Cohomology and Vector Bundles

Corrin Clarkson
REU 2008
September 28, 2008

Abstract
Vector bundles are a generalization of the cross product of a topo-
logical space with a vector space. Characteristic classes assign to the
each vector bundle a cohomology class of the base space. The Euler
class, the Thom class and the Chern classes are a few characteristic
classes of vector bundles.

The primary source for this paper is Bott and Tu’s Differential Forms in
Algebraic Topology [1].

1 Vector bundles
1.1 Definition and examples
Definition 1.1 (Vector bundle). For a field F = R or C and a positive
integer k, a rank k F -vector bundle ξ consists of a triple (E, p, B) where
E and B are topological spaces and p : E → B is a surjective continuous
map. The triple must satisfy the condition that there be a local trivialization
{φα , Uα } where {Uα } is an open cover of B and the φα are homeomorphisms
satisfying the following commutative diagram:
φα
Uα × F k / E|U
α
ss
ss
s
ss p
π
 ysss
B

1
Here π is the natural projection onto the first factor and E|Uα = p−1 (Uα ).
Each fiber p−1 (b) is a vector space and the restrictions φα |b×F k : b × F k →
p−1 (b) are vector space isomorphisms. The final property satisfied by {φα , Uα }
is that the transition functions gαβ : Uα ∩ Uβ → GLk (F ) given by gαβ (b) =
φ−1
β φα |{b}×F k be continuous.
E is called the total space, B the base space and F k the fiber space. The
preimage p−1 (b) ⊂ E of a point b ∈ B is called the fiber over b, and is
often denoted Eb . The dimension k of the fiber space is called the rank of
the vector bundle. Rank 1 vector bundles are commonly referred to as line
bundles.

Definition 1.2 (Bundle morphism). A bundle morphism between two vector

bundles (E, p, B) and (E 0 , p0 , B 0 ) is composed of a pair of functions (h, f )
where f maps between the base spaces and h maps between the total spaces.
We require that the restriction of h to any fiber be linear and that the fiber
over b ∈ B be mapped to the fiber over f (b). In other words the following
commutative diagram is satisfied:

E
h / E0
p p0
 f 
B / B0

Example 1.3 (Trivial bundle). For any topological space B and vector space
V , the cross product B × V is a vector bundle call the trivial bundle. The
projection map is simply the natural projection onto the first factor, and the
local trivialization is given by the identity map. Thus the open cylinder is a
trivial real line bundle over S 1 .

Example 1.4 (Möbius band). The Möbius band

[0, 1] × R/(0, t) ∼ (1, −t)

is a nontrivial line bundle on S 1 . Thinking of the circle as the unit interval

with zero and one associated, the projection map comes from the natural
projection onto the first factor. We take the open cover consisting of two
overlapping semicircles. The trivializations amount to choosing an orienta-
tion of R for each of the semicircles (see figure 1). The resulting transition
maps are both constant and therefore continuous.

2
 Figure 1: Möbius band

Example 1.5 (Grassmannian). One important vector bundle is the canon-

ical bundle on the Grassmannian. The Grassmannian, Gk (F n ), is the space
of k-dimensional subspaces of F n with the quotient topology defined by the
map: (v1 , . . . , vk ) 7→ Span(v1 , . . . , vk ). The infinite Grassmannian Gk (F ∞ ) =
n
S
n≥k Gk (F ) is given the topology induced from the natural inclusions

Gk (F n ) ,→ Gk (F n+1 ).

The canonical vector bundle γkn on the Grassmannian Gk (F n ) (0 < n ≤

∞) is a subbundle of the trivial bundle Gk (F n ) × F n composed of elements
of the form (V, x) where x ∈ V . Thus the fiber over a subspace V ∈ Gk (F n )
consists of the elements of V .

1.2 Transition functions

The transition functions of a vector bundle are essentially glueing maps that
describe how the trivial neighborhoods stick together to make the total space.

Theorem 1.6. Let B be a topological space and {Uα } an open cover of B.

