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 InverseMapping Theorems 6
2. Topological and Diﬀerentiable Manifolds, Diﬀeomorphisms, Immersions, Submersions
and Submanifolds 9
2.1. Topological Spaces 9
2.2. Topological Manifolds 10
2.3. Diﬀerentiable 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. GSpaces and Principal GBundles 25
4.3. The Associated Principal GBundles 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 γ be a curve in R
n
: γ : t → (γ
1
(t), γ
2
(t), . . . , γ
n
(t)).
(A curve can be described as a vectorvalued function. Converse a vectorvalued function gives a
curve in R
n
.) The tangent line at the point γ(t
0
) is given with the direction
dγ
dt
(t
0
) =
dγ
1
dt
(t
0
), . . . ,
dγ
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 diﬀerentiable function f, where x
i
refers to
ith 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 γ is a curve through P (when t = t
0
)
that lies in the level surface f(x
1
, . . . , x
n
) = c, that is
f(γ
1
(t), γ
2
(t), . . . , γ
n
(t)) = c.
By taking derivatives on both sides,
∂f
∂x
1
(P)(γ
1
)
(t
0
) + +
∂f
∂x
n
(P)(γ
n
)
(t
0
) = 0
and so the tangent line of γ is really normal (orthogonal) to ∇f. When γ 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, ﬁrst 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 ndimensional space. But we can look at in a little
detail.
Let γ be a smooth curve through P. We may assume that γ(0) = P. Let ω be another smooth
curve with ω(0) = P. γ is called to be equivalent to ω if the directives γ
(0) = ω
(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
) =
¸
P∈R
n
T
P
(R
n
), called the tangent bundle of R
n
. If S is a region of R
n
, let T(S) =
¸
P∈S
T
P
(S), called the tangent bundle of S.
Note. Each T
P
(R
n
) is an ndimensional 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 γ is a smooth curve from P to Q in R
n
, then the tangent space T
P
(R
n
) moves along γ to
T
Q
(R
n
). The direction for this moving is given γ
(t), which introduces the following important
concept.
Deﬁnition 1.1. A vector ﬁeld 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 ﬁelds 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 ﬁelds.
1.3. Operator Representations of Vector Fields. Let J be an open interval containing 0 and
let γ : J → R
n
be a smooth curve with γ(0) = P. Let f = f(x
1
, . . . , x
n
) be a smooth function
deﬁned on a neighborhood of P. Assume that the range of γ is contained in the domain of f. By
applying the chain rule to the composite T = f ◦ γ : J →R,
D
γ
(f) :=
dT
dt
=
n
¸
i=1
dγ
i
(t)
dt
∂f
∂x
i
x
i
=γ
i
(t)
Proposition 1.2.
D
γ
(af +bg) = aD
γ
(f) +bD
γ
(g), where a, b are constant.
D
γ
(fg) = D
γ
(f)g +fD
γ
(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 satisﬁes 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
). Deﬁne
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 ﬁelds:
Given a vector ﬁelds 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 ﬁeld 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 ﬁeld on an neighborhood U of P.
An integral curve to V is a smooth curve s: (−δ, ) → U, deﬁned for suitable δ, > 0, such that
s
(t) = v(s(t))
for −δ < t < .
Theorem 1.4. Let V : x →v(x) be a (smooth) vector ﬁeld 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 ﬁrst order diﬀerential 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 ﬁrst 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
x
(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
x
(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 InverseMapping Theorems.
Theorem 1.7. Let D be an open region in R
n+1
and let F be a function welldeﬁned 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
δ
(x
1
0
, . . . , x
n
0
) ⊆ R
n
, and a unique
function z = g(x
1
, x
2
, . . . , x
n
) deﬁned for (x
1
, . . . , x
n
) ∈ N
δ
(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
δ
(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
δ
(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
δ
(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 ﬁxed
(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 δ > 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
δ
(x
1
0
, . . . , x
n
0
).
Step 3. Fixed (x
1
, . . . , x
n
) ∈ N
δ
(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 deﬁnes a function z = g(x
1
, . . . , x
n
) for (x
1
, . . . , x
n
) ∈ N
δ
(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
δ
(x
1
0
, . . . , x
n
0
). Let
(x
1
1
(k), . . . , x
n
1
(k)) be any sequence in N
δ
(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
i−1
, 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
φ(t) = F(x
1
, . . . , x
i
+th, . . . , x
n
, z +tk)
for 0 ≤ t ≤ 1. Then φ(1) −φ(0) = φ
(ξ)(1−0).), where ˜ x
i
is between x
i
and x
i
+h, and ˜ z is between
z and z +k. Now
∂g
∂x
i
= lim
h→0
g(x
1
, . . . , x
i−1
, x
i
+h, x
i+1
, . . . , x
n
) −z
h
= lim
h→0
k
h
= − lim
h→0
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 welldeﬁned 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
δ
(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
)
deﬁned for (x
1
, . . . , x
m
) ∈ N
δ
(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
δ
(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
j−1
, x
j
, u
j+1
, . . . , u
n
)
for all (x
1
, . . . , x
m
) ∈ N
δ
(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
δ
(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
n−1
)
to the last equation. Consider
G
1
= F
1
(x
1
, . . . , x
m
, u
1
, . . . , u
n−1
, g
n
)
G
2
= F
2
(x
1
, . . . , x
m
, u
1
, . . . , u
n−1
, g
n
)
G
n−1
= F
n−1
(x
1
, . . . , x
m
, u
1
, . . . , u
n−1
, 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
n−1
1
¸
Then
∂F
i
∂u
j
B =
∂G
i
∂u
j
n−1,n−1
∗
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
n−1
)
∂(u
1
, . . . , u
n−1
)
.
Thus
∂(G
1
,...,G
n−1
)
∂(u
1
,...,u
n−1
)
= 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 deﬁned 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
δ
(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 welldeﬁned and has continuous partial derivatives on N
δ
(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.
Deﬁnition 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.
Deﬁnition 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 preimage 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 Hausdorﬀ space, that is, a topological spaces where any two points has
disjoint neighborhoods.
Deﬁnition 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 deﬁnition, 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 classiﬁcation of curves is known. The
topological classiﬁcation of (twodimensional) surfaces is known as well. However the topological
classiﬁcation of 3dimensional manifolds (we will learn manifolds later.) is quite open.
