TOPOLOGICAL K-THEORY AND SOME OF ITS

APPLICATIONS

SAAD SLAOUI

Abstract. This thesis is concerned with the development of topological K-

theory, with a focus on the complex case. Following a treatment of the classical
theory of principal bundles and classifying spaces and its connection to the
study of vector bundles, we introduce the complex K-group and its extension
to a generalized cohomology theory. We then proceed to construct the Chern
character and the Adams operations, two important computational tools for
this cohomology theory. We conclude by proving two classical theorems in
topology using the language of topological K-theory: the non-existence of
complex structures on S 2n for n > 3, and the resolution of the Hopf invariant
one problem.

Contents
Introduction 2
Ackowledgments 3
1. Principal G-Bundles and Classifying Spaces 4
1.1. First Principles 4
1.2. A Homotopy Classification Theorem 6
1.3. Specialization to Vector Bundles 8
1.4. Standard Constructions on Vector Bundles 10
2. Defining Topological K-Theory 11
2.1. Construction of K 0 (−), K̃ 0 (−) 12
2.2. Bott Periodicity 16
2.3. Spectra and Generalized Cohomology Theories 18
2.4. Extending K̃ 0 (−) to a Reduced Generalized Cohomology Theory 20
3. Adams Operations and the Chern Character 21
3.1. Construction of the Adams operations 21
3.2. Chern Classes and the Chern Character 25
4. Applying Topological K-Theory to Problems in Classical Topology 29
4.1. Non-Existence of Complex Structures on Spheres 30
4.2. The Hopf Invariant One Problem 31
4.3. Reduction to an Ext Computation 34
Appendix A. Constructing Classifying Spaces at the Level of Categories 40
A.1. Refresher on Simplicial Sets, Nerves and Geometric Realizations 40
A.2. Construction at the Level of Categories 42
A.3. Construction at the Level of Simplicial Sets 43
A.4. From Adjunctions to Homotopy Equivalences 44
A.5. Construction at the Level of Spaces 45
References 47
1
2 SAAD SLAOUI

Introduction
Inspired by the establishment of the Bott periodicity theorem in 1956, Michael
Atiyah and Friedrich Hierzebruch undertook the investigation of topological K-
theory as one of the first instances of a generalized cohomology theory, mirroring
the properties of singular cohomology in the exotic settings of vector bundles. This
development served as a catalyst in the exploration of other kinds of generalized
cohomology theories, a project which was later shown to have close ties to the
world of stable homotopy theory via the Brown representability theorem published
by Edgar Brown in 1962 [7]. Moreover, topological K-theory quickly proved to
be of great importance in tackling difficult problems in classical topology, offering
new insights or drastic simplifications of existing proofs. We shall conclude the
thesis with two of these success stories, after a careful development of topological
K-theory as a generalized cohomology theory and of some of its important tools.
Throughout this paper, we choose to focus on complex topological K-theory rather
than its real counterpart, chiefly due to the nature of the applications we have in
mind in Chapter 4. However, much of the elementary theory goes through nearly
identically in the real case, and we make an effort to mention the analogous state-
ments where relevant.

Chapter 1 begins with a general treatment of the elementary theory of principal

G-bundles for any topological group G, leading up to the homotopy classification
theorem. We then specialize to complex vector bundles via the equivalence of
categories given by the Borel construction, and recover the traditional homotopy
classification of rank n complex vector bundles. As the latter objects form the back-
bone of topological complex K-theory, we take some time to elaborate on the role
of the tautological n-plane bundle over the infinite complex Grassmanian Grn (C∞ )
as universal rank n complex vector bundles in Section 1.3, and devote Section 1.4
to describing a receipe for patching up standard operations at the level of vector
spaces into operations on vector bundles.

Chapter 2 takes the above preliminaries as a starting point to introduce complex

topological K-theory. We start with the definition of the complex K-group functor
K 0 (−) and its reduced analog for K̃ 0 (−) pointed spaces, and exploit the homotopy
classification theorem for complex vector bundles to obtain a representability result
for these functors. Next, we introduce the Bott periodicity theorem in Section 2.2
and use it to extend K̃ 0 (−) to a generalized reduced cohomology theory in Section
2.4, after exploring the correspondence between generalized reduced cohomology
theory and Ω-spectra afforded by the Brown representability theorem in Section
2.3.

Chapter 3 introduces some important tools in topological K-theory, namely the

Chern character and the Adams operations. Both of these operations will be seen
to extend uniquely from the prescription of a desired behavior on line bundles, to-
gether with requirements of naturality and additivity. The Adams operations on
topological K-theory may be thought of as the analogs of the Steenrod square opera-
tions in singular cohomology with Z/2 coefficients, allowing us to interpret complex
K-groups as modules over an appropriate ring. The Chern character is built from
Chern classes and provides a bridge between topological K-theory and singular
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 3

cohomology in the form of a natural transformation Ch : K 0 (−) → H ev (−; Q)

compatible with the ring structure of both sides.

Finally, Chapter 4 illustrates some of the uses that complex K-theory has found
by drawing from the language and machinery introduced in the previous chapters to
resolve two problems in classical topology: the non-existence of complex structures
on S 2n for n > 3, and the non-existence of maps f : S 2n−1 → S n of Hopf invari-
ant one for n 6= 1, 2, 4, 8. The latter problem has the interesting consequence that
Rn does not admit the structure of a division algebra for n falling outside of that
range. The treatment we shall give, inspired by Dugger [9], also has the advantage
of indicating the relevance of the Hopf invariant one problem to understanding the
stable homotopy groups of spheres - a direction first pursued by Adams [2] in 1963.

Topological K-theory was initially investigated in the late 1950s and early 1960s
by Michael Atiyah and Friedrich Hierzebruch, alongside the development of cobor-
dism theories as generalized cohomology theories. The complex versions of these
two disciplines, complex topological K-theory and complex cobordism, have the
additional feature of being complex-oriented cohomology theories, in the sense that
they are multiplicative cohomology theories admitting a notion of Chern classes
mirorring the usual notion in singular cohomology.

In the words of Jack Morava, “ordinary cohomology lies on a stratum of infinite

codimension in the moduli space of complex-oriented theories” [19]. The subject of
chromatic homotopy theory organizes this stratum into a so-called ”chromatic filtra-
tion” of complex-oriented theories, indexed by the non-negative integers. Singular
cohomology occupies level zero of that filtration, followed by complex topological
K-theory at level n = 1. Complex cobordism theory is obtained in the limit as n
tends to infinity. The investigation of the intermediary landscape remains an area
of active research, involving a fascinating interplay between ideas and methods
from algebraic topology and algebraic geometry. It is the hope of the author that
this paper, possibly complemented by the introduction to cobordism theory given
in [21], will provide an accessible account of the classical framework and thereby
assist the interested reader in learning about contemporary areas of research in
algebraic topology.

Ackowledgments
I am deeply grateful to my advisor, Mona Merling, for her unwavering support
and helpful guidance throughout the production of this thesis. I would also like
to thank Laurent Harelle for igniting in me an inexhaustible sense of curiosity for
mathematical phenomena quite a few years ago, and Herman Gluck for creating
channels through which this burgeoning passion was allowed to progressively con-
cretize into a lifelong calling. I also wish to thank Christopher Douglas, Peter May,
and André Henriques, some of the mathematicians whose teaching and mentorship
allowed me to take my first steps into the world of topology and geometry, and
whose mathematical vision serves as a constant source of inspiration. Finally, I
would like to extend my gratitude to my family for accepting and even encouraging
this odd pursuit of mine, and my close friends for tolerating my occasional bursts
of hermitage and for providing immensely valuable support - in particular, I would
4 SAAD SLAOUI

like to thank Mark Macerato for sharing his mathematical drive and camaraderie
throughout these past few years.

1. Principal G-Bundles and Classifying Spaces

1.1. First Principles. In this section, we give a streamlined treatment of the the-
ory of principal G-bundles and classifying spaces. A more comprehensive treatment
of the material of this section may be found in Mitchell [18]. Vector bundles are a
special class of such objects which play a central role in topological K-theory, and
so later in this section we specialize our attention to these structures and provide
a brief overview of standard constructions. For a classical treatment of the theory
of vector bundles, see Atiyah [5] or the more recent text by Hatcher [11].

First, we lay down some preliminary notions. Let G be a topological group. Re-
call that a map f : X → Y between two right G-spaces is said to be G-equivariant,
or a G-map, if it is compatible with the respective group actions, in the sense that
f (xg) = f (x)g for all x ∈ X, g ∈ G. Note that any space X can be viewed as a
trivial G-space under the action xg = x for all x ∈ X, g ∈ G; then, any G-map
f : Z → X into a trivial G-space X has the property that the action of G on the
fibers of f is well-defined.
π
Definition 1.1. A principal G-bundle ξ : P −
→ X consists of a G-map π : P → X
between a S
right G-space P and a trivial G-space X, together with with an open
cover X = α Uα such that for each α, Uα fits into a commutative triangle:

π −1 (Uα )
ϕα
/ Uα × G

π π1
# {
Uα
where Uα × G is viewed as a right G-space under the action (x, g)h := (x, gh),
π1 : Uα → G is the projection map onto the first factor, and ϕα : π −1 (Uα ) → Uα ×G
is a G-equivariant homeomorphism. We call P the total space and X the base space
of the bundle ξ. The open subsets Uα are called local trivializations of ξ.
π
Remark 1.2. (i) Given a principal G-bundle ξ : P − → X, for each x ∈ X, we
observe the induced action of G on the fiber π −1 (x) is free and transitive. This
follows by direct computation from the fact that, letting Uα be a trivializing open
containing x, the associated map ϕα : π −1 (Uα ) → Uα × G restricts to a homeomor-
phism ϕα |π−1 (x) : π −1 (x) → {x} × G ∼
= G.
(ii) As noted above, π : P → X is constant on G-orbits, hence by the universal
property of quotients π factors uniquely through a map π̃ : P/G → X. The fact
that the action of G on P is free implies that π̃ is in fact a homeomorphism, hence
X can be interpreted as the orbit space of P .
Examples 1.3. Principal G-bundles for some topological group G are a common
phenomenon in topology. We give a few examples:
π
• The group of deck transformations of the universal cover X̃ − → X of a
connected space is naturally isomorphic to the fundamental group π1 (X),
π
so that X̃ −
→ X has the structure of a principal π1 (X)-bundle.
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 5

π
• The Hopf fibration η : S 3 − → S 2 obtained via restriction of the natural
1 ∼ 2
projection map C → CP = S to the unit 3-sphere S 3 ⊂ C2 is a principal
2

S 1 -bundle, with action on the fibers given by phase multiplication, and

local trivializations given by the opens S 2 − {N }, S 2 − {S} obtained from
S 2 by removing the North, resp. South pole.
• Given any Lie group G and a closed subgroup H ≤ G, the projection map
π
G − → G/H has the structure of a principal H-bundle, with action on G
given by right multiplication.
Principal G-bundles naturally fit into a category where a morphism between two
p p0
→ X and η : Q −→ Y 0 consists of a G-map g : P → Q
principal G-bundles ξ : P −
and a continuous map f : X → Y 0 such that the associated square commutes:

P
g
/Q
p p0
 
X / Y.
f

When considering a morphism of bundles over the same base space X, we shall
assume unless otherwise specified that the component map X → X is the identity
map. Next, we introduce a key construction on principal G-bundles. We assume
the reader is familiar with the categorical notion of pullbacks.
π
Definition 1.4. Let ξ : P −
→ X be a principal G-bundle. Given a continuous map
π0
f : Y → X, we define the pullback of ξ under f to be the map f ∗ ξ : f ∗ P −→ Y
obtained from the pullback diagram in the category Top of topological spaces
induced by the maps π and f .
π0
Hence the pullback bundle f ∗ ξ : f ∗ P −→ Y fits into a commutative square:
f ∗P /P

π0 π
 
Y
f
/ X.
Concretely, the total space f ∗ P is given by the set:
f ∗ P = {(y, a) ∈ Y × P | f (y) = p(a)},
π0
and f ∗ P −→ Y is the projection map onto the first factor. It can be verified that
π0
f ∗ ξ : f ∗ P −→ Y has the structure of a principal G-bundle, with local trivializations
provided by open sets of the form f −1 (U ) where U is a local trivialization for the
π π
bundle ξ : P − → X. In particular, if f : A → X is an inclusion and ξ : P − → X is a
π0
principal G-bundle over X, then f ∗ ξ : f ∗ P −→ A is simply the restriction of ξ to
the subspace A, which we denote by ξ|A .

We thus have a way of associating to a space X the set PG (X) of isomorphism

classes of principal G-bundles over X, and to a map f : X → Y the induced
pullback assignment f ∗ : PG (Y ) → PG (X) at the level of bundles. Playing around
with universal properties readily yields that this assignment gives a contravariant
functor PG (−) : Topop → Set, i.e. we have that (idX )∗ = idPG (X) for any space
6 SAAD SLAOUI

X, and (f g)∗ = g ∗ f ∗ : PG (Z) → PG (X) for any composite of continuous maps

g f
X−
→Y −
→ Z.

1.2. A Homotopy Classification Theorem. Next, we prove that the functor

PG (−) that we have just defined is homotopy invariant. The following argument
follows closely the one given in [section 1.2] of Hatcher [11].
p
Proposition 1.5. Given a principal G-bundle ξ : P −→ Y and homotopic maps
f ' g : X → Y from a paracompact space X, the associated pullback bundles f ∗ P ,
g ∗ P over X are isomorphic.
Proof. Let h : X × I → Y be the provided homotopy between f and g, so that
f = hi0 and g = hi1 , where ik : X × {k} ,→ X × I, k = 0, 1 denote the canonical
inclusions. Consider the pullback of ξ under h:
h∗ E /E
ξ
 
X ×I
h / Y.
By functoriality, we have that
f ∗ E = (hi0 )∗ E ∼
= i∗0 (h∗ E),
hence f ∗ E is isomorphic to the restriction of h∗ E over X × {0}. Similarly, we
see that g ∗ E is isomorphic to (h∗ E)|X×{1} . Hence to obtain the desired result it
suffices to show that given a principal G-bundle ξ : E → Y and a continuous map
h : X × I → Y , the bundles over X obtained by restricting h∗ E to X × {0} and
X × {1} are isomorphic.

Start with an arbitrary x ∈ X, and find a family of local trivializations {Ut ×

Vt }t∈I , where each Ut × Vt is taken to be a basic open set of X × I containing
the point (x, t), and where we may further assume that each Vt ⊂ I is an open
interval. By compactness of I, the open cover {Vt }t∈I of I admits a finite subcover
{Vt1 , ..., Vtn }. We may assume that this cover is “linearly ordered”, meaning that
if we write Vti =T(ai , bi ) and Vti+1 = (ai+1 , bi+1 ), then ai < ai+1 and bi < bi+1 .
n
Now, set Ux := i=1 Uti , and observe that we may inductively exploit the local
trivializations above Ux × Vti , Ux × Vti+1 to obtain a local trivialization above
Ux × (Vti ∪ Vti+1 ). In this way, we obtain a local trivialization above Ux × I.

Hence in particular we obtain an open cover {Ux }x∈X of X. We use the fact that
if X is a paracompact space, then every open cover {Ux } of X admits an associated
countable open cover {Vi }i≥1 equipped with a partition of unity, with the property
that each Vi is a disjoint union of opens contained in Uα ’s. So let {Vi }i≥1 be as
above, with associated partition of unity {ϕi }. The local trivializations over each
Ux × I clearly induce local trivializations over each Vi × I.

Now, define functions ψn : X → I as follows: set ψ0 := 1, and inductively set

ψn := ψn−1 − ϕn for n ≥ 1. By construction, we have a well-defined function
1 − Σi≥1 ϕi = 0. For each n ≥ 0, we may consider the graph of ψn , given by:
Γψn = {(x, ψn (x)) | x ∈ X} ⊂ X × I.
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 7

For each n, taking pullbacks of the bundle h∗ E → X × I along the inclusion

ln : Γψn ,→ X × I yields corresponding restricted bundles En := ln∗ (h∗ E) over Γψn .
Our plan is to iteratively exhibit bundle isomorphisms fn : En → En+1 , whose
composite will provide us with a bundle isomorphism from E0 = (h∗ E)|X×{1} to
the limiting term (as n tends to infinity) (h∗ E)|X×{0} .
By construction, for a given n, we have that suppϕn+1 ⊂ Vn+1 , so that the
graphs of ψn and ψn+1 differ at most along the open subset Vn × I of X × I. Hence
we may define a map f : En → En+1 to act as the identity above (B − Vn ) × I, and
using the existing local trivialization above Vn × I, we may define fn above Vn × I
to be given by:
fn : (x, ψn (x), g) 7→ (x, ψn (x) − ϕn+1 (x), g) = (x, ψn+1 (x), g).
Each fn : En → En+1 is then a G-equivariant, fiber-preserving homeomorphism,
hence a bundle isomorphism,Qand local finiteness of the family {ϕi } ensures that
the infinite composite f := n≥0 fn is a well-defined bundle isomorphism from
h∗ E|X×{1} and h∗ E|X×{0} , completing the proof. 

Thus we obtain a well-defined contravariant functor:

PG : Ho(Top)op → Set
where Ho(Top) denotes the homotopy category of spaces and homotopy classes of
continuous maps. It is defined as above by sending a space X to the set PG (X) of
isomorphism classes of principal G-bundles over X.