Given a set of functions {gαβ : Uα ∩ Uβ → GLk (F )} such that gαβ gβγ = gαγ
on the triple intersection Uα ∩ Uβ ∩ Uγ . There is a rank k F -vector bundle
on B with transition functions {gαβ }.
Proof. Consider the disjoint union
[
E= Uα × F k
α

Use the functions {gαβ } to define the relation (a, x)α ∼ (b, y)β if and only
if a = b and y = gαβ (a)x. The cocycle condition gαβ gβγ = gαγ implies

3
that gβα = (gαβ )−1 . Therefore if a = b and y = gαβ (a)x, then b = a and
x = (gαβ (a))−1 y = gβα (b)y. Thus ∼ is an equivalence relation. The quotient
E/ ∼ is a vector bundle on B. Its trivialization functions result from the
inclusion maps Uα × F k ,→ E and its transition functions are exactly the
{gαβ }.
0
Definition 1.7. Two sets of transition functions {gαβ } and {gαβ } with re-
spect to the same cover {Uα } are said to be equivalent if there exists a collec-
0
tion of continuous maps {λα : Uα → GLk (F )} such that gαβ = λα ·gαβ ·(λβ )−1 .
Here · is multiplication in GLk (F ).

Theorem 1.8. Two rank k F -vector bundles ξ and η on a space B with local
trivializations relative to some cover {Uα } are isomorphic if and only if their
transition functions are equivalent.

Remark 1.9. If the transitions functions of a vector bundle are equivalent to

a set of transition functions living in some subgroup H < GLn (F ), then we
say that the structure group can be reduced to H.

Definition 1.10 (Orientable). A rank k real vector bundle is call orientable if

its transition functions are equivalent to a set of transition functions mapping
to GL+ k (F ), the linear isomorphisms with positive determinant.

Proposition 1.11. The structure group of a real vector bundle of rank k can
be reduced to SO(k) if and only if it is orientable.

1.3 Induced bundles

Definition 1.12 (Bundle morphism). A bundle morphism between two vec-
tor bundles (E, p, B) and (E 0 , p0 , B 0 ) is composed of a pair of functions (h, f )
where f maps between the base spaces and h maps between the total spaces.
We require that the restriction of h to any fiber be linear and that the fiber
over b ∈ B be mapped to the fiber over f (b). In other words the following
commutative diagram is satisfied:

E
h / E0
p p0
 f 
B / B0

4
 Figure 2: Induced bundle

Definition 1.13 (Induced bundle). Given a vector bundle ξ = (E, p, B), a

continuous map f : A → B induces a bundle f ∗ (ξ) on A that has total space
{(a, x) ∈ A × E | f (a) = p(x)} and projection map (a, x) 7→ a. The fiber
f ∗ (ξ)a is canonically identified with p−1 (b).
Example 1.14 (Figure eight). Consider the Möbius band µ as a real line
bundle over the circle and the continuous map f from the figure eight to the
circle obtained by viewing the figure eight as the wedge of two circles. The
induced bundle f ∗ (µ) is then two Möbius bands glued together at a single
fiber (see figure 2).

Example 1.15 (Projectivization). The projectivization P (V ) of a vector

space V = F n is simply the first Grassmannian G1 (F n ). In example 1.5,
two tautological bundles over this space were discussed, namely the trivial
bundle P (V ) × F n and the canonical bundle γ1n . The canonical bundle on
P (V ) is also know as the universal subbundle and is often denoted by S.
Given a vector bundle ξ = (E, p, B) with fiber space V , we can take
the projectivization P (E) by taking the projectivization of each fiber. The
resulting space is a fiber bundle over B with fiber space P (V ). A fiber
bundle is similar to a vector bundle; the primary difference is that the fiber
space need not be a vector space. The structure of this fiber bundle can
be understood in terms of its transition functions {ḡαβ } which are simply
transition functions {gαβ } of ξ modulo scalar matrices. Thus the ḡαβ map
into the projective general linear group rather than the general linear group.
The tautological vector bundles on P (E) are similar to those on the
projectivization of a vector space. The analog of the trivial bundle is the
induced bundle π ∗ (ξ) where π : P (E) → B is the projection map de-
rived from p. In this bundle, the fiber over a line lb in Eb is exactly Eb .