The famous Poicar´e conjecture is related to this problem, which states that any simply connected
3dimensional (topological) manifold is homeomorphic to the 3sphere S
3
. A space X is called simply
connected if (1) X is pathconnected (that is, given any two points, there is a continuous path joining
them) and (2) the fundamental group π
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 Hausdorﬀ space M is called a (topological) nmanifold 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 nsphere S
n
is an nmanifold. A 2dimensional manifold is called
a surface. The objects traditionally called ‘surfaces in 3space’ can be made into manifolds in
a standard way. The compact surfaces have been classiﬁed as spheres or projective planes with
various numbers of handles attached.
By the deﬁnition of manifold, the closed ndisk D
n
is not an nmanifold because it has the
‘boundary’ S
n−1
. D
n
is an example of ‘manifolds with boundary’. We give the deﬁnition of
manifold with boundary as follows.
A Hausdorﬀ space M is called an nmanifold 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 deﬁnition, there is a an open neighborhood U(x) of x and
a homeomorphism φ
x
from U(x) onto an open set in R
n
+
. The collection ¦(U(x), φ
x
)[x ∈ M¦ has
the property that 1) ¦U(x)[x ∈ M¦ is an open cover and 2) φ
x
is a homeomorphism from U(x)
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 11
onto an open set in R
n
+
. The subspace φ
x
(U
x
) in R
n
+
plays a role as a local coordinate system. The
collection ¦(U(x), φ
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, φ) such that 1) U is an open set in M and 2) φ is
a homeomorphism from U onto an open set in R
n
+
. The map
φ: U →R
n
+
can be given by n coordinate functions φ
1
, . . . , φ
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
α
, φ
α
)[α ∈ J¦ such that ¦U
α
[α ∈ J¦ is an open
cover of M.
Proposition 2.4. A Hausdorﬀ 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), φ
x
)[x ∈ M¦ is an atlas. Con
versely suppose that M has an atlas. For any x ∈ M there exists α such that x ∈ U
α
and so U
α
is
an open neighborhood of x that is homeomorphic to an open set in R
n
+
. Thus M is a manifold.
We deﬁne a subset ∂M as follows: x ∈ ∂M if there is a chart (U
α
, φ
α
) such that x ∈ U
α
and
φ
α
(x) ∈ R
n−1
= ¦x ∈ R
n
[x
n
= 0¦. ∂M is called the boundary of M. For example the boundary of
D
n
is S
n−1
.
Proposition 2.5. Let M be a nmanifold with boundary. Then ∂M is an (n−1)manifold without
boundary.
Proof. Let ¦(U
α
, φ
α
)[α ∈ J¦ be an atlas for M. Let J
⊆ J be the set of indices such that
U
α
∩ ∂M = ∅ if α ∈ J
. Then Clearly
¦(U
α
∩ ∂M, φ
α
[
U
α
∩∂M
[α ∈ 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
n−1
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
n−1
, there
is an open disk E
(x) ⊆ V and so
E
(x) ∩ R
n−1
⊆ V ∩ R
n−1
= U ∩ R
n−1
is an open (n −1)dimensional disk in R
n−1
centered at x.
2.3. Diﬀerentiable Manifolds.
Deﬁnition 2.6. A Hausdorﬀ space M is called a diﬀerential manifold of class C
k
(with boundary)
if there is an atlas of M
¦(U
α
, φ
α
)[α ∈ J¦
such that
12 JIE WU
For any α, β ∈ J, the composites
φ
α
◦ φ
−1
β
: φ
β
(U
α
∩ U
β
) →R
n
+
is diﬀerentiable of class C
k
.
The atlas ¦(U
α
, φ
α
[α ∈ J¦ is called a diﬀerential 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
α
, φ
α
)[α ∈ J¦
such that each φ
α
: U
α
→ R
n
+
is diﬀerentiable of class C
k
, then M is a diﬀerentiable manifold of
class C
k
. This is the deﬁnition of diﬀerentiable (smooth) manifolds in [6] as in the beginning they
already assume that M is a subspace of R
N
with N large. In our deﬁnition (the usual deﬁnition
of diﬀerentiable manifolds using charts), we only assume that M is a (Hausdorﬀ) topological space
and so φ
α
is only an identiﬁcation of an abstract U
α
with an open subset of R
n
+
. In this case we can
not talk diﬀerentiability of φ
α
unless U
α
is regarded as a subspace of a (large dimensional) Euclidian
space.)
Two diﬀerential atlases of class C
k
¦(U
α
, φ
α
)[α ∈ I¦ and ¦(V
β
, ψ
β
)[β ∈ J¦ are called equivalent
if
¦(U
α
, φ
α
)[α ∈ I¦ ∪ ¦(V
β
, ψ
β
)[β ∈ J¦
is again a diﬀerential atlas of class C
k
(this is an equivalence relation). A diﬀerential structure of
class C
k
on M is an equivalence class of diﬀerential atlases of class C
k
on M. Thus a diﬀerential
manifold of class C
k
means a manifold with a diﬀerential structure of class C
k
. A smooth manifold
means a diﬀerential 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.
Deﬁnition 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
α
, φ
α
[α ∈ I¦ for M
and ¦(V
β
, ψ
β
)β ∈ J¦ for N the functions
ψ
β
◦ f ◦ φ
−1
α
[
φ
α
(f
−1
(V
β
)∩U
α
)
: φ
α
(f
−1
(V
β
) ∩ U
α
) →R
n
+
are of class C
∞
.
Proposition 2.8. If f : M → N is smooth with respect to atlases
¦(U
α
, φ
α
[α ∈ I¦, ¦(V
β
, φ
β
[β ∈ J¦
for M, N then it is smooth with respect to equivalent atlases
¦(U
δ
, θ
δ
[α ∈ I
¦, ¦(V
γ
, η
γ
[β ∈ J
¦
Proof. Since f is smooth with respect with the atlases
¦(U
α
, φ
α
[α ∈ I¦, ¦(V
β
, φ
β
[β ∈ J¦,
f is smooth with respect to the smooth atlases
¦(U
α
, φ
α
[α ∈ I¦ ∪ ¦(U
δ
, θ
δ
[α ∈ I
¦, ¦(V
β
, φ
β
[β ∈ J¦ ∪ ¦(V
γ
, η
γ
[β ∈ J
¦
by look at the local coordinate systems. Thus f is smooth with respect to the atlases
¦(U
δ
, θ
δ
[α ∈ I
¦, ¦(V
γ
, η
γ
[β ∈ J
¦.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 13
Thus the deﬁnition of smooth maps between two smooth manifolds is independent of choice of
atlas.
Deﬁnition 2.9. A smooth map f : M → N is called a diﬀeomorphism if f is onetoone and onto,
and if the inverse f
−1
: N → M is also smooth.
Deﬁnition 2.10. Let M be a smooth nmanifold, 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, φ) about P in M such that
φ(U ∩ X) = φ(U) ∩ R
k
+
,
where R
k
+
⊆ R
n
+
is the standard inclusion.
Note. In the above deﬁnition, the collection ¦(U ∩X, φ[
U∩X
)¦ is an atlas for making X to a smooth
kmanifold 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 ndimensional 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
m×n
¸
¸
¸
v
1
v
2
v
n
¸
n×1
=
¸
¸
¸
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 deﬁne the tangent space to a (diﬀerentiable) manifold M at a point P as
follows:
First we consider the set
T
P
= ¦(U, φ, v) [ P ∈ U, (U, φ) is a chart v ∈ T(φ(P))(φ(U))¦.
The point is that there are possibly many charts around P. Each chart creates an ndimension vector
space. So we need to deﬁne an equivalence relation in T
P
such that, T
P
modulo these relations is
only one copy of ndimensional vector space which is also independent on the choice of charts.
Let (U, φ, v) and (V, ψ, w) be two elements in T
P
. That is (U, φ) and (V, ψ) are two charts with
P ∈ U and P ∈ V . By the deﬁnition,
ψ ◦ φ
−1
: φ(U ∩ V )