Let us say a word about functors of this kind in the realm of based spaces. Say
that a functor F : Ho(Top∗ )op → Set∗ (where Set∗ denotes the category of pointed
sets and based set maps) satisfies the wedge W
axiom if it sends coproducts
W toQ
products,
in the sense that given a wedge of spaces i Xi , the map F ( i Xi ) →W i F (Xi ),
x 7→ (F (li )(x))i induced by the natural inclusion maps li : Xi → i Xi is an
isomorphism. Next, say that F satisfies the Mayer Vietoris axiom if whenever a
space X can be expressed as a union of opens X = U ∪ V , the square induced by
the natural inclusion maps
F (X) / F (U )

 
F (V ) / F (U ∩ V )

has the property that whenever two elements x ∈ F (U ), y ∈ F (V ) restrict to

the same element in F (U ∩ V ), there exists a z ∈ F (X) such that z|U = x and
z|V = y. The following remarkable result was established by Brown [7] in his study
of cohomology theories. A modern treatment of the proof may be found in Chapter
3 of Kochman [12].
Theorem 1.6. (Brown Representability Theorem) Let F : Ho(Top∗ )op →
Set∗ be a contravariant functor on connected based spaces which satisfies the wedge
axiom and the Mayer-Vietoris axiom. Then F is representable. That is, there
exists a connected based space L and an element u ∈ F (L) such that the natural
transformation Tu : [−, L] → F (−) given by Tu (g) := F (g)(u) ∈ F (X) for any
[g] ∈ [X, L] is a natural isomorphism.
8 SAAD SLAOUI

Remark 1.7. By the Yoneda lemma, any two connected based spaces L, L0 to-
gether with classes u ∈ F (L), u0 ∈ F (L0 ) with the property that Tu : [−, L] →
F (−), Tu0 : [−, L0 ] → F (−) are natural isomorphisms must be homotopy equiva-
lent.
Now, the functor PG satisfies the Mayer-Vietoris axiom and the unpointed ver-
sion of the wedge axiom (coproducts are disjoint unions), and a sleight of hand
makes it possible to infer representability of PG from the above statement. To say
that PG is representable amounts to saying that there exists a connected space
BG and an associated principal G-bundle EG → BG such that for every space
X, the set [X, BG] of homotopy classes of maps from X into BG is in a natural
bijection with the elements of PG (X), with correspondence given by sending a map
f : X → BG to the principal G-bundle f ∗ EG → X over X obtained by pullback of
the universal principal G-bundle EG → BG. In that case, we call BG a classifying
space for principal G-bundles and refer to EG → BG as the associated universal
principal G-bundle.

We state the classification theorem in its traditional form:

Theorem 1.8. (Classification Theorem for Principal G-Bundles) Let G
be a topological group. Then there exists a classifying space BG for principal G-
bundles, unique up to homotopy equivalence, such that for every space X, there is
a natural bijection:
[X, BG] ∼= PG (X),
obtained by sending a map f : X → BG to the pullback under f of a fixed universal
bundle EG → BG.
We relegate to Appendix A the general construction of a universal principal G-
bundle EG → BG for any topological group, and spend the next section discussing
the cases G = U (n) and G = O(n), which are of central importance in complex,
resp. real topological K-theory. To close our discussion of the general case, we
note that the verification that a given principal G-bundle is universal is greatly
simplified by the following theorem, whose proof may be found in Mitchell [18]:
p
Theorem 1.9. Let ξ : E − → B be a principal G-bundle whose total space E is
contractible. Then ξ is a universal principal G-bundle and B is a model for the
classifying space BG.
This theorem also informs us about the homotopy type of BG, which is well-
defined by uniqueness as mentioned in Remark 1.7. Since exhibiting a universal
principal G-bundle as in Theorem 1.9 is always possible (as we will see in Appendix
A), we can consider the fibration G → EG → BG, where EG is weakly contractible,
so that the associated homotopy LES yields an isomorphism for all n:
(1.10) πn+1 (BG) ∼
= πn (G).

1.3. Specialization to Vector Bundles. Of particular interest to us are vector

bundles and their corresponding classifying spaces, both in the real and complex
case. We start by specializing to principal U (n)-bundles and describe their close
ties with complex vector bundles of rank n. A completely analogous argument can
be made to establish a correspondence between principal O(n)-bundles and real
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 9

vector bundles of rank n.

Let Vno (C∞ ) be the Stiefel manifold of orthonormal n-frames in C∞ , and let
Grn (C∞ ) be the Grassmanian manifold of complex n-planes in C∞ . These spaces
may be topologized as direct limits limk Vno (Cn+k ), resp. limk Grn (Cn+k ) taken
−→ −→
under inclusions, each of whose constituent spaces is readily seen to be compact
Hausdorff, the latter as the image of the former under the canonical projection map
sending an orthonormal n-frame to the n-plane it spans. The same projection map
on the direct limits:
π : Vno (C∞ ) → Grn (C∞ )
may be verified to have the structure of a principal U (n)-bundle ξ, with the U (n)
action on the fibers given by matrix multiplication. We claim that ξ is a universal
principal U (n)-bundle.
We may then prove the following:
π
Proposition 1.11. The Stiefel manifold Vno (C∞ ) is contractible, hence ξ : Vno (C∞ ) −
→
Grn (C∞ ) is a universal principal U (n)-bundle and Grn (C∞ ) is a model for BU (n).
Proof. Denote by {ei }i≥1 the standard basis for C∞ . We construct a map f :
Vno (C∞ ) → Vno (C∞ ) such that idVno (C∞ ) ' f ' cx , where x := ( √1n e1 , ..., √1n en ) and
cx is the constant map at x ∈ Vno (C∞ ). Given an arbitrary element v ∈ C∞ − {0},
we can always find a finite number corresponding to the last non-zero entry in
the expression of v as a linear combination of the ei ’s. Denote this assignment
by σ : C∞ − {0} → N. Next, let T : Vno (C∞ ) → Vno (C∞ ) be the linear operator
sending (v1 , ..., vn ) to (v10 , ..., vn0 ), where each vi0 is obtained from vi by shifting all
entries in the basis expression of vi by one unit to the right. Finally, define the
map f : Vn (C∞ ) → Vn (C∞ ) via
f (v1 , ..., vn ) := T max{σ(v1 ),...,σ(vn )} (v1 , ..., vn ),
where T k denotes the k-fold iteration of T . By√construction, we may then use
straight line homotopies normalized to live in S ∞ ( n) to get a homotopy idVno (C∞ ) '
f ' cx , as needed. Thus Vno (C∞ ) is contractible, and the second part of the propo-
sition follows from Theorem 1.9. 
π
Now, recall that a complex vector bundle ξ : E −→ X of rank n over a space X
consists of a continuous map π : E → X such that the fiber over each point has the
structure of an n-dimensional complex vector space, together with an open cover
∼
{Uα } of X admitting fiber-preserving homeomorphisms ϕα : π −1 (Uα ) −→ Uα × R n
which restrict to linear isomorphisms on the fibers of π. There exists a general
process known as the Borel construction which specializes to an equivalence of
categories between principal U (n)-bundles and rank n complex vector bundles over
a given base space. The Borel construction takes a principal U (n)-bundle E → X
to an associated rank n vector bundle E ×U (n) Cn → X, with total space given by
the quotient:
E ×U (n) Cn := E × Cn (xA, y) ∼ (x, Ay),


ranging over all x ∈ E, y ∈ Cn , A ∈ U (n).

Since classifying spaces and universal bundles were defined categorically, they
are preserved under this equivalence of categories. By investigating the structure
10 SAAD SLAOUI

of the balanced product γn∞ := Vno (C∞ ) ×U (n) Cn , we thus find that a model
for the universal rank n-vector bundle is given by the tautological vector bundle
γn∞ : EU (n) → BU (n), where
EU (n) = {(p, v) ∈ Grn (C∞ ) × C∞ | v ∈ p}.
Remark 1.12. A completely analogous treatment can be given for real vector bun-
dles, where the spaces Vno (R∞ ), resp. Grn (R∞ ) are obtained via direct limit con-
π
structions as above. In these settings, the canonical projection map ξ : Vno (R∞ ) −→
∞
Grn (R ) has the structure of a principal O(n)-bundle with respect to matrix mul-
tiplication on the fibers. Furthermore, the Stiefel manifold Vno (R∞ ) is contractible,
p
hence ξ : Vno (R∞ ) −
→ Grn (R∞ ) is a universal principal O(n)-bundle and Grn (R∞ )
is a model for BO(n). Under the Borel construction, ξ corresponds to the universal
real vector bundle γn∞ : EO(n) → BO(n) of rank n, where
EO(n) := {(p, v) ∈ Grn (R∞ ) × R∞ | v ∈ p}.
1.4. Standard Constructions on Vector Bundles. We next present a general
framework for exploiting the structure of known vector bundles over a fixed base
space in order to create new ones. The following construction boils down to one key
idea, which is to give new objects the final topology with respect to the appropriate
collection of maps so as to obtain a vector bundle structure. The following discus-
sion is inspired from Section (3.f) of Milnor [16]. We present the constructions for
complex vector bundles, but the same arguments apply in the case of real vector
bundles.

Denote by Vectiso
f the category of finite dimensional complex vector spaces and
linear isomorphisms. Observe that each morphism set in this category can be given
the structure of a topological space as a subspace of the corresponding space of
linear operators. Composition of linear isomorphisms is continuous with respect to
this topology, hence we may view Vectiso f as a category enriched over Top. The
same observation applies to the k-fold product category (Vectiso k
f ) , whose morphism
sets can be given the product topology. Say that a functor T : (Vectiso k
f ) → Vectf
iso

is continuous if the induced maps on morphism sets:

(V ectiso 0 iso 0 T / (V ectiso )(T (V1 , ..., Vk ), T (V10 , ..., V 0 ))

f )(V1 , V1 ) × ... × (V ectf )(Vk , Vk ) f k

are all continuous.

Construction 1.13. Fix a continuous functor T : (Vectiso k
f ) → Vectf
iso
on k
variables, and let ξ1 , ..., ξk be given vector bundles on a connected base space X,
so that each ξi has a well defined rank ni . We construct a new vector bundle
T (ξ1 , ..., ξk ) of rank dim C T (Cn1 , ..., Cnk ) over X as follows: for each x ∈ X, define
Fx (ξ1 , ..., ξk ) := T (Fx (ξ1 ), ..., Fx (ξk )),
where Fx (ξi ) denotes the fiber of ξi over x. Then, let
a
E(ξ1 , ..., ξk ) := Fx (ξ1 , ..., ξk ),
x∈X

and define π : E(ξ1 , ..., ξk ) → X to be the canonical projection map sending

Fx (ξ1 , ..., ξk ) to x.
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 11

For a given x ∈ X, we can find local trivializations U1 , ..., Uk about x correspond-

ing to ξ1 , ..., ξk , respectively. Set U := ∩ki=1 Ui . Now, the assignment v 7→ (x, v)
provides a linear isomorphism hi,x : Cni → Fx (ξi ) for each i = 1, ..., k. Taken as a
k-tuple, the hi,x ’s are taken by T to a linear isomorphism
∼
T (h1,x , ..., hk,x ) : T (Cn1 , ..., Cnk ) −
→ Fx (ξ1 , ..., ξk ).
Letting x range over U , the resulting maps can then be used to obtain a bijection:
h̃ : U × T (Cn1 , ..., Cnk ) → π −1 (U )
via the assignment (x, v) 7→ T (h1,x , ..., hk,x )(v). Composing each such h̃ with the
inclusion π −1 (U ) ,→ E(ξ1 , ..., ξk ) yields a family of maps:
hU : U × T (Cn1 , ..., Cnk ) → E(ξ1 , ..., ξk ),
where the left hand side can be given the product topology.

Now, give the set E(ξ1 , ..., ξk ) the final topology induced by the maps hU as
above. This is defined by the universal property that a map f : E(ξ1 , ..., ξk ) → Z is
said to be continuous if and only if each composite f hU is continuous, and it is the
finest topology on E(ξ1 , ..., ξk ) making all the maps hU continuous. In particular,
under this topology, each of the maps h̃ described above become fiber-preserving
homeomorphisms restricting to linear isomorphisms on the fibers by construction.
Thus E(ξ1 , ..., ξk ) is a valid total space for the vector bundle
π
T (ξ1 , ..., ξk ) : E(ξ1 , ..., ξk ) −
→X
over X with fibers Fb (ξ1 , ..., ξk ) and local trivializations provided by the opens U
and the maps h̃ as defined above.
Example 1.14. To give a sense of the usefulness of this construction, we list a few
important examples of new vector bundles that can be obtained from existing ones
by following the above recipe, all of which will be used at some later point. In the
following, let ξ, resp. η denote fixed complex vector bundles over a base space X
with rank n, resp. m.
• The direct sum, or Whitney sum of bundles ξ⊕η, induced by the assignment
(V, W ) 7→ V ⊕ W , with rk ξ ⊕ η = n + m;
• The tensor product of bundles ξ ⊗ η, induced by the assignment (V, W ) 7→
V ⊗ W , with rk ξ ⊗ η = n · m;
• The hom bundle Hom(ξ, η), induced by the assignment (V, W ) 7→ HomC (V, W ),
with rk Hom(ξ, η) = n · m. In particular, letting 1 denote the trivial line
bundle over X, we may construct the dual bundle ξ ∗ := Hom(ξ, 1 ) from
the assignment V 7→ HomC (V, C);
Vk Vk
• The k th exterior power bundle ξ, induced by the assignment V 7→ V,
Vk
ξ = nk .


with rk

2. Defining Topological K-Theory

From now on, unless otherwise specified, we work with compact Hausdorff spaces
and complex vector bundles over them. This chapter is devoted to the development
of topological K-theory from the building blocks laid down in the previous chap-
ter. Namely, we will associate to the functor VectC (−) : Ho(Top)op → Set closely
associated functors K 0 (−) : Ho(Top)op → Ab (resp. K̃ 0 (−) : Ho(Top∗ )op → Ab)
12 SAAD SLAOUI

taking values in the category of abelian groups, and assigning to a space X its (re-
duced) complex K-group. Then, from the homotopy classification theorem applied
to VectC (−) ∼= PU (n) together with an additional structural property enjoyed by
the spaces {BU (n)} as a result of the Bott periodicity theorem, we will be able to
extend K̃ 0 (−) to a collection of functors K̃ n (−) : Ho(Top∗ )op → Ab fitting into a
reduced generalized cohomology theory known as complex topological K-theory.
2.1. Construction of K 0 (−), K̃ 0 (−). We start our discussion by introducing a
general procedure for associating abelian groups to abelian monoids:
Construction 2.1. (Grothendieck group) Let M be an abelian monoid. Define
the Grothendieck group associated to M , denoted by Gr(M ), to be the abelian group
obtained as the quotient of the free abelian group on elements of M by the subgroup
generated by relations of the form [m] + [n] − [m + n]. That is, we define:

Gr(M ) := Z < [m] | m ∈ M > ([m] + [n] − [m + n], m, n ∈ M ).
Observe that the quotient relation enables us to express any element of Gr(M ) in
the form [m] − [n] for some m, n ∈ M , by gathering together elements of the same
sign. This expression is not unique, but has the property that if [m] − [n] = [r] − [s]
then [m + s] = [r + n].
The group Gr(M ) comes equipped with a semigroup homomorphism η : M →
Gr(M ) given by the natural inclusion m 7→ [m]. In a precise sense, Gr(M ) is
the “most general” abelian group that admits such a map. Namely, it can be
immediately verified that the pair (Gr(M ), η : M → Gr(M )) satisfies the following
universal property: given any pair (H, f : M → H) where H is an abelian group
and f a semigroup homomorphism, there exists a unique map ϕ : Gr(M ) → H
making the following triangle commute:
∀f
M / H.
;
η
∃!ϕ

Gr(M )
This assignment can in fact be extended to a covariant functor
Gr(−) : AbMon → Ab
from the category of abelian monoids to the category of abelian groups, by send-
ing a monoid homomorphism ϕ : M → N to the obvious group homomorphism
ϕ̃ : Gr(M ) → Gr(N ) given by ϕ̃([m] − [n]) := [ϕ(m)] − [ϕ(n)], well defined by the
comment made above. From there, we see that the data of the above diagram en-
codes the existence of an adjunction between the Grothendieck group functor and
the forgetful functor U : Ab → AbMon, in that we have a natural bijection, for
any abelian monoid M and any abelian group H:
Ab(Gr(M ), H) ∼= AbMon(M, U H).
Now, observe that for any space X, the set VectC (X) equipped with the Whitney
sum operation has the structure of an abelian monoid.
Definition 2.2. Given a space X, the complex K-group of X, denoted K 0 (X), is
defined to be the Grothendieck completion of the abelian monoid VectC (X):
K 0 (X) := Gr(VectC (X)).
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 13

