LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS

JIE WU
Contents
1. Tangent Spaces, Vector Fields in R
n
and the Inverse Mapping Theorem 2
1.1. Tangent Space to a Level Surface 2
1.2. Tangent Space and Vectors Fields on R
n
2
1.3. Operator Representations of Vector Fields 3
1.4. Integral Curves 5
1.5. Implicit- and Inverse-Mapping Theorems 6
2. Topological and Dierentiable Manifolds, Dieomorphisms, Immersions, Submersions
and Submanifolds 9
2.1. Topological Spaces 9
2.2. Topological Manifolds 10
2.3. Dierentiable Manifolds 11
2.4. Tangent Space 13
2.5. Immersions 15
2.6. Submersions 17
3. Examples of Manifolds 19
3.1. Open Stiefel Manifolds and Grassmann Manifolds 19
3.2. Stiefel Manifold 21
4. Fibre Bundles and Vector Bundles 22
4.1. Fibre Bundles 22
4.2. G-Spaces and Principal G-Bundles 25
4.3. The Associated Principal G-Bundles of Fibre Bundles 27
4.4. Vector Bundles 31
4.5. The Construction of Gauss Maps 31
5. Tangent Bundles and Vector Fields 38
6. Cotangent Bundles and Tensor Fields 38
7. Orientation of Manifolds 38
8. Tensor Algebras and Exterior Algebras 38
9. DeRham Cohomology 38
10. Integration on Manifolds 38
11. Stokes Theorem 38
References 38
1
2 JIE WU
1. Tangent Spaces, Vector Fields in R
n
and the Inverse Mapping Theorem
1.1. Tangent Space to a Level Surface. Let be a curve in R
n
: : t (
1
(t),
2
(t), . . . ,
n
(t)).
(A curve can be described as a vector-valued function. Converse a vector-valued function gives a
curve in R
n
.) The tangent line at the point (t
0
) is given with the direction
d
dt
(t
0
) =

d
1
dt
(t
0
), . . . ,
d
n
t
(t
0
)

.
(Certainly we need to assume that the derivatives exist. We may talk about smooth curves, that is,
the curves with all continuous higher derivatives.)
Consider the level surface f(x
1
, x
2
, . . . , x
n
) = c of a dierentiable function f, where x
i
refers to
i-th coordinate. The gradient vector of f at a point P = (x
1
(P), x
2
(P), . . . , x
n
(P)) is
f = (
f
x
1
, . . . ,
f
x
n
).
Given a vector u = (u
1
, . . . , u
n
), the directional derivative is
D
u
f = f u =
f
x
1
u
1
+ +
f
x
n
u
n
.
The tangent space at the point P on the level surface f(x
1
, . . . , x
n
) = c is the (n 1)-dimensional
(if f = 0) space through P normal to the gradient f. In other words, the tangent space is given
by the equation
f
x
1
(P)(x
1
x
1
(P)) + +
f
x
n
(P)(x
n
x
n
(P)) = 0.
From the geometric views, the tangent space should consist of all tangents to the smooth curves
on the level surface through the point P. Assume that is a curve through P (when t = t
0
)
that lies in the level surface f(x
1
, . . . , x
n
) = c, that is
f(
1
(t),
2
(t), . . . ,
n
(t)) = c.
By taking derivatives on both sides,
f
x
1
(P)(
1
)

(t
0
) + +
f
x
n
(P)(
n
)

(t
0
) = 0
and so the tangent line of is really normal (orthogonal) to f. When runs over all possible
curves on the level surface through the point P, then we obtain the tangent space at the point P.
Roughly speaking, a tangent space is a vector space attached to a point in the surface.
How to obtain the tangent space: take all tangent lines of smooth curve through this point on the
surface.
1.2. Tangent Space and Vectors Fields on R
n
. Now consider the tangent space of R
n
. Accord-
ing to the ideas in the previous subsection, rst we assume a given point P R
n
. Then we consider
all smooth curves passes through P and then take the tangent lines from the smooth curves. The
obtained vector space at the point P is the n-dimensional space. But we can look at in a little
detail.
Let be a smooth curve through P. We may assume that (0) = P. Let be another smooth
curve with (0) = P. is called to be equivalent to if the directives

(0) =

(0). The tangent

space of R
n
at P, denoted by T
P
(R
n
), is then the set of equivalence class of all smooth curves
through P.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 3
Let T(R
n
) =

PR
n
T
P
(R
n
), called the tangent bundle of R
n
. If S is a region of R
n
, let T(S) =

PS
T
P
(S), called the tangent bundle of S.
Note. Each T
P
(R
n
) is an n-dimensional vector space, but T(S) is not a vector space. In other
words, T(S) is obtained by attaching a vector space T
P
(R
n
) to each point P in S. Also S is assumed
to be a region of R
n
, otherwise the tangent space of S (for instance S is a level surface) could be a
proper subspace of T
P
(R
n
).
If is a smooth curve from P to Q in R
n
, then the tangent space T
P
(R
n
) moves along to
T
Q
(R
n
). The direction for this moving is given

(t), which introduces the following important

concept.
Denition 1.1. A vector eld V on a region S of R
n
is a smooth map (also called C

-map)
V : S T(S) P v(P).
Let V : P v(P) and W : P w(P) be two vector elds and let f : S R be a smooth
function. Then V + W : P v(P) + w(P) and fV : P f(P)v(P) give (pointwise) addition and
scalar multiplication structure on vector elds.
1.3. Operator Representations of Vector Fields. Let J be an open interval containing 0 and
let : J R
n
be a smooth curve with (0) = P. Let f = f(x
1
, . . . , x
n
) be a smooth function
dened on a neighborhood of P. Assume that the range of is contained in the domain of f. By
applying the chain rule to the composite T = f : J R,
D

(f) :=
dT
dt
=
n

i=1
d
i
(t)
dt
f
x
i

x
i
=
i
(t)
Proposition 1.2.
D

(af +bg) = aD

(f) +bD

(g), where a, b are constant.

D

(fg) = D

(f)g +fD

(g).
Let C

(R
n
) denote the set of smooth functions on R
n
. An operation D on C

(R
n
) is called a
derivation if D maps C

(R
n
) to C

(R
n
) and satises the conditions
D(af +bg) = aD(f) +bD(g), where a, b are constant.
D(fg) = D(f)g +fD(g).
Example: For 1 i n,

i
: f
f
x
i
is a derivation.
Proposition 1.3. Let D be any derivation on C

(R
n
). Given any point P in R
n
. Then there
exist real numbers a
1
, a
2
, . . . , a
n
R such that
D(f)(P) =
n

i=1
a
i

i
(f)(P)
for any f C

(R
n
), where a
i
depends on D and P but is independent on f.
4 JIE WU
Proof. Write x for (x
1
, . . . , x
n
). Dene
g
i
(x) =

1
0
f
x
i
(t(x P) +P)dt.
Then
f(x) f(P) =

1
0
d
dt
f(t(x P) +P)dt
=

1
0
n

i=1
f
x
i
(t(x P) +P) (x
i
x
i
(P))dt
=
n

i=1
(x
i
x
i
(P))

1
0
f
x
i
(t(x P) +P)dt =
n

i=1
(x
i
x
i
(P))g
i
(x).
Since D is a derivation, D(1) = D(1 1) = D(1) 1 + 1 D(1) and so D(1) = 0. It follows that
D(c) = 0 for any constant c. By applying D to the above equations,
D(f(x)) = D(f(x) f(P)) =
n

i=1
D(x
i
x
i
(P))g
i
(x) + (x
i
x
i
(P))D(g
i
(x))
=
n

i=1
D(x
i
)g
i
(x) + (x
i
x
i
(P))D(g
i
(x))
because D(f(P)) = D(x
i
(P)) = 0. Let a
i
= D(x
i
)(P) which only depends on D and P. By
evaluating at P,
D(f)(P) =
n

i=1
D(x
i
)(P)g
i
(P) + 0 =
n

i=1
a
i
g
i
(P).
Since
g
i
(P) =

1
0
f
x
i
(t(P P) +P)dt =

1
0
f
x
i
(P)dt =
f
x
i
(P) =
i
(f)(P),
D(f)(P) =
n

i=1
a
i

i
(f)(P),
which is the conclusion.
From this proposition, we can give a new way to looking at vector elds:
Given a vector elds P v(P) = (v
1
(P), v
2
(P), . . . , v
n
(P)), a derivation
D
v
=
n

i=1
v
i
(P)
i
on C

(R
n
) is called an operator representation of the vector eld P v(P).
Note. The operation v
i
(x)
i
is given as follows: for any f C

(R
n
),
D
v
(f)(P) =
n

i=1
v
i
(P)
i
(f)(P)
for any P.
From this new view, the tangent spaces T(R
n
) admits a basis
1
,
2
, . . . ,
n
.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 5
1.4. Integral Curves. Let V : x v(x) be a (smooth) vector eld on an neighborhood U of P.
An integral curve to V is a smooth curve s: (, ) U, dened for suitable , > 0, such that
s

(t) = v(s(t))
for < t < .
Theorem 1.4. Let V : x v(x) be a (smooth) vector eld on an neighborhood U of P. Then there
exists an integral curve to V through P. Any two such curves agree on their common domain.
Proof. The proof is given by assuming the fundamental existence and uniqueness theorem for sys-
tems of rst order dierential equations.
The requirement for a curve s(t) = (s
1
(t), . . . , s
n
(t)) to be an integral curve is:

ds
1
(t)
dt
= v
1
(s
1
(t), s
2
(t), . . . , s
n
(t))
ds
2
(t)
dt
= v
2
(s
1
(t), s
2
(t), . . . , s
n
(t))

ds
n
(t)
dt
= v
n
(s
1
(t), s
2
(t), . . . , s
n
(t))
with the initial conditions
s(0) = P (s
1
(0), s
2
(0), . . . , s
n
(0)) = (x
1
(P), x
2
(P), . . . , x
n
(P))
s

(0) = v(P)

ds
1
dt
(0), . . . ,
ds
n
dt
(0)

= (v
1
(P), . . . , v
n
(P)).
Thus the statement follows from the fundamental theorem of rst order ODE.
Example 1.5. Let n = 2 and let V : P v(P) = (v
1
(P), v
2
(P)), where v
1
(x, y) = x and v
2
(x, y) =
y. Given a point P = (a
1
, a
2
), the equation for the integral curve s(t) = (x(t), y(t)) is

(t) = v
1
(s(t)) = x(t)
y

(t) = v
2
(s(t)) = y(t)
with initial conditions (x(0), y(0)) = (a
1
, a
2
) and (x

