An Introduction to Complex K-Theory

May 23, 2010

Jesse Wolfson

Abstract
Complex K-Theory is an extraordinary cohomology theory defined
from the complex vector bundles on a space. This essay aims to provide
a quick and accessible introduction to K-theory, including how to cal-
culate with it, and some of its additional features such as characteristic
classes, the Thom isomorphism and Gysin maps.

1
Contents
1 Generalized Cohomology Theories 3
1.1 K-Theory Take 1 . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 A Quick Note on K-classes . . . . . . . . . . . . . . . . . . . . 5
1.3 K-Theory Take 2 . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Computational Tools 8
2.1 A Quick Recap of Spectral Sequences . . . . . . . . . . . . . . 8
2.2 The Atiyah-Hirzebruch Spectral Sequence . . . . . . . . . . . 9
2.3 The Chern Character . . . . . . . . . . . . . . . . . . . . . . . 10

3 Sample Computations 10

4 Theoretical Tools 10
4.1 Splitting Principles . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Adams Operations . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3 The Thom Isomorphism and Applications . . . . . . . . . . . 13
4.3.1 Orientability in K-theory . . . . . . . . . . . . . . . . 13
4.3.2 The Isomorphism . . . . . . . . . . . . . . . . . . . . . 14
4.3.3 Gysin Maps . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3.4 Thom Classes and Characteristic Classes . . . . . . . 15
4.4 Constructing the Chern Character . . . . . . . . . . . . . . . 16

5 Hirzebruch-Riemann-Roch 17
5.1 The Todd Class . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 The Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Riemann-Roch for Curves . . . . . . . . . . . . . . . . . . . . 19

6 Conclusion 20

2
1 Generalized Cohomology Theories
We begin with the definition of ordinary cohomology due to Eilenberg and
Steenrod:
Def. 1. An ordinary cohomology theory is a collection {H i }i2Z such that:
• For each n 2 Z, H n is a contravariant functor from the category of
pairs of spaces to abelian groups.1

• (Homotopy Invariance) If f ' g through maps of pairs, then H n f =

H n g for all n.
` Q
• (Preserves Products) H n ( X↵ ) = H n (X↵ ) for all n.

• (LES of the Pair) For each pair (X, A), there exists a long exact se-
quence

· · · ! H i (X, A) ! H i (X) ! H i (A) ! H i+1 (X, A) ! · · ·

such that the boundary map is natural.

• (Excision) If Z ⇢ A ⇢ X and Z ⇢ Int(A) then the induced map

H i (X, A) ! H i (X Z, A Z)

is an isomorphism for each i.

• (Dimension) H i (⇤) = 0 for i 6= 0.

An extraordinary cohomology theory satifies all of the above except the
dimension axiom. Complex K-theory was one of the first extraordinary
cohomology theories to be discovered and studied in depth. My aim here
is to present it as such and develop some of the key structures of K-theory
as a cohomology theory. Whenever going through the gory details would
obscure this development, I’ll refrain and refer interested readers to other
sources instead.

1.1 K-Theory Take 1

As a disclaimer, assume all spaces X are compact and Hausdor↵.
Def. 2. Given a space X, let V ectC (X) denote the semiring of isomorphism
classes of (finite dimensional) complex vector bundles over X with addition
given by and multiplication by ⌦.
Def. 3. We define K 0 (X) to be the group completion of V ectC (X).
1
The assignment X 7! (X, ;) makes this into a functor on spaces as well, and this is
what is meant by H i (X).

3
Example 1. All vector bundles over a point are trivial, so V ectC (⇤) = N
and K 0 (⇤) = Z.

Let X⇤ denote a space with basepoint ⇤. For any space X, let X +

denote the union of X with a disjoint basepoint. Let S n (X⇤ ) denote the
n-fold reduced suspension of X⇤ . With this notation, we define the negative
K-groups as follows:

Def. 4. Letting i : ⇤ ! X⇤ denote the inclusion of basepoint, we define

e 0 (X⇤ ) := ker(i⇤ : K(X⇤ ) ! K(⇤))
K
e 0 (X + ). Now,
This is called the reduced K-group. Observe that K 0 (X) = K
for n 2 N, let,
e
K e 0 (S n (X⇤ ))
(X⇤ ) := K
n

K n (X) := K e 0 (S n (X + ))
e 0 (S n (X/Y ))
K n (X, Y ) := K

To extend our definition of the K-groups to the positive integers, we use

Bott Periodicity.

Theorem 1. (Bott Periodicity v. 1) Let [H] denote the class of the canoni-
cal bundle in K 0 (CP1 ). Then, identifying CP1 with S 2 , and letting ⇤ denote
the reduced exterior product, the map
e 0 (X⇤ ) ! K
K e 0 (S 2 (X⇤ ))
[E] 7! ([H] 1) ⇤ [E]

is an isomorphism for all compact, Hausdor↵ spaces X. We call [H] 1 the

Bott class.