5
When restricted to any fiber π −1 (b) = P (Eb ), this bundle is trivial i.e.
π ∗ (ξ)|P (E)b = P (Eb ) × Eb . The analogue of the universal subbundle is
the subbundle S = {(lb , x) ∈ π ∗ (ξ) | x ∈ lb }. The fiber in S over a line
lb ∈ P (E) consists of the points in lb , and the restriction of S to any fiber
π −1 (b) = P (Eb ) is the universal subbundle of P (Eb ).
Proposition 1.16 (Functoriality). Given a vector bundle ξ over a space
g f
B and two continuous maps X /Y / B , the bundle induced by the
composition f g is isomorphic to that induced from g ∗ (ξ) by f . In other words,
the following relation is satisfied: g ∗ (f ∗ (ξ)) ∼
= (f g)∗ (ξ).
Theorem 1.17 (Classification of vector bundles). Every rank k, F -vector
bundle on a paracompact space B is isomorphic to a bundle induced by a map
from the base space to the Grassmannian Gk (F ∞ ).
Proof. Let ξ = (E, p, B) be a rank k, F -vector bundle on a paracompact
space B. Also let {(φα , Uα )} be a local trivialization of ξ. As B is paracom-
pact we can assume without loss of generality that {Uα } is locally finite and
countable. Let {εα } be a partion ofP unity subordinate to {Uα }.
Now consider the map g : E → α F k defined by:

X
g= gα such that gα |p−1 (Uα ) = (εα p) · (π2 φ−1
α ) and gα = 0 everywhere else.
α

Here π2 : Uα × F k → F k is the natural projection onto the second factor.

The composition εα p assigns a scalar from the interval [0,1] to each point
in E|Uα . The function gα then multiplies this scalar by the vector given by
π2 φ−1
α . The scalar given by the partition of unity ensures that the extension
of gα by zero is continuous. The sum of the gα makes sense, because {Uα } is
locally finite and countable. The continuity of g follows from that of the gα .
The gα : E → F k restrict to linear injections on each fiber p−1 (b) where
εα (b) is nonzero. This follows from the fact that both π2 and φ−1 α restrict
to linear injections on fibers. This implies g restricts to linear
P injections on
k
fibers as the gα map to linearly independant subspaces of α F .
Thus the map g associates each fiber of E to a k-dimensional subspace of
∞
F in a continuous way. This association can be used to define a map from
B to Gk (F ∞ ) in the following way:

f : B → Gk (F ∞ ) such that f (b) = g(p−1 (b)).

6
 In order to see that f is continuous, consider p−1 as a function from B to
the fibers of E. Locally −1 maps Uα to the fibers of E|Uα . The trivializing
map φα is a fiber preserving homeomorphism between E|Uα and Uα × F k .
The projection π1 : Uα × F k → Uα gives a homeomorphism between Uα
and the space of fibers of Uα × F k . Thus p = φ−1 α π1 can be viewed as a
homeomorphism between the space of fibers of E|Uα and Uα . This implies
that p−1 is a continuous map. Therefore f = gp−1 must also be continuous.
All that remains to be shown is that f ∗ (γk∞ ) is isomorphic to ξ.
From the definition of γk∞ and the properties of induced bundles, we
have that f ∗ (B) is a subbundle of the trivial bundle B × F ∞ . The map
(p.g) : E → B × F ∞ is clearly continuous. Furthermore its image is exactly
f ∗ (γk∞ ) and the restriction of g to any fiber Eb is a linear isomorphism to the
fiber f (b) = g(p−1 (b)) in f ∗ (γk∞ ). This implies that f ∗ (γk∞ ) is isomorphic to
ξ.

Theorem 1.18. Let B be a paracompact space. Two continuous maps f, g :

B → Gk (F ∞ ) induce isomorphic bundles on B if and only if they are homo-
topic.

The proof of this theorem is too long to be included here, but can be
found in Husemöller’s Fibre Bundles [2].
By the previous two theorems, there is a correspondence between isomor-
phism classes of vector bundles on B and homotopy classes of maps from B
to Gk (F ∞ ).

2 Cohomology
2.1 Differential forms
A more thorough explanation of this subject can be found in Munkres’s
Analysis on Manifolds [3].
This section focuses on real manifolds, but the analogous theorems and
definitions hold for complex manifolds.