(0), y

(0)) = v(a
1
, a
2
) = (a
1
, a
2
). Thus the
solution is
s(t) = (a
1
e
t
, a
2
e
t
).
Example 1.6. Let n = 2 and let V : P v(P) = (v
1
(P), v
2
(P)), where v
1
(x, y) = x and v
2
(x, y) =
y. Given a point P = (a
1
, a
2
), the equation for the integral curve s(t) = (x(t), y(t)) is

(t) = v
1
(s(t)) = x(t)
y

(t) = v
2
(s(t)) = y(t)
with initial conditions (x(0), y(0)) = (a
1
, a
2
) and (x

(0), y

(0)) = v(a
1
, a
2
) = (a
1
, a
2
). Thus the
solution is
s(t) = (a
1
e
t
, a
2
e
t
).
6 JIE WU
1.5. Implicit- and Inverse-Mapping Theorems.
Theorem 1.7. Let D be an open region in R
n+1
and let F be a function well-dened on D with
continuous partial derivatives. Let (x
1
0
, x
2
0
, . . . , x
n
0
, z
0
) be a point in D where
F(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
) = 0
F
z
(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
) = 0.
Then there is a neighborhood N

(z
0
) R, a neighborhood N

(x
1
0
, . . . , x
n
0
) R
n
, and a unique
function z = g(x
1
, x
2
, . . . , x
n
) dened for (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
) with values z N

(z
0
) such
that
1) z
0
= g(x
1
0
, x
2
0
, . . . , x
n
0
) and
F(x
1
, x
2
, . . . , x
n
, g(x
1
, . . . , x
n
)) = 0
for all (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
).
2) g has continuous partial derivatives with
g
x
i
(x
1
, . . . , x
n
) =
F
x
i (x
1
, . . . , x
n
, z)
F
z
(x
1
, . . . , x
n
, z)
for all (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
) where z = g(x
1
, . . . , x
n
).
3) If F is smooth on D, then z = g(x
1
, . . . , x
n
) is smooth on N

(x
1
0
, . . . , x
n
0
).
Proof. Step 1. We may assume that
F
z
(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
) > 0. Since F
z
is continuous, there
exists a neighborhood N

(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
) in which F
z
is continuous and positive. Thus for xed
(x
1
, . . . , x
n
), F is strictly increasing on z in this neighborhood. It follows that there exists c > 0
such that
F(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
c) < 0 F(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
+c) > 0
with
(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
c), (x
1
0
, x
2
0
, . . . , x
n
0
, z
0
+c) N

(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
).
Step 2. By the continuity of F, there exists a small > 0 such that
F(x
1
, x
2
, . . . , x
n
, z
0
c) < 0 F(x
1
, x
2
, . . . , x
n
, z
0
+c) > 0
with
(x
1
, x
2
, . . . , x
n
, z
0
c), (x
1
, x
2
, . . . , x
n
, z
0
+c) N

(x
1
0
, x
2
0
, . . . , x
n
0
, z
0
)
for (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
).
Step 3. Fixed (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
), F is continuous and strictly increasing on z. There is
a unique z, z
0
c < z < z
0
+c, such that
F(x
1
, . . . , x
n
, z) = 0.
This denes a function z = g(x
1
, . . . , x
n
) for (x
1
, . . . , x
n
) N

(x
1
0
, . . . , x
n
0
) with values z (z
0

c, z
0
+c).
Step 4. Prove that z = g(x
1
, . . . , x
n
) is continuous. Let (x
1
1
, . . . , x
n
1
) N

(x
1
0
, . . . , x
n
0
). Let
(x
1
1
(k), . . . , x
n
1
(k)) be any sequence in N

(x
1
0
, . . . , x
n
0
) converging to (x
1
1
, . . . , x
n
1
). Let A be any
subsequential limit of z
k
= g(x
1
1
(k), . . . , x
n
1
(k)), that is A = lim
s
z
k
s
. Then, by the continuity of
F,
0 = lim
s
F(x
1
1
(k
s
), . . . , x
n
1
(k
s
), z
k
s
)
= F( lim
s
x
1
1
(k
s
), . . . , lim
s
x
n
1
(k
s
), lim
s
z
k
s
)
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 7
= F(x
1
1
, . . . , x
n
1
, A).
By the unique solution of the equation, A = g(x
1
1
, . . . , x
n
1
). Thus z
k
converges g(x
1
1
, . . . , x
n
1
) and
so g is continuous.
Step 5. Compute the partial derivatives
z
x
i
. Let h be small enough. Let
z +k = g(x
1
, . . . , x
i1
, x
i
+h, x
i+1
, . . . , x
n
),
that is
F(x
1
, . . . , x
i
+h, . . . , x
n
, z +k) = 0
with z
0
c < z +k < z
0
+c. Then
0 = F(x
1
, . . . , x
i
+h, . . . , x
n
, z +k) F(x
1
, . . . , x
n
, z)
= F
x
i (x
1
, . . . , x
i
, . . . , x
n
, z)h +F
z
(x
1
, . . . , x
i
, . . . , x
n
, z)k
by the mean value theorem (Consider the function
(t) = F(x
1
, . . . , x
i
+th, . . . , x
n
, z +tk)
for 0 t 1. Then (1) (0) =

()(10).), where x
i
is between x
i
and x
i
+h, and z is between
z and z +k. Now
g
x
i
= lim
h0
g(x
1
, . . . , x
i1
, x
i
+h, x
i+1
, . . . , x
n
) z
h
= lim
h0
k
h
= lim
h0
F
x
i (x
1
, . . . , x
i
, . . . , x
n
, z)
F
z
(x
1
, . . . , x
i
, . . . , x
n
, z)
. =
F
x
i
F
z
,
where z z as h 0 because g is continuous (and so k 0 as h 0).
Step 6. Since F
z
is not zero in this small neighborhood, g
x
i
is continuous for each i. If F is smooth,
then all higher derivatives of g are continuous and so g is also smooth.
Theorem 1.8 (Implicit Function Theorem). Let D be an open region in R
m+n
and let F
1
, F
2
, . . . , F
n
be functions well-dened on D with continuous partial derivatives. Let (x
1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
)
be a point in D where

F
1
(x
1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
) = 0
F
2
(x
1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
) = 0

F
n
(x
1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
) = 0
and the Jacobian
J =
(F
1
, F
2
, . . . , F
n
)
(u
1
, u
2
, . . . , u
n
)
= det

F
i
u
j

= 0
at the point (x
1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
). Then there are neighborhoods N

(x
1
0
, . . . , x
m
0
), N

1
(u
1
0
),
N

2
(u
2
0
), . . ., N

n
(u
n
0
), and unique functions

u
1
= g
1
(x
1
, x
2
, . . . , x
m
)
u
2
= g
2
(x
1
, x
2
, . . . , x
m
)

u
n
= g
n
(x
1
, x
2
, . . . , x
m
)
dened for (x
1
, . . . , x
m
) N

(x
1
0
, . . . , x
m
0
) with values u
1
N

1
(u
1
0
), . . . , u
n
N

n
(u
n
0
) such that
8 JIE WU
1) u
i
0
= g
i
(x
1
0
, x
2
0
, . . . , x
m
0
) and
F
i
(x
1
, x
2
, . . . , x
n
, g
i
(x
1
, . . . , x
m
)) = 0
for all 1 i n and all (x
1
, . . . , x
m
) N

(x
1
0
, . . . , x
m
0
).
2) Each g
i
has continuous partial derivatives with
g
i
x
j
(x
1
, . . . , x
m
) =
1
J

(F
1
, . . . , F
n
)
(u
1
, u
2
, . . . , u
j1
, x
j
, u
j+1
, . . . , u
n
)
for all (x
1
, . . . , x
m
) N

(x
1
0
, . . . , x
m
0
) where u
i
= g
i
(x
1
, . . . , x
m
).
3) If each F
i
is smooth on D, then each u
i
= g
i
(x
1
, . . . , x
m
) is smooth on N

(x
1
0
, . . . , x
m
0
).
Sketch of Proof. The proof is given by induction on n. Assume that the statement holds for n 1
with n > 1. (We already prove that the statement holds for n = 1.) Since the matrix

F
i
u
j

is invertible at the point P = (x

1
0
, x
2
0
, . . . , x
m
0
, u
1
0
, u
2
0
, . . . , u
n
0
) (because the determinant is not zero),
we may assume that
F
n
u
n
(P) = 0.
(The entries in the last column can not be all 0 and so, if
F
i
u
n
(P) = 0, we can interchange F
i
and
F
n
.)
From the previous theorem, there is a solution
u
n
= g
n
(x
1
, . . . , x
m
, u
1
, . . . , u
n1
)
to the last equation. Consider

G
1
= F
1
(x
1
, . . . , x
m
, u
1
, . . . , u
n1
, g
n
)
G
2
= F
2
(x
1
, . . . , x
m
, u
1
, . . . , u
n1
, g
n
)

G
n1
= F
n1
(x
1
, . . . , x
m
, u
1
, . . . , u
n1
, g
n
).
Then
G
i
u
j
=
F
i
u
j
+
F
i
u
n

g
n
u
j
for 1 i, j n 1, where
F
n
u
j
+
F
n
u
n

g
n
u
j
= 0.
Let
B =

1 0 0 0 0
0 1 0 0 0
0 0 1 0 0

0 0 0 1 0
g
n
u
1
g
n
u
2
g
n
u
3

g
n
u
n1
1

Then

F
i
u
j

B =

G
i
u
j

n1,n1

0
F
n
u
n

.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 9
By taking the determinant,
J =
(F
1
, . . . , F
n
)
(u
1
, . . . , u
n
)
=
F
n
u
n

(G
1
, . . . , G
n1
)
(u
1
, . . . , u
n1
)
.
Thus
(G
1
,...,G
n1
)
(u
1
,...,u
n1
)
= 0 at P and, by induction, there are solutions
u
i
= g
i
(x
1
, . . . , x
m
)
for 1 i n 1.
Theorem 1.9 (Inverse Mapping Theorem). Let D be an open region in R
n
. Let

x
1
= f
1
(u
1
, . . . , u
n
)
x
2
= f
2
(u
1
, . . . , u
n
)

x
n
= f
n
(u
1
, . . . , u
n
)
be functions dened on D with continuous partial derivatives. Let (u
1
0
, . . . , u
n
0
) D satisfy x
i
0
=
f
i
(u
1
0
, . . . , u
n
0
) and the Jacobian
(x
1
, . . . , x
n
)
(u
1
, . . . , u
n
)
= 0 at (u
1
0
, . . . , u
n
0
).
Then there are neighborhood N

(x
1
0
, . . . , x
n
0
) and N

(u
1
0
, . . . , u
n
0
) such that

u
1
= f
1
1
(x
1
, . . . , x
n
)
u
2
= f
1
2
(x
1
, . . . , x
n
)

u
n
= f
1
n
(x
1
, . . . , x
n
)
is well-dened and has continuous partial derivatives on N

(x
1
0
, . . . , x
n
0
) with values in N

(u
1
0
, . . . , u
n
0
).
Moreover if each f
i
is smooth, then each f
1
i
is smooth.
Proof. Let F
i
= f
i
(u
1
, . . . , u
n
) x
i
. The assertion follows from the Implicit Function Theorem.
2. Topological and Differentiable Manifolds, Diffeomorphisms, Immersions,
Submersions and Submanifolds
2.1. Topological Spaces.
Denition 2.1. Let X be a set. A topology | for X is a collection of subsets of X satisfying
i) and X are in |;
ii) the intersection of two members of | is in |;
iii) the union of any number of members of | is in |.
The set X with | is called a topological space. The members U | are called the open sets.
Let X be a topological space. A subset N X with x N is called a neighborhood of x if there
is an open set U with x U N. For example, if X is a metric space, then the closed ball D

(x)
and the open ball B

(x) are neighborhoods of x. A subset C is said to closed if X ` C is open.