Periodicity allows us to define the positive K-groups inductively, setting

K n ( ) := K n 2 ( ), and similarly for the reduced groups.
In order to verify that K-theory gives a cohomology theory, we need two
last facts:

Prop. 2. If X is compact and Hausdor↵, and E is any vector bundle on

Y , then a homotopy of maps f ' g : X ! Y induces an isomorphism of
bundles f ⇤ E ⇠
= g ⇤ E.2
Prop. 3. To every pair of compact, Hausdor↵ spaces (X, Y ), there exists
an infinite exact sequence
... / K n (X, Y ) / K n (X) / K n (Y ) / K n+1 (X, Y ) / ...

2
The proof is an application of the Tietze extension theorem formulated for vector
bundles. c.f. Atiyah [2] L.1.4.3.

4
 which is natural in the usual sense.3

Now, checking our definitions against the axioms:

• Pullback of bundles makes the K-groups into contravariant functors

so Axiom 1 is satisfied.

• Homotopy invariance follows from the proposition we just stated.

• For products, a quick check shows that the map

a Y
V ectC ( X↵ ) ! V ectC (X↵ )

given by pullback along the inclusions is an isomorphism, and that

this isomorphism is preserved under group completion.

• The LES of the pair was given above.

• Excision is satisfied because X/Y ⇠

= (X Z)/(Y Z) for any Z ⇢
Y ⇢ X.

so we see that K-theory is indeed a cohomology theory.

1.2 A Quick Note on K-classes

From the definitions we’ve given, every K-class is an element of K 0 (X) for
some compact space X. We can say more than this:

1. Two vector bundles E and F define the same K-class if there exists a
trivial bundle ✏n such that E ✏n ⇠ = F ✏n . This is known as stable
isomorphism, so we see K 0 (X) is the group completion of the semiring
of vector bundles modulo stable isomorphism.

2. Every K-class can be written as [H] [✏n ] for some vector bundle H
over X.
e 0 (X) if and only if
3. A vector bundle E is in the kernel of K 0 (X) ! K
it is stably isomorphic to a trivial bundle.

The upshot of this is that when we want to make arguments in K-theory,

we can actually make arguments using vector bundles and then check that
these arguments behave well when we pass to K-classes. This is one of the
main techniques for making constructions in K-theory.
These conclusions follow from two facts:
3
The proof of this requires the most work, after Bott periodicity, in setting up K-
theory as a cohomology theory. Both Atiyah [2] (P.2.4.4) and Hatcher[7] provide a detailed
construction, but I recommend just taking this as a given when first getting a handle on
K-theory, and coming back to the details later.

5
Prop. 4. Every vector bundle on a compact space is a direct summand of
a trivial bundle.
This follows from a partition of unity argument, and the finiteness of the
cover; in particular, this can fail for paracompact spaces. See Hatcher [7]
P.1.4.
Prop. 5. Given a commutative monoid A, with group completion K(A),
K(A) ⇠
= A ⇥ A/ (A) and x 7! (x, 0) gives the canonical map A ! K(A).
Since K(A) is defined by a universal property (that it’s a left adjoint to
the forgetful functor from groups to monoids), it’s sufficient (and straight-
forward) to check that A ! A ⇥ A/ (A) satisfies the universal property.
Putting these together, we see every K-class is of the form [E] [F ] for
two bundles E and F . 1 follows because [E] = [F ] , 9 G s.t. E G ⇠ = F G.
Given such a G, let G0 be a bundle such that G G0 ⇠ = ✏n for some n. Then
[E] = [F ] , E G G0 ⇠
=F G G0
i.e. E ✏n ⇠ = F ✏n . The proofs of 2 and 3 are similarly straightforward
applications of the two propositions above.

1.3 K-Theory Take 2

We can give another characterization of K-theory that is frequently use-
ful, and which illuminates the definitions. Recall that a spectrum is a se-
quence of spaces (CW-complexes) {E(n)} and connecting maps fn : E(n) !
⌦E(n+1). A loop spectrum is one where the connecting maps are homotopy
equivalences.
Theorem 6. (Brown Representability) Every reduced cohomology theory
h̃ on the category of pointed CW complexes has a representing loop spec-
trum {H(n)}, unique up to homotopy, such that h̃n (X) = [X, H(n)]⇤ (where
[ , ]⇤ denotes based maps up to homotopy).4
Since we recover an unreduced theory h by adding in a disjoint basepoint,
i.e. h⇤ (X) := h̃⇤ (X + ), we see Brown also says that unreduced cohomology
theories correspond to unbased maps up to homotopy.
We can start to identify the spectrum KU of complex K-theory using
the following:
Prop. 7. For any compact, Hausdor↵ space X⇤ ,
e 0 (S n (X⇤ ) ⇠
K = [X⇤ , U]
where the unitary group U := lim
! U (n).
5

4
For a fuller discussion and proof of Brown Representability, see Hatcher [6] § 4.E.
5
The proof follows from considering a clutching argument and that if X is compact,
[X,-] preserves filtered colimits (e.g. direct limits).

