nLab
topological K-theory

Context

Cohomology

cohomology

Special and general types

Special notions

Variants

Extra structure

Operations

Theorems

Contents

Idea

What is called topological K-theory is a collection of generalized (Eilenberg-Steenrod) cohomology theories whose cocycles in degree 0 on a space X can be represented by pairs of vector bundles, real or complex ones, on X modulo a certain equivalence relation.

Notice that “ordinary cohomology” is the generalized (Eilenberg-Steenrod) cohomology that is represented by the Eilenberg-MacLane spectrum which, as a stably abelian infinity-groupoid is just the additive group

\mathbb{Z}

of integers.

To a large extent K-theory is the cohomology theory obtained by categorifying this once:

somethinglikeVect.\mathbb{Z} \;\; \mapsto something like \mathbf{Vect} \,.

Motivational example: “nonabelian K-cohomology”

To see how this works, first consider the task of generalizing the “nonabelian cohomology” or cohomotopy theory given by the coefficient object , the additive semi-group of of natural numbers.

This does have arbitrarily high deloopings in the context of omega-categories, but not in the context of infinity-groupoids. So for the purposes of cohomology is just the monoidal 0-groupoid which as a coefficient object induces a very boring cohomology theory: the -cohomology of anything connected is just the monoidal set itself. While we cannot deloop it, we can categorify it and do obtain an interesting nonabelian cohomology theory:

Namely the category Core(Vect) of finite dimensional vector spaces with invertible linear maps between them would serve as a categorification of : isomorphism classes of finite dimensional vector spaces V are given by their dimension d(V), and direct sum of vector spaces corresponds to addition of these numbers.

If we want to use the category Core(Vect) as the coefficient for a cohomology theory, we should for greater applicability equip it with its natural topological or smooth structure, so that it makes sense to ask what the Vect-cohomology of a topological space or a smooth space would be. The canonical way to do this is to regard Vect as a generalized smooth space called a smooth infinity-stack and consider it as the assignment

Vect:DiffGrpd\mathbf{Vect} : Diff \to \infty Grpd
UCore(VectBund(U))U \mapsto Core(VectBund(U))

that sends each smooth test space U (a smooth manifold, say), to groupoid of smooth vector bundles over U with bundle isomorphisms betweem them. We regard here a vector bundle VU as a smooth U-parametrized family of vector spaces (the fibers over each point) and thus as a smooth probe or plot of the category Core(Vect).

The nonabelian cohomology theory with coefficients in Vect has no cohomology groups, but at least cohomology monoids

H(X,Vect):=π 0H diff(X,Vect).H(X,\mathbf{Vect}) := \pi_0 \mathbf{H}_{diff}(X, \mathbf{Vect}) \,.

It is equivalent to the nonabelian cohomology with coefficients the delooping BU of the stable unitary group U:=colim nU(n).

K-theory as a groupoidification of Vect

The integers are obtained from the natural numbers by including “formal inverses” to all elements under the additive operation. Another way to think of this is that the delooped groupoid B is obtained from B by groupoidification (under the nerve operation this is fibrant replacement in the model structure on simplicial sets).

The idea of K-cohomology is essentially to apply this groupoidification process to not just to , but to its categorification Vect.

Just as an integer k=nm may be regarded as an equivalence class of natural numbers (n,m)× under the relation

[(n,m)]=[(n+r,m+r)]r[(n,m)] = [(n+r, m+r)] \;\; \forall r \in \mathbb{N}

one can similarly look at equivalence classes of pairs (V,W)Vect(U)×Vect(U) of vector bundles.

This perspective on K-theory was originally realized by Atiyah and Hirzebruch. The resulting cohomology theory is usually called topological K-theory.

As one of several variations, it is useful to regard a pair of vector bundles as a single 2-graded vector bundle.

One version of 2-graded vector bundles, which lead to a description of twisted K-theory are vectorial bundles.

Definition

Let X be a compact Hausdorff topological space. Write k for either the field of real numbers or of complex numbers C . By a vector space we here mean a vector space over k of finite dimension. By a vector bundle we mean a topological k-vector bundle. We write I nX for the trivial vector bundle I n=k n×X over X of rank n.