Denition 2.2. A function f : X Y between two topological spaces is said to be continuous if
for every open set U of Y the pre-image f
1
(U) is open in X.
10 JIE WU
A continuous function from a topological space to a topological space is often simply called a
map. A space means a Hausdor space, that is, a topological spaces where any two points has
disjoint neighborhoods.
Denition 2.3. Let X and Y be topological spaces. We say that X and Y are homeomorphic if
there exist continuous functions f : X Y, g : Y X such that f g = id
Y
and g f = id
X
. We
write X

= Y and say that f and g are homeomorphisms between X and Y .
By the denition, a function f : X Y is a homeomorphism if and only if
i) f is a bijective;
ii) f is continuous and
iii) f
1
is also continuous.
Equivalently f is a homeomorphism if and only if 1) f is a bijective, 2) f is continuous and 3) f is
an open map, that is f sends open sets to open sets. Thus a homeomorphism between X and Y is
a bijective between the points and the open sets of X and Y .
A very general question in topology is how to classify topological spaces under homeomorphisms.
For example, we know (from complex analysis and others) that any simple closed loop is homeo-
morphic to the unit circle S
1
. Roughly speaking topological classication of curves is known. The
topological classication of (two-dimensional) surfaces is known as well. However the topological
classication of 3-dimensional manifolds (we will learn manifolds later.) is quite open.
The famous Poicare conjecture is related to this problem, which states that any simply connected
3-dimensional (topological) manifold is homeomorphic to the 3-sphere S
3
. A space X is called simply
connected if (1) X is path-connected (that is, given any two points, there is a continuous path joining
them) and (2) the fundamental group
1
(X) is trivial (roughly speaking, any loop can be deformed
to be the constant loop in X). The manifolds are the objects that we are going to discuss in this
course.
2.2. Topological Manifolds. A Hausdor space M is called a (topological) n-manifold if each
point of M has a neighborhood homeomorphic to an open set in R
n
. Roughly speaking, an n-
manifold is locally R
n
. Sometimes M is denoted as M
n
for mentioning the dimension of M.
(Note. If you are not familiar with topological spaces, you just think that M is a subspace of
R
N
for a large N.)
For example, R
n
and the n-sphere S
n
is an n-manifold. A 2-dimensional manifold is called
a surface. The objects traditionally called surfaces in 3-space can be made into manifolds in
a standard way. The compact surfaces have been classied as spheres or projective planes with
various numbers of handles attached.
By the denition of manifold, the closed n-disk D
n
is not an n-manifold because it has the
boundary S
n1
. D
n
is an example of manifolds with boundary. We give the denition of
manifold with boundary as follows.
A Hausdor space M is called an n-manifold with boundary (n 1) if each point in M has a
neighborhood homeomorphic to an open set in the half space
R
n
+
= (x
1
, , x
n
) R
n
[x
n
0.
Manifold is one of models that we can do calculus locally. By means of calculus, we need local
coordinate systems. Let x M. By the denition, there is a an open neighborhood U(x) of x and
a homeomorphism
x
from U(x) onto an open set in R
n
+
. The collection (U(x),
x
)[x M has
the property that 1) U(x)[x M is an open cover and 2)
x
is a homeomorphism from U(x)
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 11
onto an open set in R
n
+
. The subspace
x
(U
x
) in R
n
+
plays a role as a local coordinate system. The
collection (U(x),
x
)[x M is somewhat too large and we may like less local coordinate systems.
This can be done as follows.
Let M be a space. A chart of M is a pair (U, ) such that 1) U is an open set in M and 2) is
a homeomorphism from U onto an open set in R
n
+
. The map
: U R
n
+
can be given by n coordinate functions
1
, . . . ,
n
. If P denotes a point of U, these functions are
often written as
x
1
(P), x
2
(P), . . . , x
n
(P)
or simply x
1
, x
2
, . . . , x
n
. They are called local coordinates on the manifold.
An atlas for M means a collection of charts (U

)[ J such that U

[ J is an open
cover of M.
Proposition 2.4. A Hausdor space M is a manifold (with boundary) if and only if M has an
atlas.
Proof. Suppose that M is a manifold. Then the collection (U(x),
x
)[x M is an atlas. Con-
versely suppose that M has an atlas. For any x M there exists such that x U

and so U

is
an open neighborhood of x that is homeomorphic to an open set in R
n
+
. Thus M is a manifold.
We dene a subset M as follows: x M if there is a chart (U

) such that x U

and

(x) R
n1
= x R
n
[x
n
= 0. M is called the boundary of M. For example the boundary of
D
n
is S
n1
.
Proposition 2.5. Let M be a n-manifold with boundary. Then M is an (n1)-manifold without
boundary.
Proof. Let (U

)[ J be an atlas for M. Let J

J be the set of indices such that

U

M = if J

. Then Clearly
(U

M,

[
U

M
[ J

can be made into an atlas for M.

Note. The key point here is that if U is open in R
n
+
, then U R
n1
is also open because: Since U
is open in R
n
+
, there is an open subset V of R
n
such that U = V R
n
+
. Now if x U R
n1
, there
is an open disk E

(x) V and so
E

(x) R
n1
V R
n1
= U R
n1
is an open (n 1)-dimensional -disk in R
n1
centered at x.
2.3. Dierentiable Manifolds.
Denition 2.6. A Hausdor space M is called a dierential manifold of class C
k
(with boundary)
if there is an atlas of M
(U

)[ J
such that
12 JIE WU
For any , J, the composites

(U

) R
n
+
is dierentiable of class C
k
.
The atlas (U

[ J is called a dierential atlas of class C

k
on M.
(Note. Assume that M is a subspace of R
N
with N >> 0. If M has an atlas (U

)[ J
such that each

: U

R
n
+
is dierentiable of class C
k
, then M is a dierentiable manifold of
class C
k
. This is the denition of dierentiable (smooth) manifolds in [6] as in the beginning they
already assume that M is a subspace of R
N
with N large. In our denition (the usual denition
of dierentiable manifolds using charts), we only assume that M is a (Hausdor) topological space
and so

is only an identication of an abstract U

with an open subset of R

n
+
. In this case we can
not talk dierentiability of

unless U

is regarded as a subspace of a (large dimensional) Euclidian

space.)
Two dierential atlases of class C
k
(U

)[ I and (V

)[ J are called equivalent

if
(U

)[ I (V

)[ J
is again a dierential atlas of class C
k
(this is an equivalence relation). A dierential structure of
class C
k
on M is an equivalence class of dierential atlases of class C
k
on M. Thus a dierential
manifold of class C
k
means a manifold with a dierential structure of class C
k
. A smooth manifold
means a dierential manifold of class C

.
Note: A general manifold is also called topological manifold. Kervaire and Milnor [4] have shown
that the topological sphere S
7
has 28 distinct oriented smooth structures.
Denition 2.7. let M and N be smooth manifolds (with boundary) of dimensions m and n re-
spectively. A map f : M N is called smooth if for some smooth atlases (U

[ I for M
and (V

) J for N the functions

f
1

(f
1
(V

)U

)
:

(f
1
(V

) U

) R
n
+
are of class C

.
Proposition 2.8. If f : M N is smooth with respect to atlases
(U

[ I, (V

[ J
for M, N then it is smooth with respect to equivalent atlases
(U

[ I

, (V

[ J

Proof. Since f is smooth with respect with the atlases

(U

[ I, (V

[ J,
f is smooth with respect to the smooth atlases
(U

[ I (U

[ I

, (V

[ J (V

[ J

by look at the local coordinate systems. Thus f is smooth with respect to the atlases
(U

[ I

, (V

[ J

LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 13

Thus the denition of smooth maps between two smooth manifolds is independent of choice of
atlas.
Denition 2.9. A smooth map f : M N is called a dieomorphism if f is one-to-one and onto,
and if the inverse f
1
: N M is also smooth.
Denition 2.10. Let M be a smooth n-manifold, possibly with boundary. A subset X is called a
properly embedded submanifold of dimension k n if X is a closed in M and, for each P X, there
exists a chart (U, ) about P in M such that
(U X) = (U) R
k
+
,
where R
k
+
R
n
+
is the standard inclusion.
Note. In the above denition, the collection (U X, [
UX
) is an atlas for making X to a smooth
k-manifold with boundary X = X M.
If M = , by dropping the requirement that X is a closed subset but keeping the requirement
on local charts, X is called simply a submanifold of M.
2.4. Tangent Space. Let S be an open region of R
n
. Recall that, for P S, the tangent space
T
P
(S) is just the n-dimensional vector space by putting the origin at P. Let T be an open region
of R
m
and let f = (f
1
, . . . , f
m
): S T be a smooth map. Then f induces a linear transformation
Tf : T
P
(S) T
f(P)
(T)
given by
Tf(v) =

f
i
x
j

mn

v
1
v
2

v
n

n1
=

v
1

1
(f
1
) +v
2

2
(f
1
) + +v
n

n
(f
1
)
v
1

1
(f
2
) +v
2

2
(f
2
) + +v
n

n
(f
2
)

v
1

1
(f
m
) +v
2

2
(f
m
) + +v
n

n
(f
m
)