6
 This together with periodicity and the suspension-loop adjunction shows
that
e 0 (X⇤ ) = K
K e 2 (X⇤ )
=Ke 0 (S 2 (X⇤ ))
= [S(X⇤ ), U]
= [X⇤ , ⌦U]

Thus, we can give a homotopy theoretic definition of complex K-theory

as
e
K n
(X⇤ ) = [X⇤ , ⌦n+1 U]
and Brown Representability, plus periodicity, again shows that this gives a
cohomology theory.
In particular, periodicity can be restated as

Theorem 8. (Bott Periodicity v.2) ⌦U ' BU ⇥ Z, and since G ' ⌦BG

for any topological group G, we see ⌦2 U ' U. Moreover, For all n 2 N,

⇡2n+1 (U) = Z
⇡2n (U) = 0

This is in fact the original form in which Bott proved it.6

We can view this version as a calculation of the reduced K-groups of
spheres. Passing to unreduced, we see that the values of K-theory for spheres
are:

K 0 (S 2n ) = Z Z
K (S ) = 0
1 2n

and

K 0 (S 2n+1 ) = Z
K 1 (S 2n+1 ) = Z

Since the spectrum of a cohomology theory is only specified up to homo-

topy, it’s possible to give several equivalent spectra, each of which can shed
light on the theory.
For example we can interpret K-theory in terms of operators on a sepa-
rable infinite dimensional Hilbert space H. Recall that a bounded operator
6
Both homotopy equivalences have concrete implementations which Bott was able to
formulate and study using Morse theory. The interested reader should see Milnor [11] for
the full proof. Alternatively, both Hatcher [7] and Atiyah [2] provide proofs of version 1
in terms of bundle constructions.

7
on a Hilbert space is Fredholm if it has a closed image, and its kernel and
cokernel are finite. To each operator T , we can assign an index

Index(T ) := dim ker(T ) dim coker(T )

It turns out that this is the restriction to a point of an isomorphism

index : [X, F red(H)] ! K 0 (X)

Appendix A to Atiyah [2] spells this out in detail. Note that this interpreta-
tion of the spectrum K provides a link between K-theory and index theory
of elliptic operators. Other representing spectra also exist and they illumi-
nate deep connections between complex K-theory and areas of interest to
mathematical physics and analysis,7 but this is beyond my scope right now.

2 Computational Tools
As with any cohomology theory, we have the usual computational tools of
Mayer-Vietoris sequences, and the LES of the pair. However, these are of-
ten not very useful in K-theory, because periodicity means we rarely have
enough zero entries to reduce the long exact sequences to a series of iso-
morphisms. However, K-theory, and in fact any extraordinary cohomology
theory, comes with two additional tools which relate its values to those of
ordinary cohomology. These are the Atiyah-Hirzebruch spectral sequence,
which is a special instance of a generalized Serre spectral sequence, and the
Chern character, which relates K-theory to rational cohomology.

2.1 A Quick Recap of Spectral Sequences

Spectral sequences can seem quite daunting at first, at least they did to me.8
However, once you get comfortable using them, they open up an incredible
number of calculations which look nearly impossible without them. Recall
that a spectral sequence is an infinite sequence of “pages” consisting of a
grid of groups and of di↵erentials between them. We write Erp,q for the
(p,q)th entry of the rth page, and dr : Erp,q ! Erp+r,q r+1 for the di↵erential.
You pass from one page to the next by taking the homology with respect
to the di↵erentials, and cooking up a new set of di↵erentials from the data
of the previous ones. A spectral sequence converges if for for each (p,q),
Erp,q = Er+1
p,q
for r p,q
0; we write E1 for this stable group.
7
For example, see Atiyah-Bott-Shapiro [3].
8
If you haven’t worked with them before, I highly recommend Ch.1 of Hatcher [8],
available online. I recommend taking the construction of spectral sequences as a given, and
focusing first on how use them. Hatcher has a great set of examples, sample computations
and exercises and I really enjoyed this.

