fA),C)fKKcj(Γ0(X,A),C). Here f:Kjtop(Y,fA)Kjtop(X,A) is defined by (M,φ,σ)(M,fφ,σ) on the level of K-cycles.

Paul has of­ten enun­ci­ated the un­twis­ted ver­sion of the fol­low­ing prob­lem.

The Gen­er­al Twis­ted In­dex Prob­lem: When X is a loc­ally fi­nite CW-com­plex and A is a twist­ing datum on X then the gen­er­al twis­ted in­dex prob­lem is: giv­en an ele­ment ξKKc(Γ0(X,A),C), ex­pli­citly con­struct a twis­ted-by-A K-cycle (M,φ,σ) for (X,A) with η(M,φ,σ)=ξ.

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 X.

The­or­em 10: Let X be a loc­ally fi­nite CW-com­plex with a giv­en twist­ing datum A. Then the nat­ur­al map ηX:Kjtop(X,A)KKcj(Γ0(X,A),C),j=0,1, is an iso­morph­ism of abeli­an groups.

5.2.3. Normal bordism

One of the main in­nov­a­tions in [4] was the elu­cid­a­tion of the fun­da­ment­al role of nor­mal bor­d­ism. Let (M,φ,σ) be a twis­ted-by-A K-cycle on X. A nor­mal bundle for M is an real vec­tor bundle ν on M to­geth­er with a giv­en short ex­act se­quence 0TMM×Rqν0 of real vec­tor bundles on M. Here q is a pos­it­ive in­teger and M×Rq is the trivi­al real vec­tor bundle with fibre Rq. The two-out-of-three prin­ciple im­plies that ν is a Spinc vec­tor bundle. If ν has even-di­men­sion­al fibres, then the modi­fic­a­tion of (M,φ,σ) by ν will be de­noted by ν#(M,φ,σ).

Defin­i­tion: Two twis­ted-by-A K-cycles (M,φ,σ) and (M,φ,σ) are nor­mally bord­ant, de­noted by (M,φ,σ)N(M,φ,σ), if there ex­ist nor­mal bundles with even-di­men­sion­al fibres ν and ν for M and M re­spect­ively such that ν#(M,φ,σ) is bord­ant to ν#(M,φ,σ).

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 K-cycles (M,φ,σ) and (M,φ,σ) are equi­val­ent if and only if they are nor­mally bord­ant.

A twis­ted-by-A K-cycle (M,φ,σ) is zero in Ktop(X,A) if and only if (M,φ,σ) is nor­mally bord­ant to a twis­ted-by-A K-cycle which bounds.

5.2.4. The group of D-cycles

In this sec­tion, we in­tro­duce an­oth­er no­tion of K-cycles for (X,A), called D-cycles, which are closely re­lated to the no­tion of D-branes in string the­ory. We be­gin with the fol­low­ing:

Defin­i­tion: A D-cycle for (X,A) is a 4-tuple (M,E,φ,S) such that

  1. M is a closed ori­ented C Rieman­ni­an man­i­fold;

  2. E is a com­plex vec­tor bundle on M;

  3. φ is a con­tinu­ous map from M to X;

  4. S is a spinor bundle for CliffC+(TM)φAop.

Re­mark:

  1. If M is even-di­men­sion­al, then CliffC+(TM) is the twist­ing datum on M whose fibre at pM is the com­plexi­fied Clif­ford al­gebra CRCliff(TpM). As usu­al, TpM is the tan­gent space to M at p. On the oth­er hand, if M is odd-di­men­sion­al, giv­en pM, choose a pos­it­ively ori­ented or­thonor­mal basis e1,e2,,en for TpM. Set n=2r+1 and define ω(p)CRCliff(TpM) by ω(p)=ir+1e1e2en. Then ω(p) does not de­pend on the choice of pos­it­ively ori­ented or­thonor­mal basis for TpM. Also ω(p) is in the centre of CliffC(TpM)=CRCliff(TpM) and ω(p)2=1.