ψ(U ∩ V )
is diﬀeomorphism and so it induces an isomorphism of vector spaces
T(ψ ◦ φ
−1
): T
φ(P)
(φ(U ∩ V ))

T
ψ(P)
(ψ(U ∩ V )).
Now (U, φ, v) is called equivalent to (V, ψ, w), denoted by (U, φ, v) ∼ (V, ψ, w), if
T(ψ ◦ φ
−1
)(v) = w.
14 JIE WU
Deﬁne T
P
(M) to be the quotient
T
P
(M) = T
P
/ ∼ .
Exercise 2.1. Let M be a diﬀerentiable nmanifold and let P be any point in M. Prove that
T
P
(M) is an ndimensional vector space. [Hint: Fixed a chart (U, φ) and deﬁned
a(U, φ, v) +b(U, φ, w) := (U, φ, av +b w).
Now given any (V, ψ, x), (
˜
V ,
˜
ψ, y) ∈ T
P
, consider the map
φ ◦ ψ
−1
: ψ(U ∩ V ) → φ(U ∩ V ) φ ◦
˜
ψ
−1
:
˜
ψ(U ∩
˜
V ) → φ(U ∩
˜
V )
and deﬁne
a(V, ψ, x) +b(
˜
V ,
˜
ψ, y) = (U, φ, aT(φ ◦ ψ
−1
)(x) +bT(φ ◦
˜
ψ
−1
)(y)).
Then prove that this operation gives a welldeﬁned 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, φ)
with P ∈ U, there is a unique isomorphism
T
φ
: T
P
(M) → T
φ(P)
(φ(U)).
by choosing (U, φ, v) as representatives for its equivalence class. If (V, ψ) is another chart with
P ∈ V , then there is a commutative diagram
(1)
T
P
(M)
T
φ
∼
=

T
φ(P)
(φ(U ∩ V ))
T
P
(M)
¸
¸
¸
¸
¸
¸
¸
¸
¸
T
ψ
∼
=

T
ψ(P)
(ψ(U ∩ V )),
T(ψ ◦ φ
−1
)
?
where T(ψ ◦ φ
−1
) is the linear isomorphism induced by the Jacobian matrix of the diﬀerentiable
map ψ ◦ φ
−1
: φ(U ∩ V ) → ψ(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
φ
∼
=

T
φ(P)
(φ(U))
T
f(P)
(N)
Tf
?
T
ψ
∼
=

T
ψ(f(P))
(ψ(V ))
T(ψ ◦ f ◦ φ
−1
)
?
commutes for any chart (U, φ) with P ∈ U and any chart (V, ψ) with f(P) ∈ V . [First ﬁx a choice
of (U, φ) with P ∈ U and (V, ψ) with f(P) ∈ V , the linear transformation Tf is uniquely deﬁned
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, φ) about P and (V, ψ) about f(P) such that the diagram
U
f[
U

V
R
m
φ(P) = 0 φ
?
⊂
canonical coordinate inclusion

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

V
φ(U)
φ
?
g = ψ ◦ f ◦ φ
−1

ψ(V )
ψ
?
R
m
?
∩
R
n
.
?
∩
By the assumption,
Tg : T
0
(φ(U))

T
0
(ψ(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 ﬁrst m rows
forms an invertible matrix A
m×m
at the origin.
Deﬁne a function
h = (h
1
, h
2
, . . . , h
n
): φ(U) R
n−m

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
m×m
0
m×(n−m)
B
(n−m)×m
I
n−m
¸
,
where B is taken from (m+ 1)st row to nth row in the matrix (2). Thus the Jacobian of h is not
zero at the origin. By the Inverse Mapping Theorem, h is an diﬀeomorphism 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
φ(
˜
U)
φ[
˜
U
?
g = ψ ◦ f ◦ φ
−1

ψ(
˜
V )
ψ[
˜
V
?
φ(
˜
U) 0
¸
¸
¸
¸
¸
¸
¸
¸
¸
⊂ 
φ(
˜
U) U
2
∼
= h
−1
?
R
m
= R
m
0
?
∩
⊂ 
R
n
,
?
∩
where U
2
is a small neighborhood of the origin in R
n−m
.
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 onetoone 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 diﬀeomorphism.
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), ψ[
f(U)
)¦ gives an atlas for f(M) such that f(M) is
a submanifold of M. Now f : M → f(M) is a diﬀeomorphism because it is locally diﬀeomorphism
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, φ) about P and (V, ψ) about f(P) such that the diagram
U
f[
U

V
R
m
φ(P) = 0 φ
?
⊂
canonical coordinate proj.

R
n
ψ(f(P)) = 0 ψ
?
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 preimage of Q.
Theorem 2.14 (Preimage 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),
φ[
f
−1
(Q)
: f
−1
(Q) ∩ U

R
m−n
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, ψ) of N about f(P) such
that V = V
1
V
2
with V
1
= V ∩ Z and (V
1
, ψ[
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 Preimage 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 ktuples 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 kdimensional subspaces of R
n
, that is, all kplanes
through the origin. Let
π:
˜
V
k,n
→ G
k,n
be the quotient by sending ktuples of linearly independent vectors to the kplanes spanned by k
vectors. The topology in G
k,n
is given by quotient topology of π, namely, U is an open set of G
k,n
if and only if π
−1
(U) is open in
˜
V
k,n
.
For (v
1
, . . . , v
k
)
T
∈
˜
V
k,n
, write 'v
1
, . . . , v
k
` for the kplane spanned by v
1
, . . . , v
k
. Observe that
two ktuples (v
1
, . . . , v
k
)
T
and ( w
1
, . . . , w
k
)
T
spanned the same kplane 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 identiﬁcation 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 Hausdorﬀ. If k = n, then G
n,n
is only one point. So we assume that
k < n. Given an kplane X and w ∈ R
n
, let ρ
w
be the square of the Euclidian distance from w to
X. Let ¦e
1
, . . . , e
k
¦ be the orthogonal basis for X, then
ρ
w
(X) = w w −
k
¸
j=1
( w e
j
)
2
.
Fixing any w ∈ R
n
, we obtain the continuous map
ρ
w
: G
k,n