Following the discussion in Construction 2.1, this assignment yields a contravariant

functor:
K 0 (−) : Ho(Top)op → Ab.
Remark 2.3. We call an element of K 0 (X) a virtual bundle over X. By the
discussion in Construction 2.1, any virtual bundle admits an expression in the form
x − y for some x, y ∈ VectC (X). We shall see in Corollary 2.7 that a little more is
true. By convention, we sometimes denote the class of the trivial rank q bundle by
q ≡ q, thereby obtaining a copy of Z inside each K 0 (X).
As a trivial first computation, since finite-dimensional vector spaces are uniquely
characterized up to isomorphism by their dimension, we have over the one-point
space that VectC (pt) ∼ = N, and hence that K 0 (pt) = Gr(N) ∼ = Z. This leads us to
the following:
Definition 2.4. Given a space X, the reduced complex K-theory group of X, de-
noted by K̃ 0 (X), is defined to be the kernel of the map l : pt ,→ X induced by the
inclusion of the basepoint:
K̃ 0 (X) := ker(l∗ : K 0 (X) → K 0 (pt) = Z).
Remark 2.5. Note that by definition of pullbacks, the map l∗ : K 0 (X) → K 0 (pt)
sends a bundle ξ to the dimension of its fiber over the basepoint, i.e. the rank of the
restriction of ξ to the connected component of the basepoint of X. In particular, if
X is connected, l∗ (ξ) equals the rank of ξ.
Since K 0 (pt) ∼
= Z, we always have a short exact sequence of abelian groups:
/ K̃ 0 (X) / K 0 (X) l∗ /Z /0,
0
and hence, since Z is projective, an isomorphism:
K 0 (X) ∼
= K̃ 0 (X) ⊕ Z.
We next record a few useful facts about vector bundles, which are treated in
detail in Section 9 of Dugger [9]:
Proposition 2.6. Any short exact sequence of vector bundles splits, and any vector
bundle x ∈ VectC (X) may be expressed as a direct summand of a trivial bundle, i.e.
x ⊕ y = q for some y ∈ VectC (X), q ∈ N.
In particular, we get the following convenient fact about virtual bundles (in the
notation of Remark (2.3)):
Corollary 2.7. Any virtual bundle x−y ∈ K 0 (X) may be expressed as x−y = z−q
for some z ∈ VectC (X), q ∈ N.
Proof. By Proposition 2.6 above, we can write y⊕w = q for some w ∈ VectC (X), q ∈
N. Then, letting z = x ⊕ w ∈ VectC (X), we find that x − y = x + w − q = z − q
in K 0 (X), as needed. 
We now outline an argument showing that the functors K 0 (−) and K̃ 0 (−) are
representable, following the argument given in Chapter 24 of May [13]. Intuitively,
from the natural isomorphisms VectnC (X) ∼ = [X, BU (n)] discussed in Chapter 1,
we expect the representing space of K 0 (−) to be closely related to the colimit
BU := limn BU (n). Here, the direct system structure is induced from the system
−→
14 SAAD SLAOUI

of inclusions U (n) ,→ U (n + 1) given by matrix inclusion A 7→ diag(A, 1), on which

we apply the classifying space functor. Explicitly, each map BU (n) → BU (n + 1)
can be described as the classifying map associated to the rank (n+1) bundle γn∞ ⊕1
over BU (n), where γn∞ denotes the universal rank n bundle. Accounting for the
Grothendieck group construction involved in defining K 0 (−), it turns out that the
desired representing space is BU × Z.
Construction 2.8. Given a space X, let X+ denote the based space obtained from
X by adding a disjoint basepoint. We construct a map
g : K 0 (X) → [X+ , BU × Z],
as follows: given an element x − y ∈ K 0 (X), we can write x − y = z − q for some
z ∈ VectC (X), q ∈ N by Corollary 2.7 above. Let f : X → BU (n) be a classifying
map for z. Then, composing f with the canonical inclusion BU (n) → BU to view
it as a map X → BU , we define g(z − q) := (f, rkz − q) : X+ → BU × Z.
We readily check that g is well-defined: suppose x − y admits different decom-
positions x − y = z − q = w − r for some z, w ∈ VectC (X), q, r ∈ N. Without loss
of generality, assume q ≥ r, so that we may write z = w ⊕ q−r . Then it follows
from the above description of the direct system structure on limn BU (n) that the
−→
classifying maps of z and w viewed as maps X → BU are homotopic. Next, we
may also write z ⊕ r = w ⊕ q , so that
rkz + r = rk(z ⊕ r ) = rk(w ⊕ q ) = rkw + q,
hence g(z − q) = g(w − r), as needed.
Now, we claim that g gives a natural isomorphism K 0 (−) ∼
= [−, BU × Z], so that
the complex K-group functor is representable by the space BU × Z.
Theorem 2.9. The map g : K 0 (X) → [X+ , BU ×Z] defines a natural isomorphism
K 0 (X) ∼
= [X+ , BU × Z]
between the functors K 0 (−), [−+ , BU × Z] : Ho(Top)op → Ab.
Proof. We have already verified that the map g is well-defined. We check naturality:
starting with a map of spaces h : X → Y , we wish to show that the associated square
h∗ / K 0 (X)
K 0 (Y )
g g
 h∗

[Y+ , BU × Z] / [X+ , BU × Z]

commutes. So let ξ − q be an arbitrary element in K 0 (Y ). Let f : Y → BU (n) be a

classifying map for the bundle ξ over Y , so that ξ = f ∗ γn∞ . Then by functoriality,
we see that (f h)∗ (γn∞ ) = h∗ f ∗ γn∞ = h∗ ξ, hence f h is a valid classifying map for
h∗ ξ. We also have that the rank of a bundle is invariant under pullback, so that
g(h∗ (ξ − q)) = (f h, rk(h∗ ξ) − q) = (f h, rkξ − q) = h∗ (f, rkξ − q),
as needed.
Next, let f : X+ → BU × Z be an arbitrary map given by the pair (f˜, q). By
compactness of X, the map f˜ : X → BU factors through a map X → BU (n)
for some n ≥ 0, which corresponds under pullback to a rank n bundle ξ over X.
Then, consider the virtual bundle ξ − (n − q) ∈ K 0 (X). Either q ≤ n, in which
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 15

case g(ξ − (n − q)) = (f˜, rkξ − n + q) = (f˜, q) by construction, or q ≥ n, so that

ξ − (n − q) = ξ ⊕ q−n . In this case, by a similar argument to the one given in
Construction 2.8, the classifying map for ξ⊕q−n is homotopic to f˜ when viewed as a
map X → BU , hence g(ξ −(n−q)) = (f˜, rk(ξ ⊕q−n )) = (f˜, q), proving surjectivity.
Finally, suppose z − q lies in the kernel of q, so that g(z − q) is homotopic to the
pair (c∗ , 0), where c∗ : X → BU is the constant map on the basepoint. Then by
construction, z is obtained by the pullback of some universal bundles γk∞ by c∗ ,
hence z = k for some k, and we also have that rkz − q = 0, so that z = q , thus
z − q = q − q = 0, completing the proof that g : K 0 (X) → [X+ , BU × Z] defines
a natural isomorphism. 
It is then possible to formally deduce from the above natural isomorphism the
associated result in the reduced case:
Theorem 2.10. There is a natural isomorphism:
(2.11) K̃ 0 (X) ∼
= [X, BU × Z]
between the functors K̃ 0 (−), [−, BU × Z] : Ho(Top∗ )op → Ab.
Proof. Given a based space X, we may add a new, disjoint basepoint to obtain
the associated based space X+ = X t {∗}. Then, the based inclusion l : S 0 → X+
sending the second point of S 0 to the original basepoint of X is a cofibration, giving
rise to a cofiber sequence:

S0
l / X+ j
/ X+ /S 0 ∼
= X.
Applying the functor [−, BU × Z] to the associated Puppe sequence yields a LES:
j∗ l∗
... / [ΣS 0 , BU × Z] / [X, BU × Z] / [X+ , BU × Z] / [S 0 , BU × Z],

where the leftmost term is π1 (BU ×Z) = π1 (BU ) = π0 (U ) = 0 since U is connected,

so that by exactness the map j ∗ is injective and ker l∗ ∼
= [X, BU × Z]. Now, under
the identification S 0 = {pt}+ and by naturality of the isomorphism K 0 (X) ∼ =
[X+ , BU × Z], the rightmost map of the above LES fits into a commutative square:
l∗ / [S 0 , BU × Z]
[X+ , BU × Z]
' '
 
l∗ / K 0 (pt),
K 0 (X)

and hence Equation 2.11 follows by definition of K̃ 0 (X) := ker l∗ . 

[Interlude] (Connection to algebraic K-theory) The content of Proposi-
tion 2.6 suggests that the category of vector bundles over a compact Hausdorff space
has properties analogous to the category of finitely generated projective modules
over a commutative ring. Under consideration of the right objects, this impression
can in fact be turned into a rigorous equivalence of categories. Indeed, given a
π
vector bundle E − → X we may look at the family of global sections, given by:
Γ(X, E) := {s : X → E | πs = idX }.
This is a C-vector space under pointwise addition and scalar multiplication, and
it may furthermore be given the structure of a module over the ring C(X) of
16 SAAD SLAOUI

continuous complex-valued functions on X via the assignment (f ·s)(x) := f (x)s(x)

for f ∈ C(X), s ∈ Γ(X, E). It can be shown that Γ(X, E) is always a finitely
generated projective C(X)-module, and we have the following result:
Theorem 2.12. (Swan’s theorem) For any compact Hausdorff space X, the
functor
Γ : VectC (X) → Pf g (C(X))
π
sending a vector bundle E − → X to the C(X)-module of global sections Γ(X, E)
gives an equivalence of categories between VectC (X) and the category Pf g (C(X))
of finitely-generated projective C(X)-modules.
The proof of this result consists of a series of straightforward checks, for which
we refer the interested reader to Section 2.10 of Dugger [9]. For now, we simply
observe that this equivalence of categories reinterprets K 0 (X) as the Grothendieck
completion of the commutative monoid Pf g (C(X)) under direct sums of modules,
and thereby motivates the study of Gr(Pf g (R)) for arbitrary commutative rings.
This question leads to the vast subject of algebraic K-theory, a treatment of which
lies outside the scope of this paper. For an integrated treatment of these two
viewpoints rich in examples, we refer the interested reader to Ravenel [20].
2.2. Bott Periodicity. We next introduce the Bott periodicity theorem, which
will allow us to extend the K-group functor to a generalized cohomology theory.
Before stating the theorem, we review some preliminaries on clutching functions
associated to vector bundles over spheres.
π
Let E −→ S n be a given vector bundle of rank k. Let D+ n
, resp. D− n
denote
n
small open neighborhoods of the northern, resp southern hemisphere of S , so that
n n ∼ n−1 n n
D+ ∩ D− =S × (−, ) for some  > 0. Both D+ and D− are contractible, hence
we can find trivializations of the restrictions E|D+n , resp. E|D− n , given by fiber-

n
preserving homeomorphisms ϕ˜+ : E|D+n → D+ × Ck , resp. ϕ˜− : E|D− n → D
n
− ×C
k

which restrict to linear isomorphisms on the fibers. These maps induce continuous
n n
maps ϕ+ : D+ → GLk (C), resp. ϕ− : D− → GLk (C) sending a given point x to
'
the corresponding linear isomorphism Fx E − → Ck . In particular, letting S n−1 ⊂
n n n
D+ ∩ D− denote the equator of S , we may construct a map:
f : S n−1 → GLk (C),
x 7→ ϕ−1
+ (x)ϕ− (x)

We call f as above the clucthing function associated to the bundle E. It can be

checked that isomorphic bundles give rise to homotopic maps, and in fact that this
assignment gives rise to a bijection (c.f. Section 1.3 of Cohen [8] for a proof):
Theorem 2.13. The assignment VectkC (S n ) → [S n−1 , GLk (C)] sending a rank k
vector bundle over S n to its associated clutching function is a bijection.
As an exercise in manipulating clutching functions which will prove useful later
in this section, we establish the following result:
Lemma 2.14. Given clutching functions f, g : S 1 → GLn (C), the associated rank
n vector bundles over S 2 satisfy the following relation:
(2.15) Ef ⊕ Eg ∼ = Ef g ⊕ n .
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 17

Proof. It follows directly from the definition that the clutching function associated
to the left hand side, resp. right hand side of Lemma 2.14 are given by the maps z 7→
diag(f (z), g(z)), resp. z 7→ diag(f (z)g(z), In ), where In denotes the n by n identity
matrix. Hence by Theorem 2.13, it suffices to show that these two assignments are
homotopic as maps S 1 → GL2n (C). Now, GL2n (C) is path-connected (which can
be see for instance via the polar decomposition), hence in particular we can
find a
path γ : I → GL2n (C) from the identity matrix to the matrix P := I0n I0n . Then,
the map H : S 1 × I → GL2n (C) given by:
H(z, t) := diag(f (z), In ) · γ(t) · diag(In , g(z)) · γ(t)
gives a homotopy H : diag(f (z), g(z)) ' diag(f (z)g(z), In ), as needed. 

Of particular interest to us is the so-called Bott element β ∈ K̃ 0 (S 2 ), defined

π
to be β := L − 1, where L − → S2 ∼ = CP 1 denotes the tautological line bundle. By
Lemma 2.14, since product of clutching functions corresponds to tensor product of
bundles for line bundles, we have the relation L ⊕ L ∼ = L2 ⊕ 1 . In K 0 (S 2 ), this
implies that
(2.16) β 2 = (L − 1)2 = L2 + 1 − 2L = 0.
The statement of the Bott periodicity theorem which we choose to give relies
on the existence of a product structure on K-groups associated to different spaces,
analogous to the cross product in singular cohomology:
Definition 2.17. Given spaces X and Y , define the external product on K 0 (X)
and K 0 (Y ) to be the bilinear map:

K 0 (X) ⊗ K 0 (Y )
−×−
/ K 0 (X × Y )
π π
given by x × y := π1∗ (x)π2∗ (y), where X × Y −→
1 2
X, X × Y −→ Y are the standard
projections.
Remark 2.18. The external product is clearly natural in both factors. In par-
ticular, fixing an element z ∈ K 0 (Y ) allows us to define a natural transformation
−×z
K 0 (−) −−−→ K 0 (− × Y ) given by the composite:

K 0 (X)
id⊗z
/ K 0 (X) ⊗ K 0 (Y ) −×−
/ K 0 (X × Y ).

Lemma 2.19. Given spaces X and Y , the external product restricts to a reduced
external product:
K̃ 0 (X) ⊗ K̃ 0 (Y )
−×−
/ K̃ 0 (X ∧ Y )

In the same way as in Remark 2.18, any element z ∈ K̃ 0 (Y ) induces a natural

−×z
transformation K̃ 0 (−) −−−→ K̃ 0 (− ∧ Y ). We are now in a position to state the
main theorem, a proof of which is given in Chapter 2 of Hatcher [11] via a detailed
analysis of clutching functions:
Theorem 2.20. (Bott Periodicity) The map
−×β
K̃ 0 (X) / K̃ 0 (X ∧ S 2 ) = K̃ 0 (Σ2 X)

induced by taking external products with the Bott element is a natural isomorphism.
18 SAAD SLAOUI

As an immediate application, since K̃ 0 (pt) ∼ = Z, we find that K̃ 0 (S 2 ), and

0 2n
more generally K̃ (S ) for any even dimensional sphere is isomorphic to a copy
of Z generated by β ×n , with ring structure given by (β ×n )2 = (β 2 )×n = 0 by
Equation 2.16. Next, as observed in the proof of Theorem 2.10, using Equation 1.10,
we find that:
∼ [S 1 , BU × Z] =
K̃ 0 (S 1 ) = ∼ π1 (BU × Z) = π1 (BU ) = π0 (U ) = 0,
since U is path-connected. It then immediately follows from Theorem 2.20 that
K̃ 0 (S 2n+1 ) = 0 for all odd-dimensional spheres.

We now take a moment to deduce the homotopy-theoretic version of the Bott

periodicity theorem as expressed in Bott’s original paper [6], as this is the version
which we will mostly be needing in the next few sections. Combining Theorem 2.10,
the Bott periodicity theorem and the loop-suspension adjunction leads us to a chain
of natural isomorphisms:
[X, Ω2 (BU × Z)] =∼ [Σ2 X, BU × Z] =
∼ K̃ 0 (Σ2 X) =
∼ K̃ 0 (X) =
∼ [X, BU × Z].
Thus, by the Yoneda lemma, it follows that we have a homotopy equivalence:
BU × Z ' Ω2 (BU × Z).
Also, by compatibility of the based hom functor with products, we readily see that:
Ω(BU × Z) = ∼ ΩBU × ΩZ = ∼ ΩBU,
where the last identification follows from the fact that ΩZ = {pt} since Z is a
discrete space. Hence the Bott periodicity theorem as stated in Theorem 2.20 also
gives a homotopy equivalence:
BU × Z ' Ω2 BU.
2.3. Spectra and Generalized Cohomology Theories. Equipped with Bott
periodicity, we are in a position to formally extend the reduced complex K-group
functor defined in Section 2.1 to a reduced generalized cohomology theory. We
start by recalling some facts about generalized cohomology theories, an instance of
which is reduced singular cohomology with coefficients in an abelian group G.