Lemma

For every vector bundle EX (with X compact Hausdorff) there exists a vector bundle EX such that

EEI rankE+rankE.E \oplus E' \simeq I^{rank E + rank E'} \,.
Proof

One invokes a partition of unity relative to an open cover on which E trivializes, constructs E locally and glues.

For details see for instance (Hatcher, prop. 1.4) or (Friedlander, prop. 3.1).

Definition

Define an equivalence relation on the set of finite-rank vector bundles EX over X by declaring that E 1E 2 if there exists k,l such that there is an isomorphism of vector bundles between the (fiberwise) direct sum of E 1 with I k and of E 2 with I l

(E 1E 2):(E 1I kE 2I l).(E_1 \sim E_2) :\Leftrightarrow \exists (E_1 \oplus I^k \simeq E_2 \oplus I^l) \,.

Write

K˜(X):=Vect(X) \tilde K(X) := Vect(X)_\sim

for the quotient set of equivalence classes.

Proposition

With X compact Hausdorff as in the assumption, we have that fiberwise direct sum of vector bundles equips K˜(X) with the structure of an abelian group. Together with the fiberwise tensor product of vector bundles this yields a ring.

Therefore K(X) is called the topological K-theory ring of X or just the K-theory group or even just the K-theory of X, for short.

Proof

The non-trivial part of the statement is that in K(X) there is an inverse to the operation of direct sum of vector bundles. Because in Vect(X) direct sum acts by addition of the ranks of vector bundles, it clearly has no inverse in Vect(X).

On the other hand, clearly the K-class [I n] of any trivial bundle I n is the neutral element in K(X)

[I n]=0[I^n] = 0

for all n, because by definition I nI 0. Therefore an inverse of a class [E 1] is given by a vector bundle E 2 with the property that the direct sum

E 1E 2I nE_1 \oplus E_2 \simeq I^n

is isomorphic to a trivial bundle for some n. This is the case by lemma 1.

Proposition

K˜(X) is isomorphic to the Grothendieck group of (Vect(X),).

However, def. 1 is more directly related to the definition of K-theory by a classifying space, hence by a spectrum. This we discuss below.

Properties

Classifying space

Definition

For n write U(n) for the unitary group in dimension n and O(n) for the orthogonal group in dimension n, both regarded as topological groups in the standard way. Write BU(n),BO(n) Top $ for the corresponding classifying space.

Write

[X,BO(n)]:=π 0Top(X,BO(n))[X, B O(n)] := \pi_0 Top(X, B O(n))

and

[X,BU(n)]:=π 0Top(X,BU(n))[X, B U(n)] := \pi_0 Top(X, B U(n))

for the set of homotopy-classes of continuous functions XBU(n).

Proposition

This is equivalently the set of isomorphism classes of O(n)- or U(n)-principal bundles on X as well as of rank-n real or complex vector bundles on X, respectively:

[X,BO(n)]O(n)Bund(X)Vect (X,n),[X, B O(n)] \simeq O(n) Bund(X) \simeq Vect_{\mathbb{R}}(X,n) \,,
[X,BU(n)]U(n)Bund(X)Vect (X,n).[X, B U(n)] \simeq U(n) Bund(X) \simeq Vect_{\mathbb{C}}(X,n) \,.
Definition

For each n let

U(n)U(n+1)U(n) \to U(n+1)

be the inclusion of topological groups given by inclusion of n×n matrices into (n+1)×(n+1)-matrices given by the block-diagonal form

[g][1 [0] [0] [g]].\left[g\right] \mapsto \left[ \array{ 1 & [0] \\ [0] & [g] } \right] \,.

This induces a corresponding sequence of morphisms of classifying spaces, def. 2, in Top

BU(0)BU(1)BU(2).B U(0) \hookrightarrow B U(1) \hookrightarrow B U(2) \hookrightarrow \cdots \,.

Write

BU:=lim nBU(n)B U := {\lim_{\to}}_{n \in \mathbb{N}} B U(n)

for the homotopy colimit (the “homotopy direct limit”) over this diagram (see at homotopy colimit the section Sequential homotopy colimits).