8
 In a wonderful variety of situations, we can cook up a spectral sequence
such that E2 page starts with something that we know, and the E1 page is
closely related to something we want to calculate, e.g. the cohomology ring
of a space X. As a shorthand, we say the spectral sequence converges to the
thing we want, and we write something like

E2p,q ) hp+q (X)

However, this notation is shorthand and should not be taken literally. Spec-
tral sequences do not converge to the groups written on the right hand side
of the arrow, they converge instead to the associated graded objects of a
filtraton of these groups. Whether or not we can recover the groups we care
about depends on an extension problem, and is often nontrivial.
Alright, with these disclaimers in place, let’s lay out the main tools:

2.2 The Atiyah-Hirzebruch Spectral Sequence

Recall that given a fibration F ! E ! B, we have the Serre spectral
sequence:
H p (B, H q (F )) ) H p+q (E)
In fact, the proof and construction carry over to any cohomology theory h
giving us a generalized Serre spectral sequence:

H p (B, hq (F )) ) hp+q (E)

Taking the trivial fibration id : X ! X, we get the Atiyah-Hirzebruch

spectral sequence
H p (X, hq (⇤)) ) hp+q (X)
This spectral sequence should be seen as reiterating for generalized coho-
mology what we already know from ordinary cohomology: namely, that
cohomology theories are largely determined by their values on the point.9
As I’ll show in a moment, this spectral sequence, along with the generalized
Serre SS for K-theory, provides one of the main tools for computing K ⇤ (X).
Note also, that the generalized Serre SS for K-theory allows us to prove
a K-theory version of Kunneth (as the same proof in ordinary cohomology
using the Serre SS carries over here), and its formulation is precisely the one
we’re used to.
9
More precisely, two cohomology theories h and h0 are equivalent if there exists a
natural transformation ↵ : h ! h0 such that ↵ induces an isomorphism h(⇤) ⇠
= h0 (⇤). c.f.
Adams [1].

9
2.3 The Chern Character
The Chern Character is a ring homomorphism

ch : K ⇤ ( ) ! H ⇤ ( ; Q) ⌦ K ⇤ (⇤)

induces an isomorphism K ⇤ ( ) ⌦ Q ⇠
= H ⇤ ( ; Q) ⌦ K ⇤ (⇤).10 In other words,
for any space X,
M
K 0 (X) ⌦ Q ⇠
= H n (X; Q) ⌦ K n (⇤)
n2N
M
⇠
= H 2n (X; Q)
n2N

and similarly, M
K 1 (X) ⌦ Q ⇠
= H 2n+1 (X; Q)
n2N

3 Sample Computations
The following set of spaces are suggestions for good examples to apply these
tools to calculate the K-theory of spaces.

• Riemann Surfaces

• CP n

• SO(3)

• O(4)

The first two can be computed immediately from the Atiyah-Hirzebruch

SS. For SO(3), you’ll need to combine the isomorphism from the Chern
character with the Atiyah-Hirzebruch SS. You can use this to calculate the
K-theory of O(3) and then use this plus Kunneth to calculate the K-theory
of O(4).

4 Theoretical Tools
As we observed above, since every K-class can be represented as formal dif-
ference of actual bundles, we can usually make our arguments for K-theory
in terms of actual bundles, and then observe that these arguments behave
well when we pass to K-classes. Frequently these constructions involve re-
ducing the structure group, for example, in decomposing a bundle as sum of
10
Such an isomorphism is actually a general property of cohomology theories, and the
map h ! H ⇤ ( ; Q) ⌦ K ⇤ (⇤) is called the character of the theory.

10
line bundles, or reducing its dimension by one. I sketch the general process
in the next section; we will use it frequently. Following this, the theoretical
tools discussed are:

• Adams operations

• The Thom Isomorphism and Applications

• A construction of the Chern character

These are largely independent, and you should feel free to tackle them
in any order.

4.1 Splitting Principles

In the most general case, a splitting principle refers to an operation in which
a G-bundle E ! X is pulled-back along a map f : F ! X, such that f ⇤ E
has a reduced structure group, and the map induced by f on cohomology is
injective.
The generic splitting principle arises from a fiber sequence

H ,! G ! G/H

where H and G are topological groups.

Give a G bundle E ! X, the splitting principle is the action of pulling
back the G-bundle along the projection of the associated bundle

E ⇥G G/H ! X

This pullback gives us a bundle with structure group H, and for nice enough
fiber sequences, the projection induces the desired injective map on coho-
mology.11
Some common examples of splitting principles for complex bundles are:

• SU (n) ! U (n) ! S 1 which corresponds to orienting the vector bun-

dle.
11
The cleanest way to see that this construction does reduce the structure group is to
pass to universal bundles over classifying spaces. Recall that over paracompact spaces X,
isomorphism classes of bundles E correspond to homotopy classes of maps fE : X ! BG,
where BG is the classifying space of G-bundles. We recover the bundle E on X by pulling
back the universal G-bundle EG ! BG along fE . Up to homotopy, EG is defined as
a contractible space with a free G-action, and BG is defined as EG/G. Now, given an
injection H ! G, EG also has a free H-action, so BH ' EG/H. But then BH ! BG is
the universal G-bundle with fibre G/H. Forming an associated bundle E ⇥G G/H over X
corresponds to pulling back the universal bundle BH along the map fE , and pulling back
EG along fE⇤ (BH) ! BH ! BG gives the “splitting” of E. See Segal [12] for the basics
of classifying spaces.