Definition 2.1 (Differential form). A differential form ω on a smooth k-

dimensional manifold M is a collection of forms ωα on Rk corresponding to
an atlas {(φα , Uα )} such that each ωα is a form on φα (Uα ). The ωα ’s agree
on the intersections Uα ∩ Uβ in the following sense. The maps induced from

7
  
the inclusions Uα ∩ Uβ  i / Uα and Uα ∩ Uβ 
j
/ Uβ take ωα and ωβ to
the same form i.e. i∗ ωα = j ∗ ωβ
The wedge product of forms on M is derived from that of forms on Rk in
the following way: (ω ∧ ν)α = ωα ∧ να . Similarly the differential operator d
is obtained from that on forms on Rk by defining (dω)α = dωα . The familiar
properties of the differential operator still hold.
Proposition 2.2. Let ω and ν be differential forms on a manifold M . The
following relations hold:
d(d(ω)) = 0
d(ω + ν) = dω + dν
d(ω ∧ ν) = dω ∧ dν

2.2 de Rham cohomology

Definition 2.3 (Closed and exact forms). A differential form ω is called
closed if dω = 0. It is called exact if there exists a form η such that dη = ω.
Definition 2.4 (de Rham cohomology). The q th de Rham cohomology of a
manifold H q (M ) is obtained by taking the quotient of the closed q-forms on
M by the exact q-forms.

H q (M ) = {closed q-forms}/{exact q-forms}

This construction results in the same cohomology as the standard singular

cohomology with R (or C) coefficients used in algebraic topology and thus
also results in a contravariant functor. The key difference between these two
contsructions is that the de Rham cohomology only works for smooth spaces
and smooth maps. Here are a few important theorems regarding the basic
properties of singular cohomology.

2.3 Properties of cohomology

Theorem 2.5. Continuous maps f : A → B between spaces induce ho-
momorphisms of the cohomologies f ∗ : H ∗ (B) → H ∗ (A). Furthermore the
homomorphism induced by a composition f g is exactly the composition g ∗ f ∗ .
Theorem 2.6. Homotopic maps between spaces induce the same homomor-
phism on their cohomology groups.

8
Definition 2.7 (Compact vertical support). A vector bundle (E, p, M ) over
a manifold M has a collection of forms Ω∗cv (E) that have compact support on
each fiber of E i.e. the restriction of a form ω ∈ Ω∗cv (E) to a fiber Ep = p−1
has compact support.
Definition 2.8 (Thom map). Let (E, p, M ) be an orientable rank k real
vector bundle on a manifold M and {(φα , Uα )} be a local trivialization. The
q-forms on E are locally of the form p∗ ωα ∧ (π2 φ−1 ∗
α ) να where π2 : Uα × R →
k
k
R is the natural projection onto the second factor, the ωα are m-forms on
the Uα and the να are n-forms on Rk such that m + n = q. Formally we write
ωα ⊗ να rather than p∗ ωα ∧ (π2 φ−1 ∗
α ) να .
The Thom map takes forms on (E, p, M ) to forms on M . More precisely
it maps from Ω∗cv (E) to Ω∗ (M ). The map is given locally by integration over
fibers and is defined as follows:
Z
ωα ⊗ να 7→ 0 if deg(να ) < k and ωα ⊗ να 7→ ωα · να if deg(να ) = k
Rk

Here integration is done on each fiber giving a map from M to R.

Theorem 2.9. The local definition of the Thom map extends to a global
definition.
Proof. Let η = {ωα ⊗ να } be a global q-form on (E, p, M ), and π∗ denote
the Thom map. If the να have degree less than k, then π∗ (η) = 0 as each of
its components maps to zero. In the case where the να have degree k, they
can be written as να = fα (x, v)dv where fα : Uα × Rk → R is a smooth map.
Restricting to Uα ∩ Uβ × Rk we have that
R
π∗ (ωα ⊗ να ) = ωα · RRk να by definition
= ωα · RRk fα (x, v)dv by definition
= ωα · RRk fα (x, gβα (x)w)detgβα (x)dw change of variables φ−1
α φβ
= ωα · Rk fα (x, gβα (x)w)dw because WLOG detgβα (x) = 1
R due to orientability
= ωβ · Rk fβ (x, w)dw because η is a global form
= π∗ (ωβ ⊗ νβ ) by definition

Thus the components π∗ (ωβ ⊗νβ ) piece together into a global form π∗ (η).
Theorem 2.10 (Thom isomorphism). The Thom map induces an isomor-
∼
∗+n
phism of cohomology rings: Hcv (E)
= / H ∗ (M ) .