Note

The topological space BU is not equivalent to BU(), where U() is the unitary group on a separable infinite-dimensional Hilbert space . In fact the latter is contractible, hence has a weak homotopy equivalence to the point

BU()*B U(\mathcal{H}) \simeq *

while BU has nontrivial homotopy groups in arbitrary higher degree (by Kuiper's theorem).

But there is the group U() 𝒦U() of unitary operators that differ from the identity by a compact operator. This is essentially U=ΩBU. See below.

Proposition

Write for the set of integers regarded as a discrete topological space.

The product spaces

BO×,BU×B O \times \mathbb{Z}\,,\;\;\;\;\;B U \times \mathbb{Z}

are classifying spaces for real and complex K-theory, respectively: for every compact Hausdorff topological space X, we have an isomorphism of groups

K˜(X)[X,BU].\tilde K(X) \simeq [X, B U ] \,.
K(X)[X,BU×].K(X) \simeq [X, B U \times \mathbb{Z}] \,.

See for instance (Friedlander, prop. 3.2) or (Karoubi, prop. 1.32, theorem 1.33).

Proof

First consider the statement for reduced cohomology K˜(X):

Since a compact topological space is a compact object in Top (and using that the classifying spaces BU(n) are (see there) paracompact topological spaces, hence normal, and since the inclusion morphisms are closed inclusions (…)) the hom-functor out of it commutes with the filtered colimit

Top(X,BU) =Top(X,lim nBU(n)) lim nTop(X,BU(n)).\begin{aligned} Top(X, B U) &= Top(X, {\lim_\to}_n B U(n)) \\ & \simeq {\lim_\to}_n Top(X, B U (n)) \end{aligned} \,.

Since [X,BU(n)]U(n)Bund(X), in the last line the colimit is over vector bundles of all ranks and identifies two if they become isomorphic after adding a trivial bundle of some finite rank.

For the full statement use that by prop. \ref{missing} we have

K(X)H 0(X,)K˜(X).K(X) \simeq H^0(X, \mathbb{Z}) \oplus \tilde K(X) \,.

Because H 0(X,)[X,] it follows that

H 0(X,)K˜(X)[X,]×[X,BU][X,BU×].H^0(X, \mathbb{Z}) \oplus \tilde K(X) \simeq [X, \mathbb{Z}] \times [X, B U] \simeq [X, B U \times \mathbb{Z}] \,.

There is another variant on the classifying space

Definition

Let

U 𝒦={gU()gid𝒦}U_{\mathcal{K}} = \left\{ g \in U(\mathcal{H}) | g - id \in \mathcal{K} \right\}

be the group of unitary operators on a separable Hilbert space which differ from the identity by a compact operator.

Palais showed that

Proposition

U 𝒦 is a homotopy equivalent model for BU. It is in fact the norm closure? of the evident model of BU in U().

Moreover U 𝒦U() is a Banach Lie? normal subgroup.

Since U() is contractible, it follows that

BU 𝒦U()/U 𝒦B U_{\mathcal{K}} \coloneqq U(\mathcal{H})/U_{\mathcal{K}}

is a model for the classifying space of reduced K-theory.

Spectrum

Being a generalized (Eilenberg-Steenrod) cohomology theory, toplogical K-theory is represented by a spectrum: the K-theory spectrum.

The degree-0 part of this spectrum, i.e. the classifying space for degree 0 topological K-theory is modeled in particular by the space Fred of Fredholm operators.

Relation to algebraic K-theory

The topological K-theory over a space X is not identical with the algebraic K-theory of the ring of functions on X, but the two are closely related. See for instance (Paluch, Rosenberg).

References

Introductions are in

A discussion of the topological K-theory of classifying spaces of Lie groups is in

  • Stefan Jackowski and Bob Oliver, Vector bundles over classifying spaces of compact Lie groups (pdf)

Relations to algebraic K-theory are discussed in

  • Michael Paluch, Algebraic K-theory and topological spaces K-theory 0471 (web)
  • Jonathan Rosenberg, Comparison Between Algebraic and Topological K-Theory for Banach Algebras and C *-Algebras, (pdf)
Revised on June 8, 2012 12:32:22 by Urs Schreiber (131.130.239.35)