11
 • U (n 1) ! U (n) ! S 2n 1 which allows us to make arguments/constructions
by inducting on the dimension of the bundle.

• Tn ! U (n) ! F ln which corresponds to reducing a vector bundle to

a direct sum of line bundles.12
This last principle is the one most commonly referred to as the splitting
principle, and we will use it repeatedly going forward. In fact, the moral for
basic K-theory constructions seems to be:
• Define the construction for (direct sums of) line bundles.

• Extend to arbitrary bundles.

• Use the splitting principle to show that properties which hold for line
bundles hold in general.

4.2 Adams Operations

The Adams operations { k } are cohomology operations on complex K-
theory, analogous to Steenrod Squares in mod 2 cohomology.
Their basic properties are:
Prop. 9. For each compact, Hausdor↵ space X, and each k 2 N, there
exists a ring homomorphism
k
: K 0 (X) ! K 0 (X)

satisfying:
1. Naturality, i.e. kf ⇤ = f⇤ k for all maps f : X ! Y .

2. For any line bundle L, k ([L]) = [L]k .

3. k l = kl .

4. p (x) ⌘ xp (modp).
Property 2 characterizes the Adams operations, and we can use this,
plus L
the splitting principle, to give a general construction. Observe that if
E = ni=0 Li , then property 2 and being a homomorphism says
n
X
k
(E) = Lki
i=0

We can extend this to a general definition using exterior powers. Given any
bundle E, let i E denote the ith exterior power of E. From linear algebra,
we know
12
F ln denotes the complete flag variety over Cn .

12
 k (E
Lk
• E0) = i=0
i (E) ⌦ k i (E 0 )

• 0 (E) = 1, the trivial line bundle.

• 1 (E) = E.

• k (E) = 0 for k > dimE.

There exists a useful class of integral polynomials called the Newton

polynomials,Ldenoted sk (for k 2 N).PWith a little work,13 one can show
that if E = ni=0 Li as above, then ni=0 Lki = sk ( 1 (E), . . . , k (E)). We
can then take
k
(E) := sk ( 1 (E), . . . , k (E)
as the general definition. By applying the splitting principle associated with
Tn ! U (n) ! F ln , it’s enough to verify for (direct sums of) line bundles
that the Adams operations satisfy the specified properties, and this is a
straightforward check from our definitions.14

4.3 The Thom Isomorphism and Applications

Given the dependence of K-theory on vector bundles, we might expect that
those features of ordinary cohomology related to vector bundles also arise
in K-theory (e.g. the Thom Isomorphism, characteristic classes, and Gysin
maps). All of these rely on the orientability of the vector bundles in ordi-
nary cohomology, and formulating these for K-theory will similarly require
a suitable notion of K-orientable bundle and manifold.

4.3.1 Orientability in K-theory

Orientations for vector bundles or manifolds are defined formally in any
cohomology theory analogously to how they are defined in ordinary coho-
mology. First, some notation: Given a bundle V on X, let B(V ) denote the
unit ball bundle, and S(V ) the unit sphere bundle (under any metric).

Def. 5. A bundle V is orientable in the cohomology theory E ⇤ if there exists

a class ! 2 E ⇤ (B(V ), S(V )) such that !|p is a generator of E ⇤ (B(V )p , S(V )p )
as a module over E ⇤ (⇤) for all p 2 X.

Despite the formal similarity between ordinary orientability and ori-

entability in a general theory, we should not expect these to be closely
related. First, while orientability in ordinary integral cohomology is equiva-
lent to orientability in a (di↵erential) geometric sense, there is no guarantee
that this is the case in general, or that integral orientability is in any way
13
See Hatcher [7] §2.3 for the details of this construction.
14
For the proof of this splitting principle, see Hatcher [7] §2.3.

13
related to E ⇤ orientability. In general, the question of whether an inte-
grally orientable bundle or manifold is E ⇤ -orientable, for some theory E,
is a question of whether the orientation class survives to the E1 page of
the Atiyah-Hirzebruch SS and then whether we can recover the cohomology
from the associated graded. In general, this is not trivial, and it still leaves
us with the question of interpreting the meaning of the E-orientability of a
space.
In complex K-theory, Atiyah-Bott-Shapiro [3] shows that a vector bun-
dle is K-orientable if and only if it admits a spinc structure. This has
applications for mathematical physics, but is beyond the scope of my intro-
duction. On the other hand, every complex bundle admits a spinc structure,
and is thus K-orientable. More directly, we can construct the orientation
class of the bundle from its exterior algebra. Thus, for the remainder of this
section, I will assume that any bundle or manifold is almost complex, and
avoid worrying about the more general case.