Recall the Eilenberg-Steenrod axioms for a reduced generalized cohomology the-

ory:
Definition 2.21. A reduced generalized cohomology theory consists of a sequence
of functors h̃n : Ho(Top∗ )op → Ab satisfying the following axioms for all n:
i
• (i) (exactness) Given any cofibration A →− X, the associated cofiber se-
i
quence A →− X → Ci gives rise to an exact sequence:
h̃n (Ci ) → h̃n (X) → h̃n (A);
• (ii) (suspension) There exists a natural isomorphism of functors:
'
Σ : h̃n (X) −
→ h̃n+1 (ΣX);
i l
• (iii) (additivity) given a wedge sum X = ∨i Xi , the inclusion maps Xi −
→
X give rise to an isomorphism:
'
Y
h̃n (X) −
→ h̃n (Xi );
i
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 19

f
• (iv) (weak equivalence) given a weak equivalence X −
→ Y , the induced
∗
f
map h̃n (Y ) −→ h̃n (X) is an isomorphism.
Remark 2.22. Any reduced generalized cohomology theory determines and is
determined by an unreduced generalized cohomology theory, as defined in Chapter
18 of May [13]. In particular, the construction we present in Section 2.4 for reduced
topological K-theory as a reduced generalized cohomology theory extending K̃ 0 (−)
in degree 0 corresponds to an unreduced theory extending K 0 (−).
Now, we may view each h̃n as taking values in the category of sets. By defini-
tion, h̃n is then a homotopy invariant, contravariant functor which commutes with
coproducts. It can further be checked that the Mayer-Vietoris axiom is satisfied as
a result of the above axioms. Hence, by the Brown representability theorem stated
in Section 1.2, each functor h̃n is representable, i.e. admits a natural isomorphism:
∼
→ h̃n (X),
[X, Tn ] −
for some based space Tn which is unique up to homotopy equivalence.

Now using the loop-suspension adjunction together with the suspension axiom,
we get a sequence of natural isomorphisms:
[X, ΩTn+1 ] ∼
= [ΣX, Tn+1 ] ∼
= h̃n+1 (ΣX) ∼
= h̃n (X),
implying that the functor h̃n is also represented by the space ΩTn+1 . It follows by
uniqueness that we must have a homotopy equivalence:
∼
(2.23) Tn −
→ ΩTn+1 ,
corresponding under the loop-suspension adjunction to a unique map (up to homo-
topy) ΣTn → Tn+1 .

Such a sequence of based spaces {Tn }n≥0 equipped with structure maps σn :
ΣEn → En+1 for each n is known as a spectrum. Spectra constitute a central ob-
ject of study in stable homotopy theory. In what follows, we are mainly concerned
with the notion of Ω-spectra: we say that a spectrum {Tn }n≥0 is an Ω-spectrum
if the adjoint map σ̃n : Tn → ΩTn+1 of each structure map is a weak homotopy
equivalence.

Thus, the Brown representability theorem implies that any reduced generalized
cohomology theory is represented by an Ω-spectrum. A converse to the above
statement can readily be established, once one knows what kind of structure to
look for. Namely, the sequence of functors h̃n (−) := [−, En ] : Topop ∗ → Set
associated with any given Ω-spectrum gives rise to a valid generalized cohomology
theory. First observe that these functors are clearly homotopy invariant, and that
the isomorphism
h̃n (X) = [X, En ] ∼
= [X, Ω2 En+2 ]
for all n ≥ 0 guarantees that each h̃n (X) has an abelian group structure, hence
that we in fact get a sequence of functors h̃n : Ho(Top∗ )op → Ab.

To check that these functors do indeed determine a reduced cohomology theory,

we make use of the fact, essentially due to the CW approximation theorems, that
20 SAAD SLAOUI

a reduced generalized cohomology theory as in Definition 2.21 determines and is

determined by its restriction to based CW-complexes, where the weak equivalence
axiom is trivially satisfied by the Whitehead theorem and for which the exactness
axiom is replaced by the requirement that for any pair of CW-complexes (X, A)
and for all n ≥ 0, the resulting sequence
h̃n (X/A) → h̃n (X) → h̃n (A)
is exact (see Chapter 19 of May [13] for more details). All other verifications are
more or less immediate: the exactness axiom holds by general exactness of the
functor [−, Z] applied to a cofiber sequence for any based space Z, the suspension
axiom follows from naturality of the loop-suspension adjunction, and the additivity
axiom follows from compatibility of the hom functor with coproducts.

Therefore, there is a bijective correspondence between Ω-spectra and reduced

generalized cohomology theories given by the assignment:
 
{En }n≥0 7→ h̃n : Ho(Top∗ )op → Ab, X 7→ [X, En ] .
n≥0

Remark 2.24. (i) It should be noted that, under the proper notion of morphisms
of Ω-spectra and cohomology theories, this bijection on objects does not produce
an equivalence of categories. This is due to the existence of hyperphantom maps
between spectra producing the zero map on the associated cohomology theories.
(ii) Some of the existing literature requires that a cohomology theory be defined
for negative n as well, in which case we may extend the above correspondence via
the assignment X 7→ [X, Ω−n E0 ] for n < 0.
2.4. Extending K̃ 0 (−) to a Reduced Generalized Cohomology Theory. We
are now ready to introduce the Ω-spectrum corresponding to (complex) topological
K-theory. Recall that Bott periodicity gives us a homotopy equivalence Ω2 BU =
Ω2 (BU × Z) ' BU × Z. We may then formulate the following definition.
Definition 2.25. The complex K-theory spectrum KU is defined to be the Ω-
spectrum whose components are given by:
(
BU × Z if n is even,
KUn :=
ΩBU if n is odd.
For even n, the adjoint of the structure map σ̃n : KUn → ΩKUn+1 is defined to be
'
the Bott homotopy equivalence BU × Z − → Ω2 BU , and for odd n it is defined to
'
be the homeomorphism ΩBU − → Ω(BU × Z).
We then define complex topological K-theory to be the reduced generalized co-
homology theory associated to this Ω-spectrum, by defining for all n ≥ 0 and based
space X:
K̃ n (X) := [X, KUn ].
The 0th component of this cohomology theory coincides with our earlier definition
of the reduced complex K-theory group, under the natural isomorphism K̃ 0 (X) ∼ =
[X, BU ×Z] discussed in Section 2.1. Next, for n = 1, we find by the loop-suspension
adjunction that:
K̃ 1 (X) = [X, ΩBU ] ∼
= [X, Ω(BU × Z)] ∼ = [ΣX, BU × Z] = K̃ 0 (ΣX).
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 21

Now, complex K-theory is a 2-periodic cohomology theory, in the sense that, for
all n:
(2.26) K̃ n+2 (X) = [X, KUn+2 ] = [X, KUn ] = K̃ n (X).
Hence, the above two cases suffice to determine the behavior of this cohomology
theory for all n.

3. Adams Operations and the Chern Character

In this chapter, we spend some time developing tools that will allow us to get a
better grasp on topological K-theory as a generalized cohomology theory, eventu-
ally leading to the resolution of the Hopf invariant one problem in Chapter 4. In
Section 3.1, we construct the Adams operations, which are cohomology operations
on topological K-theory which can be viewed as the analogues of the Steenrod
operations in singular cohomology. Next, in Section 3.2, we use Chern classes to
define the Chern character, which will give us a way to translate data from topolog-
ical K-theory into the world of singular cohomology via a natural transformation
compatible with the Adams operations.
3.1. Construction of the Adams operations. By a cohomology operation on
K-theory, we mean a natural transformation F : K 0 (−) → K 0 (−) from the K-
group functor to itself, viewed as a functor K 0 (−) : Ho(Top∗ )op → Set (that is, we
do not necessarily require the component maps FX : K 0 (X) → K 0 (X) to be group
homomorphisms). In order to carry out the construction of the Adams operations,
we will be needing a key lemma whose proof may be found in Chapter 2 of Hatcher
[11]:
Lemma 3.1. (Splitting principle) Let ξ : E → X be a vector bundle of rank
n over a compact Hausdorff base space. Then there exists a map p : F (E) → X
from a compact Hausdorff space F (E) such that the induced map on K-groups
p∗ : K 0 (X) → K 0 (F (E)) is injective, andLsuch that the image of ξ under p splits as
n
a direct sum of lines bundles, i.e. p∗ E = i=1 Li for some line bundles L1 , ..., Ln ∈
0
K (F (E)).
Remark 3.2. Notice that the splitting principle may be slightly extended so as
π π
to account for a pair of vector bundles ξ : E − → X, ξ 0 : E 0 −
→ X. Namely, starting
with the map p : F (E) → X provided by applying the splitting principle to E, with
p∗ injective and p∗ ξ splitting as a sum of line bundles, we may apply the splitting
principle again with respect to the vector bundle p∗ ξ 0 ∈ K 0 (F (E)) to obtain a map
q : F (F (E)) → F (E) such that q ∗ is injective and q ∗ p∗ ξ 0 = (pq)∗ ξ 0 splits as a
sum of line bundles. Then the composite (pq)∗ is injective, and by compatibility of
pullbacks with taking Whitney sums, (pq)∗ ξ also splits as a sum of line bundles in
K 0 (F (F (E))), as needed.
Theorem 3.3. (Adams Operations) There exists a unique collection of coho-
mology operations on K-theory (ψ k : K 0 (−) → K 0 (−))k≥0 satisfying the following
properties:
• (i) Each ψ k : K 0 (X) → K 0 (X) is a group homomorphism;
• (ii) For any line bundle L, ψ k (L) = Lk ;
• (iii) Each ψ k : K 0 (X) → K 0 (X) is a ring homomorphism;
• (iv) ψ k ψ l = ψ l ψ k = ψ kl , for all k, l ≥ 0;
22 SAAD SLAOUI

• (v) For all primes p, x ∈ K 0 (X), we have ψ p (x) ≡ xp (mod p), where by
x ≡ y(mod p) we mean that x = y + pz for some z ∈ K(X);
• (vi) For all m, k ≥ 0, x ∈ K̃ 0 (S 2m ) ∼
= Z, we have ψ k (x) = k m x.
We call ψ k : K 0 (−) → K 0 (−) the k th Adams operation.
Remark 3.4. Part (iv) of the above theorem relies on the fact that naturality of
the Adams operations with respect to the inclusion map l : {pt} ,→ X guarantees
that we obtain induced natural transformations:
ψ k : K̃ 0 (−) → K̃ 0 (−).
We readily observe that properties (i) and (ii) already suffice to characterize the
Adams operations uniquely. For suppose that ψ k , ϕk : K 0 (−) → K 0 (−) are two
natural transformations satisfying these two properties. Then agreement of ψ k , ϕk
is immediate on line bundles, and by additivity we also get that
ψ k (⊕ki=1 Li ) = ⊕ki=1 ψ k (Li ) = ⊕ki=1 Lki = ⊕ki=1 ϕk (Li ) = ϕk (⊕ki=1 Li ),
giving agreement of ψ k and ϕk for any finite sum of line bundles. Finally, for an
π
arbitrary vector bundle ξ : E − → X, we may use the splitting principle to find a
space F (E) and a map p : F (E) → X as in Lemma 3.1, such that p∗ ξ = ⊕ni=1 Li
for some line bundles L1 , ..., Ln ∈ K 0 (F (E)), so that by naturality of ψ k and ϕk
we see that
p∗ ψ k (ξ) = ψ k (p∗ ξ) = ψ k (⊕ni=1 Li ) = ϕk (⊕ni=1 Li ) = ϕk (p∗ ξ) = p∗ ϕk (ξ);
from there, injectivity of p∗ allows us to conclude that ψ k (ξ) = ϕk (ξ), hence that
ψ k = ϕk , as claimed. Also note that once existence is established, property (iv)
follows immediately from the fact that, without loss of generality, the composite
ψ k ψ l is a natural transformation, a group homomorphism on each component, and
on line bundles satisfies ψ k ψ l (L) = ψ k (Ll ) = (Ll )k = Llk , hence by uniqueness we
must have that ψ k ψ l = ψ kl , resp. ψ l ψ k = ψ kl .

For the remainder of this section, we set out to prove the existence of Adams
operations satisfying all of the above properties. Our approach is inspired by the
elegant treatment in Chapter 11 of Wirthmüller [23]. The essential ingredient in the
following construction is the fact that each K-theory ring K 0 (X) has the structure
of a λ-ring induced from the exterior power operations E 7→ Λk (E) induced for
each k ≥ 0 from the k th exterior power operation at the level of vector spaces
following Construction 1.13. These operations satisfy certain properties following
from the properties of the standard exterior product on vector spaces: Λ0 (E) = 1,
Λk (E) = 0 whenever k > rk E, and on direct sums
M
(3.5) Λk (E ⊕ F ) = Λi (E) ⊗ Λj (F ).
i+j=k

Construction 3.6. Observe that the collection of exterior power operations may
be used to associate to any vector bundle on a space X a formal power series with
values in K(X). Namely, we define
λt : Vect(X) → K 0 (X)[[t]],
X
λt (E) := Λk (E)tk .
k≥0
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 23

Since the constant coefficient of λt (E) always equals 1, this assignment in fact maps
into the multiplicative group of formal power series with coefficients in K 0 (X):
λt : VectC (X) → K 0 (X)[[t]]× ,
where the product on K 0 (X)[[t]]× is given by formal product of power series. Then,
we see from Equation 3.5 above that λt is in fact a semigroup homomorphism, hence
factors through a unique group homomorphism
λt : K 0 (X) → K 0 (X)[[t]]×
by the defining universal property of the Grothendieck group. We also denote
this homomorphism by λt , so that on virtual bundles we have λt (E − F ) =
λt (E)λt (F )−1 .
Equipped with this homomorphism, consider the following power series:

ψt : K 0 (X) → K 0 (X)[[t]]
d
(3.7) ψt (E) := ψ 0 (E) − t logλ−t (E),
dt
where ψ 0 (E) denotes the trivial bundle over X whose rank coincides with the rank
of E over each connected component of X. We then define the Adams opera-
tions (ψ k : K(−) → K(−))k≥0 to be the coefficients of the resulting power series
expansion: X
ψt (E) := ψ k (E)tk .
k≥0
The properties of the Adams operations listed in Theorem 3.3 may then be ver-
ified through direct manipulations of formal power series, together with the use of
the splitting principle stated in Lemma 3.1. We first observe that naturality of λt
with respect to pullbacks implies naturality of the Adams operations by construc-
tion.

Starting with a line bundle L ∈ K 0 (X), we directly compute that:

d d
ψt (L) = 1 − t log(1 − Lt) = 1 − t(1 − Lt)−1 (1 − Lt)
dt X X X dt
= 1 + Lt( Lk tk ) = Lk tk = ψ k (L)tk ,
k≥0 k≥0 k≥0

verifying property (i).

Next, given two line bundles L, P ∈ K 0 (X), using the properties of the loga-
rithm, we find that:
d d
ψt (L ⊕ P ) = 2 − t logλ−t (L ⊕ P ) = 2 − t log(λ−t (L)λ−t (P ))
dt dt
d d
= (1 − t logλ−t (L)) + (1 − t logλ−t (P ))
dt dt
X X
= (ψ k (L) + ψ k (P ))tk = ψ k (L ⊕ P )tk ,
k≥0 k≥0

and likewise for finite sums of line bundles. Then, if E, E 0 ∈ K 0 (X) are arbitrary
vector bundles, we may use the extended splitting principle discussed in Remark 3.2
24 SAAD SLAOUI

to find a space Y and a map p : Y → X such that p∗ : K 0 (X) → K 0 (Y ) is injective

and such that p∗ E = ⊕ni=1 Li , p∗ E 0 = ⊕m
j=1 Pj both split as direct sums of line
bundles. We then have by naturality:
(3.8) p∗ (ψ k (E ⊕ E 0 )) = ψ k (p∗ E ⊕ p∗ E 0 ) = ψ k ((⊕ni=1 Li ) ⊕ (⊕m
j=1 Pj ))
k ∗ k ∗ 0
= ψ k (⊕ni=1 Li ) ⊕ ψ k (⊕m
j=1 Pj ) = ψ (p E) ⊕ ψ (p E )

= p∗ (ψ k (E) ⊕ ψ k (E 0 )),

whence it follows from injectivity of p∗ that ψ k (E ⊕E 0 ) = ψ k (E)⊕ψ k (E 0 ), verifying

property (ii). As observed earlier, property (iv) now also follows by uniqueness.

Similarly, starting with two line bundles L, P ∈ K 0 (X), since L ⊗ P is also a

line bundle, we compute:
X X X X
ψt (L ⊗ P ) = (LP )k tk = Lk P k tk = ψ k (L)ψ k (P )tk = ψ k (L ⊗ P )tk ,
k≥0 k≥0 k≥0 k≥0

giving multiplicativity of the Adams operations on line bundles. From there, we

find that on finite sums of line bundles:
n,m
X
n,m
ψt ((⊕ni=1 Li ) ⊗ (⊕m
j=1 Pj )) = ψt (⊕i,j=1 (Li ⊗ Pj )) = ψt (Li ⊗ Pj )
i,j=1
n,m X
X X Xn,m
= ψ k (Li )ψ k (Pj )tk = ( ψ k (Li )ψ k (Pj ))tk
i,j=1 k≥0 k≥0 i,j=1
X X
k
= ψ (⊕ni=1 Li )ψ k (⊕m
j=1 Pj )t
k
= ψ k ((⊕ni=1 Li ) ⊗ (⊕m k
j=1 Pj ))t ,
k≥0 k≥0

as needed. The verification of property (iii) for the general case then follows from
the splitting principle in the same way as in Equation 3.8 above.

To establish property (v), we first observe that for p: prime and ⊕ni=1 Li a sum
of line bundles, we immediately have:
ψ p (⊕ni=1 Li ) = ⊕ni=1 Lpi ≡ (⊕ni=1 Li )p (mod p),
and the general case follows from the splitting principle.