,
namely Tf is obtained by taking directional derivatives of (f
1
, . . . , f
m
) along vector v for any
v T
P
(S).
Now we are going to dene the tangent space to a (dierentiable) manifold M at a point P as
follows:
First we consider the set
T
P
= (U, , v) [ P U, (U, ) is a chart v T((P))((U)).
The point is that there are possibly many charts around P. Each chart creates an n-dimension vector
space. So we need to dene an equivalence relation in T
P
such that, T
P
modulo these relations is
only one copy of n-dimensional vector space which is also independent on the choice of charts.
Let (U, , v) and (V, , w) be two elements in T
P
. That is (U, ) and (V, ) are two charts with
P U and P V . By the denition,

1
: (U V )
-
(U V )
is dieomorphism and so it induces an isomorphism of vector spaces
T(
1
): T
(P)
((U V ))
-
T
(P)
((U V )).
Now (U, , v) is called equivalent to (V, , w), denoted by (U, , v) (V, , w), if
T(
1
)(v) = w.
14 JIE WU
Dene T
P
(M) to be the quotient
T
P
(M) = T
P
/ .
Exercise 2.1. Let M be a dierentiable n-manifold and let P be any point in M. Prove that
T
P
(M) is an n-dimensional vector space. [Hint: Fixed a chart (U, ) and dened
a(U, , v) +b(U, , w) := (U, , av +b w).
Now given any (V, , x), (

V ,

, y) T
P
, consider the map

1
: (U V ) (U V )

1
:

(U

V ) (U

V )
and dene
a(V, , x) +b(

V ,

, y) = (U, , aT(
1
)(x) +bT(

1
)(y)).
Then prove that this operation gives a well-dened vector space structure on T
P
, that is, independent
on the equivalence relation.]
The tangent space T
P
(M), as a vector space, can be described as follows: given any chart (U, )
with P U, there is a unique isomorphism
T

: T
P
(M) T
(P)
((U)).
by choosing (U, , v) as representatives for its equivalence class. If (V, ) is another chart with
P V , then there is a commutative diagram
(1)
T
P
(M)
T

=
-
T
(P)
((U V ))
T
P
(M)

=
-
T
(P)
((U V )),
T(
1
)
?
where T(
1
) is the linear isomorphism induced by the Jacobian matrix of the dierentiable
map
1
: (U V ) (U V ).
Exercise 2.2. Let f : M N be a smooth map, where M and N need not to have the same
dimension. Prove that there is a unique linear transformation
Tf : T
P
(M)
-
T
f(P)
(N)
such that the diagram
T
P
(M)
T

=
-
T
(P)
((U))
T
f(P)
(N)
Tf
?
T

=
-
T
(f(P))
((V ))
T( f
1
)
?
commutes for any chart (U, ) with P U and any chart (V, ) with f(P) V . [First x a choice
of (U, ) with P U and (V, ) with f(P) V , the linear transformation Tf is uniquely dened
by the above diagram. Then use Diagram (1) to check that Tf is independent on choices of charts.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 15
2.5. Immersions. A smooth map f : M N is called immersion at P if the linear transformation
Tf : T
P
(M) T
f(P)
(M)
is injective.
Theorem 2.11 (Local Immersion Theorem). Suppose that f : M
m
N
n
is immersion at P. Then
there exist charts (U, ) about P and (V, ) about f(P) such that the diagram
U
f[
U
-
V
R
m
(P) = 0
?

canonical coordinate inclusion

-
R
n
(f(P)) = 0
?
commutes.
Proof. We may assume that (P) = 0 and (f(P)) = 0. (Otherwise replacing and by (P)
and (f(P)), respectively.)
Consider the commutative diagram
U
f[
U
-
V
(U)

?
g = f
1
-
(V )

?
R
m
?

R
n
.
?

By the assumption,
Tg : T
0
((U))
-
T
0
((V ))
is an injective linear transformation and so
rank(Tg) = m
16 JIE WU
at the origin. The matric for Tg is
(2)

g
1
x
1
g
1
x
2

g
1
x
m
g
2
x
1
g
2
x
2

g
2
x
m

g
m
x
1
g
m
x
2

g
m
x
m

g
n
x
1
g
n
x
2

g
n
x
m

.
By changing basis of R
n
(corresponding to change the rows), we may assume that the rst m rows
forms an invertible matrix A
mm
at the origin.
Dene a function
h = (h
1
, h
2
, . . . , h
n
): (U) R
nm
-
R
n
by setting
h
i
(x
1
, . . . , x
m
, x
m+1
, . . . , x
n
) = g
i
(x
1
, . . . , x
m
)
for 1 i m and
h
i
(x
1
, . . . , x
m
, x
m+1
, . . . , x
n
) = g
i
(x
1
, . . . , x
m
) +x
i
for m+ 1 i n. Then Jacobian matrix of h is

A
mm
0
m(nm)
B
(nm)m
I
nm

,
where B is taken from (m+ 1)-st row to n-th row in the matrix (2). Thus the Jacobian of h is not
zero at the origin. By the Inverse Mapping Theorem, h is an dieomorphism in a small neighborhood
of the origin. It follows that there exist open neighborhoods

U U of P and

V V of f(P) such
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 17
that the following diagram commutes

U
f[

U
-
V
(

U)
[

U
?
g = f
1
-
(

V )
[

V
?
(

U) 0

-
(

U) U
2

= h
1
?
R
m
= R
m
0
?

-
R
n
,
?

where U
2
is a small neighborhood of the origin in R
nm
.
Theorem 2.12. Let f : M N be a smooth map. Suppose that
1) f is immersion at every point P M,
2) f is one-to-one and
3) f : M f(M) is a homeomorphism.
Then f(M) is a smooth submanifold of M and f : M f(M) is a dieomorphism.
Note. In Condition 3, we need that if U is an open subset of M, then there is an open subset V of
N such that V f(M) = f(U).
Proof. For any point P in M, we can choose the charts as in Theorem 2.11. By Condition 3, f(U)
is an open subset of f(M). The charts (f(U), [
f(U)
) gives an atlas for f(M) such that f(M) is
a submanifold of M. Now f : M f(M) is a dieomorphism because it is locally dieomorphism
and the inverse exists.
Condition 3 is important in this theorem, namely an injective immersion need not give a dif-
feomorphism with its image. (Construct an example for this.) An injective immersion satisfying
condition 3 is called an embedding.
2.6. Submersions. A smooth map f : M N is called submersion at P if the linear transforma-
tion
Tf : T
P
(M) T
f(P)
(M)
is surjective.
18 JIE WU
Theorem 2.13 (Local Submersion Theorem). Suppose that f : M
m
N
n
is submersion at P.
Then there exist charts (U, ) about P and (V, ) about f(P) such that the diagram
U
f[
U
-
V
R
m
(P) = 0
?

canonical coordinate proj.

-
R
n
(f(P)) = 0
?
commutes.
For a smooth map of manifolds f : M N, a point Q N is called regular if Tf : T
P
(M)
T
Q
(N) is surjective for every P f
1
(Q), the pre-image of Q.
Theorem 2.14 (Pre-image Theorem). Let f : M N be a smooth map and let Q N such
that f
1
(Q) is not empty. Suppose that Q is regular. Then f
1
(Q) is a submanifold of M with
dimf
1
(Q) = dimM dimN.
Proof. From the above theorem, for any P f
1
(Q),
[
f
1
(Q)
: f
1
(Q) U
-
R
mn
gives a chart about P.
Let Z be a submanifold of N. A smooth map f : M N is said to be transversal to Z if
Im(Tf : T
P
(M) T
f(P)
(N)) +T
f(P)
(Z) = T
f(P)
(N)
for every x f
1
(Z).
Theorem 2.15. If a smooth map f : M N is transversal to a submanifold Z N, then f
1
(Z)
is a submanifold of M. Moreover the codimension of f
1
(Z) in M equals to the codimension of Z
in N.
Proof. Given P f
1
(Z), since Z is a submanifold, there is a chart (V, ) of N about f(P) such
that V = V
1
V
2
with V
1
= V Z and (V
1
, [
V
1
) is a chart of Z about f(P). By the assumption,
the composite
f
1
(V )
f|
f
1
(V )
-
V
proj.
-
V
2
is submersion. By the Pre-image Theorem, f
1
(V ) f
1
(Z) is a submanifold of the open subset
f
1
(V ) of M and so there is a chart about P such that Z is a submanifold of M.
With respect to the assertion about the codimensions,
codim(f
1
(Z)) = dimV
2
= codim(Z).

Consider the special case that both M and Z are submanifolds of N. Then the transversal
condition is
T
P
(M) +T
P
(Z) = T
P
(N)
for any P M Z.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 19
Corollary 2.16. The intersection of two transversal submanifolds of N is again a submanifold.
Moreover
codim(M Z) = codim(M) + codim(Z)
in N.
3. Examples of Manifolds
3.1. Open Stiefel Manifolds and Grassmann Manifolds. The open Stiefel manifold is the
space of k-tuples of linearly independent vectors in R
n
:

V
k,n
= (v
1
, . . . , v
k
)
T
[ v
i
R
n
, v
1
, . . . , v
k
linearly independent,
where

V
k,n
is considered as the subspace of k n matrixes M(k, n)

= R
kn
. Since

V
k,n
is an open
subset of M(k, n) = R
kn
,

V
k,n
is an open submanifold of R
kn
.
The Grassmann manifold G
k,n
is the set of k-dimensional subspaces of R
n
, that is, all k-planes
through the origin. Let
:

V
k,n
G
k,n
be the quotient by sending k-tuples of linearly independent vectors to the k-planes spanned by k
vectors. The topology in G
k,n
is given by quotient topology of , namely, U is an open set of G
k,n
if and only if
1
(U) is open in

V
k,n
.
For (v
1
, . . . , v
k
)
T

V
k,n
, write 'v
1
, . . . , v
k
` for the k-plane spanned by v
1
, . . . , v
k
. Observe that
two k-tuples (v
1
, . . . , v
k
)
T
and ( w
1
, . . . , w
k
)
T
spanned the same k-plane if and only if each of them
is basis for the common plane, if and only if there is nonsingular k k matrix P such that
P(v
1
, . . . , v
k
)
T
= ( w
1
, . . . , w
k
)
T
.
This gives the identication rule for the Grassmann manifold G
k,n
. Let GL
k
(R) be the space of
general linear groups on R
k
, that is, GL
k
(R) consists of k k nonsingular matrixes, which is an
open subset of M(k, k) = R
k
2
. Then G
k,n
is the quotient of