R
because ρ
w
◦ π:
˜
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 ◦ π 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 ρ
w
(X) = ρ
w
(Y ). Let V
1
and V
2
be disjoint
open subsets of R such that ρ
w
(X) ∈ V
1
and ρ
w
(Y ) ∈ V
2
. Then ρ
−1
w
(V
1
) and ρ
−1
w
(V
2
) are two open
subset of G
k,n
that separate X and Y , and so G
k,n
is Hausdorﬀ.
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 α such that U
X
∼
= R
k(n−k)
.
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 kmatrix. Fixing Q, deﬁne
X
α
= ¦(P
k
, B
k,n−k
)Q [ det(P
k
) = 0, B
k,n−k
∈ M(k, n −k)¦ ⊆
˜
V
k,n
.
Then E
X
is an open subset of
˜
V
k,n
. Let U
X
= π(E
X
) ⊆ G
k,n
. Since
π
−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`
φ
−1
X

U
X
,
π
?
U
X
is homeomorphic to M(k, n −k) = R
k(n−k)
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)
φ
−1
X

U
X
A → '(I
k
, A)Q`
M(k, n −k)
φ
−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
φ
X
(U
X
∩ U
Y
)
φ
−1
X

U
X
∩ U
Y
φ
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
n−1
. From the above, RP
n−1
is a manifold of dimension n −1.
3.2. Stiefel Manifold. The Stiefel manifold, denoted by V
k,n
, is deﬁned 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 Preimage 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
s→0
f(A+sB) −f(A)
s
= lim
s→0
(A+sB)(A+sB)
T
−AA
T
s
= lim
s→0
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 identiﬁcation 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)). Deﬁne
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 Preimage
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(n−1)
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
n−1
which is manifold of dimension n −1.
When k = n −1, then V
n−1,n
is a manifold of dimension
(n−1)n
2
. One can check that
V
n−1,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 kfold
Cartesian product of S
n−1
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 Hausdorﬀ space
is compact. The composite
V
k,n
⊂  ˜
V
k,n
π

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 π 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 identiﬁ
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 ﬁbre of the bundle over b ∈ B.
Intuitively, a bundle can be thought as a union of ﬁbres f
−1
(b) for b ∈ B parameterized by B
and glued together by the topology of the space E. Usually a Greek letter ( ξ, η, ζ, λ, etc) is used to
denote a bundle; then E(ξ) denotes the total space of ξ, and B(ξ) denotes the base space of ξ.
A morphism of bundles (φ,
¯
φ): ξ → ξ
is a pair of (continuous) maps φ: E(ξ) → E(ξ
) and
¯
φ: B(ξ) → B(ξ
) such that the diagram
E(ξ)
φ

E(ξ
)
B(ξ)
p(ξ)
?
¯
φ

B(ξ
)
p(ξ
)
?
commutes.
The trivial bundle is the projection of the Cartesian product:
p: B F → B, (x, y) → x.
Roughly speaking, a ﬁbre bundle p: E → B is a “locally trivial” bundle with a “ﬁxed ﬁbre” 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 φ
U
: p
−1
(U) → U F such that the diagram
U F
φ
x

p
−1
(U)
U
π
1
?
========= U
p
?
commutes, that is, p(φ(x
, y)) = x
for any x
∈ U and y ∈ F.
Similar to manifolds, we can use “chart” to describe ﬁbre bundles. A chart (U, φ) for a bundle
p: E → B is (1) an open set U of B and (2) a homeomorphism φ: U F → p
−1
(U) such that
p(φ(x
, y)) = x
for any x
∈ U and y ∈ F. An atlas is a collection of charts ¦(U
α
, φ
α
)¦ such that
¦U
α
¦ is an open covering of B.
Proposition 4.1. A bundle p: E → B is a ﬁbre bundle with ﬁbre F if and only if it has an atlas.
Proof. Suppose that p: E → B is a ﬁbre bundle. Then the collection ¦(U(x), φ
x
)[x ∈ B¦ is an atlas.
Conversely suppose that p: E → B has an atlas. For any x ∈ B there exists α such that x ∈ U
α
and so U
α
is an open neighborhood of x with the property that p[
p
−1
(U)
: p
−1
(U
α
) → U
α
is a trivial
bundle. Thus p: E → B is a ﬁbre bundle.
Let ξ be a ﬁbre bundle with ﬁbre F and an atlas ¦(U
α
, φ
α
)¦. The composite
φ
−1
α
◦ φ
β
: (U
α
∩ U
β
) F
φ
β

p
−1
(U
α
∩ U
β
)
φ
−1
α

(U
α
∩ U
β
) F
has the property that
φ
−1
α
◦ φ
β
(x, y) = (x, g
αβ
(x, y))
for any x ∈ U
α
∩ U
β
and y ∈ F. Consider the continuous map g
αβ
: U
αβ
F → F. Fixing any x,
g
αβ
(x, −): F → F, y → g
αβ
(x, y) is a homeomorphism with inverse given by g
βα
(x, −). This gives
a transition function
g
αβ
: U
α
∩ U
β