Finally, we verify property (vi). Recall that by the Bott periodicity theorem,
∼ Z is generated by the rth external product β ×r of the Bott element
K̃ 0 (S 2r ) =
β = (1 − L) ∈ K̃ 0 (S 2 ). Now, since L is a line bundle, we have that
k−1
X
ψ k (L − 1) = ψ k (L) − 1 = Lk − 1 = (L − 1)( Lj ) = k(L − 1),
j=0

j j
since each L = ((L − 1) + 1) ≡ 1(mod L − 1), and hence it follows from the fact
that ψ k is a ring homomorphism and a natural transformation that
ψ k ((L − 1)×r ) = (ψ k (L − 1))×r = k r (L − 1)×r ,
completing the proof.
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 25

3.2. Chern Classes and the Chern Character. We now set out to introduce
the Chern character, which provides us with a useful connection between complex
K-groups and singular cohomology. The Chern character will be defined as a
“polynomial over the Chern classes”, and so we take a moment to recall the main
properties of the latter:
π
Theorem 3.9. (Chern classes) Given a rank n complex vector bundle ξ : E − →
X, there is a unique family of cohomology classes ci (E) ∈ H 2i (X; Z), satisfying the
following properties:
• c0 (E) = 1, the unit in H ∗ (B; Z), and ci (E) = 0 for i > n;
• (Naturality) Chern classes commute with pullbacks: given a map f : Y → X
and a bundle E over X, we have that ci (f ∗ E) = f ∗ (ci (E));
• (Whitney product formula) Given vector bundles E and E 0 over the same
base space, we have:
X
ci (E ⊕ E 0 ) = cp (E) ∪ cq (E 0 );
p+q=i

• Letting γ1∞ denote the tautological line bundle over CP ∞ , we have that
c1 (γ1 ) generates H 2 (CP ∞ ).
∞

We refer to ci (E) ∈ H 2i (X; Z) as the ith Chern class of the bundle ξ.

Remarks 3.1. (i) The naturality axiom ensures that each ci (−) specifies the data
of a natural transformation ci (−) : VectC (−) → H ∗ (−; Z), taking values in even
degree cohomology groups, i.e. factoring through H ev (−; Z) = ⊕i∈Z H 2i (−; Z).
π
(ii) Given a rank n bundle ξ : E − → X, we may view the formal sum c(E) =
1+c1 (E)+...+cn (E) as an element in the multiplicative group of units H ∗ (X; Z)× ,
which we refer to as the total Chern class of the bundle ξ. Then, the Whitney
product formula for vector bundles E, E 0 over X simplifies to:
c(E ⊕ E 0 ) = c(E)c(E 0 ).
(iii) Given two line bundles L, L0 , we also have the useful relation:
c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 ).
In constructing the Chern character, it will be useful to keep in mind the explicit
π
construction of the Chern classes of a vector bundle. Let γk∞ : EU (k) − → BU (k)
∞
denote the universal rank k vector bundle, where BU (k) = Grk (C ). Under the
canonical isomorphism (C∞ )k ∼ = C∞ , we have an inclusion map S : (CP ∞ )k →
∞ k ∼
Grk ((C ) ) = BU (k) sending k-lines to the k-plane they span in (C∞ )k . Then, it
can be checked that this map appears in the pullback square:
(3.10) ⊕ki=1 πi∗ (L) / EU (k)

 
(CP ∞ )k
S / Grk ((C∞ )k ) = BU (k),

where πi : (CP ∞ )k → CP ∞ are the projection maps for i = 1, ..., k and L → CP ∞

is the tautological line bundle over CP ∞ .

Next, S induces a map in cohomology:

(3.11) S ∗ : H ∗ (BU (k); Z) → H ∗ ((CP ∞ )k ; Z) ∼
= Z[x1 , ..., xk | |xi | = 2],
26 SAAD SLAOUI

where the isomorphism H ∗ ((CP ∞ )k ; Z) ∼ = Z[x1 , ..., xk | |xi | = 2] is obtained from

the Künneth theorem together with the standard fact that H ∗ (CP ∞ ; Z) ∼ = Z[x | |xi | =
2], so that each generator xi in Equation 3.11 is obtained via pullback of the gen-
erator of the cohomology ring of the ith component of (CP ∞ )k , i.e. xi = πi∗ c1 (L).
Next, consider the standard action of Σk , the symmetric group on k letters, on
the product (CP ∞ )k , given by permutation of the variables. Passing to integral
cohomology, we get an induced action of Σk on H ∗ ((CP ∞ )k ; Z) permuting the
generators. We then have the identification:
H ∗ ((CP ∞ )k ; Z)Σk ∼
= Z[σ1 , ..., σk ],
where H ∗ ((CP ∞ )k ; Z)Σk is the subgroup of Σk -fixed points, and where each σi
denotes the ith elementary symmetric polynomial on k variables. Now, for each
θ ∈ Σk , since the inclusion S ◦ θ : (CP ∞ )k → BU (k) produces a bundle isomorphic
to ⊕ki=1 πi∗ (L) under pullback of γk∞ , we get by the homotopy classification theorem
that Sθ ' S, and hence that θ∗ S ∗ = S ∗ at the level of cohomology. Therefore, the
map S ∗ : H ∗ (BU (k); Z) → H ∗ ((CP ∞ )k ; Z) factors through a map:
(3.12) S ∗ : H ∗ (BU (k); Z) → H ∗ ((CP ∞ )k ; Z)Σk ∼
= Z[σ1 , ..., σk ],
One may then show that the above map is an isomorphism, so that H ∗ (BU (k); Z)
is isomorphic to the polynomial algebra over the elementary symmetric polynomials
in k variables. Under this isomorphism, we may then define the ith Chern class of
γk∞ for i = 1, ..., k via the relation:
ci (γk∞ ) := (S ∗ )−1 (σi ).
Note that this is well-defined since each xj has degree 2, so that σi has degree 2i.
Next, given an arbitrary rank k vector bundle E → X, let f : X → BU (k) be a
classifying map and define
ci (E) := f ∗ (ci (γk∞ )),
which is well-defined by the homotopy classification theorem and homotopy invari-
ance of cohomology. This construction may then be verified to satisfy the properties
of Theorem 3.9.

Equipped with the notion of Chern classes taking values in singular cohomology,
we start moving towards a natural transformation ϕ : K 0 (−) → H ∗ (−), which we
require to be a ring homomorphism in each component. Intuitively, we start with
Chern classes and look for an appropriate polynomial over them. In particular, we
expect ϕ to factor through H ev (−; Q) = ⊕i∈Z H 2i (−; Z). ThePrelation c1 (L ⊗ L0 ) =
0 1
c1 (L) + c1 (L ) motivates the definition ϕ(L) = exp(c1 (L)) = r≥0 r! c1 (L)r on line
bundles, in which case we must allow ϕ to take values in H ev (−; Q). We then have
the following:
Theorem 3.13. (Chern Character) There exists a unique natural transforma-
tion Ch : K 0 (−) → H ev (−; Q) satisfying the following properties:
• (i) Each Ch : K 0 (X) → H ev (X; Q) is a group homomorphism;
• (ii) For any line bundle L, Ch(L) = exp(c1 (L));;
• (iii) Each Ch : K 0 (X) → H ev (X; Q) is a ring homomorphism;
• (iv) For all m ≥ 0, Ch : K̃ 0 (S 2m ) → H̃ ev (S 2m ; Q) maps K̃ 0 (S 2m ) isomor-
phically onto H ev (S 2m ; Z) ⊂ H ev (S 2m ; Q).
We refer to Ch : K 0 (−) → H ev (−; Q) as the Chern character.
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 27

Remark 3.14. (i) Part (iv) of the above theorem relies on the fact that the Chern
character induces a natural transformation:
Ch : K̃ 0 (−) → H̃ ev (−; Q)
in the same way as in Remark 3.4 (ii) Via the suspension isomorphism and Bott
periodicity, the Chern character may also be extended to complex topological K-
theory in every degree in the form of a map
Ch : K ∗ (−) → ⊕p∈Z H ∗+2p (−).

We shall see that Chern character is given on an arbitrary rank k vector bundle
E → X by:
X 1
(3.15) Ch(E) = sr (E),
r!
r≥0

where sr (E) := Sr (c1 (E), ..., ck (E)) is defined to be the rth Newton polynomial
evaluated on the Chern classes of E (recall that the rth Newton polynomial is
the unique polynomial Sr ∈ Z[σ1 , ..., σ k ] satisfying Sr (σ1 , ..., σk ) = xr1 + ... + xrk ,
where σi ∈ Z[x1 , ..., xk ] is the ith elementary symmetric polynomial). This as-
signment at the level of VectC (X) factors by additivity through the desired map
Ch : K 0 (X) → H ev (X; Q).

As with the Adams operations, specifying the behavior of such a map on line
bundles and requiring it to be additive suffices to give uniqueness. Indeed, suppose
that we disposed of such a natural transformation ϕ : K 0 (−) → H ev (−; Q). Then
the inclusion map S : (CP ∞ )k ,→ BU (k) as in Equation 3.10 fits into a commutative
square:
S∗
K 0 ((CP ∞ )k ) o K 0 (BU (k))
ϕ ϕ
 ∗ 
H ev ((CP ∞ )k ; Q) o
S
H ev (BU (k); Q).
As mentioned earlier, the bottom horizontal map is an isomorphism onto the sub-
group H ev ((CP ∞ )k ; Q)Sk of Sk -fixed points, hence in particular it is a monomor-
phism. From the pullback square in Equation 3.10, we also have that S ∗ γk∞ =
⊕ki=1 πi∗ (L). Hence, by additivity and naturality of ϕ, we find that:
k
X
S ∗ ϕ(γk∞ ) = ϕ(S ∗ γk∞ ) = ϕ(⊕ki=1 πi∗ (L)) = πi∗ (ϕ(L))
i=1
k
X k
X X
= πi∗ (exp(c1 (L))) = exp(πi∗ c1 (L)) = (xr1 + ... + xrk ),
i=1 i=1 r≥0

where we set xi := πi∗ c1 (L) for i = 1, ..., k as in the notation of Equation 3.11.
Recall that the Chern classes of γk∞ were defined via the relation S ∗ ci (γk∞ ) = σi
under the isomorphism
'
→ H ∗ ((CP ∞ )k ; Z) ∼
S ∗ : H ∗ (BU (k); Z) − = Z[σ1 , ..., σk ].
28 SAAD SLAOUI

Thus, we have that:

X  X
1 1 ∗
S∗ sr (γk∞ ) = S (Sr (c1 (γk∞ ), ..., ck (γk∞ ))
r! r!
r≥0 r≥0
X 1 X
= Sr (σ1 , ..., σk ) = (xr1 + ... + xrk ).
r!
r≥0 r≥0
∗
Hence, by injectivity of S , we get from the above two computations that the
behavior of ϕ on the universal rank k-bundle is forced to be:
X 1
ϕ(γk∞ ) = sr (γk∞ ).
r!
r≥0

Next, for an arbitrary vector bundle E → X, let f : X → BU (k) be an associated

classifying map, so that f ∗ γk∞ = E. Also recall that the Chern classes of E were
defined by the relation ci (E) := f ∗ ci (γk∞ ). Then, by naturality of ϕ, we see that:
X 
1
ϕ(E) = ϕ(f ∗ γk∞ ) = f ∗ (ϕ(γk∞ )) = f ∗ sr (γk∞ )
r!
r≥0
X 1 X 1
= Sr (f ∗ c1 (γk∞ ), ..., f ∗ ck (γk∞ )) = sr (E),
r! r!
r≥0 r≥0

completing the verification of uniqueness.

Hence it suffices to verify that the map Ch : K 0 (−) → H ev (−; Q) given by

Equation 3.15 is natural, and that it is compatible with sums and products of
vector bundles. Naturality follows readily from naturality of the Chern classes:
f
given a map Y − → X and a rank k bundle E → X, we have that:
X  X
1 1
f ∗ Ch(E) = f ∗ Sr (c1 (E), ..., ck (E)) = Sr (c1 (f ∗ E), ..., ck (f ∗ E)) = Ch(f ∗ E).
r! r!
r≥0 r≥0

Next, for additivity, first observe that by the Whitney product formula, given
line bundles L1 , ..., Lk , we have that:
k
Y k
Y k
X
c(⊕ki=1 Li ) = c(Li ) = (1 + c1 (Li )) = σi (c1 (L1 ), ..., c1 (Lk )),
i=1 i=1 i=1

i.e. cj (⊕ki=1 Li ) = σj (c1 (L1 ), ..., c1 (Lk )). Hence, on sums of line bundles, we find
that:
X 1 X 1
Ch(⊕ki=1 Li ) = sr (⊕ki=1 Li ) = Sr (σ1 , ..., σk )
r! r!
r≥0 r≥0

X 1 Xk Xk
= (c1 (L1 )r + ... + c1 (Lk )r ) = exp(c1 (Li )) = Ch(Li ),
r! i=1 i=1
r≥0

as needed. Next, given arbitrary bundles E, E 0 → X, we may use the splitting

principle stated in the previous section to find a map q : F → X such that
q ∗ : K 0 (X) → K 0 (F ) is injective and q ∗ E = ⊕ki=1 Li , q ∗ E 0 = ⊕lj=1 Pj both split
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 29

as sums of line bundles. Then, by naturality, we get that:

q ∗ (Ch(E ⊕ E 0 )) = Ch(q ∗ E ⊕ q ∗ E 0 ) = Ch(⊕ki=1 Li ⊕ ⊕lj=1 Pj )
= Ch(⊕ki=1 Li ) + Ch(⊕lj=1 Pj ) = q ∗ (Ch(E) + Ch(E 0 )),
hence by injectivity it follows that Ch(E ⊕ E 0 ) = Ch(E) + Ch(E).

Finally, to check multiplicativity, recall the formula c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 )

for the first Chern class of the tensor product of two line bundles, which yields:
Ch(L ⊗ L0 ) = exp(c1 (L ⊗ L0 )) = exp(c1 (L) + c1 (L0 ))
= exp(c1 (L))exp(c1 (L0 )) = Ch(L)Ch(L0 )
Then, on sums of line bundles ⊕ki=1 Li , ⊕lj=1 Pj , we get that, by additivity:
X
Ch(⊕ki=1 Li ⊗ ⊕lj=1 Pj ) = Ch(⊕i,j (Li ⊗ Pj )) = Ch(Li ⊗ Pj )
i,j

X k
X l
 X 
= Ch(Li )Ch(Pj ) = Ch(Li ) Ch(Pj ) .
i,j i=1 j=1

The result for arbitrary bundles then follows from the splitting principle in a similar
way as above.

Finally, we verify the claim made in Theorem 3.13 (iv): first recall that the Bott
π
element β = 1 − L, where L − → CP 1 ∼= S 2 is the tautological line bundle, is a
generator of K̃ (S ), and that by Bott periodicity the mth external power β ×m is
0 2

a generator for K̃ 0 (S 2m ). Now, by looking at the pullback under the inclusion map
CP 1 ,→ CP ∞ , it follows from Theorem 3.9 that c1 (L) is a generator of H̃ ev (S 2 ; Z).
Next, since H̃ ev (S 2 ; Z) = Z(2) , higher powers of c1 (L) vanish, and so:
Ch(β) = Ch(1 − L) = Ch(1) − Ch(L) = 1 − exp(c1 (L)) = −c1 (L),
hence Ch maps K̃ 0 (S 2 ) isomorphically onto H̃ ev (S 2 ; Z) ⊂ H̃ ev (S 2 ; Q). The general
case follows from naturality and multiplicativity of the Chern character, from which
we see that:
Ch(β ×m ) = (Ch(β))×m = (−1)m c1 (L)×m ,

where ±c1 (L)m is a generator of H̃ ev (S 2m ; Z), as needed.

4. Applying Topological K-Theory to Problems in Classical

Topology
We now begin to explore some of the bearings the theory we have developed so
far has on problems in classical topology. The first section is devoted to proving
that no sphere of dimension greater than 6 can be given the structure of a complex
manifold, and the following two sections are concerned with the Hopf invariant
one problem. Both proofs are quite simple in essence with the tools we have at
hand. We should note that the applications presented here only scratch the surface
of what can be done - perhaps the most famous application we are omitting is
the establishment of a sharp bound on the maximal possible number of pointwise
linearly independent vector fields on spheres, first proven by Adams in 1961 [4]
30 SAAD SLAOUI

4.1. Non-Existence of Complex Structures on Spheres. Starting with the

observation that the 2-sphere S 2 can be realized as a complex 1-manifold under
the homeomorphism S 2 ∼ = CP 1 , one may wonder whether this is true for spheres of
different (necessarily even) dimensions - i.e, for which n can the smooth manifold
structure on S 2n be enriched into that of a complex manifold? In this section, we
answer this question in the negative for n > 3. It turns out that the same conclu-
sion holds for n = 2; the case n = 3 is still an open problem.