V
k,n
by the action of GL
k
(R).
First we prove that G
k,n
is Hausdor. If k = n, then G
n,n
is only one point. So we assume that
k < n. Given an k-plane X and w R
n
, let
w
be the square of the Euclidian distance from w to
X. Let e
1
, . . . , e
k
be the orthogonal basis for X, then

w
(X) = w w
k

j=1
( w e
j
)
2
.
Fixing any w R
n
, we obtain the continuous map

w
: G
k,n
-
R
because
w
:

V
k,n
R is continuous and G
k,n
given by the quotient topology. (Here we use
the property of quotient topology that any function f from the quotient space G
k,n
to any space
is continuous if and only if f from

V
k,n
to that space is continuous.) Given any two distinct
points X and Y in G
k,n
, we can choose a w such that
w
(X) =
w
(Y ). Let V
1
and V
2
be disjoint
open subsets of R such that
w
(X) V
1
and
w
(Y ) V
2
. Then
1
w
(V
1
) and
1
w
(V
2
) are two open
subset of G
k,n
that separate X and Y , and so G
k,n
is Hausdor.
Now we check that G
k,n
is manifold of dimension k(n k) by showing that, for any X in G
k,n
,
there is an open neighborhood U
X
of such that U
X

= R
k(nk)
.
20 JIE WU
Let X G
k,n
be spanned by (v
1
, . . . , v
k
)
T
. There exists a nonsingular n n matrix Q such that
(v
1
, . . . , v
k
)
T
= (I
k
, 0)Q,
where I
k
is the unit k k-matrix. Fixing Q, dene
X

= (P
k
, B
k,nk
)Q [ det(P
k
) = 0, B
k,nk
M(k, n k)

V
k,n
.
Then E
X
is an open subset of

V
k,n
. Let U
X
= (E
X
) G
k,n
. Since

1
(U
X
) = E
X
is open in

V
k,n
, U
X
is open in G
k,n
with X U
X
. From the commutative diagram
GL
k
(R) M(k, n k)
(P, A) (P, PA)Q

=
-
E
X
M(k, n k)
proj.
?
A '(I
k
, A)Q`

1
X
-
U
X
,

?
U
X
is homeomorphic to M(k, n k) = R
k(nk)
and so G
k,n
is a (topological) manifold.
For checking that G
k,n
is a smooth manifold, let X and Y G
k,n
be spanned by (v
1
, . . . , v
k
)
T
and ( w
1
, . . . , w
k
)
T
, respectively. There exists nonsingular n n matrixes Q and

Q such that
(v
1
, . . . , v
k
)
T
= (I
k
, 0)Q, ( w
1
, . . . , w
k
)
T
= (I
k
, 0)

Q.
Consider the maps:
M(k, n k)

1
X
-
U
X
A '(I
k
, A)Q`
M(k, n k)

1
Y
-
U
Y
A '(I
k
, A)

Q`.
If Z U
X
U
Y
, then
Z = '(I
k
, A
Z
)Q` = '(I
k
, B
Z
)

Q`
for unique A, B M(k, n k). It follows that there is a nonsingular k k matrix P such that
(I
k
, B
Z
)

Q = P(I
k
, A
Z
)Q (I
k
, B
Z
) = P(I
k
, A
Z
)Q

Q
1
.
Let
T = Q

Q
1
=

T
11
T
12
T
21
T
22

.
Then
(I
k
, B
Z
) = (P, PA
Z
)T = (PT
11
+PA
Z
T
21
, PT
12
+PA
Z
T
22
)

I
k
= P(T
11
+A
Z
T
21
)
B
Z
= P(T
12
+A
Z
T
22
).
It follows that
Z U
X
U
Y
if and only if det(T
11
+A
Z
T
21
) = 0 (that is, T
11
+A
Z
T
21
is invertible).
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 21
From the above, the composite

X
(U
X
U
Y
)

1
X
-
U
X
U
Y

Y
-
M(k, n)
is given by
A (T
11
+AT
21
)
1
(T
12
+AT
22
) ,
which is smooth. Thus G
k,n
is a smooth manifold.
As a special case, G
1,n
is the space of lines (through the origin) of R
n
, which is also called
projective space denoted by RP
n1
. From the above, RP
n1
is a manifold of dimension n 1.
3.2. Stiefel Manifold. The Stiefel manifold, denoted by V
k,n
, is dened to be the set of k orthog-
onal unit vectors in R
n
with topology given as a subspace of

V
k,n
M(k, n). Thus
V
k,n
= A M(k, n) [ A A
T
= I
k
.
We prove that V
k,n
is a smooth submanifold of M(k, n) by using Pre-image Theorem.
Let S(k) be the space of symmetric matrixes. Then S(k)

= R
(k+1)k
2
is a smooth manifold of
dimension. Consider the map
f : M(k, n) S(k) A AA
T
.
For any A M(k, n), Tf
A
: T
A
(M(k, n)) T
f(A)
(S(k)) is given by setting Tf
A
(B) is the directional
derivative along B for any B T
A
(M(k, n)), that is,
Tf
A
(B) = lim
s0
f(A+sB) f(A)
s
= lim
s0
(A+sB)(A+sB)
T
AA
T
s
= lim
s0
AA
T
+sAB
T
+sBA
T
+s
2
BB
T
AA
T
s
= AB
T
+BA
T
.
We check that Tf
A
: T
A
(M(k, n)) T
f(A)
(S(k)) is surjective for any A f
1
(I
k
).
By the identication of M(k, n) and S(k) with Euclidian spaces, T
A
(M(k, n)) = M(k, n) and
T
f(A)
(S(k)) = S(k)). Let A f
1
(I
k
) and let C T
f(A)
(S(k)). Dene
B =
1
2
CA T
A
(M(k, n)).
Then
Tf
A
(B) = AB
T
+BA
T
=
1
2
AA
T
C
T
+
1
2
CAA
T
=======
AA
T
=I
k 1
2
C
T
+
1
2
C =====
C=C
T
C.
Thus Tf : T
A
(M(k, n)) T
f(A)
(S(k)) is onto and so I
k
is a regular value of f. Thus, by Pre-image
Theorem, V
k,n
= f
1
(I
k
) is a smooth submanifold of M(k, n) of dimension
kn
(k + 1)k
2
=
k(2n k 1)
2
.
Special Cases: When k = n, then V
n,n
= O(n) the orthogonal group. From the above, O(n) is a
(smooth) manifold of dimension
n(n1)
2
. (Note. O(n) is a Lie group, namely, a smooth manifold
plus a topological group such that the multiplication and inverse are smooth.)
When k = 1, then V
1,n
= S
n1
which is manifold of dimension n 1.
When k = n 1, then V
n1,n
is a manifold of dimension
(n1)n
2
. One can check that
V
n1,n

= SO(n)
22 JIE WU
the subgroup of O(n) with determinant 1. In general case, V
k,n
= O(n)/O(n k).
As a space, V
k,n
is compact. This follows from that V
k,n
is a closed subspace of the k-fold
Cartesian product of S
n1
because V
k,n
is given by k unit vectors (v
1
, . . . , v
k
)
T
in R
n
that are
solutions to v
i
v
j
= 0 for i = j, and the fact that any closed subspace of compact Hausdor space
is compact. The composite
V
k,n
-
V
k,n

--
G
k,n
is onto and so the Grassmann manifold G
k,n
is also compact. Moreover the above composite is a
smooth map because is smooth and V
k,n
is a submanifold. This gives the diagram
V
k,n

submanifold
-
M(k, n)
submersion at I
k
A AA
T
--
S(k)
G
k,n
smooth
?
?
Note. By the construction, G
k,n
is the quotient of V
k,n
by the action of O(k). This gives identi-
cations:
G
k,n
= V
k,n
/O(k) = O(n)/(O(k) O(n k)).
4. Fibre Bundles and Vector Bundles
4.1. Fibre Bundles. A bundle means a triple (E, p, B), where p: E B is a (continuous) map.
The space B is called the base space, the space E is called the total space, and the map p is called
the projection of the bundle. For each b B, p
1
(b) is called the bre of the bundle over b B.
Intuitively, a bundle can be thought as a union of bres f
1
(b) for b B parameterized by B
and glued together by the topology of the space E. Usually a Greek letter ( , , , , etc) is used to
denote a bundle; then E() denotes the total space of , and B() denotes the base space of .
A morphism of bundles (,

):

is a pair of (continuous) maps : E() E(

) and

: B() B(

) such that the diagram

E()

-
E(

)
B()
p()
?

-
B(

)
p(

)
?
commutes.
The trivial bundle is the projection of the Cartesian product:
p: B F B, (x, y) x.
Roughly speaking, a bre bundle p: E B is a locally trivial bundle with a xed bre F.
More precisely, for any x B, there exists an open neighborhood U of x such that p
1
(U) is a trivial
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 23
bundle, in other words, there is a homeomorphism
U
: p
1
(U) U F such that the diagram
U F

x
-
p
1
(U)
U

1
?
========= U
p
?
commutes, that is, p((x

, y)) = x

for any x

U and y F.
Similar to manifolds, we can use chart to describe bre bundles. A chart (U, ) for a bundle
p: E B is (1) an open set U of B and (2) a homeomorphism : U F p
1
(U) such that
p((x

, y)) = x

for any x

U and y F. An atlas is a collection of charts (U

) such that
U

is an open covering of B.
Proposition 4.1. A bundle p: E B is a bre bundle with bre F if and only if it has an atlas.
Proof. Suppose that p: E B is a bre bundle. Then the collection (U(x),
x
)[x B is an atlas.
Conversely suppose that p: E B has an atlas. For any x B there exists such that x U

and so U

is an open neighborhood of x with the property that p[

p
1
(U)
: p
1
(U

) U

is a trivial
bundle. Thus p: E B is a bre bundle.
Let be a bre bundle with bre F and an atlas (U

). The composite

: (U

) F

-
p
1
(U

)

1

-
(U

) F
has the property that

(x, y) = (x, g

(x, y))
for any x U

and y F. Consider the continuous map g

: U

F F. Fixing any x,
g

(x, ): F F, y g

(x, y) is a homeomorphism with inverse given by g

(x, ). This gives

a transition function
g

: U

-
Homeo(F, F),
where Homeo(F, F) is the group of all homeomorphisms from F to F.
Exercise 4.1. Prove that the transition functions g

satisfy the following equation

(3) g

(x) g

(x) = g

(x) x U

.
By choosing = = , g

(x) g

(x) = g

(x) and so
(4) g

(x) = x x U

.
By choosing = , g

(x) g

(x) = g

(x) = x and so
(5) g

(x) = g

(x)
1
x U

.
We need to introduce a topology on Homeo(F, F) such that the transition functions g

are
continuous. The topology on Homeo(F, F) is given by compact-open topology briey reviewed as
follows:
Let X and Y be topological spaces. Let Map(X, Y ) denote the set of all continuous maps from
X to Y . Given any compact set K of X and any open set U of Y , let
W
K,U
= f Map(X, Y ) [ f(K) U.
24 JIE WU
Then the compact-open topology is generated by W
K,U
, that is, an open set in Map(X, Y ) is an
arbitrary union of a nite intersection of subsets with the form W
K,U
.
Map(F, F) be the set of all continuous maps from F to F with compact open topology. Then
Homeo(F, F) is a subset of Map(F, F) with subspace topology.
Proposition 4.2. If Homeo(F, F) has the compact-open topology, then the transition functions
g

: U

Homeo(F, F) are continuous.