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
αβ
¦ satisfy the following equation
(3) g
αβ
(x) ◦ g
βγ
(x) = g
αγ
(x) x ∈ U
α
∩ U
β
∩ U
γ
.
By choosing α = β = γ, g
αα
(x) ◦ g
αα
(x) = g
αα
(x) and so
(4) g
αα
(x) = x x ∈ U
α
.
By choosing α = γ, g
αβ
(x) ◦ g
βα
(x) = g
αα
(x) = x and so
(5) g
βα
(x) = g
αβ
(x)
−1
x ∈ U
α
∩ U
β
.
We need to introduce a topology on Homeo(F, F) such that the transition functions g
αβ
are
continuous. The topology on Homeo(F, F) is given by compactopen topology brieﬂy 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 compactopen topology is generated by W
K,U
, that is, an open set in Map(X, Y ) is an
arbitrary union of a ﬁnite 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 compactopen topology, then the transition functions
g
αβ
: U
α
∩ U
β
→ Homeo(F, F) are continuous.
Proof. Given W
K,U
, we show that g
−1
αβ
(W
K,U
) is open in U
α
∩ U
β
. Let x
0
∈ U
α
∩ U
β
such that
g
αβ
(x
0
) ∈ W
(K,U)
. We need to show that there is a neighborhood V is x
0
such that g
αβ
(V ) ⊆ W
K,U
,
or g
αβ
(V K) ⊆ U. Since U is open and g
αβ
: (U
α
∩ U
β
) F → F is continuous, g
−1
(U) is an
open set of (U
α
∩U
β
) F with x
0
K ⊆ g
−1
αβ
(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
αβ
(U). Since ¦N(y) [ y ∈ K¦ is an open cover
of the compact set K, there is a ﬁnite cover ¦N(y
1
), . . . , N(y
n
)¦ of K. Let V =
n
¸
j=1
V (y
j
). Then
V K ⊆ g
−1
αβ
(U) and so g
αβ
(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
◦ g
∈ 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 Hausdorﬀ, 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
compactopen topology only fails in the continuity of g
−1
. A modiﬁcation on compactopen topology
eliminates this defect [1].
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 25
4.2. GSpaces and Principal GBundles. Let G be a topological group and let X be a space.
A right Gaction on X means a(continuous) map µ: 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) Gspace. Let X and Y be (right) Gspaces.
A continuous map f : X → Y is called a Gmap if f(x g) = f(x) g for any x ∈ X and g ∈ G. Let
X/G be the set of Gorbits xG, x ∈ X, with quotient topology.
Proposition 4.5. Let X be a Gspace.
1) For ﬁxing any g ∈ G, the map x → x g is a homeomorphism.
2) The projection π: 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,
π
−1
(π(U)) =
¸
g∈G
U g
is open because it is a union of open sets, and so π(U) is open by quotient topology. Thus π is an
open map.
We are going to ﬁnd some conditions such that π: X → X/G has canonical ﬁbre bundle structure
with ﬁbre G. Given any point ¯ x ∈ X/G, choose x ∈ X such that π(x) = ¯ x. Then
π
−1
(¯ x) = ¦x g [ g ∈ G¦ = G/H
x
,
where H
x
= ¦g ∈ G [ x g = x¦.
For having constant ﬁbre G, we need to assume that the Gaction 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 Gspace.
Since a ﬁbre 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 π: X →
X/G has local crosssections. 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 π ◦ s
¯ x
= id
U(¯ x)
.
(Note. For every point ¯ x, we can always choose a preimage of π, the local crosssection means
the preimages 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 Gspace with local crosssections to π: X → X/G. Let ¯ x be any
point in X/G. Let U(¯ x) be a neighborhood of ¯ x with a (continuous) crossesection s
¯ x
: U(¯ x) → X.
Deﬁne
φ
¯ x
: U(¯ x) G

π
−1
(U(¯ x)) (¯ y, g)

s
¯ x
(¯ y) g
for any y ∈ U(¯ x).
Exercise 4.2. Let X be a (right) free Gspace with local crosssections to π: X → X/G. Then the
continuous map φ
¯ x
: U(¯ x) G → π
−1
(U(¯ x)) is onetoone and onto.
We need to ﬁnd the third condition such that φ
¯ x
is a homeomorphism. Let
X
∗
= ¦(x, x g) [ x ∈ X, g ∈ G¦ ⊆ X X.
26 JIE WU
A function
τ : X
∗

G
such that
x τ(x, x
) = x
for all (x, x
) ∈ X
is called a translation function. (Note. If X is a free Gspace, 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 deﬁnition,
τ(x, x
) = g.)
Proposition 4.6. Let X be a (right) free Gspace with local crosssections to π: X → X/G. Then
the following statements are equivalent each other:
1) The translation function τ : X
∗
→ G is continuous.
2) For any ¯ x ∈ X/G, the map φ
¯ x
: U(¯ x) G → π
−1
(U(¯ x)) is a homeomorphism.
3) There is an atlas ¦(U
α
, φ
α
¦ of X/G such that the homeomorphisms
φ
α
: U
α
G

π
−1
(U
α
)
satisfy the condition φ
α
(¯ y, gh) = φ
α
(¯ y, g) h, that is φ
α
is a homeomorphism of Gspaces.
Proof. (1) =⇒ (2). Consider the (continuous) map
θ: π
−1
(U(¯ x))

U(¯ x) G z → (π(z), τ(s
¯ x
(π(z)), z)).
Then
θ ◦ φ
¯ x
(¯ y, g) = θ(s
¯ x
(¯ y) g) = (¯ y, τ(s
¯ x
(¯ y), s
¯ x
(¯ y) g)) = (¯ y, g),
φ
¯ x
◦ θ(z) = φ
¯ x
(π(z), τ(s
¯ x
(π(z)), z)) = s
¯ x
(π(z)) τ(s
¯ x
(π(z)), z) = z.
Thus φ
¯ x
is a homeomorphism.
(2) =⇒ (3) is obvious.
(3) =⇒ (1). Note that the translation function is unique for free Gspaces. It suﬃces to show
that the restriction
τ(X): X
∗
∩ (π
−1
(U
α
) π
−1
(U
α
)) =
π
−1
(U
α
)
∗

G
is continuous. Consider the commutative diagram
(U
α
G)
∗
φ
∗
α
∼
=

π
−1
(U
α
)
∗
G
τ(U
α
G)
?
=========== G.
τ(X)
?
Since
τ(U
α
G)((¯ y, g), (¯ y, h)) = g
−1
h
is continuous, the translation function restricted to
π
−1
(U
α
)
∗
τ(X) = τ(U
α
G) ◦ ((φ
α
)
∗
)
−1
is continuous for each α and so τ(X) is continuous.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 27
Now we give the deﬁnition. A principal Gbundle is a free Gspace X such that
π: X → X/G
has local crosssections and one of the (equivalent) conditions in Proposition 4.6 holds.
Example. Let Γ be a topological group and let G be a closed subgroup. Then the action of G
on Γ given by (a, g) → ag for a ∈ Γ and g ∈ G is free. Then translation function is given by
τ(a, b) = a
−1
b, which is continuous. Thus Γ → Γ/G is principal Gbundle if and only if it has local
crosssections.
4.3. The Associated Principal GBundles of Fibre Bundles. We come back to look at ﬁbre
bundles ξ given by p: E → B with ﬁbre F. Let ¦(U
α
, φ
α
)¦ be an atlas and let
g
αβ
: U
α
∩ U
β

Homeo(F, F)
be the transition functions. A topological group G is called a group of the bundle ξ if
1) There is a group homomorphism
θ: G

Homeo(F, F).
2) There exists an atlas of ξ such that the transition functions g
αβ
lift to G via θ, that is,
there is commutative diagram
U
α
∩ U
β
g
αβ

Homeo(F, F)
U
α
∩ U
β
¸
¸
¸
¸
¸
¸
¸
¸
¸
g
αβ

G,
θ
6
(where we use the same notation g
αβ
.)
3) The transition functions
g
αβ
: U
α
∩ U
β

G
are continuous.
4) The Gaction on F via θ is continuous, that is, the composite
GF
θ×id
F