We start with an observation that follows relatively quickly from Theorem 3.13:
S2 n
Lemma 4.1. Given any rank n ≥ 2 complex vector bundle E −−→, we have that
(n − 1)! | cn (E)
inside H̃ 2n
(S 2n
; Z) ∼
= Z.
Proof. Recall from part (iv) of Theorem 3.13 that the image of the Ch : S 2n →
H̃ ev (S 2n ; Q) lives entirely in H̃ ev (S 2n ; Z), i.e.
X 1
(4.2) Ch(E) = Qr (c1 (E), ..., cn (E)) ∈ H̃ ev (S 2n ; Z).
r!
r≥0

Next, by degree considerations, since H̃ ev (S 2n ; Z) is concentrated in degree 2n, all

1
terms in the above sum vanish except for n! Qn (c1 (E), ..., cn (E)). Now, it can be
computed inductively that, for n ≥ 2:
n−1
X
Qn (c1 (E), ..., cn (E)) = ci (E)Qn−i (c1 (E), ..., cn (E)) + (−1)n+1 ncn (E)
i=1

(see Lemma B.1 in Dugger [9] for a quick verification of this fact). Since ci (E) ∈
H̃ 2i (S 2n ; Z) must vanish for i < n, the above equation simplifies to:
Qn (c1 (E), ..., cn (E)) = (−1)n+1 ncn (E).
(−1)n+1
Hence, by Equation 4.2 above, we get that (n−1)! cn (E) ∈ H̃ 2n (S 2n ; Z), and the
result follows. 
Equipped with this lemma, the proof of the main result boils down to an exercise
in characteristic classes:
π
Theorem 4.3. If n > 3, then there is no complex vector bundle T − → S 2n of rank
n whose underlying real vector bundle of rank 2n is the tangent bundle induced from
a smooth manifold structure on S 2n .
π
Proof. Suppose we had such a bundle T − → S 2n . Then the associated Euler class
2n 2n
e(T ) ∈ H̃ (S ; Z) satisfies the equation
2n
X
< e(T ), [S 2n ] >= χ(S 2n ) = (−1)i rk H i (S 2n ; Z) = 2,
i=0
2n 2n
where [S ] ∈ H2n (S ; Z) denotes the fundamental class. But we also know that
the Euler class equals the top Chern class of T (for a quick proof up to sign, use
naturality w.r.t the classifying map of T and the fact that cn (γn∞ ) ∈ H ∗ (BU (n); Z)
is a generator by construction to get that e(T ) = λcn (T ) for some λ ∈ Z, and use
naturality again together with the Gysin sequence in degree 0 associated to the
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 31

universal rank n bundle to see that cn (T ) = µe(T ) for some µ ∈ Z. The result
follows by combining the two relations). From there, by Lemma 4.1, it follows that:
2 =< e(T ), [S 2n ] >=< cn (T ), [S 2n ] >= (n − 1)!k
2
for some k ∈ Z. Hence (n−1)! must be an integer, which requires that n ≤ 3,
completing the proof. 

4.2. The Hopf Invariant One Problem. In 1928, Heinz Hopf first defined the
Hopf invariant of a smooth map f : S 3 → S 2 in terms of the notion of linking
numbers of f , and used it to demonstrate that the Hopf fibration η : S 3 → S 2
mentioned in Examples 1.3 is not null-homotopic. It was later realized that the
notion of Hopf invariant could be defined for a wider range of maps, and had a
bearing on classical problems in algebra and topology, as specified by the following
proposition, a proof of which can be found in Section 2.3 of Hatcher [11]:
Proposition 4.4. The following are equivalent:
• (i) There exists a map f : S 2n−1 → S n with Hopf invariant one;
• (ii) The vector space Rn admits a division algebra structure;
• (iii) The sphere S n−1 admits an H-space structure.
Thus, determining the non-existence of maps f : S 2n−1 → S n of Hopf invariant
one is equivalent to proving the non-existence of division algebra structures on Rn ,
as well as the non-existence of H-space structures on S n−1 (where the data of an
H-space structure on a space X consists of a continuous map µ : X × X → X and a
distinguished element e ∈ x such that µ(−, e) = µ(e, −) = idX ). We already know
that maps of Hopf invariant one exist for n = 1, 2, 4, 8, corresponding respectively
to the usual real division algebras R, C, H, and O. Strikingly, this list of admissible
dimensions turns out to be exhaustive:
Theorem 4.5. There exists a map f : S 2n−1 → S n with Hopf invariant one if and
only if n ∈ {1, 2, 4, 8}.
This section introduces the notion of Hopf invariant and explores some of its
properties, and the next section builds towards a proof of the above theorem.

We start by presenting two definitions for the Hopf invariant of a map f : S 2n−1 →
n
S , and verify that they coincide. The first formulation is homotopy theoretic in
nature, while the second is inspired from the language of differential forms. As
mentioned in the opening of this section, a more geometric definition may also be
given; we choose to omit it from this discussion, and refer the interested reader to
a treatment by Steenrod [22].
Construction 4.6. Let f : S 2n−1 → S n be any continuous map. We may form
the mapping cone associated to f , denoted by Xf and obtained via the following
pushout square:

S 2n−1
f
/ Sn

l α
 
D2n / Xf ,
β
32 SAAD SLAOUI

where we identify the cone on S 2n−1 with a 2n-disk D2n . Viewing S n as a CW-
complex with one 0-cell and one n-cell, the above diagram immediately gives Xf
the structure of a CW-complex with one cell each in dimensions 0, n, and 2n.
Cellular cohomology then tells us that the cohomology ring of Xf with integral
coefficients consists of a copy of Z in dimensions 0, n, 2n. Hence, if we let a, resp.
b denote generators for H n (Xf ; Z), resp. H 2n (Xf ; Z), then the product structure
of H ∗ (Xf ; Z) as a graded ring is fully determined by an integer, which we denote
by H(f ), such that:
a ∪ a := H(f )b.
We observe a few things: H(f ) as above is well-defined up to sign, and the latter
behavior can be controlled by making a consistent choice of orientation. Since
homotopic maps f ' g : S 2n−1 → S n yield homotopy equivalent mapping cones
Xf ' Xg , the integer H(f ) only depends on f up to homotopy, hence induces
an assignment H(−) : π2n −1 (S n ) → Z, which we will later verify to be a group
homomorphism. Also observe that for odd n, graded commutativity of the cup
product implies that
2
H(f )b = a ∪ a = (−1)|a| a ∪ a = −H(f )b,
and hence that H(f ) = 0. Thus the value of H(f ) can only be non-zero for maps
f into even-dimensional spheres.
Construction 4.7. Starting again with a continuous map f : S 2n−1 → S n , let [x]
denote a generator of the top integral cohomology group of S n . Then the pullback
of x under f yields an element f ∗ (x) ∈ H n (S 2n−1 ; Z) = 0, so that we may express
f ∗ (x) = δy as the coboundary of some (n − 1)-cochain y ∈ C n−1 (S 2n−1 ). Letting
[S 2n−1 ] ∈ H2n−1 (S 2n−1 ; Z) denote the fundamental class of S 2n−1 , we may then
obtain an integer, which we denote by H̃(f ), via the following dual pairing:
H̃(f ) :=< [S 2n−1 ], y ∪ δy > .
Again, we observe that H̃(f ) is well-defined up to sign, independently of the choices
of x and y made above.
Passing to singular cohomology with real coefficients via the UCT and applying
de Rham’s theorem, we see that this construction translates into a definition in
the language of differential forms. Namely, if we let [ω] ∈ HdR (S n ) be a top
degree form such that S n ω = 1 and write the pullback form f ∗ ω = dη for some
R

η ∈ Ωn−1 (S 2n−1 ) by exactness, we have that

Z
H̃(f ) = η ∧ dη.
S 2n−1

Proposition 4.8. Given any continuous map f : S 2n−1 → S n , the integers H(f )
and H̃(f ) defined above are equal. We define the Hopf invariant of the map f to
be the integer H(f ).
Proof. Let us recall the pushout square from Construction 4.6:

(4.9) S 2n−1
f
/ Sn

l α
 
D2n / Xf .
β
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 33

As above, denote by a, resp. b chosen generators for H n (Xf ; Z), resp. H 2n (Xf ; Z).
Since [β ∗ a] ∈ H n (D2n ) = 0, we may write β ∗ a = δz for some cochain z ∈
C n−1 (D2n ). We then compute:

H(f ) = H(f ) < [D2n ], β ∗ b >

=< [D2n ], β ∗ (a ∪ a) > by definition of H(f)
2n ∗ ∗
=< [D ], β a ∪ β a) > by naturality of the cup product
2n
=< [D ], δz ∪ δz) >
=< [D2n ], δ(z ∪ δz) >
=< ∂[D2n ], z ∪ δz >
=< l∗ [S 2n−1 ], z ∪ δz >
=< [S 2n−1 ], l∗ z ∪ δl∗ z > by duality and naturality.
Now, by commutativity of the diagram in Equation 4.9, if we set y := l∗ z ∈
C n−1 (S 2n−1 ), then we get that
δy = δl∗ z = l∗ δz = l∗ β ∗ a = f ∗ (α∗ a),
where α∗ a is a valid generator for H n (S n ; Z), hence by Construction 4.6 we get
that:
H̃(f ) =< [S 2n−1 ], δy ∪ y >=< [S 2n−1 ], l∗ z ∪ δl∗ z >= H(f ),
as needed. 
We establish a few elementary facts about the Hopf invariant:
Proposition 4.10. The map H(−) : π2n−1 (S n ) → Z sending a map f to its Hopf
invariant is a group homomorphism.
Proof. First recall that the group structure on π2n−1 (S n ) is given by sending two
maps f, g : S 2n−1 → S n to the composite
f ∨g
f + g : S 2n−1 → S 2n−1 ∨ S 2n−1 −−→ S n ,
where the first map is the pinch map collapsing the equator of S 2n−1 to a point.
Now, we may consider the intermediate mapping cone Xf ∨g , obtained by using the
data of the map f ∨ g : S 2n−1 ∨ S 2n−1 → S n to glue two 2n-cells to S n . We then
get a quotient map
q : Xf +g → Xf ∨g
obtained by gluing together all of the equators of the (2n − 1)-spheres forming the
2n-cell of Xf +g , corresponding to an equatorial (2n − 1)-disk. The map q restricts
to a homeomorphism on the n-cells of Xf +g and Xf ∨g , so that q ∗ (af ∨g ) = af +g in
the usual notation. Next, considering the pullback maps induced by the inclusions
Xf ,→ Xf ∨g , Xg ,→ Xf ∨g tells us that:
a2f ∨g = H(f )bf + H(g)bg ,
where bf , bg denote the generators of the H 2n (Xf ∨g ; Z). It follows by definition of
the Hopf invariant of f + g and linearity of the pullback that:
H(f + g)bf +g = a2f +g = q ∗ (a2f ∨g ) = H(f )q ∗ (bf ) + H(g)q ∗ (bg )
= (H(f ) + H(g))bf +g ,
34 SAAD SLAOUI

as needed. 
Proposition 4.11. For any even n, there exists a map f : S 2n−1 → S n with
H(f ) = 2. Hence, for even n, the image of H(−) : π2n−1 (S n ) → Z always contains
a copy of Z as a direct summand.
Proof. Given an even integer n, we construct a map f : S 2n−1 → S n as follows:
identifying S n with the quotient I n /∂I n of the unit cube in Rn , we may view the
product S n × S n as the quotient of I 2n where we make opposite side identifications
on ∂I n ×I and on I n ×∂I n . Under this quotient, ∂I 2n is homeomorphic to S n ∨S n .
Hence, viewing S 2n−1 ⊂ I 2n ⊂ R2n , we may compose the radial projection map
S 2n−1 → ∂I 2n with the restricted quotient map ∂I 2n → S n × S n to obtain a map
S 2n−1 → S n ∨ S n . Composing further with the fold map S n ∨ S n → S n results in
a map:
f : S 2n−1 → S n ∨ S n → S n .
One can then verify that the mapping cone of f is given by the James reduced
product Xf = S n × S n /(x, ∗) ∼ (∗, x) with respect to the basepoint ∗ ∈ S n . Now,
consider the projection map q : S n × S n → Xf , inducing a map in cohomology
q ∗ : H ∗ (Xf ; Z) → H ∗ (S n × S n ; Z). By the Künneth formula, we know that
H ∗ (S n × S n ; Z) ∼
= Z(0) ⊕ Z2 ⊕ Z(2n)
(n)

has two generators a1 , a2 in degree n and one generator b1 in degree 2n, with cup
product structure given by a21 = a22 = 0, a1 a2 = b1 . Furthermore, q takes each
cell of S n × S n homeomorphically onto a cell of Xf . Thus, letting a ∈ H n (Xf ),
b ∈ H 2n (Xf ), we have by cellular cohomology that q ∗ (a) = a1 + a2 , q ∗ (b) = b1 , so
that:
H(f )a1 a2 = H(f )b1 = H(f )q ∗ (b) = q ∗ (a2 )
= (a1 + a2 )2 = a21 + a1 a2 + a2 a1 + a22 = 2a1 a2 ,
since n is even, so that H(f ) = 2, as needed. The second part of the proposition
follows from the fact that H(−) : π2n−1 (S n ) → Z is a group homomorphism. 
4.3. Reduction to an Ext Computation. We now start working towards a res-
olution of the Hopf invariant one problem. The reader acquainted with elements of
stable homotopy theory may take the initiative of probing the situation by studying
the structure of the cohomology ring H ∗ (Xf ; Z/2) defined in Construction 4.6 as a
module over the mod 2 Steenrod algebra A. This approach leads to the following
realization: in the notation of Construction 4.6, if the map f has Hopf invariant
one, then we get that
Sq n (a) = a ∪ a = b.
It follows from the fact that there are no non-trivial cohomology classes in H i (Xf ; Z/2)
for n < i < 2n that the element Sq n ∈ A must be indecomposable. But the inde-
k
composable elements of A consist precisely of Steenrod squares of the form Sq 2
for some k, whence we get the following:
Theorem 4.12. If a map f : S 2n−1 → S n has Hopf invariant one then n must be
a power of two.
Restricting the range of possible values of n any further becomes a very dif-
ficult problem if one continues to work exclusively with singular cohomology. In
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 35

1960, Adams published a complex proof along those lines involving secondary co-
homology operations and spectral sequence arguments [1]. On the other hand, it is
possible to view the problem through the lens of topological K-theory by means of
the Chern character, in which case the use of Adams operations makes the problem
much more tractable. This approach was first presented in a much shorter paper
of Adams and Atiyah in 1964 [3]. In this section, inspired by the treatment given
in Chapter 5 of Dugger [9], we aim to present the K-theoretic approach in slightly
different settings leading to a more general problem in homological algebra.

Having verified in Construction 4.6 that the Hopf invariant one problem is only
interesting for maps into even dimensional spheres, we assume from now on that n
f
is even and write n = 2r. Let S 2n−1 −
→ S n → Xf be the cofiber sequence associated
with a given map f : S 2n−1 → S n . Then the associated Puppe sequence
S 4r−1 / S 2r / Xf / S 4r / S 2r+1 / ΣXf / ...

induces a long exact sequence of reduced K-groups:

... / K̃ 0 (S 2r+1 ) / K̃ 0 (S 4r ) / K̃ 0 (Xf ) / K̃ 0 (S 2r ) / K̃ 0 (S 4r−1 ).

Now, we saw in Chapter 2 that K̃ 0 vanishes on odd-dimensional spheres, so that

we obtain a short exact sequence:
(4.13) 0 / K̃ 0 (S 4r ) / K̃ 0 (Xf ) / K̃ 0 (S 2r ) / 0.

The above SES may in fact be interpreted in the category of modules over an ap-
propriate ring. Indeed, if we let B := Z[ψ 2 , ψ 3 , ψ 5 , ...] denote the monoid ring over
Z generated by the Adams operations subject to the relations of Theorem 3.3 (iv),
we see that each K-group K̃ 0 (X) comes equipped with a B-module structure as
a result of the fact that the ψ k ’s are group homomorphisms. Then, naturality of
the Adams operations implies that the maps in Equation 4.13 are valid B-module
homomorphisms.

Let Z(r) denote the B-module consisting of a copy of Z generated by an element

x, with the property that ψ k (x) = k r x for all k ≥ 0. Then by Theorem 3.3 (vi),
K̃ 0 (S 2r ) = Z(r) as B-modules. Hence K̃ 0 (Xf ) fits into a SES of B-modules:

0 / Z(2r) / K̃ 0 (Xf ) / Z(r) / 0.

By the usual correspondence between Ext1B (A, B) and isomorphism classes of ex-
tensions of B by A, it follows that we may associate to K̃ 0 (Xf ) a unique element
of the group Ext1B (Z(2r), Z(r)).

We are thus naturally led to the slightly more general problem of computing the
element of Ext1B (Z(s), Z(r)) associated to a SES of B-modules

(4.14) 0 / Z(s) i /X j
/ Z(r) / 0.

In a later discussion, we will clarify the utility of this generalization in the context
of stable homotopy theory. For now, we start an investigation of this problem which
will rather quickly lead us to a resolution of the Hopf invariant one problem.
36 SAAD SLAOUI

Choose generators b ∈ Z(s), ã ∈ Z(r), and let a ∈ X be a preimage of ã under

the surjective map j. For simplicity of notation, we identify b with its image in X
under the injective map i. Let k ≥ 0, and observe that by naturality of ψ k we have
that
j(ψ k a) = ψ k ã = k r ã = j(k r a),
hence by exactness it follows that ψ k a − k r a = Pk b, i.e. that
ψ k a = k r a + Pk b
for some integer Pk ∈ Z. In particular, the structure of X as a B-module is fully
determined by a tuple of integers (P2 , P3 , ...) defined via the above relations. By
Theorem 3.3 (vi), we have that ψ k ψ l (a) = ψ l ψ k (a) for all k, l ≥ 0. Now, a quick
computation yields:
ψ k ψ l (a) = ψ k (lr a + Pl b) = lr ψ k (a) + Pl ψk (b)
= lr (k r a + Pk b) + Pl k s b,

ψ l ψ k (a) = ψ l (k r a + Pk b) = k r ψ l (a) + Pk ψ l (b)

= k r (lr a + Pl b) + Pk ls b.
Equating the two terms, we find that
(k r − k s )Pl = (lr − ls )Pk .
In particular, setting k = 2, we see that specifying the value of P2 determines the
entire tuple via the relations
lr − ls
(4.15) Pl = P2 .
2r − 2s
It follows that a valid B-module structure on X as in Equation 4.14 is determined
by any choice of integer P2 ∈ Zr,s , where we define
lr − ls
Zr,s := {P ∈ Z | P ∈ Z for all l}.
2r − 2s
Note that Zr,s is an ideal of Z, non-empty since 2r − 2s ∈ Zr,s . We thus get a
surjective mapping
ϕ : Zr,s / Ext1B (Z(s), Z(r)) ,

obtained by sending an element P ∈ Zr,s to the B-module XP fitting in a SES as

r
−3s
in Equation 4.14 with B-module structure given by the tuple (P, 23r −2s P, ...).