4.3.2 The Isomorphism

In preparation for the Thom isomorphism, we reformulate our test space for
orientations of bundles as follows:
Def. 6. Given a vector bundle E on X, we define its Thom space, X E , to
be the one point compactification of E. 15
Notice that (X E /1) ⇠ = B(E)/S(E), so we can replace K ⇤ (B(E), S(E))
with K ⇤ (X E in our definition of orientability above.
Prop. 10. Every complex vector bundle E is K-orientable, with a canonical
e 0 (X E ) satisfying
orientation class E 2 K
1. Naturality, i.e. f ⇤E = f⇤ E

2. Sums, i.e. E E0 = (⇡1⇤ E) [ (⇡2⇤ E0 )

For the construction of E from the exterior algebra of the bundle E,

and a verification that it satisfies the desired properties, see Atiyah [2], §2.6,
p. 98-99. E is known as the Thom class of the bundle. We can now state:
Theorem 11. (Thom Isomorphism) If E is a complex vector bundle over
X, then Ke ⇤ (X E ) is a free module of rank 1 over K ⇤ (X) with generator !E .
In other words, the map

K
e ⇤ (X E )
: K ⇤ (X) ! K
given by K (x) = E · x is an isomorphism.
15
Equivalently, if E is a complex vector bundle, we can define X E := P (E 1)/P (E).
A quick check will show these are the same. Namely, P (E 1) amounts to compactifying
the fibers of E by gluing in projective hyperplanes at 1, and modding out by P (E) sends
all these hyperplanes to a point.

14
 The theorem can be proven via showing it in the case for (direct sums
of) line bundles and then using the splitting principle associated to Tn !
U (n) ! F ln as usual. Though this proof, as Atiyah [2] presents it, relies on
Bott periodicity, Bott periodicity is in fact equivalent to the Thom isomor-
phism theorem. Indeed, the Thom space of the trivial bundle X ⇥ C is the
reduced suspension S 2 (X + ). The Thom isomorphism then gives

K 0 (X) ⇠
=Ke 0 (S 2 (X + ))

i.e. Bott periodicity holds.

4.3.3 Gysin Maps

Gysin, or wrong-way, maps, are a useful tool in cohomology theories. Using
either Duality or the Thom Isomorphism, we can associate covariant maps
to orientation preserving maps between manifolds, in addition to the con-
travariant mappings guaranteed by the theory. As a matter of notation,
given such a map M ! M 0 , let NM denote the normal bundle of M in M 0 ,
and ✏M denote a tubular neighborhood of M in M 0 isomorphic to B(NM )
with boundary @(✏M ) isomorphic to S(NM ).16 Let ✏M denote the interior
of ✏M , i.e. ✏M = ✏M @(✏M ).
Def. 7. (Gysin Maps) Given an immersion f : M ! M 0 , of almost complex
manifolds, we define f! : K ⇤ (M ) ! K ⇤ (M 0 ) as the composite:
e ⇤ (M NM ) ! K ⇤ (✏M , @(✏M )) ! K ⇤ (M 0 , M 0
K ⇤ (M ) ! K ✏M ) ! K ⇤ (M 0 )

where the first map is the Thom isom., the second is the isom. (B(NM ), S(NM )) ⇠
=
(✏M , @(✏M )) given by the tubular neighborhood theorem, the third is the
isom. due to excision, and the last map corresponds to the canonical map
(M 0 , ;) ! (M 0 , M 0 \ ✏M ).
Note that in ordinary cohomology, the Gysin map raises the degree by
dim(M 0 ) dim(M ). However, if M and M 0 are almost complex, they are of
even dimension, so the di↵erence is even as well. The periodicity of K-theory
then ensures that Gysin maps between almost complex manifolds preserve
degree.

4.3.4 Thom Classes and Characteristic Classes

We use the Thom class to construct characteristic classes in K-theory. There
are two equivalent constructions for these classes, one using a splitting prin-
ciple and the other using the Adams operations. I will sketch both and the
interested reader can check that they are equivalent by doing the calculations
with the universal bundle on BU .
16
The existence of such a neighborhood is an immediate consequence of the usual tubular
neighborhood theorem.