Homeo(F, F) F
evaluation

F
is continuous.
We write
¯
ξ = ¦(U
α
, g
αβ
)¦ for the set of transition functions to the atlas ¦(U
α
, φ
α
)¦.
Note. In Steenrod’s deﬁnition [11, p.7], θ is assume to be a monomorphism (equivalently, the
Gaction on F is eﬀective, that is, if y g = y for all y ∈ F, then g = 1.).
We are going to construct a principal Gbundle π: E
G
→ B. Then prove that the total space
E = F
G
E
G
and p: E → B can be obtained canonically from π: E
G
→ B. In other words,
all ﬁbre bundles can obtained through principal Gbundles through this way. Also the topological
group G plays an important role for ﬁbre bundles. Namely, by choosing diﬀerent topological groups
G, we may get diﬀerent properties for the ﬁbre bundle ξ. For instance, if we can choose G to be
trivial (that is, g
αβ
lifts to the trivial group), then ﬁbre bundle is trivial. We will see that the bundle
group G for ndimensional 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
¯
ξ is the set of transition functions for the space B and topological group G,
then there is a principal Gbundle ξ
G
given by
π: E
G

B
and an atlas ¦(U
α
, φ
α
)¦ such that
¯
ξ is the set of transition functions to this atlas.
Proof. The proof is given by construction. Let
¯
E =
¸
α
U
α
Gα,
that is
¯
E is the disjoint union of U
α
G. Now deﬁne a relation on
¯
E by
(b, g, α) ∼ (b
, g
, β) ⇐⇒ b = b
, g = g
αβ
(b)g
.
This is an equivalence relation by Equations (3)(5). Let E
G
=
¯
E/ ∼ with quotient topology and
let ¦b, g, α¦ for the class of (b, g, α) in E
G
. Deﬁne π: E
G
→ B by
π¦b, g, α¦ = b,
then π is clearly welldeﬁned (and so continuous). The right Gaction on E
G
is deﬁned by
¦b, g, α¦ h = ¦b, gh, α¦.
This is welldeﬁned (and so continuous) because if (b
, g
, β) ∼ (b, g, α), then
(b
, g
h, β) = (b, (g
αβ
(b)g)h, β) = (b, g
αβ
(b)(gh), β) ∼ (b, gh, α).
Deﬁne φ
α
: U
α
G → π
−1
(U
α
) by setting
φ
α
(b, g) = ¦b, g, α¦,
then φ
α
is continuous and satisﬁes π ◦ φ
α
(b, g) = b and
φ
α
(b, g) = ¦b, 1 g, α¦ = ¦b, 1, α¦ g
for b ∈ U
α
and g ∈ G. The map φ
α
is a homeomorphism because, for ﬁxing α, the map
¸
β
(U
α
∩ U
β
) Gβ

U
α
G (b, g
, β) → (b, g
αβ
(b)g
)
induces a map π
−1
(U
α
) → U
α
G which the inverse of φ
α
. Moreover,
φ
α
(b, g
αβ
(b)g) = ¦b, g
αβ
(b)g, α¦ = ¦b, g, β¦ = θ
β
(b, g)
for b ∈ U
α
∩ U
β
and g ∈ G. Thus the ¦(U
α
, g
αβ
)¦ is the set of transition function to the atlas
¦(U
α
, φ
α
)¦.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 29
Let X be a right Gspace and let Y be a left Gspace. The product over G is deﬁned by
X
G
Y = X Y/(xg, y) ∼ (x, gy)
with quotient topology. Note that the composite
X Y
π
X

X
π

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 ∈ π
−1
(¯ x) ⊆ X, then
p
−1
(¯ x) = π
−1
(¯ x)
G
Y = x Y/H
x
,
where H
x
= ¦g ∈ G [ xg = x¦. Thus if X is a free right Gspace, then the projection p: X
G
Y →
X/G has the constant ﬁbre Y .
Proposition 4.8. Let π: X → X/G be a (right) principal Gbundle and let Y be any left Gspace.
Then
p: X
G
Y

X/G
is a ﬁbre bundle with ﬁbre Y .
Proof. Consider a chart (U
α
, φ
α
) for π: X → X/G. Since the homeomorphism φ
α
: U
α
G →
π
−1
(U
α
) is a Gmap, there is a commutative diagram
U
α
Y === (U
α
G)
G
Y
φ
α
∼
=

π
−1
(U
α
)
G
Y === p
−1
(U
α
)
U
α
π
U
α
?
=========== U
α
π
U
α
?
============= U
α
p
?
========== U
α
p
?
and hence the result.
Let ξ be a (right) principal Gbundle given by π: X → X/G. Let Y be any left Gspace. Then
ﬁbre bundle
p: X
G
Y

X/G
is called induced ﬁbre bundle of ξ, denoted by ξ[Y ].
Now let p: E → B is a ﬁbre bundle with ﬁbre 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
θ: G → Homeo(F, F).
A bundle morphism
E(ξ)
φ

E(ξ
)
B(ξ)
p(ξ)
?
¯
φ

B(ξ
)
p(ξ
)
?
30 JIE WU
is call an isomorphism if both φ and
¯
φ are homeomorphisms. (Note. this means that (φ
−1
, (
¯
φ)
−1
)
are continuous.) In this case, we write ξ
∼
= ξ
.
Theorem 4.9. Let ξ be a ﬁbre bundle given by p: E → B with ﬁbre F and bundle group G. Let ξ
G
be the principal Gbundle constructed in Proposition 4.7 according to a set of transitions functions
to ξ Then ξ
G
[F]
∼
= ξ.
Proof. Let ¦(U
α
, φ
α
)¦ be an atlas for ξ. We write
˜
φ
α
for φ
α
in the proof of Proposition 4.7. Consider
the map θ
α
given by the composite:
π
−1
(U
α
)
G
F
˜
φ
α
id
F
∼
=
(U
α
Gα)
G
F === U
α
F
φ
α
∼
=

p
−1
(U
α
).
From the commutative diagram
((U
α
∩ U
β
) Gβ)
G
F
(b, g
, y) → (b, g
, g
αβ
(b)(y))
((b, g
, β), y) → ((b, g
αβ
(b)g
, α), y)

((U
α
∩ U
β
) Gα)
G
F
(U
α
∩ U
β
) F
¸
¸
¸
¸
¸
¸
¸
¸
¸
(b, y) → (b, g
αβ
(b, y))

(U
α
∩ U
β
) F
¸
¸
¸
¸
¸
¸
¸
¸
¸
p
−1
(U
α
∩ U
β
)
∼
= φ
β
?
=========================================== p
−1
(U
α
∩ U
β
),
∼
= φ
α
?
the map θ
α
induces a bundle map
E
G
G
F
θ

E(ξ)
B(ξ)
?
====== B(ξ).
?
This is an bundle isomorphism because θ is onetoone and onto, and θ is a local homeomorphic by
restricting each chart. The assertion follows.
This theorem tells that any ﬁbre bundle with a bundle group G is an induced ﬁbre bundle of
a principal Gbundle. Thus, for classifying ﬁbre bundles over a ﬁxed base space B, it suﬃces to
classify the principal Gbundles 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 ﬁbre bundles is also called (unstable) Ktheory, which is
one of important applications of homotopy theory to geometry. Rough introduction to this theory
is as follows:
There exists a universal Gbundle ω
G
as π: EG → BG. Given any principal Gbundle ξ over B,
there exists a (continuous) map f : B → BG such that ξ, as a principal Gbundle, is isomorphic to
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 31
the pullback bundle f
∗
ω
G
given by
E(f
∗
ω
G
) = ¦(x, y) ∈ B EG [ f(x) = π(y)¦