Proposition 4.16. The map ϕ : Zr,s → Ext1B (Z(s), Z(r)) defined above is a group
homomorphism.
Proof. First recall that the abelian group structure on Ext1B (Z(s), Z(r)) viewed as
isomorphism classes of extensions is given by the Baer sum operation: starting with
two short exact sequences of B-modules
f g
0→A−
→B−
→ C → 0,
f0 g0
0 → A −→ B 0 −→ C → 0,
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 37

we first construct B ×C B 0 from the pullback square

B ×C B 0 /B
g
 
B0 / C,
g0

then define the Baer sum to be the B-module YB,B 0 obtained as the coequalizer of
the diagram:
(f,0)
/ 0 / XB,B 0 .
A / B ×C B
0
(0,f )

Concretely, we then have that:

XB,B 0 = {(b, b0 ) ∈ B ⊕ B 0 | g(b) = g 0 (b0 )} (f (a), −f 0 (a)) | a ∈ A .


In particular, under the maps

f¯
→ XB,B 0 , a 7→ (f (a), 0) = (0, f 0 (a0 )),
A−
ḡ
→ C, (b, b0 ) 7→ g(b) = g 0 (b0 ),
XB,B 0 −
XB,B 0 fits into the SES:
f¯
0 /A / XB,B 0 ḡ
/C /0

In our settings, we are interested in the Baer sum of the short exact sequences

0 / Z(s) iP
/ XP jP
/ Z(r) / 0,

iQ jQ
0 / Z(s) / XQ / Z(r) / 0,

corresponding to the B-modules XP , XQ as defined above, for given elements P, Q ∈

Zr,s . Denote this extension by XP,Q , so that we have a short exact sequence:
i¯P j¯P
0 /A / XP,Q /C /0.

Then as a set, we have that:


XP,Q = {(x, y) ∈ XP ⊕ XQ | jP (x) = jQ (y)} (iP (x), −iQ (x)) | x ∈ Z(s) .
In particular, we may look at the class of the element (aP , aQ ), belonging to XP,Q
since jP (aP ) = jQ (aQ ) = ã by construction. The isomorphism class of XP,Q
in Ext1B (Z(s), Z(r)) is then determined by the integer N such that ψ 2 (aP , aQ ) =
2r (aP , aQ ) + N (b, 0). We verify that:
ψ 2 (aP , aQ ) = (ψ 2 (aP ), ψ 2 (aQ )) = (2r aP + P b, 2r aQ + Qb)
= 2r (aP , 0) + 2r (0, aQ ) + P (b, 0) + Q(0, b)
= 2r (aP , aQ ) + (P + Q)(b, 0),
since (0, b) = (b, 0) in XP,Q . It then readily follows that XP,Q ∼
= XP +Q , as needed.

38 SAAD SLAOUI

The work we have done so far actually suffices to resolve the Hopf invariant one
problem, and so we choose to halt this slightly more general discussion and conclude
the proof of the Hopf invariant one problem. It is worth remarking that the Ext
computation in the general case is not only of algebraic but also of topological
interest, in that it allows us to learn something about the stable homotopy groups
of spheres. Indeed, by a similar reasoning as above, for any k ∈ N, the mapping
cone of a map f : S 2r+2k−1 → S 2r fits into a SES of B-modules:
0 / K̃ 0 (S 2r+2k ) / K̃ 0 (Xf ) / K̃ 0 (S 2r ) / 0,

and hence corresponds to an element of Ext1B (Z(r + k), Z(r)). We thus obtain a
group homomorphism:
A(−) : π2k−1+2r (S 2r ) → Ext1B (Z(r + k), Z(r)).
Acquiring more information about the target group and the kernel of this map can
thus provide insight into the structure of the stable homotopy group π2k−1 (S). For
a more in-depth discussion of the general computation of Ext1B (Z(s), Z(r)), we refer
the interested reader to Section 31 of Dugger [9].

Let us come back to the SES of B-modules associated with a given map f : S 2n−1 →
n
S :

(4.17) 0 / Z(2r) i / K̃ 0 (Xf ) j

/ Z(r) / 0.

As before, choose generators b ∈ Z(2r), ã ∈ Zr , identify ib ≡ b in K̃ 0 (Xf ), and

let a ∈ K̃ 0 (Xf ) be such that j(a) = ã. Then from the ring structure of K̃ 0 (S 2r )
described in Section 2.2, we see that j(a2 ) = (j(a))2 = 0, so that by exactness
a2 = hb for some integer h ∈ Z. Let us verify that this integer is well-defined up
to sign. Injectivity of i ensures that the resulting b ∈ K̃ 0 (Xf ) is unique up to sign.
Now, suppose we had chosen another a0 ∈ K̃ 0 (Xf ) such that j(a0 ) = ã. Then by
exactness we have that a0 = a + ηb for some η ∈ Z, so that
a02 = a2 + 2ηab + η 2 b2 = hb + 2ηab,
by graded-commutativity and the fact that b2 vanishes in K̃ 0 (S 4r ). Further, we
note that j(ab) = j(a)ji(b) = 0, hence by exactness ab = νb for some ν ∈ Z.
Applying a to both sides gives that
0 = hb2 = a2 b = νab,
hence ν = 0 and a02 = hb, as needed.

We now claim that h = H(f ) corresponds to the Hopf invariant of f . To see this,
we have recourse to the Chern character introduced in Section 4.4. Recall that the
Chern character provides a natural transformation Ch : K̃ 0 (−) → H̃ ev (−; Q). We
may apply the Chern character to the SES in Equation 4.13, so that by naturality,
we obtain a commutative diagram:

0 / K̃ 0 (S 4r ) i / K̃ 0 (Xf ) j
/ K̃ 0 (S 2r ) /0

Ch Ch Ch
  
0 / H̃ ev (S 4r ; Q) i / H̃ ev (Xf ; Q) j
/ H̃ ev (S 2r ; Q) /0
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 39

Now, by Theorem 3.13 (iv), the Chern character is injective on even dimensional
spheres, and takes the generator of K̃ 0 (S 2r ) to the generator of H̃ 2r (S 2r ; Z) ⊂
H̃ ev (S 2r ; Q). Hence the 5-lemma applied to the factorization of the above dia-
gram through H̃ ev (−; Z) implies that the middle vertical map induces an isomor-
'
phism Ch : K̃ 0 (Xf ) − → H̃ ev (Xf ; Z). It follows by commutativity that Ch(b) ∈
H (Xf ; Z), Ch(a) ∈ H 2r (Xf ; Z) are valid generators, where b, a ∈ K̃ 0 (Xf ) are
4r

as above. Hence by definition of the Hopf invariant as in Construction 4.6, we

have that H(f ) satisfies the relation Ch(a)2 = H(f )Ch(b), and since the Chern
character is a ring homomorphism we get that
H(f )Ch(b) = Ch(a)2 = Ch(a2 ) = Ch(hb) = hCh(b),
whence it follows that h = H(f ), as claimed.

Therefore, if the map f has Hopf invariant one, we get that

ψ 2 (a) ≡ a2 = hb = H(f )b ≡ b (mod 2).
Next, if we let P ∈ Z2r,r be the integer determining the B-module structure of
K̃ 0 (Xf ) via ψ 2 (a) = 2r a + P b, it follows that
P b ≡ ψ 2 (a) ≡ b (mod 2),
hence P must be odd. Then, Equation 4.15 implies that for all odd primes l, we
must have:
l2r − lr lr lr − 1
2r r
P = r r P ∈ Z,
2 −2 2 2 −1
and thus necessarily
2r | lr − 1
for all odd primes l. Specializing to the case l = 3 suffices to complete the proof of
Theorem 4.5 using elementary modular arithmetic:
Lemma 4.18. Suppose r ∈ Z≥0 satisfies 2r | 3r − 1. Then r ∈ {0, 1, 2, 4}.
Proof. Write r = 2l m for m odd. We proceed by induction on l and show that the
highest power of 2 dividing 3r − 1 is 2l+2 for l ≥ 1, and 2 for l = 0. From there, we
see that 2r | 3r − 1 implies that
2r ≤ 2l+2 ≤ 4 · 2l m = 4r,
and hence that r ≤ 4. The result then follows by manual verification of the cases
r = 0, 1, 2, 3, 4.
So first suppose l = 0, so that r = m is odd. Note that 2 | 3 − 1, so 2 is a
valid candidate. Now, write m = 2k + 1 for some k. Then we find that 3m − 1 ≡
(32 )k · 3 − 1 ≡ 2 (mod 4), hence 22 cannot divide 3m − 1. Next, for l = 1, r = 2m,
so that 3r − 1 = (3m − 1)(3m + 1). The first factor is divisible by at most 2
by the above reasoning; next, we compute that, writing m = 2k + 1, we have
3m + 1 = (32 )k · 3 + 1 ≡ 4 (mod 8), so that the second factor is divisible by at most
22 , and therefore the highest power of 2 that can divide 3r − 1 is 23 , as needed. For
the general case, we have that
l+1 l l
32 m
− 1 = (32 m − 1)(32 m + 1).
The first factor is divisible by at most 2l+2 by induction hypothesis, and we compute
l l−1
that 32 m + 1 ≡ ((32 )2 )m + 1 ≡ 2 (mod 4), hence the second factor is divisible
40 SAAD SLAOUI

by at most 2, and therefore the highest power of 2 that can divide 3r − 1 is 2l+3 ,
completing the proof. 

Appendix A. Constructing Classifying Spaces at the Level of

Categories
The purpose of this appendix is to describe a remarkably simple conceptual
formulation of a model for the classifying space BG of principal G-bundles for any
topological group G, which we introduced in Chapter 1. The approach we adopt is
close in spirit to the bar construction, and we refer the interested reader to Milnor
[16] for a treatment via the join construction.
Throughout this exposition, we shall make use of simplicial sets as a bridge
allowing us to transfer data from the settings of small categories to the settings of
spaces. This translation process is allowed by the existence of functors fitting into
the diagram:

N |.|
+ +
(A.1) Cat k sSet k Top,
Π S

subject to the adjunctions:

Cat(ΠK, C ) ∼
= sSet(K, N C ),

Top(|K|, X) ∼
= sSet(K, S X).
In the above, N , Π, |.|, S refer to the nerve, fundamental category, geometric re-
alization and singular functors, respectively, A more in-depth discussion of these
adjunctions may be found in [14]. We shall only consider the forward direction, i.e.
going from categories to spaces under the composite |N |. However, we will occa-
sionally rely on the existence of these adjunction pairs to obtain desirable properties
of the functors under consideration.

A.1. Refresher on Simplicial Sets, Nerves and Geometric Realizations.

For completeness, we provide a streamlined treatment of the necessary notions
about simplicial sets. For building intuition about these objects, we refer the reader
to the introductory treatment by Friedman [10].
One impetus for considering simplicial sets is to provide a flexible yet tractable
combinatorial framework in which to do homotopy theory. A concise way to arrive
to this framework is by first defining the category ∆ whose objects are the totally
ordered finite sets [n] = {0, ..., n}, and whose morphisms are given by monotonic
increasing set maps µ : [n] → [m], i.e. those satisfying i < j =⇒ µ(i) ≤ µ(j). This
is a small category whose morphism set is generated by two families of monotonic
maps called the face and degeneracy maps, denoted respectively by δi : [n] → [n+1]
and σi : [n + 1] → [n], indexed over all n ≥ 0 and 0 ≤ i ≤ n. The ith face map in
degree n is the inclusion of [n] into [n + 1] which skips the element i ∈ [n + 1], and
the ith degeneracy map in degree n is the projection of [n + 1] onto [n] which sends
both elements i, (i + 1) ∈ [n + 1] to the element i ∈ [n] and subsequently sends each
j > i in [n + 1] to (j − 1) ∈ [n]. Further, these maps satisfy certain relations under
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 41

composition which we will only need in the formulation of some definitions:

δi δj = δj−1 δi , i < j;

(A.2) σj−1 δi , i < j

δi σj = id , i = j or i = (j + 1) .

σj δi−1 , (j + 1) < i


One can then show that every monotonically increasing map can be expressed as a
composite of face maps and degeneracy maps.

Now, we define a simplicial set K to be a contravariant functor from the category

∆ to the category of sets, i.e. K : ∆op → Set. On objects, such a functor
corresponds to a choice of set Kn = K[n] for each n ≥ 0, to which we refer as
the n-simplices of K. On morphisms, by the above discussion, it suffices to specify
associated set maps di : Kn+1 → Kn for each face map δi : [n] → [n + 1] and
si : Kn → Kn+1 for each degeneracy map σi : [n + 1] → [n] for all n ≥ 0, 0 ≤ i ≤ n,
in such a way that the relations in Equation A.2 are satisfied in the image. The
action of K on the remaining morphisms in ∆ is then determined by functoriality.
We call an n-simplex k ∈ Kn degenerate if it lies in the image of a degeneracy map;
otherwise, we say k is a non-degenerate n-simplex.
We may then consider the category sSet with simplicial sets as objects and
natural transformations between them as morphisms. Once again, it is enough to
require an assignment (fn : Kn → Ln )n≥0 to commute with face maps and degen-
eracy maps in order for it to be a valid natural transformation.

There is a systematic way to associate in a functorial way a simplicial set N C to

any small category C . Composing this map with the geometric realization functor
will allow us to associate a topological space to a given small category C .

Construction A.3. Let C be a small category. Define the nerve of C to be the

simplicial set whose 0-simplices are given by the set obC , and whose n-simplices
are given by n-tuples of composable morphisms in C . That is, to every chain
f1 f2 f2 fn
of morphisms a0 −→ a1 −→ a3 −→ ... −→ an in C , we associate an n-simplex
(f1 |...|fn ) ∈ N Cn . Next, define the face maps di : N Cn → N Cn−1 and degeneracy
maps si : N Cn → N Cn+1 as follows:
si : (f1 |...|fn ) 7→ (f1 |...|fi |id|fi+1 |...|fn ),

(f2 |...|fn ) , i = 0

di : (f1 |...|fn ) 7→ (f1 |...|fi+1 fi |...|fn ) , 1 ≤ i ≤ (n − 1)

(f1 |...|fn−1 ) , i = n.


In the above definition for si , id refers to the identity morphism on domain of fi+1
(or equivalently the codomain of fi ). It can be verified that these maps satisfy
the relations in Equation A.2, and hence that they generate a valid simplicial set
structure on N C .
Further, given a functor F : C → D between categories, we may obtain an
induced simplicial map N F : N C → N D given in degree n by the assignment:
N Fn : (f1 |...|fn ) 7→ (F (f1 )|...|F (fn )).
42 SAAD SLAOUI

One may check that functoriality of F makes the map N F commute with face and
degeneracy maps, so that N F defines a valid simplicial map. It readily appears
that the resulting assignment N : Cat → sSet at the level of categories is in fact
a covariant functor, as desired.
Remark A.4. Specializing to a one object groupoid category (as we shall do in the
next section), the simplicial set structure of the corresponding nerve corresponds
to what is known as the bar construction, a closely related approach to achieving
the construction of classifying spaces for principal bundles.
Next, we come to the second step indicated in the diagram in Equation A.1.
Namely, we shall specify a way to functorially map simplicial sets to topological
spaces. The construction is very similar to the analogous geometric realization
functor for simplicial complexes, with which the reader may already be familiar.
Construction A.5. For each n ≥ 0, denote by ∆tn the standard topological n-
simplex, i.e. the subset of Rn+1 characterized as the convex hull of the standard
orthonormal basis in Rn+1 . That is,
n
X
∆tn := {(t0 , ..., tn ) ∈ Rn+1 | ti ≥ 0, ti = 1}.
i=0

Together, these objects fit into a covariant functor ∆t∗ : ∆ → Top taking the ith
face map δi : [n] → [n + 1] to the map δi : ∆tn → ∆tn+1 given by δi : (t0 , ..., tn ) 7→
(t0 , ..., ti−1 , 0, ti , ..., tn ), resp. the ith degeneracy map σi : [n] → [n − 1] to the map
σi : ∆tn → ∆tn−1 given by σi : (t0 , ..., tn ) 7→ (t0 , ..., ti−1 , ti + ti+1 , ti+1 , ..., tn ). Now,
let K be a simplicial set, with face and degeneracy maps denoted by di , resp. si .
Viewing each Kn as a topological space endowed with the discrete topology, we
define the geometric realization of K to be the space |K| given by:
a
|K| := Kn × ∆tn /(∼),
n≥0

where (di k, v) ∼ (k, δi v) and (si k, v) ∼ (k, σi v) wherever it makes sense. In par-
ticular, thinking of the discrete set Kn as an indexing set for as many copies of
the topological n-simplex ∆tn as there are elements in Kn , this equivalence rela-
tion ensures that every equivalence class in T K contains a unique point lying in
the interior of a topological n-simplex indexed by a non-degenerate element of Kn .
Further, it can be verified that a simplicial map f : K → L induces a continuous
map |f | : |K| → |L| under the assignment (|f |)[(k, x)] := [f (k), x]. This assignment
turns geometric realization into a covariant functor |.| : sSet → Top.
A.2. Construction at the Level of Categories. We introduce the functor be-
tween categories which we claim will correspond to a universal principal G-bundle
after application of the composite functor |N | : Cat → Top. The striking simplic-
ity of the construction at this level will hopefully help motivate this rather abstract
approach to obtaining classifying spaces.