15
Def. 8. (The Euler Class) Given a complex vector bundle E on X, let
⇣ denote the 0-section. Then the Euler class, e(E) 2 K 0 (X), is given by
e(E) = ⇣ ⇤ E .
Def. 9. (Chern Classes) Given a complex vector bundle E on X, we define
the Chern classes classes c1 (E), . . . , cn (E) 2 K 0 (X) inductively as follows:
1. cn (E) = 0 if n > rk(E)
2. cn (E) := e(E)
b where E
3. ci (E) := ⇡⇤ 1 (ci (E)) b is the vector bundle corresponding to
the splitting principle given by U (n 1) ! U (n) ! S 2n 1 .17
We can also define ci (E) by ci (E) := K1 i(
E ) where
i is the ith

Adams operation.
An obvious question to ask is how the Chern classes in K-theory behave
relative to those in ordinary integral cohomology. Recall that in integral
cohomology, if L and L0 are line bundles, then c1 (L ⌦ L0 ) = c1 (L) + c1 (L0 ).18
In K-theory, a similar calculation shows that c1 (L ⌦ L0 ) = c1 (L) + c1 (L0) +
c1 (L)c1 (L0 ). Jacob Lurie includes a nice discussion of this in [10]. In partic-
ular, he introduces the notion of formal group laws, and observes that in this
language, ordinary chern classes are governed by the formal additive group,
whereas K-theoretic chern classes are governed by the formal multiplicative
group. We can exploit this to motivate both the definition and the existence
of the Chern character.

4.4 Constructing the Chern Character

I asserted the existence of the Chern character above, and this section out-
lines a concrete construction of it. Morally, the Chern character for K-
theory arises from the observation that, over Q, exponentiation gives an
isomorphism between the formal additive group and the formal multiplica-
tive group. We can make this concrete as follows:
Given a K-class [L] represented by a line bundle L ! X, define
c1 (L)2 c1 (L)k
ch([L]) := exp(c1 (L)) = 1 + c1 (L) + + ... + + ...
2 k!
where c1 (L) is the first Chern class
L of L in ordinary cohomology. Given a
K-class [E] represented by E = ni=0 Li , define
n
X tk1 + . . . tkn
ch([E]) := ch(Li ) = n + (t1 + . . . tn ) + . . . + + ...
k!
i=0
17
We show this to be a splitting principle via the Gysin sequence coming from the Thom
isomorphism.
18
We show this by doing the computations with the chern class of the universal line
bundle over CPn ).

16
where ti = c1 (Li ). By the same algebra that we used in the construction
of the Adams operations, we can convert this expression into one purely in
terms of the ordinary chern classes of E.19 If cj denotes the j th Chern class
cj (E) of E, and sk denotes the k th Newton polynomial, then we can rewrite
this as
X sk (c1 , . . . , ck )
ch(E) = dim(E) +
k!
k>0

and take this as a general definition for arbitrary bundles. The interested
reader can check that this is well defined and that it lands in the correct
dimensions in H ⇤ (X; Q). Moreover, a straightforward check from our defi-
nitions shows
P thatPfor any line bundles L and L0 , ch(L ⌦ L0 ) = ch(L)ch(L0 )
and ch( i Li ) = i ch(Li ), so ch gives a ring homomorphism as desired.
Now apply the splitting principle associated to Tn ! U (n) ! F ln to con-
clude that any cohomology relation which holds for line bundles under this
reduction also holds for arbitrary bundles.20 This completes the construc-
tion of the Chern character. For the proof that it induces the isomorphism
claimed above, see Hatcher [7], Ch.4, P.4.3 and 4.5.

5 Hirzebruch-Riemann-Roch
The Hirzebruch-Riemann-Roch Theorem was the first in a series of gener-
alizations of the classical Riemann-Roch theorem, which eventually culmi-
nated in the Grothendieck-Hirzebruch-Riemann-Roch Theorem in algebraic
geometry, and in the Atiyah-Singer Index Theorem in the di↵erential case.
I include it here, both for historical purposes, and because some of the
deeper applications of K-theory occur in its implications for index theory,
and Hirzebruch-Riemann-Roch is a first sign of these results.
Before we can state the theorem, we need to define another characteristic
class T d⇤ ( ) called the Todd class. We can think of the Todd class as a
formal reciprocal of the Chern character, and its construction is similar.
19
From the axioms of characteristic classes, the total chern class

c(E) = (1 + t1 ) . . . (1 + tn ) = 1 + 1 + ... + n

th
where j denotes the j symmetric polynomial in the ti , and considering cohomological
degrees, we see cj (E) = j . As noted in the construction of the Adams operations, there
exists a class of polynomials sk called the Newton polynomials with the property that tk1 +
. . . tkn = sk ( 1 , . . . , k ). This allows us to rewrite the formula above as ch([E]) = dim(E)+
P sk ( 1 ,..., k )
k>0 k!
. See Hatcher [7] §.2.3 for more details on the Newton polynomials.
20
This is why we need the injectivity of the map p : F l(E) ! X on cohomology. For a
proof of this splitting principle, see Hatcher [7], §3.