EG
B
(x, y) → x
?
f

BG.
π
?
Moreover, for continuous maps f, g : B → BG, f
∗
ω
G
∼
= g
∗
ω
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 onetoone correspondent to the set of
isomorphic classes of principal Gbundles over G.
Seminar Topic: The classiﬁcation of principal Gbundles and ﬁbre bundles. (References: for
instance [8, pp.4858] Or [9, 10].)
4.4. Vector Bundles. Let F denote R, C or Hthe real, complex or quaternion numbers. An
ndimensional Fvector bundle is a ﬁbre bundle ξ given by p: E → B with ﬁbre F
n
and an atlas
¦(U
α
, φ
α
)¦ in which each ﬁbre p
−1
(b), b ∈ B, has the structure of vector space over F such that
each homeomorphism φ
α
: U
α
F
n
→ p
−1
(U
α
) has the property that
φ
α
[
{b}×F
n : ¦b¦ F
n

p
−1
(b)
is a vector space isomorphism for each b ∈ U
α
.
Let ξ be a vector bundle. From the composite
(U
α
∩ U
β
) F
n
φ
β

p
−1
(U
α
∩ U
β
)
φ
−1
α

(U
α
∩ U
β
) F
n
,
the transition functions
g
αβ
: U
α
∩ U
β

Homeo(F
n
, F
n
)
have that property that, for each x ∈ U
αβ
,
g
αβ
(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 ξ be an ndimensional Fvector space over B. Then there exists a principal
GL
n
(F)bundle ξ
GL
n
(F)
over B such that ξ
∼
= ξ
GL
n
(F)
[F
n
]. Conversely, for any principal GL
n
(F)
bundle over B, ξ
GL
n
(F)
[F
n
] is an ndimensional Fvector bundle over B.
In other words, the total spaces of all vector bundles are just given by E(ξ
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 Fsubspaces of F
m
, that is, all nFplanes 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(γ
m
n
) = ¦(V, x) ∈ G
n,m
(F) F
m
[ x ∈ V ¦.
32 JIE WU
Exercise 4.3. Show that
p: E(γ
m
n
) → G
n,m
(F) (V, x) → V
is an ndimensional Fvector bundle, denoted by γ
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(γ
m
n
) = V
n,m
(F)
O(n,F)
F
n
.]
A Gauss map of an ndimensional Fvector bundle in F
m
(n ≤ m ≤ ∞) is a (continuous) map
g : E(ξ) →F
m
such that g restricted to each ﬁbre is a linear monomorphism.
Example. The map
q : E(γ
m
n
) →F
m
(V, x) → x
is a Gauss map.
Proposition 4.11. Let ξ be an ndimensional Fvector bundle.
1) If there is a vector bundle morphism
E(ξ)
u

E(γ
m
n
)
B(ξ)
p(ξ)
?
f

G
n,m
(F)
p(γ
m
n
)
?
that is an isomorphism when restricted to any ﬁbre of ξ, then q ◦ u: E(ξ) →F
m
is a Gauss
map.
2) If there is a Gauss map g : E(ξ) →F
m
, then there is a vector bundle morphism (u, f): ξ →
γ
m
n
such that qu = g.
Proof. (1) is obvious. (2). For each b ∈ B(ξ), g(p(ξ)
−1
(b)) is an ndimensional Fsubspace of F
m
and so a point in G
n,m
(F). Deﬁne the functions
f : B(ξ) → G
n,m
(F) f(b) = g(p(ξ)
−1
(b)),
u: E(ξ) → E(γ
m
n
) u(z) = (f(p(z)), g(z)).
The functions f and u are welldeﬁned. For checking the continuity of f and u, one can look at a
local coordinate of ξ and so we may assume that ξ is a trivial bundle, namely, g : B(ξ) F
n
→F
m
restricted to each ﬁbre is a linear monomorphism. Let ¦e
1
, . . . , e
n
¦ be the standard Fbases 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 ﬁbre 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(ξ)
u

E(γ
m
n
)
E(ξ) E(ξ)
∆
?
(f ◦ p(ξ)) g

G
n,m
(F) F
m
?
∩
is continuous. This ﬁnishes the proof.
Let ξ be a vector bundle and let f : X → B(ξ) be a (continuous) map. Then the induced vector
f
∗
ξ is the pullback
E(f
∗
ξ) = ¦(x, y) ∈ X E(ξ) [f(x) = p(y)¦

E(ξ)
X
?
f

B(ξ).
p
?
Proposition 4.12. There exists a Gauss map g : E(ξ) →F
m
(n ≤ m ≤ ∞) if and only if
ξ
∼
= f
∗
(γ
m
n
)
over B(ξ) for some map f : B(ξ) → G
n,m
(F).
Proof. ⇐= is obvious.
=⇒ Assume that ξ has a Gauss map g. From Part (2) of Proposition 4.11, there is a commutative
diagram
E(ξ)
u

E(γ
m
n
)
B(ξ)
p(ξ)
?
f

G
n,m
(F).
p(γ
m
n
)
?
Since E(f
∗
γ
m
n
) is deﬁned to be the pullback, there is commutative diagram
E(ξ)
˜ u