Proof. Given W
K,U
, we show that g
1

(W
K,U
) is open in U

. Let x
0
U

such that
g

(x
0
) W
(K,U)
. We need to show that there is a neighborhood V is x
0
such that g

(V ) W
K,U
,
or g

(V K) U. Since U is open and g

: (U

) F F is continuous, g
1
(U) is an
open set of (U

) F with x
0
K g
1

(U). For each y K, there exist open neighborhoods

V (y) of x and N(y) of y such that V (x) N(y) g
1

(U). Since N(y) [ y K is an open cover

of the compact set K, there is a nite cover N(y
1
), . . . , N(y
n
) of K. Let V =
n

j=1
V (y
j
). Then
V K g
1

(U) and so g

(V ) W
K,U
.
Proposition 4.3. If F regular and locally compact, then the composition and evaluation maps
Homeo(F, F) Homeo(F, F)
-
Homeo(F, F) (g, f) f g
Homeo(F, F) F
-
F (f, y) f(y)
are continuous.
Proof. Suppose that f g W
K,U
. Then f(g(K)) U, or g(K) f
1
(U), and the latter is open.
Since F is regular and locally compact, there is an open set V such that
g(K) V

V f
1
(U)
and the closure

V is compact. If g

W
K,V
and f

W
V ,U
, then f

W
K,U
. Thus W
K,V
and
W
V ,U
are neighborhoods of g and f whose composition product lies in W
K,U
. This implies that
Homeo(F, F) Homeo(F, F) Homeo(F, F) is continuous.
Let U be an open set of F and let f
0
(y
0
) U or y
0
f
1
0
(U). Since F is regular an locally
compact, there is a neighborhood V of y
0
such that

V is compact and y
0
V

V g
1
0
(U).
If g W
V ,U
and y V , then g(y) U and so the evaluation map Homeo(F, F) F F is
continuous.
Proposition 4.4. If F is compact Hausdor, then the inverse map
Homeo(F, F)
-
Homeo(F, F) f f
1
is continuous.
Proof. Suppose that g
1
0
W
K,U
. Then g
1
0
(K) U or K g
0
(U). It follows that
F K F g
0
(U) = g
0
(F U)
because g
0
is a homeomorphism. Note that F U is compact, F K is open and g
0
W
FU,FK
.
If g W
FU,FK
, then, from the above arguments, g
1
W
K,U
and hence the result.
Note. If F is regular and locally compact, then Homeo(F, F) is a topological monoid, namely
compact-open topology only fails in the continuity of g
1
. A modication on compact-open topology
eliminates this defect [1].
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 25
4.2. G-Spaces and Principal G-Bundles. Let G be a topological group and let X be a space.
A right G-action on X means a(continuous) map : X G X, (x, g) x g such that x 1 = x
and (x g) h = x (gh). In this case, we call X a (right) G-space. Let X and Y be (right) G-spaces.
A continuous map f : X Y is called a G-map if f(x g) = f(x) g for any x X and g G. Let
X/G be the set of G-orbits xG, x X, with quotient topology.
Proposition 4.5. Let X be a G-space.
1) For xing any g G, the map x x g is a homeomorphism.
2) The projection : X X/G is an open map.
Proof. (1). The inverse is given by x x g
1
.
(2) If U is an open set of X,

1
((U)) =

gG
U g
is open because it is a union of open sets, and so (U) is open by quotient topology. Thus is an
open map.
We are going to nd some conditions such that : X X/G has canonical bre bundle structure
with bre G. Given any point x X/G, choose x X such that (x) = x. Then

1
( x) = x g [ g G = G/H
x
,
where H
x
= g G [ x g = x.
For having constant bre G, we need to assume that the G-action on X is free, namely
x g = x = g = 1
for any x X. This is equivalent to the property that
x g = x h = g = h
for any x X. In this case we call X a free G-space.
Since a bre bundle is locally trivial (locally Cartesian product), there is always a local cross-
section from the base space to the total space. Our second condition is that the projection : X
X/G has local cross-sections. More precisely, for any x X/G, there is an open neighborhood U( x)
with a continuous map s
x
: U( x) X such that s
x
= id
U( x)
.
(Note. For every point x, we can always choose a pre-image of , the local cross-section means
the pre-images can be chosen continuously in a neighborhood. This property depends on the
topology structure of X and X/G.)
Assume that X is a (right) free G-space with local cross-sections to : X X/G. Let x be any
point in X/G. Let U( x) be a neighborhood of x with a (continuous) crosse-section s
x
: U( x) X.
Dene

x
: U( x) G
-

1
(U( x)) ( y, g)
-
s
x
( y) g
for any y U( x).
Exercise 4.2. Let X be a (right) free G-space with local cross-sections to : X X/G. Then the
continuous map
x
: U( x) G
1
(U( x)) is one-to-one and onto.
We need to nd the third condition such that
x
is a homeomorphism. Let
X

= (x, x g) [ x X, g G X X.
26 JIE WU
A function
: X

-
G
such that
x (x, x

) = x

for all (x, x

) X
is called a translation function. (Note. If X is a free G-space, then translation function is unique
because, for any (x, x

) X

, there is a unique g G such that x

= x g, and so, by denition,

(x, x

) = g.)
Proposition 4.6. Let X be a (right) free G-space with local cross-sections to : X X/G. Then
the following statements are equivalent each other:
1) The translation function : X

G is continuous.
2) For any x X/G, the map
x
: U( x) G
1
(U( x)) is a homeomorphism.
3) There is an atlas (U

of X/G such that the homeomorphisms

: U

G
-

1
(U

)
satisfy the condition

( y, gh) =

( y, g) h, that is

is a homeomorphism of G-spaces.
Proof. (1) = (2). Consider the (continuous) map
:
1
(U( x))
-
U( x) G z ((z), (s
x
((z)), z)).
Then

x
( y, g) = (s
x
( y) g) = ( y, (s
x
( y), s
x
( y) g)) = ( y, g),

x
(z) =
x
((z), (s
x
((z)), z)) = s
x
((z)) (s
x
((z)), z) = z.
Thus
x
is a homeomorphism.
(2) = (3) is obvious.
(3) = (1). Note that the translation function is unique for free G-spaces. It suces to show
that the restriction
(X): X

(
1
(U

)
1
(U

)) =

1
(U

-
G
is continuous. Consider the commutative diagram
(U

G)

=
-

1
(U

G
(U

G)
?
=========== G.
(X)
?
Since
(U

G)(( y, g), ( y, h)) = g

1
h
is continuous, the translation function restricted to

1
(U

(X) = (U

G) ((

)
1
is continuous for each and so (X) is continuous.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 27
Now we give the denition. A principal G-bundle is a free G-space X such that
: X X/G
has local cross-sections and one of the (equivalent) conditions in Proposition 4.6 holds.
Example. Let be a topological group and let G be a closed subgroup. Then the action of G
on given by (a, g) ag for a and g G is free. Then translation function is given by
(a, b) = a
1
b, which is continuous. Thus /G is principal G-bundle if and only if it has local
cross-sections.
4.3. The Associated Principal G-Bundles of Fibre Bundles. We come back to look at bre
bundles given by p: E B with bre F. Let (U

) be an atlas and let

g

: U

-
Homeo(F, F)
be the transition functions. A topological group G is called a group of the bundle if
1) There is a group homomorphism
: G
-
Homeo(F, F).
2) There exists an atlas of such that the transition functions g

lift to G via , that is,

there is commutative diagram
U

-
Homeo(F, F)
U

-
G,

6
(where we use the same notation g

.)
3) The transition functions
g

: U

-
G
are continuous.
4) The G-action on F via is continuous, that is, the composite
GF
id
F
-
Homeo(F, F) F
evaluation
-
F
is continuous.
We write

= (U

, g

) for the set of transition functions to the atlas (U

).
Note. In Steenrods denition [11, p.7], is assume to be a monomorphism (equivalently, the
G-action on F is eective, that is, if y g = y for all y F, then g = 1.).
We are going to construct a principal G-bundle : E
G
B. Then prove that the total space
E = F
G
E
G
and p: E B can be obtained canonically from : E
G
B. In other words,
all bre bundles can obtained through principal G-bundles through this way. Also the topological
group G plays an important role for bre bundles. Namely, by choosing dierent topological groups
G, we may get dierent properties for the bre bundle . For instance, if we can choose G to be
trivial (that is, g

lifts to the trivial group), then bre bundle is trivial. We will see that the bundle
group G for n-dimensional vector bundles can be chosen as the general linear group GL
n
(R). The
28 JIE WU
vector bundle is orientable if and only if the transition functions can left to the subgroup of GL
n
(R)
consisting of n n matrices whose determinant is positive. If n = 2m, then GL
m
(C) GL
2m
(R).
The vector bundle admits (almost) complex structure if and only if the transition functions can left
to GL
m
(C). (For manifolds, one can consider the structure on the tangent bundles. For instance,
an oriented manifold means its tangent bundle is oriented.)
Proposition 4.7. If

is the set of transition functions for the space B and topological group G,
then there is a principal G-bundle
G
given by
: E
G
-
B
and an atlas (U

) such that

is the set of transition functions to this atlas.
Proof. The proof is given by construction. Let

E =

G,
that is

E is the disjoint union of U

G. Now dene a relation on

E by
(b, g, ) (b

, g

, ) b = b

, g = g

(b)g

.
This is an equivalence relation by Equations (3)-(5). Let E
G
=

E/ with quotient topology and
let b, g, for the class of (b, g, ) in E
G
. Dene : E
G
B by
b, g, = b,
then is clearly well-dened (and so continuous). The right G-action on E
G
is dened by
b, g, h = b, gh, .
This is well-dened (and so continuous) because if (b

, g

, ) (b, g, ), then
(b

, g

h, ) = (b, (g

(b)g)h, ) = (b, g

(b)(gh), ) (b, gh, ).