    Now set CliffC+(TpM)={aCliffC(TpM)ω(p)a=a}. Then CliffC+(TM) is the twist­ing datum on M whose fibre at pM is CliffC+(TpM).

  2. The ex­ist­ence of a spinor bundle S for CliffC+(TM)φAop im­plies DD(CliffC+(TM))=φ(DD(A)). By stand­ard al­geb­ra­ic to­po­logy, DD(CliffC+(TM)) is the third (in­teg­ral) Stiefel–Whit­ney class of M, so the ex­ist­ence of S im­plies W3(M)=φ(DD(A)), which is the Freed–Wit­ten an­om­aly can­cel­la­tion con­di­tion for Type IIB D-branes as ex­plained in [e33].

Defin­i­tion: Keep­ing (X,A) fixed, de­note by {(M,E,φ,S)} the col­lec­tion of all D-cycles for (X,A). On this col­lec­tion im­pose the equi­val­ence re­la­tion gen­er­ated by the three ele­ment­ary steps

  • bor­d­ism,

  • dir­ect sum - dis­joint uni­on,

  • vec­tor bundle modi­fic­a­tion.

These three ele­ment­ary moves can be pre­cisely defined as fol­lows.

Iso­morph­ism: Two D-cycles (M,E,φ,S), (M,E,φ,S) for (X,A) are iso­morph­ic if there is an ori­ent­a­tion-pre­serving iso­met­ric dif­feo­morph­ism f:MM such that the dia­gram shown to the right com­mutes, and fEE, fSS.

Bor­d­ism: Two D-cycles (M0,E0,φ0,S0), (M1,E1,φ1,S1) for (X,A) are bord­ant if there ex­ists a 4-tuple (W,E,Φ,S) such that W is a com­pact ori­ented Rieman­ni­an man­i­fold with bound­ary, E is a com­plex vec­tor bundle on W, Φ is a con­tinu­ous map from WX and (W,E|W,Φ|W,S(+)|W)(M0,E0,φ0,S0)(M1,E1,φ1,S1). When W is of odd di­men­sion, CliffC(+)(TW)|WCliffC(T(W). Then S(+)=S. When W is of even di­men­sion, S(+) is the pos­it­ive part of S, the (+1)-ei­gen­bundle of the chir­al­ity sec­tion of CliffC(+)(TW).

Dir­ect sum-dis­joint uni­on: Let (M,E,φ,S) be a D-cycle for (X,A) and let E be a com­plex vec­tor bundle on M; then (M,E,φ,S)(M,E,φ,S)(M,EE,φ,S).

Vec­tor bundle modi­fic­a­tion: Let (M,E,φ,S) be a D-cycle for (X,A) and let F be a Spinc vec­tor bundle on M with even-di­men­sion­al fibres. Then, with SF be­ing the Spinor bundle for the Spinc vec­tor bundle F, (M,E,φ,S)(S(F1R),βπE,φπ,πSFπS), with the oth­er nota­tion in this equi­val­ence be­ing the same as that in­tro­duced pre­vi­ously. The group of D-cycles for (X,A), de­noted by Kgeo(X,A), is the set of equi­val­ence classes of these D-cycles un­der the equi­val­ence defined above in terms of the three re­la­tions. Ad­di­tion in Kgeo(X,A) is dis­joint uni­on of D-cycles: (M,E,φ,S)+(M,E,φ,S)=(MM,EE,φφ,SS). The equi­val­ence re­la­tion on D-cycles (M,E,φ,S) pre­serves the di­men­sion of M mod­ulo 2 so that we have Kgeo(X,A)=K0geo(X,A)K1geo(X,A), where Kjgeo(X,A) is the sub­group of Kgeo(X,A) gen­er­ated by those (X,A) D-cycles (M,E,φ,S) such that every con­nec­ted com­pon­ent of M has di­men­sion j mod­ulo 2, j=0,1.