9
Definition 2.11 (Thom class). Let (E, p, M ) be a rank k real vector bundle
∗+n
on a manifold M . The Thom class of E is the cohomology class ω ∈ Hcv (E)
that is mapped to 1 ∈ H 0 (M ) ∼
= R by the Thom isomorphism.

Lemma 2.12 (Projection formula). Let (E, p, M ) be a rank n real vector

bundle over a manifold M . Also let τ be a form on M and ω be a form on
E with compact vertical support. The following relation is satisfied by π∗ the
Thom map described in definition 2.8.

π∗ (p∗ τ ∧ ω) = τ ∧ π∗ ω

Proof. ω is locally p∗ µ ∧ j ∗ ν where µ is a form on M , ν a form on Rn and

j is the composition of the trivialization map with the natural projection on
to Rn . There are two cases to consider: deg(ν) < n and deg(ν) = n. In the
first case, we have that

π∗ (p∗ τ ∧ ω) = π∗ (p∗ τ ∧ p∗ µ ∧ j ∗ ν)
= π∗ (p∗ (τ ∧ µ) ∧ j ∗ ν) because p∗ is a homomorphism
= 0 by the definition of the Thom map
= τ ∧ π∗ ω by the definition of the Thom map

In the second case, we have that

π∗ (p∗ τ ∧ ω) = π∗ (p∗ τ ∧ p∗ µ ∧ j ∗ ν)
= π∗ (p∗ (τ ∧R µ) ∧ j ∗ ν) because p∗ is a homomorphism
= (τ ∧ µ) ·R Rn ν by the definition of the Thom map
= τ ∧ (µ · Rn ν) property of differential forms
= τ ∧ π∗ ω by the definition of the Thom map

Thus the projection formula holds locally. Because the Thom map is derived
from a local definition the formula also holds globally.

Definition 2.13 (Poincaré dual). Let M be an oriented m-dimensional man-

ifold and S ⊆ M a closed oriented submanifold of dimension k. The Poincaré
dual of S in M is the unique cohomology class η̄S ∈ H n−k (M ) such that
Z Z
∗
iω= ω ∧ ηS for all closed k-forms ω with compact support.
S M

Here i : S ,→ M is the inclusion map.

10
 This definition only makes sense if such a form exists. The following
theorem proves the existance of such a form and relates the the Poincaré
dual of a submanifold to the Thom class of its normal bundle.
Theorem 2.14. The Poincaré dual of a closed oriented submanifold S in an
oriented manifold M and the Thom class of the normal bundle of S can be
represented by the same form.
Proof. Let M be an n dimensional oriented real manifold, S ⊆ M be a k
dimensional closed oriented submanifold and i : S ,→ M be the inclusion
map. Also let T ⊆ M be a tubular neighborhood of S. There exists a map
p : T → S such that (T, p, S) is isomorphic to the normal bundle on S. Let
α : S ,→ T be the inclusion of S into T . Note that this is map is equivalent
to taking the zero section of the normal bundle. Forms with compact vertical
support on T (viewed as the normal bundle) can be extended by zero to forms
∗
on M as their support lies in the interior of T . Let j∗ : Hcv (T ) → H ∗ (M ) be
the map obtained by extending by zero. Finally let Φ be the Thom class of
T and ω be a k-form on M with compact support.
because the support of j ∗ Φ lies in T
R
M
ω ∧ j∗ Φ = intT ω̄ ∧ j∗ Φ
Here the bar denotes the restriction to T
= T (p∗ α∗ ω̄ + dτ ) ∧ Φ because p∗ and α∗ are inverse isomor-
R

R ∗ ∗ phisms in cohomology
= T (p α ω̄) ∧ Φ by Stokes theorem and the fact that
R ∗ Supp(Φ) lies in the interior of T
= S α ω̄ ∧ π∗ Φ by the projection formula lemma 2.12.
Here π∗ is the Thom map.
= RS α∗ ω̄
R
because π∗ Φ = 1 by definition
∗
= Si ω
Thus the Thom class of T satisfies the condition for the Poincaré dual.