Dene

: U

G
1
(U

) by setting

(b, g) = b, g, ,
then

is continuous and satises

(b, g) = b and

(b, g) = b, 1 g, = b, 1, g
for b U

and g G. The map

is a homeomorphism because, for xing , the map

(U

) G
-
U

G (b, g

, ) (b, g

(b)g

)
induces a map
1
(U

) U

G which the inverse of

. Moreover,

(b, g

(b)g) = b, g

(b)g, = b, g, =

(b, g)
for b U

and g G. Thus the (U

, g

) is the set of transition function to the atlas

(U

).
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 29
Let X be a right G-space and let Y be a left G-space. The product over G is dened by
X
G
Y = X Y/(xg, y) (x, gy)
with quotient topology. Note that the composite
X Y

X
-
X

-
X/G
(x, y) x x
factors through X
G
Y . Let p: X
G
Y X/G be the resulting map. For any x X/G, choose
x
1
( x) X, then
p
1
( x) =
1
( x)
G
Y = x Y/H
x
,
where H
x
= g G [ xg = x. Thus if X is a free right G-space, then the projection p: X
G
Y
X/G has the constant bre Y .
Proposition 4.8. Let : X X/G be a (right) principal G-bundle and let Y be any left G-space.
Then
p: X
G
Y
-
X/G
is a bre bundle with bre Y .
Proof. Consider a chart (U

) for : X X/G. Since the homeomorphism

: U

1
(U

) is a G-map, there is a commutative diagram

U

Y === (U

G)
G
Y

=
-

1
(U

)
G
Y === p
1
(U

)
U

?
=========== U

?
============= U

p
?
========== U

p
?
and hence the result.
Let be a (right) principal G-bundle given by : X X/G. Let Y be any left G-space. Then
bre bundle
p: X
G
Y
-
X/G
is called induced bre bundle of , denoted by [Y ].
Now let p: E B is a bre bundle with bre F and bundle group G. Observe that the action
of Homeo(F, F) on F is a left action because (f g)(x) = f(g(x)). Thus G acts by left on F via
: G Homeo(F, F).
A bundle morphism
E()

-
E(

)
B()
p()
?

-
B(

)
p(

)
?
30 JIE WU
is call an isomorphism if both and

are homeomorphisms. (Note. this means that (
1
, (

)
1
)
are continuous.) In this case, we write

=

.
Theorem 4.9. Let be a bre bundle given by p: E B with bre F and bundle group G. Let
G
be the principal G-bundle constructed in Proposition 4.7 according to a set of transitions functions
to Then
G
[F]

= .
Proof. Let (U

) be an atlas for . We write

for

in the proof of Proposition 4.7. Consider

the map

given by the composite:

1
(U

)
G
F

id
F

=
(U

G)
G
F === U

=
-
p
1
(U

).
From the commutative diagram
((U

) G)
G
F
(b, g

, y) (b, g

, g

(b)(y))
((b, g

, ), y) ((b, g

(b)g

, ), y)
-
((U

) G)
G
F
(U

) F

(b, y) (b, g

(b, y))
-
(U

) F

p
1
(U

?
=========================================== p
1
(U

),

?
the map

induces a bundle map

E
G

G
F

-
E()
B()
?
====== B().
?
This is an bundle isomorphism because is one-to-one and onto, and is a local homeomorphic by
restricting each chart. The assertion follows.
This theorem tells that any bre bundle with a bundle group G is an induced bre bundle of
a principal G-bundle. Thus, for classifying bre bundles over a xed base space B, it suces to
classify the principal G-bundles over B. The latter is actually done by the homotopy classes from
B to the classifying space BG of G. (There are few assumptions on the topology on B such as B is
paracompact.) The theory for classifying bre bundles is also called (unstable) K-theory, which is
one of important applications of homotopy theory to geometry. Rough introduction to this theory
is as follows:
There exists a universal G-bundle
G
as : EG BG. Given any principal G-bundle over B,
there exists a (continuous) map f : B BG such that , as a principal G-bundle, is isomorphic to
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 31
the pull-back bundle f

G
given by
E(f

G
) = (x, y) B EG [ f(x) = (y)
-
EG
B
(x, y) x
?
f
-
BG.

?
Moreover, for continuous maps f, g : B BG, f

G

= g

G
if and only if f g, that is, there is a
continuous map (called homotopy) F : B[0, 1] BG such that F(x, 0) = f(x) and F(x, 1) = g(x).
In other words, the set of homotopy classes [B, BG] is one-to-one correspondent to the set of
isomorphic classes of principal G-bundles over G.
Seminar Topic: The classication of principal G-bundles and bre bundles. (References: for
instance [8, pp.48-58] Or [9, 10].)
4.4. Vector Bundles. Let F denote R, C or H-the real, complex or quaternion numbers. An
n-dimensional F-vector bundle is a bre bundle given by p: E B with bre F
n
and an atlas
(U

) in which each bre p

1
(b), b B, has the structure of vector space over F such that
each homeomorphism

: U

F
n
p
1
(U

) has the property that

[
{b}F
n : b F
n
-
p
1
(b)
is a vector space isomorphism for each b U

.
Let be a vector bundle. From the composite
(U

) F
n

-
p
1
(U

)

1

-
(U

) F
n
,
the transition functions
g

: U

-
Homeo(F
n
, F
n
)
have that property that, for each x U

,
g

(x): F
n
-
F
n
is a linear isomorphism. It follows that the bundle group for a vector bundle can be chosen as
the general linear group GL
n
(F). By Theorem 4.9, we have the following.
Proposition 4.10. Let be an n-dimensional F-vector space over B. Then there exists a principal
GL
n
(F)-bundle
GL
n
(F)
over B such that

=
GL
n
(F)
[F
n
]. Conversely, for any principal GL
n
(F)-
bundle over B,
GL
n
(F)
[F
n
] is an n-dimensional F-vector bundle over B.
In other words, the total spaces of all vector bundles are just given by E(
GL
n
(F)
)
GL
n
(F)
F
n
.
4.5. The Construction of Gauss Maps. The Grassmann manifold G
n,m
(F) is the set of n-
dimensional F-subspaces of F
m
, that is, all n-F-planes through the origin, with the topology de-
scribed as in the topic on the examples of Manifolds. (If m = , F

j=1
F.) Let
E(
m
n
) = (V, x) G
n,m
(F) F
m
[ x V .
32 JIE WU
Exercise 4.3. Show that
p: E(
m
n
) G
n,m
(F) (V, x) V
is an n-dimensional F-vector bundle, denoted by
m
n
. [Hint: By reading the topic on the examples of
manifolds, check that V
n,m
(F) G
n,m
(F) is a principal O(n, F), where O(n, R) = O(n), O(n, C) =
U(n) and O(n, H) = Sp(n). Then check that E(
m
n
) = V
n,m
(F)
O(n,F)
F
n
.]
A Gauss map of an n-dimensional F-vector bundle in F
m
(n m ) is a (continuous) map
g : E() F
m
such that g restricted to each bre is a linear monomorphism.
Example. The map
q : E(
m
n
) F
m
(V, x) x
is a Gauss map.
Proposition 4.11. Let be an n-dimensional F-vector bundle.
1) If there is a vector bundle morphism
E()
u
-
E(
m
n
)
B()
p()
?
f
-
G
n,m
(F)
p(
m
n
)
?
that is an isomorphism when restricted to any bre of , then q u: E() F
m
is a Gauss
map.
2) If there is a Gauss map g : E() F
m
, then there is a vector bundle morphism (u, f):

m
n
such that qu = g.
Proof. (1) is obvious. (2). For each b B(), g(p()
1
(b)) is an n-dimensional F-subspace of F
m
and so a point in G
n,m
(F). Dene the functions
f : B() G
n,m
(F) f(b) = g(p()
1
(b)),
u: E() E(
m
n
) u(z) = (f(p(z)), g(z)).
The functions f and u are well-dened. For checking the continuity of f and u, one can look at a
local coordinate of and so we may assume that is a trivial bundle, namely, g : B() F
n
F
m
restricted to each bre is a linear monomorphism. Let e
1
, . . . , e
n
be the standard F-bases for F
n
.
Then the map
h: B
-
F
m
F
m
b (g(b, e
1
), g(b, e
2
), . . . , g(b, e
n
))
is continuous. Since g restricted to each bre is a monomorphism, the vectors
g(b, e
1
), g(b, e
2
), . . . , g(b, e
n
)
are linearly independent and so
(g(b, e
1
), g(b, e
2
), . . . , g(b, e
n
))

V
n,m
(F)
for each b, where V
n,m
(F) is the open Stiefel manifold over F. Thus
h: B
-
V
n,m
(F) b (g(b, e
1
), g(b, e
2
), . . . , g(b, e
n
))
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 33
is continuous and so the composite
f : B
h
-
V
n,m
(F)
quotient
--
G
n,m
(F)
is continuous. The function u is continuous because the composite
E()
u
-
E(
m
n
)
E() E()

?
(f p()) g
-
G
n,m
(F) F
m
?

is continuous. This nishes the proof.

Let be a vector bundle and let f : X B() be a (continuous) map. Then the induced vector
f

is the pull-back
E(f

) = (x, y) X E() [f(x) = p(y)

-
E()
X
?
f
-
B().
p
?
Proposition 4.12. There exists a Gauss map g : E() F
m
(n m ) if and only if

= f

(
m
n
)
over B() for some map f : B() G
n,m
(F).
Proof. = is obvious.
= Assume that has a Gauss map g. From Part (2) of Proposition 4.11, there is a commutative
diagram
E()
u
-
E(
m
n
)
B()
p()
?
f
-
G
n,m
(F).
p(
m
n
)
?
Since E(f

m
n
) is dened to be the pull-back, there is commutative diagram
E()
u
-
E(f

m
n
)
B()
p()
?
====== B(),
p
?
where u restricted to each bre is a linear isomorphism because both vector-bundle has the same
dimension and the Gauss map g restricted to each bre is a monomorphism. It follows that
34 JIE WU
u: E() E(f

m
n
) is one-to-one and onto. Moreover u is a homeomorphism by considering a
local coordinate.
We are going to construct a Gauss map for each vector bundle over a paracompact space. First,
we need some preliminary results for bundles over paracompact spaces. (For further information on
paracompact spaces, one can see [3, 162-169].
A family of ( = C

[ J of subsets of a space X is called locally nite if each x X admits a

neighborhood W
x
such that W
x
C

= for only nitely many indices J. Let | = U

and
1 = V

be two open covers of X. 1 is called a renement of | if for each , V

for some .
A Hausdor space X is called paracompact if it is regular and if every open cover of X admits a
locally nite renement.
Let | = U

[ J be an open cover of a space X. A partition of unity, subordinate to |, is

a collection

[ J of continuous functions

: X [0, 1] such that

1) The support
supp(

) U

for each , where the support

supp(

) = x X [

(x) = 0
is the closure of the subset of X on which

= 0;
2) for each x X, there is a neighborhood W
x
of x such that

[
W
x
0 for only nitely
many indices J. (In other words, the supports of

s are locally nite.)