17
5.1 The Todd Class
In the usual manner, we define the Todd class for direct sums of line bundles,
use some algebra to massage this into a formula for general bundles, and
then prove it has the desired properties by using the splitting principle. We
can characterize the Todd class as follows:

Prop. 12. Given a vector bundle E ! X, there exists a unique class

T d⇤ (E) 2 H ⇤ (X; Q) satisfying:

1. Naturality, i.e. f ⇤ T d⇤ (E) = T d⇤ (f ⇤ E).

2. T d⇤ (E F ) = T d⇤ (E)T d⇤ (F )
c1 (L)
3. T d⇤ (L) = 1 e c1 (L)
for all line bundles L.

Using the Bernoulli numbers {B2i }1

i=1 , we can expand the righthand side
of number 3 as a formal power series. Explicitly:
1
x x X B2i 2i
Q(x) := =1+ + x
1 e x 2 (2i)!
i=1

For any finite dimensional basespace X, number 3 describes a polynomial

in the chern classes with rational coefficients, and thus it does specify an
element of H ⇤ (X; Q). This, together with number 2 then dictate the formula
for direct sums of line bundles, and though we are unable to massage out a
general expression using the Newton polynomials, there exists another class
which does the job and allows us to get get a general expression for T d⇤ (E)
as a power series in the Chern classes of E. The first few terms of this
expansion are

1 1 1 1 4
T d⇤ (E) = 1+ c1 + (c21 2c2 )+ c1 c2 (c 4c21 c2 3c22 c1 c3 +c4 )+. . .
2 12 24 720 1
where the ci are the Chern classes of E. With this definition, naturality
is a consequence of the naturality of the chern classes, and the splitting
principle associated to Tn ! U (n) ! F ln shows that 2 is satisfied in general.

5.2 The Theorem

Recall that the euler characteristic of a coherent sheaf F on a space X is
the alternating sum of its Betti numbers, i.e.
X
(F ) := ( 1)i rk(H i (X, F ))

18
Theorem 13. (Hirzebruch-Riemann-Roch) Given a vector bundle E on a
compact complex manifold M , let E denote the sheaf of holomorphic sections
of E. Then Z
(E ) = ch(E) · T d⇤ (T M )
M
I don’t give the proof here, but if you’re interested, see Hirzebruch [9],
or for a nice motivational discussion, see Griffiths and Harris [4] §3.4.
As an illustration of the theorem, we can quickly derive the classical
Riemann-Roch formula for curves, which I go through below. There’s a
similarly nice derivation of Noether’s Theorem for surfaces, and the inter-
ested reader should Hartshorne’s [5], Appendix A.

5.3 Riemann-Roch for Curves

Let X be a curve of genus g and let L(D) be a line bundle X corresponding
to divisor D. We need to quickly recall several facts from algebraic geometry:
• (OX ) = 1 g.
• The first Chern class gives the isomorphism between divisors and line
bundles, so c1 (L) = D and ch(L) = 1 + D.
• The tangent bundle T X is the dual of the canonical bundle of X, and
so if K is the canonical divisor of X, T X corresponds to K, and
T d⇤ (T X) = 1 12 K.
• Since D and K define elements of H 2 (X; Z), DK = 0 for dimension
reasons.
R
• For a divisor D, M D = degD (this is tautological from the definitions,
but worth recalling).
. Putting these together, we see
1
ch(L)T d⇤ (T X) = (1 + D)(1 K)
2
1
=1+D K
2
and plugging this into the formula from the theorem, we get
Z
1
(L (D)) = (1 + D K)
M 2
1
= deg(D K)
2
Now, taking D = 0, this says 1 g = (OX ) = 1
2 degK, and plugging this
in we get the classical formula:
(L (D) = degD + 1 g

19
6 Conclusion
There is a lot more to say about complex K-theory, but I will leave o↵ here
for now. I hope these notes prove an accessible and easy to use introduction
to the main features of K-theory, and at the very least, that they make it
easier to navigate the various texts and articles which lay out the theory.

References
[1] J.F. Adams, Stable Homotopy and Generalized Homology

[2] M.F. Atiyah, K-Theory

[3] M.F. Atiyah, R. Bott, A. Shapiro, “Cli↵ord Modules”, (1964)

[4] P. Griffiths and J. Harris, Principles of Algebraic Geometry

[5] R. Hartshorne, Algebraic Geometry

[6] A. Hatcher, Algebraic Topology

[7] A. Hatcher, Vector Bundles and K-Theory

[8] A. Hatcher, Spectral Sequences in Algebraic Topology

[9] F. Hirzebruch, Topological Methods in Algebraic Geometry

[10] J. Lurie, “Survey of Elliptic Cohomology”

[11] J. Milnor, Morse Theory

[12] G. Segal, “Classifying Spaces and Spectral Sequences” (1968)

20