Let (M,E,φ,S) be a D-cycle on (X,A) such that every con­nec­ted com­pon­ent of M has its di­men­sion con­gru­ent to j mod­ulo 2. Let DE de­note the Dir­ac ele­ment of M tensored with E. It can be de­scribed in terms of the 12-sig­na­ture op­er­at­or on M. Then as in [e8], DE yields an ele­ment in the Kas­parov K-ho­mo­logy group [DE]KKj(C(M,CliffC(+)(M)),C). The iso­morph­ism ΨS defined in Lemma 9 in Sec­tion 5 reads as Ψ:KKj(C(M,CliffC(+)(M)),C)KKj(Γ(M,φA),C). The map of C-al­geb­ras Γ(X,A)Γ(M,φA) in­duces a ho­mo­morph­ism of abeli­an groups φ:KKj(Γ(M,φA),C)KKj(Γ(X,A),C). Then (M,E,φ,S)φ[DE] yields a ho­mo­morph­ism of abeli­an groups, de­noted by μ:Kgeo(X,A)KK(Γ(X,A),C). We call this the twis­ted in­dex map in twis­ted K-ho­mo­logy of (X,A).

5.2.5. The charge map Kgeo(X,A)Ktop(X,A)
We use nor­mal bundle modi­fic­a­tion to define a ho­mo­morph­ism of abeli­an groups h:Kgeo(X,A)Ktop(X,A). Here h should be viewed as the map which sends a D-cycle (M,E,φ,S) to its charge.

Defin­i­tion: Giv­en a D-cycle (M,E,φ,S) choose a nor­mal bundle ν for M, with even-di­men­sion­al fibres. Then h(M,E,φ,S):=ν#(M,E,φ,S)=(S(ν1R),φπ,σ). Note that S(ν1R) is a Spinc man­i­fold be­cause its tan­gent bundle is stably trivi­al­ized. Here σ is the ele­ment in K0(Γ(S(ν1R,(φπ)Aop)) ob­tained from E as fol­lows:

  1. De­note by s:K0(M)K0(Γ(S(ν1R),πCliffC(ν))) the Gys­in ho­mo­morph­ism as­so­ci­ated to the ca­non­ic­al sec­tion s:MS(ν1R).

  2. Ob­serve that, as twist­ing data on M, a trivi­al­isa­tion of TMν gives an equi­val­ence CliffC+(TM)φAopCliffC(ν)opφAop, a stable iso­morph­ism of bundles of ele­ment­ary C-al­geb­ras. Hence the giv­en spinor bundle S for CliffC+(TM)φAop de­term­ines a spinor bundle S~ for CliffC(ν)opφAop. Then πS~ is a spinor bundle for πCliffC(ν)op(φπ)Aop and there­fore yields an iso­morph­ism of abeli­an groups χ:K0(Γ(S(ν1R),πCliffC(ν)))K0(Γ(S(ν1R),(φπ)Aop)).

  3. σ:=χ(s[E]).

Ques­tion: Is the map h:Kjgeo(X,A)Kjtop(X,A) an iso­morph­ism for any loc­ally fi­nite CW-com­plex?

Giv­en a pos­it­ive an­swer to this ques­tion then we would have the fol­low­ing co­rol­lary:

Con­sequence: Let X be a loc­ally fi­nite CW-com­plex with a twist­ing datum A defined by a prin­cip­al PU(H)-bundle P, and Kgeo(X,P) be the twis­ted geo­met­ric K-ho­mo­logy of (X,P) in [e33]. Then the twis­ted in­dex map μ:Kgeo(X,P)KKc(Γ(X,A),C) is an iso­morph­ism.

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 Spinc-bor­d­ism of a CW-com­plex de­term­ines its K-ho­mo­logy by a simple al­geb­ra­ic tensor product. That is, the af­firm­at­ive an­swer to the ques­tion posed above that is es­tab­lished in [5] rests on a twis­ted ver­sion of this the­or­em of Hop­kins and Hovey.