3) The equation

(x) = 1
for all x X, where the summation is well-dened for each given x because there are only
nitely many non-zeros.
We give the following well-known theorem without proof. One may read a proof in [2, pp.17-20].
Theorem 4.13. If X is a paracompact space and | = U

is an open cover of X, then there

exists a partition of unity subordinate to |.
Lemma 4.14. Let be a bre bundle over a paracompact space B. Then admits an atlas with
countable charts.
Proof. Let (U

[ J be an atlas for . We are going to nd another atlas with countable

charts.
By Theorem 4.13, there is a partition of unity

[ J subordinate to U

[ J. Let
V

=
1

(0, 1] = b B [

(b) > 0.
Then, by the denition of partition of unity, V

. For each b B, let

S(b) = J [

(b) > 0.
Then, by the denition of partition of unity, S(b) is a nite subset of J.
Now for each nite subset S of J dene
W(S) = b B [

(b) >

(b) for each S and S

LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 35
=

S
S
(

)
1
(0, 1].
Then W(S) is open because for each b W(S), by denition of partition of unity, there exists a
neighborhood W
b
of b such that there are nitely many supports intersect with W
b
; and so the above
(possibly innite) intersection of open sets restricted to W
b
is only a nite intersection of open sets.
Let S and S

be two subsets of J such that S = S

and [S[ = [S

[ = m > 0, where [S[ is the

number of elements in S. Then there exist S S

and S

S because S = S

but S and
S

has the same number elements. We claim that

W(S) W(S

) = .
Otherwise there exists b W(S) W(S

). By denition W(S),

(b) >

(b) because S and

S. On the other hand,

(b) >

(b) because b W(S

), S

and S

.
Now dene
W
m
=

b B
[S[ = m
W(S(b))
for each m 1. We prove that (1) W
m
[ m = 1, . . . is an open cover of B; and (2) restricted to
W
m
is a trivial bundle for each m. (Then W
m
induces an atlas for .)
To check W
m
is an open cover, note that each W
m
is open. For each b B, S(b) is a nite set
and b W(S(b)) because

(b) = 0 for S(b) and

(b) > 0 for S(b). Let m = [S(b)[, then

b W
m
and so W
m
is an open cover of B.
Now check that restricted to W
m
is trivial. From the above, W
m
is a disjoint union of W(S(b)).
It suces to check that restricted to each W(S(b)) is trivial. Fixing S(b), for any x W(S(b)),
then

(x) >

(x)
for any S. In particular,

(x) > 0 for any x W(S(b)). It follows that W(S(b)) V

.
Since restricted to U

is trivial, restricted to W(S(b)) is trivial. This nishes the proof.

Note. From the proof, if for each b B there are at most k sets U

with b U

, then B admits
an atlas of nite (at most k) charts. [In this case, check that W
j
= for j > k.]
Theorem 4.15. Any n-dimensional F-vector bundle over a paracompact space B has a Gauss map
g : E() F

. Moreover, if has an atlas of k charts, then has a Gauss map g : E() F

kn
.
Proof. Let (U
i
,
i
)
1ik
be an atlas of with countable or nite charts, where k is nite or innite.
Let
i
be the partition of unity subordinate to U
i
. For each i, dene the map g
i
: E() F
n
as follows: g
i
restricted to p()
1
(U
i
) is given by
g
i
(z) =
i
(z)(p
2

1
i
(z)),
where p
2

1
i
is the composite
p()
1
(U
i
)

1
i
-
U
i
F
n
projection
p
2
-
F
n
;
36 JIE WU
and g
i
restricted to the outside of p()
1
(U
i
) is 0. Since the closure of
1
i
(0, 1] is contained in U
i
,
g
i
is a well-dened (continuous) map. Now dene
g : E()
k

i=1
F
n
= F
kn
g(z) =
k

i=1
g
i
(z).
This a well-dened (continuous) map because for each z, there is a neighborhood of z such that
there are only nitely many g
i
are not identically zero on it.
Since each g
i
: E() F
n
is a monomorphism (actually isomorphism) on the bres of E() over
b with
i
(b) > 0, and since the images of g
i
are in complementary subspaces of F
kn
, the map g is a
Gauss map.
This gives the following classication theorem:
Corollary 4.16. Every vector bundle over a paracompact space B is isomorphic to an induced vector
bundle f

n
) for some map f : B G
n,
(F). Moreover every vector bundle over a paracompact
space B with an atlas of nite charts is isomorphic to an induced vector bundle f

(
m
n
) for some m
and some map f : B G
n,m
(F).
Remarks: It can be proved that f

m
n

= g

m
n
if and only if f g : B G
n,m
(F). From this,
one get that the set of isomorphism classes of n-dimensional F-vector bundles over a paracompact
space B is isomorphic to the set of homotopy classes [B, G
n,
(F)].
For instance, if n = 1 and F = R, G
1,
(R) BO(1) BZ/2 RP

, where BG is so-called
the classifying space of the (topological) group G, and [B, RP

] = H
1
(B, Z/2), which states that
all line bundles are by the rst cohomology with coecients in Z/2Z. In particular, any real line
bundles over a simply connected space is always trivial.
If n = 1 and F = C, then G
1,
(C) BU(1) BS
1
CP

, and [B, CP

] = H
2
(B, Z), which
states that all complex line bundles are by the second integral cohomology.
If n = 1 and F = H, then G
1,
(H) BSp(1) BS
3
HP

, and so [B, G
1,
(H)] = [B, HP

].
However, the determination of [B, HP

] is very hard problem even when B are spheres. If B = S

n
,
then [B, HP

] =
n1
(S
3
) that is only known for n less than 66 or so, by a lot of computations
through many papers. Some people even believe that it is impossible to compute the general
homotopy groups
n
(S
3
).
Seminar Topic: Gauss Maps and the Classication of Vector Bundles. (References: for instance [8,
pp.26-29,31-33].)
A vector bundle is called of nite type if it has an atlas with nite charts. Given two vector
bundles and over B, the Whitney sum is dened to be the pull-back:
E( )
-
E() E()
B
?

diagonal
-
B B.
p() p()
?
Intuitively, is just the brewise direct sum.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 37
Proposition 4.17. For a vector bundle over a paracompact space B, the following statement are
equivalent:
1) The bundle is of nite type.
2) There exists a map f : B G
n,m
(F) such that is isomorphic to f

m
n
.
3) There exists a vector bundle such that the Whitney sum is trivial.
Proof. (1) = (2) follows from Corollary 4.16. (2) = (1) It is an exercise to check that
m
n
is
of nite type by using the property that the Grassmann manifold G
n,m
(F) = O(m, F)/O(n, F)
O(m n, F) is compact, where O(n, R) = O(n), O(n, C) = U(n) and O(n, H) = Sp(n). It follows
that

= f

m
n
is of nite type.
(2) = (3). Let (
m
n
)

be the vector bundle given by

E((
m
n
)

) = (V, v) G
n,m
(F) F
m
[ v V
with canonical projection E((
m
n
)

) G
n,m
(F). Then
m
n
(
m
n
)

is an m-dimensional trivial
F-vector bundle. It follows that
f

(
m
n
(
m
n
)

) = f

(
m
n
) f

((
m
n
)

)
is trivial. Let = f

((
m
n
)

). Then is trivial.
(3) = (2). The composite
E()
-
E( ) = B F
m
-
F
m
is a Gauss map into nite dimensional vector space, where m = dim( ). By Proposition 4.12,
there is a map f : B G
n,m
such that

= f

m
n
.
Corollary 4.18. Let be a F-vector bundle over a compact (Hausdor) space B. Then there is a
F-vector bundle such that is trivial.
In the view of (stable) K-theory, the Whitney sum is an operation on vector bundles over a (xed)
base-space, where the trivial bundles (of dierent dimensions) are all regarded as 0. In this sense,
the Whitney sum plays as an addition (that is associative and commutative with 0). The bundle
with property that is trivial for some means that is invertible. Those who are interested
in algebra can push notions in algebra to vector bundles by doing constructions brewisely. More
general situation possibly is the sheaf theory (by removing the locally trivial condition) that is pretty
useful in algebraic geometry. In algebraic topology, people also study the category whose objects are
just continuous maps f : E B with xed space B, or even more general category whose objects
are diagrams over spaces. In the terminology of bre bundles, a map f : E B is called a bundle
(without assuming locally trivial).
38 JIE WU
5. Tangent Bundles and Vector Fields
6. Cotangent Bundles and Tensor Fields
7. Orientation of Manifolds
8. Tensor Algebras and Exterior Algebras
9. DeRham Cohomology
10. Integration on Manifolds
11. Stokes Theorem
References
[1] R. Arens, Topologies for homeomorphism groups, Amer. Jour. Math. 68 (1946), 593-610.
[2] Lawrence Conlon, Dierentiable Manifolds, second Edition, Birkh auser Boston [2001].
[3] J. Dugundji, Topology, Allyn and Bacon, Boston, MA [1966].
[4] M. A. Kervaire and J. W. Milnor, Groups of homotopy spheres I, Ann. Math. 77 504-537, (1963).
[5] Serge Lang, Dierential Manifolds, Addison-Wesley Publishing Company, Inc. [1972].
[6] Victor Guillemin and Alan Pollack, Dierential Topology, Prentice Hall, Inc, Englewood Clis, New Jersey
[1974].
[7] Dominic G. B. Edelen, Applied Exterior Calculus, John Wiley & Sons, Inc. [1985].
[8] Dale Husemoller, Fibre Bundles, second edition, Graduate Texts in Mathematics Springer-Verlag [1966].
[9] J. Milnor, Construction of universal bundles I, Ann. Math. 63 (1956), 272-284.
[10] J. Milnor, Construction of universal bundles II, Ann. Math. 63 (1956), 430-436.
[11] Norman Steenrod, The Topology of Fibre Bundles, Princeton University Press, Princeton, New Jersey, Ninth
Printing, [1974].