3 Characteristic Classes
For the entirety of this section let F be R or C and k be a positive integer.
Also let B be paracompact space and BG = Gk (F ∞ ) be the Grassmann
variety of k-dimensional subspaces of F ∞ . Finally let γ be the canonical
vector bundle on the Grassmannian BG. All vector bundles will be rank
k F -vector bundles.

11
3.1 The definition
The classification of vector bundles provides a correspondence between iso-
morphism classes of vector bundles on B and homotopy classes of maps from
B to BG. The cohomology functor associates a homomorphism of cohomol-
ogy groups, H ∗ (BG) → H ∗ (B), to each homotopy class of maps from B to
BG. Thus the classification of vector bundles in combination with cohomol-
ogy theory can be used to assign to each vector bundle on B a subgroup
of the cohomology of B, namely the image of the induced map between the
cohomology groups H ∗ (BG) and H ∗ (B) (see figure 3).
By looking at the image of a specific class in H ∗ (BG) rather than that of
the entire group, this same technique can also be used to assign a cohomology
class in H ∗ (B) to each vector bundle on B.
Remark 3.1. The following notation provides a concise way of talking about
this assignment: given a cohomology class α ∈ H ∗ (BG) and a vector bundle
ξ induced by a map f : B → BG, let α(ξ) = f ∗ (α) be the image of α under
the induced cohomology homomorphism f ∗ : H ∗ (BG) → H ∗ (B).

Definition 3.2 (Characteristic class). A characteristic class α assigns to each

vector bundle η on A a cohomology class α(η) ∈ H ∗ (A). This assignment
commutes with induced maps in the following way: for all continuous maps
f : B → A we have that

f ∗ (α(η)) = α(f ∗ (η)).

Here f ∗ (η) is the induced bundle on B and f ∗ is the induced cohomology

homomorphism.

Theorem 3.3. Each cohomology class of the Grassmannian corresponds to

a unique characteristic class and all characteristic classes correspond to a
cohomology class of the Grassmannian.
Proof. Let A and B be paracompact spaces and ξ be a vector bundle on B.
Also let g : A → B be a continuous map and α ∈ H ∗ (BG) be a cohomology

Figure 3: Quick summary

vector bundles Classification continous
/ maps Cohomology / homomorphisms
on B B → BG H ∗ (BG) → H ∗ (B)

12
class of the Grassmannian. By the classification theorem there exists a map
f : B → BG that induces ξ from the canonical bundle on BG i.e. ξ = f ∗ (γ).
The assignment α() defined in remark 3.1 is a characteristic class because we
have that:
α(g ∗ (ξ)) = α((f g)∗ (γ)) by the functoriality of induced bundles
= (f g)∗ (α) by the definition of α()
= g ∗ (f ∗ (α)) because Cohomology is a functor
= g ∗ (α(f ∗ (γ))) by definition of α()
= g ∗ (α(ξ)) by definition of f

Clearly distinct cohomology class of the Grassmannian must correspond

to different characteristic classes as they assign different classes to the canon-
ical bundle on the Grassmannian.
All that remains is to show that all characteristic classes correspond to a
cohomology class of the Grassmannian. Let α be a characteristic class. This
implies α(γ) = β for some β ∈ H ∗ (BG). Now let ξ be a vector bundle on
B induced by the map f : B → BG. The following equation follows directly
from the definitions:

β(ξ) = f ∗ (β) = f ∗ (α(ξ)) = α(f ∗ (γ)) = α(ξ)

Thus β(ξ) = α(ξ) for all vector bundles ξ. Therefore all characteristic classes
correspond to a cohomology class of the Grassmannian.

Example 3.4 (Euler class). Let M be an n-dimensional manifold and ξ =

(E, p, M ) be an oriented rank 2 real vector bundle on M with atlas {(φα , Uα )}.
Also let {εα } be a partition of unity on M subordinate to {Uα }.
By proposition 1.11 we can assume without loss of generality that the
transition functions {gαβ } of ξ map to SO2 (R), thus gαβ assigns an angle of ro-
tation to each point in the intersection Uα ∩Uβ . Because ξ is orientable, coun-
terclockwise is globally defined. Let ψαβ : Uα ∩Uβ → [0, 2π) such that ψαβ (x)
is the angle of rotation assigned by gαβ (x) measured in the counterclockwise
direction. Because gαβ gβγ = gαγ , we have that ψαγ − ψβγ = ψαβ + 2nπ For
some integer n. P
On each Uα , define the 1-form ηα = γ εα dψαγ . Here d is the differential
operator and the functions dψαγ are extended by zero outside of Uα ∩ Uγ . On
the intersection Uα ∪ Uβ we have that:

13
 P
ηα − ηβ = ε dψ − εγ dψβγ
Pγ γ αγ
= ε d(ψαγ − ψβγ )
Pγ γ
= γ εγ dψαβ
= dψαβ
This implies that dηα − dηβ = d(ηα − ηβ ) = d(dψαβ ) = 0 on Uα ∪ Uβ Thus
the 2-forms {dηα } piece together to form a global 2-form on M . This form
is clearly closed and thus corresponds to a cohomology class. This class is
called the Euler class and is a characteristic class.

Proposition 3.5. The pullback of the Thom class to M by the zero section
is the Euler class.

3.2 Chern classes

Chern classes are some of the most important characteristic classes on com-
plex vector bundles. The following two theorems provide a means of defining
the Chern classes.

Theorem 3.6 (Leray-Hirsch). Let E be a fiber bundle on B. If there are

global cohomology classes {x1 , . . . , xn } on E that when restricted to each fiber
freely generate the cohomology of that fiber, then H ∗ (E) is a free module over
H ∗ (B) with basis {x1 , . . . , xn }.

Lemma 3.7. Let P (Cn ) be the projectivization of Cn and S be the universal

subbundle of P (Cn ) × Cn . We have that

H ∗ (P (Cn )) = R[x]/(xn ) where − x = e(S) the Euler class of S.

Example 3.8 (Chern classes). Let ξ = (E, p, B) be a complex vector bundle

on B. Also let P (E) be the projectivization of E and S be the universal
subbundle on P (E) (see example 1.15). Finally let x = −e(S) be the negative
of the Euler class of S viewed as a rank 2 real vector space. As the restriction
of S to any fiber of P (E) is the universal subbundle Se on the fiber and e()
is a characteristic class, x restricts to e(S)
e on each fiber of P (E).
This implies that the global classes {1, x, . . . , xn−1 } restrict to genera-
tors of cohomology of each fiber by theorem 3.7. Therefore the cohomology
H ∗ (P (E)) of P (E) is a free module over H ∗ (B) with basis {1, x, . . . , xn−1 }
by Leray-Hirsch theorem 3.6.

14
 This implies that xn can be written as a linear combination of 1, x, . . . , xn−1
with coefficients in H ∗ (B) i.e. there exists {c1 (ξ), . . . , cn (ξ)} ⊂ H ∗ (B) such
that
xn + c1 (ξ)xn−1 + · · · + cn (ξ) = 0
The ci () are called the Chern classes and are characteristic classes on complex
vector bundles. Their sum
c(ξ) = 1 + c1 (ξ) + · · · + cn (ξ)
is also a characteristic class and is called the total Chern class. By definition
the higher Chern classes of a rank n vector bundle are zero i.e. ci (ξ) = 0 for
i > n.
Theorem 3.9. The first Chern class c1 (ξ) of a complex line bundle ξ =
(E, p, B) is the same as the Euler class e(ξR ) of the corresponding rank 2 real
bundle on B.
Proof. The projectivization P (E) of a line bundle is isomorphic to the base
space B, because there is exactly one line in each fiber. Similarly the universal
subbundle S of P (E) is isomorphic to the original line bundle ξ. Therefore
x = −e(S) = −e(ξR ), and x + c1 (ξ) = 0 implies that e(ξ) = c1 (ξ).
Study of the cohomology of complex Grassmannians yields the following
important result.
Theorem 3.10. Every characteristic class of complex vector bundles over a
paracompact manifold can be written as a polynomial in the Chern classes.

References
[1] Bott, Raoul, and Loring W. Tu. Differential Forms in Algebraic Topology.
New York: Springer-Verlag, 1982.
[2] Husemöller, Dale. Fibre Bundles. 3rd ed. New York: Springer-Verlag,
1994.
[3] Munkres, James. Analysis on Manifolds. Boulder, CO: Westview Press,
1991.
[4] Novikov, Sergei Petrovich. Topology II. Berlin; New York: Springer-
Verlag, 1996.

15