Given a group G, we may consider the associated one object groupoid category.
This is the category G with a unique object {∗} and unique morphism set the
underlying set of G, with composition law given by the group structure of G and
id
identity morphism the arrow ∗ −→ ∗ corresponding to the identity element of G.
Next, define a small category E as follows: E has the underlying elements of G as
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 43

objects, and for each pair of objects g, h in E , the hom set E (g, h) is a one-point
set. Observe that by uniqueness of the morphisms in E , the data of an n-simplex
(g1 → g2 → ... → gn → gn+1 ) in the associated nerve N E may be captured by an
(n + 1)-tuple (g1 |...|gn+1 ), where the components denote elements of the underly-
ing category, rather than the conventional n-tuple of morphisms introduced in the
previous section.

Now, define a functor F : E → G acting on objects of E as the one point

projection to the unique object ∗ in G, and sending the unique morphism in E (g, h)
to the morphism hg −1 in G(∗, ∗) for each pair of objects g, h in E . Then, the
composite functor |N | : Cat → Top takes F to a continuous map between spaces

|N F | : |N E | → |N G|.

The remainder of these notes will be devoted to studying this map and the inter-
mediary simplicial map N F : N E → N G, with the intention to show that |N F | is
a universal principal G-bundle and hence that |N G| is a valid model for BG.

A.3. Construction at the Level of Simplicial Sets. The nerve functor N :

Cat → sSet assigns to the functor F : E → G a simplicial map N F : N E →
N G. The behavior of this map on n-simplices is given component-wise by N F :
(g1 |...|gn+1 ) 7→ (g2 g1−1 |...|gn+1 gn−1 ). Now, we may ask about the fiber under N F
of a given n-simplex (g1 |...|gn ) in N G. Since N F is a simplicial map, any element
in the pre-image of (g1 |...|gn ) must lie in N En . So suppose (x1 |...|xn+1 ) ∈ N En
satisfies
(x2 x−1 −1
1 |...|xn+1 xn ) = (g1 |...|gn+1 ).

Then we get equations x2 = g1 x1 , x3 = g2 g1 x1 , ..., xn+1 = gn ...g1 x1 which fully

determine the entries x2 , ..., xn+1 once x1 has been given. Now, notice that any
value h ∈ G for x1 ∈ obE = G determines a distinct element (h|g1 h|...|gn ...g1 h) in
the fiber of (g1 |...|gn ) under N F . Thus the fiber of (g1 |...|gn ) looks precisely like
a copy of G. Further, any such fiber can be given a free right G-action under the
assignment:
(N F )−1 (g1 |...|gn ) × G → (N F )−1 (g1 |...|gn )
((h|g1 h|...|gn ...g1 h), k) 7→ (hk|g1 (hk)|...|gn ...g1 (hk)).

We highlight one more restriction on elements in the preimage of a non-degenerate

n-simplex. This will be helpful later in studying the fibers at the level of geometric
realizations:

Lemma A.6. The simplicial map N F takes non-degenerate simplices to non-

degenerate simplices.

Proof. Suppose an n-simplex in the image of N F is degenerate. Then without loss

e
of generality, we may assume it contains an identity morphism ∗ − → ∗ at the ith
position and is the image of the n-simplex (g1 |...|gn+1 ) in N E . Hence, we get that
gi+1 gi−1 = e, i.e. gi = gi+1 , so that the ith morphism in (g1 |...|gn+1 ) corresponds
id
to the unique identity morphism gi −→ gi , whence (g1 |...|gn+1 ) was degenerate to
begin with. 
44 SAAD SLAOUI

A.4. From Adjunctions to Homotopy Equivalences. We now come to a use-

ful feature of the passage from small categories to spaces via the composite functor
|N | : Cat → Top, namely that natural transformations between functors in Cat
induce homotopic maps at the level of spaces, and that as a result adjoint pairs of
functors between categories C and D induce homotopy equivalences between the
geometric realizations of the corresponding nerves |N C | ' |N D|.

To speak of a homotopy between geometric realizations of nerves of categories,

it is convenient to formulate an analog of the unit interval at the level of categories.
Denote by I the category with two objects [0], [1] and a unique non-identity mor-
phism α : [0] → [1]. We claim that |N I| is homeomorphic to the unit interval
I ⊂ R. Indeed, all but one 1-simplex `(α) and two 0-simplices [0], [1] are degenerate,
hence upon taking the quotient of n N In × ∆tn by face and degeneracy maps we
are left precisely with a copy of ∆t1 ∼
= I. We proceed to prove:
Proposition A.7. Let F, G : C → D be two functors related by a natural trans-
formation η : F → G. Then there exists a homotopy h : |N C | × I → |N D| between
the corresponding maps |N F | and |N G|.
Proof. Let I be the category with two objects and one non-identity morphism as
above. Consider the product category C × I. Then C × I can be thought of as two
disjoint copies of C linked to each other by morphisms of the form (f, α), where α is
the unique non-trivial morphism of I and f is some morphism in C . We may then
define a functor h̃ : C ×I → D by letting h̃|C ×{[0]} := F , h|C ×{[1]} := G, and setting
h̃(idx ×α) := ηx for each x ∈ obC . Finally, we extend functorially the action of h̃ on
composite morphisms of the form (f ×α) = (f ×id[1] )(idx ×α) = (idy ×α)(f ×id[0] )
for each f : x → y in C .
From functoriality of F and G and commutativity of the induced diagrams:

f ×id[0] h̃(f ×id[0] )

(x, [0]) / (y, [0]) 7−→ F (x) / F (y)

idx ×α idy ×α ηx ηy
 (   ' 
(x, [1]) / (y, [1]) G(x) / G(y),
f ×id[1] h̃(f ×id[1] )

provided by naturality of η, it follows that the resulting assignment h̃ : C × I → D

is a well-defined functor.
Now, we have that |N (C × I)| ∼ = |N C | × |N I| ∼
= |N C | × I, so that at the level
of spaces h̃ induces a map h : |N C | × I → |N D| satisfying h||N C |×{0} = |N F | and
h||N C |×{1} = |N G|, thereby providing the desired homotopy. 

Corollary A.8. Suppose there exists an adjoint pair of functors C o

F / D . Then
G
the spaces |N C | and |N D| are homotopy equivalent.
Proof. One way to express this adjunction is in the form of two natural transforma-
tions η : GF → IdD , γ : F G → IdC , where IdC , IdD denote the identity functors
on the respective categories. Hence, assuming that natural transformations be-
tween functors provide homotopies between the associated continuous maps, and
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 45

using functoriality of the assignment |.| ◦ N : Cat → Top, we get homotopies:

'
|N G| ◦ |N F | −
→ id|N C | ,
'
|N F | ◦ |N G| −
→ id|N D| .
This corresponds precisely to the data of a homotopy equivalence between the
spaces |N C | and |N D|. 

In particular, we get the following result “for free”:

Corollary A.9. Let C be a category with either an initial object or a terminal
object. Then |N C | is a contractible space.
Proof. Assume C has an initial object 1 ∈ obC (a dual argument applies when C
has a terminal object). For all x ∈ obC , we have that C (1, x) is a one-point set.
Now, let 1 denote the category with one object and one morphism (say with unique
object z), and consider the functors F : C → 1, l : 1 → C , where F projects all
objects of C to z and all morphisms to idz , and l is the inclusion functor sending
z to 1 ∈ obC (and idz to id1 ). Since 1(z, F (x)) = 1(z, z) is a one point set for all
objects x of C , we get a (trivially natural) bijection
1(z, F (x)) ∼
= C (l(z), x)
for all x ∈ obC .
Thus l and F form a pair of adjoint functors, and it follows that |N C | is ho-
motopy equivalent to |N 1|, which is readily seen to be the one point space. Hence
|N C | is contractible. 

Coming back to our original context, the category E under consideration has
the property that every hom set is a one point set, hence every object of E is both
initial and final. It follows from the above that |N E | is a contractible space.

A.5. Construction at the Level of Spaces. So far, we have seen that the functor
F : E → G described in section 3 induces a continuous map |N F | : |N E | → |N G|
whose domain is contractible. Further, this map is obtained by application of the
geometric realization functor from a simplicial map N F : N E → N G with the
property that each fiber is a free G-space. It remains to exploit the latter fact to
verify that the map |N F | is indeed a principal G-bundle, after which the sought-
after result will follow from Theorem 1.9.

Given an element [x] of |N G|, find the unique representative x which lies in
the interior of a non-degenerate topological n-simplex, labeled by some element
k ∈ N Gn . Because N F takes non-degenerate simplices to non-degenerate sim-
plices by Lemma A.6 and restricts to a homeomorphism on the interior of topo-
logical n-simplices, every element in the preimage of x under |N F | must also lie
in the interior of a non-degenerate n-simplex. The distinct n-simplices forming a
copy of G in the preimage of k under N F then pick out as many distinct points
in the corresponding topological n-simplices mapping down to x, on which G acts
freely on the right. Hence each fiber under the map |N F | appears to be a free
right G-space. The verification of local triviality is more delicate, and we refer to
Theorem 8.2 of May [15] for a proof.
46 SAAD SLAOUI

Finally, we show that if G is an abelian group, then |N G| can be given the struc-
ture of a topological group. Rather than exhibiting a topological group structure
at the level of spaces, we resort to the following observation:
Lemma A.10. Let C, D be categories with finite products and terminal objects,
and let F : C → D be a functor which commutes with taking finite products and
terminal objects. Then F takes group objects in C to group objects in D.
The verification is just a matter of running through the definitions. Now, it
can be shown that the geometric realization functor satisfies the hypothesis of the
above lemma (see for instance [14] for details). Thus, if we can equip N G with the
structure of a group object at the level of simplicial sets, it will follow that |N G| is
a group object in Top, i.e. that it is a topological group.
Construction A.11. Let N G denote the nerve of the one object groupoid category
associated to the abelian group G, as before. Define maps
µ : NG × NG → NG
i : NG → NG
given on each degree n by the assignments:
µn : ((g1 |...|gn+1 ), (h1 |...|hn+1 )) 7→ (g1 h1 |...|gn+1 hn+1 )
in : (g1 |...|gn+1 ) 7→ (g1−1 |...|gn+1
−1
).
We claim that µ and i are valid simplicial maps. It suffices to show that they both
commute with faces and degeneracies in every degree. In both cases, commutativity
with degeneracies is immediate. However, the fact that G is abelian is essential in
showing that either map commutes with face maps. For instance, for a given n ≥ 2
and a given 1 ≤ i ≤ n − 1, we get that
µn di ((g1 |...|gn+1 ), (h1 |...|hn+1 )) = µn ((g1 |...|gi+1 gi |...|gn+1 ), (h1 |...|hi+1 hi |...|hn+1 ))
= (g1 h1 |...|gi+1 gi hi+1 hi |...|gn hn ),
while
di µn ((g1 |...|gn+1 ), (h1 |...|hn+1 )) = di ((g1 h1 |...|gn+1 hn+1 ))
= (g1 h1 |...|gi+1 hi+1 gi hi |...|gn hn ),
so that since G is abelian the two equations coincide. A similar argument applies
to showing that the map i : N G → N G commutes with face maps. It now follows
immediately from the underlying group structure of G that µ and i endow N G
with the structure of a group object in sSet, as needed.
Remark A.12. The fact that G is abelian is essential to obtaining a topological
group structure on BG = |N G|. Indeed, we saw in Section 1.2 that π1 (BG) = G.
Further, the fundamental group functor π1 : Top → Grp preserves finite products
and final objects, hence it takes topological groups to group objects in the category
Grp by Lemma A.10. But group objects in Grp are precisely abelian groups,
as a result of requiring the multiplication map µ : G × G → G to be a group
homomorphism. Thus we see that BG being a topological group implies that
G = π1 (BG) is an abelian group.
Putting everything together, we have proven the following theorem:
 TOPOLOGICAL K-THEORY AND SOME OF ITS APPLICATIONS 47

Theorem A.13. Let G be a discrete group. Then a model for the classifying space
BG is provided by the space |N G|, together with the map:
|N F | : |N E | → |N G|,
where G is the one object groupoid category with hom set G and composition induced
by the group structure of G, E is the category with obE = G and hom sets E (g, h) =
{∗} for all g, h ∈ G, and F : E → G is the functor given by the one point projection
at the level of objects and the assignment (g → h) 7→ hg −1 at the level of morphisms.
Further, the classifying space BG can be given the structure of a topological group
if and only if G is abelian.

References
[1] Adams, J. F. On the Non-Existence of Elements of Hopf Invariant One. Annals of Mathematics.
Vol. 72. 1960. 20-104. (On p. 35)
[2] Adams, J. F. On the Groups J(X) I-IV. Topology 2 (1963), 181-195; Topology 3 (1965),
137-171; Topology 3 (1965), 193-222; Topology 5 (1966), 21-71. (On p. 3)
[3] Adams, J. F. and Atiyah, M. F. K-Theory and the Hopf Invariant. Quarterly Journal of
Mathematics. Oxford. Vol. 17. 1966. 31-38. (On p. 35)
[4] Adams, J. F. Vector Fields on Spheres. Annals of Mathematics. Vol. 75. 1961. p. 603-632. (On
p. 29)
[5] Atiyah, M. F. Anderson D. W. K-Theory. Advanced Book Classics. 1964. (On p. 4)
[6] Bott, Raoul. An Application of the Morse Theory to the Topology of Lie Groups. Bulletin de
la Socit Mathmatique de France, Volume 84 (1956), p. 251-281. (On p. 18)
[7] Brown, Edgar. Cohomology Theories. Annals of Mathematics. Vol. 75. 1962. 467-484. (On
pp. 2 and 7)
[8] Cohen, Ralph. The Topology of Fibre Bundles (Lecture Notes). AMS Open Math Notes.
Available for download at ams.org/open-math-notes/omn-view-listing?listingId=110706 (On
p. 16)
[9] Dugger, Daniel. A Geometric Introduction to K-Theory. Available to download at
pages.uoregon.edu/ddugger/kgeom.pdf (On pp. 3, 13, 16, 30, 35, and 38)
[10] Friedman, Greg. An Elementary Illustrated Introduction to Simplicial Sets. Available for
download at arxiv.org/pdf/0809.4221.pdf (On p. 40)
[11] Hatcher, Allen. Vector Bundles and K-Theory. Available for download at
pi.math.cornell.edu/˜ hatcher/VBKT/VB.pdf (On pp. 4, 6, 17, 21, and 31)
[12] Kochman, Stanley O. Bordism, Stable Homotopy and Adams Spectral Sequence. Fields In-
stitute Monographs, American Mathematical Society. 1996. (On p. 7)
[13] May, Peter. A Concise Course in Algebraic Topology. Available for download at
math.uchicago.edu/˜ may/CONCISE/ConciseRevised.pdf (On pp. 13, 19, and 20)
[14] May, Peter. Finite Spaces and Larger Contexts. Available for download at
math.uchicago.edu/˜ may/REU2018/FINITEBOOK.pdf (On pp. 40 and 46)
[15] May, Peter. Classifying Spaces and Fibrations. Memoirs of the American Mathematical So-
ciety. 1975. (On p. 45)
[16] Milnor, J. W. Construction of Universal Bundles II. Annals of Mathematics. Vol. 63. No 3.
1955. (On pp. 10 and 40)
[17] Milnor, J. W. Stasheff, James D. Characteristic Classes. Princeton University Press and
University of Tokyo Press. 1974. (Not cited.)
[18] Mitchell, Stephen A. Notes on Principal Bundles and Classifying Spaces. Available for down-
load at nd.edu/˜ mbehren1/18.906/prin.pdf (On pp. 4 and 8)
[19] Morava, Jack. Complex Cobordism and Algebraic Topology. Available for download at
arxiv.org/pdf/0707.3216.pdf (On p. 3)
[20] Ravenel, Douglas. The Grothendieck Group K0 . Available for download at
web.math.rochester.edu/people/faculty/doug/otherpapers/Kbook.II.pdf (On p. 16)
[21] Slaoui, Saad. Building Up to the Pontryagin-Thom Theorem and Computation of π∗ (M O).
Available for download at http://math.uchicago.edu/ may/REU2018/REUPapers/Slaoui.pdf
(On p. 3)
48 SAAD SLAOUI

[22] Steenrod, N. E. Cohomology Invariants of Mappings. Annals of Mathematics. Vol. 50. No 4.

1949. (On p. 31)
[23] Wirthmüller, Klaus. Vector Bundles and K-Theory. Skeleton Lecture Notes 2011/2012 (On
p. 22)