Re­call that giv­en a CW-com­plex X with a twist­ing α:XK(Z,3), there are two ap­proaches to define to twis­ted Spinc-bor­d­ism the­ory of (X,α) in [e33]. The first ap­proach is the geo­met­ric defin­i­tion us­ing a smooth man­i­fold M with a con­tinu­ous map ι:MX and a ho­mo­topy real­ising the Freed–Wit­ten con­di­tion (see [e19]) (5)ι([α])+W3(M)=0, where W3(M) is the third in­teg­ral Stiefel–Whit­ney class of M and [α] de­notes the ho­mo­topy class of α in [X,K(Z,3)]H3(X,Z). The α-twis­ted Spinc-bor­d­ism group of X, de­noted by ΩSpinc(X,α), is defined to be the set of all iso­morph­ism classes of closed α-twis­ted Spinc man­i­folds over X mod­ulo null-bor­d­ism, with the sum giv­en by the dis­joint uni­on. The twis­ted Hop­kins–Hovey the­or­em in [5] gives a pre­cise re­la­tion­ship between this former group and the D-cycle mod­el for twis­ted K-ho­mo­logy.

This res­ult is the key to prov­ing that the geo­met­ric mod­el for twis­ted K-ho­mo­logy in terms of D-cycles, (stud­ied in one form by Bai-Ling Wang and re­for­mu­lated by us as ex­plained in [4]) is in fact iso­morph­ic to Kas­parov’s ana­lyt­ic twis­ted K-ho­mo­logy. Moreover these D-cycles are a math­em­at­ic­al ver­sion of the phys­i­cists’ D-branes.

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.

Works

[1] P. Baum and R. G. Douglas: “In­dex the­ory, bor­d­ism, and K-ho­mo­logy,” pp. 1–​31 in Op­er­at­or al­geb­ras and K-the­ory (San Fran­cisco, 7–8 Janu­ary 1981). Edi­ted by R. G. Douglas and C. Schochet. Con­tem­por­ary Math­em­at­ics 10. Amer­ic­an Math­em­at­ic­al So­ci­ety (Provid­ence, RI), 1982. MR 658506 Zbl 0507.​55004 incollection

[2] P. Baum and R. G. Douglas: “K ho­mo­logy and in­dex the­ory,” pp. 117–​173 in Op­er­at­or al­geb­ras and ap­plic­a­tions (King­ston, ON, 14 Ju­ly–2 Au­gust 1980), part 1. Edi­ted by R. V. Kadis­on. Pro­ceed­ings of Sym­po­sia in Pure Math­em­at­ics 38. Amer­ic­an Math­em­at­ic­al So­ci­ety (Provid­ence, RI), 1982. MR 679698 Zbl 0532.​55004 incollection

[3] P. Baum, N. Hig­son, and T. Schick: “On the equi­val­ence of geo­met­ric and ana­lyt­ic K-ho­mo­logy,” pp. 1–​24 in Spe­cial is­sue: In hon­or of Robert D. MacPh­er­son, Part 3, published as Pure Ap­pl. Math. Q. 3 : 1. In­ter­na­tion­al Press (Som­merville, MA), 2007. MR 2330153 Zbl 1146.​19004 incollection

[4] P. Baum, A. Carey, and B.-L. Wang: “K-cycles for twis­ted K-ho­mo­logy,” pp. 69–​98 in Nanjing spe­cial is­sue on K-the­ory, num­ber the­ory and geo­metry, published as J. K-The­ory 12 : 1. Issue edi­ted by X. Guo, H. Qin, and G. Tang. Cam­bridge Uni­versity Press, August 2013. MR 3126635 Zbl 1300.​19003 incollection

[5] P. Baum, M. Joachim, M. Khorami, and T. Schick: Twis­ted spinc-bor­d­ism, D-branes, and twis­ted K-ho­mo­logy. In pre­par­a­tion. techreport