E(f
∗
γ
m
n
)
B(ξ)
p(ξ)
?
====== B(ξ),
p
?
where ˜ u restricted to each ﬁbre is a linear isomorphism because both vectorbundle has the same
dimension and the Gauss map g restricted to each ﬁbre is a monomorphism. It follows that
34 JIE WU
˜ u: E(ξ) → E(f
∗
γ
m
n
) is onetoone 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, 162169].
A family of ( = ¦C
α
[ J¦ of subsets of a space X is called locally ﬁnite if each x ∈ X admits a
neighborhood W
x
such that W
x
∩ C
α
= ∅ for only ﬁnitely many indices α ∈ J. Let  = ¦U
α
¦ and
1 = ¦V
β
¦ be two open covers of X. 1 is called a reﬁnement of  if for each β, V
β
⊆ U
α
for some α.
A Hausdorﬀ space X is called paracompact if it is regular and if every open cover of X admits a
locally ﬁnite reﬁnement.
Let  = ¦U
α
[ α ∈ J¦ be an open cover of a space X. A partition of unity, subordinate to , is
a collection ¦λ
α
[ α ∈ J¦ of continuous functions λ
α
: X → [0, 1] such that
1) The support
supp(λ
α
) ⊆ U
α
for each α, where the support
supp(λ
α
) = ¦x ∈ X [ λ
α
(x) = 0¦
is the closure of the subset of X on which λ
α
= 0;
2) for each x ∈ X, there is a neighborhood W
x
of x such that λ
λ
α
[
W
x
≡ 0 for only ﬁnitely
many indices α ∈ J. (In other words, the supports of λ
α
’s are locally ﬁnite.)
3) The equation
¸
α∈J
λ
α
(x) = 1
for all x ∈ X, where the summation is welldeﬁned for each given x because there are only
ﬁnitely many nonzeros.
We give the following wellknown theorem without proof. One may read a proof in [2, pp.1720].
Theorem 4.13. If X is a paracompact space and  = ¦U
α
¦ is an open cover of X, then there
exists a partition of unity subordinate to .
Lemma 4.14. Let ξ be a ﬁbre bundle over a paracompact space B. Then ξ admits an atlas with
countable charts.
Proof. Let ¦(U
α
, φ
α
[ α ∈ J¦ be an atlas for ξ. We are going to ﬁnd another atlas with countable
charts.
By Theorem 4.13, there is a partition of unity ¦λ
α
[ α ∈ J¦ subordinate to ¦U
α
[ α ∈ J¦. Let
V
α
= λ
−1
α
(0, 1] = ¦b ∈ B [ λ
α
(b) > 0¦.
Then, by the deﬁnition of partition of unity, V
α
⊆
¯
V
α
⊆ U
α
. For each b ∈ B, let
S(b) = ¦α ∈ J [ λ
α
(b) > 0¦.
Then, by the deﬁnition of partition of unity, S(b) is a ﬁnite subset of J.
Now for each ﬁnite subset S of J deﬁne
W(S) = ¦b ∈ B [λ
α
(b) > λ
β
(b) for each α ∈ S and β ∈ S¦
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 35
=
¸
α ∈ S
β ∈ S
(λ
α
−λ
β
)
−1
(0, 1].
Then W(S) is open because for each b ∈ W(S), by deﬁnition of partition of unity, there exists a
neighborhood W
b
of b such that there are ﬁnitely many supports intersect with W
b
; and so the above
(possibly inﬁnite) intersection of open sets restricted to W
b
is only a ﬁnite 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 α ∈ S S
and β ∈ 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 deﬁnition W(S), λ
α
(b) > λ
β
(b) because α ∈ S and
β ∈ S. On the other hand, λ
β
(b) > λ
α
(b) because b ∈ W(S
), β ∈ S
and α ∈ S
.
Now deﬁne
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) ξ restricted to
W
m
is a trivial bundle for each m. (Then ¦W
m
¦ induces an atlas for ξ.)
To check ¦W
m
¦ is an open cover, note that each W
m
is open. For each b ∈ B, S(b) is a ﬁnite set
and b ∈ W(S(b)) because λ
β
(b) = 0 for β ∈ S(b) and λ
α
(b) > 0 for α ∈ S(b). Let m = [S(b)[, then
b ∈ W
m
and so ¦W
m
¦ is an open cover of B.
Now check that ξ restricted to W
m
is trivial. From the above, W
m
is a disjoint union of W(S(b)).
It suﬃces to check that ξ restricted to each W(S(b)) is trivial. Fixing α ∈ S(b), for any x ∈ W(S(b)),
then
λ
α
(x) > λ
β
(x)
for any β ∈ S. In particular, λ
α
(x) > 0 for any x ∈ W(S(b)). It follows that W(S(b)) ⊆ V
α
⊆ U
α
.
Since ξ restricted to U
α
is trivial, ξ restricted to W(S(b)) is trivial. This ﬁnishes the proof.
Note. From the proof, if for each b ∈ B there are at most k sets U
α
with b ∈ U
α
, then B admits
an atlas of ﬁnite (at most k) charts. [In this case, check that W
j
= ∅ for j > k.]
Theorem 4.15. Any ndimensional Fvector bundle ξ over a paracompact space B has a Gauss map
g : E(ξ) →F
∞
. Moreover, if ξ has an atlas of k charts, then ξ has a Gauss map g : E(ξ) →F
kn
.
Proof. Let ¦(U
i
, φ
i
)¦
1≤i≤k
be an atlas of ξ with countable or ﬁnite charts, where k is ﬁnite or inﬁnite.
Let ¦λ
i
¦ be the partition of unity subordinate to ¦U
i
¦. For each i, deﬁne the map g
i
: E(ξ) → F
n
as follows: g
i
restricted to p(ξ)
−1
(U
i
) is given by
g
i
(z) = λ
i
(z)(p
2
◦ φ
−1
i
(z)),
where p
2
◦ φ
−1
i
is the composite
p(ξ)
−1
(U
i
)
φ
−1
i

U
i
F
n
projection
p
2

F
n
;
36 JIE WU
and g
i
restricted to the outside of p(ξ)
−1
(U
i
) is 0. Since the closure of λ
−1
i
(0, 1] is contained in U
i
,
g
i
is a welldeﬁned (continuous) map. Now deﬁne
g : E(ξ) →
k
i=1
F
n
= F
kn
g(z) =
k
¸
i=1
g
i
(z).
This a welldeﬁned (continuous) map because for each z, there is a neighborhood of z such that
there are only ﬁnitely many g
i
are not identically zero on it.
Since each g
i
: E(ξ) →F
n
is a monomorphism (actually isomorphism) on the ﬁbres of E(ξ) over
b with λ
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 classiﬁcation theorem:
Corollary 4.16. Every vector bundle over a paracompact space B is isomorphic to an induced vector
bundle f
∗
(γ
∞
n
) for some map f : B → G
n,∞
(F). Moreover every vector bundle over a paracompact
space B with an atlas of ﬁnite charts is isomorphic to an induced vector bundle f
∗
(γ
m
n
) for some m
and some map f : B → G
n,m
(F).
Remarks: It can be proved that f
∗
γ
m
n
∼
= g
∗
γ
m
n
if and only if f · g : B → G
n,m
(F). From this,
one get that the set of isomorphism classes of ndimensional Fvector 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 socalled
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 ﬁrst cohomology with coeﬃcients 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
∞
] = π
n−1
(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 π
n
(S
3
).
Seminar Topic: Gauss Maps and the Classiﬁcation of Vector Bundles. (References: for instance [8,
pp.2629,3133].)
A vector bundle is called of ﬁnite type if it has an atlas with ﬁnite charts. Given two vector
bundles ξ and η over B, the Whitney sum ξ ⊕η is deﬁned to be the pullback:
E(ξ ⊕η)

E(ξ) E(η)
B
?
∆
diagonal

B B.
p(ξ) p(η)
?
Intuitively, ξ ⊕η is just the ﬁbrewise direct sum.
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 37
Proposition 4.17. For a vector bundle ξ over a paracompact space B, the following statement are
equivalent:
1) The bundle ξ is of ﬁnite type.
2) There exists a map f : B → G
n,m
(F) such that ξ is isomorphic to f
∗
γ
m
n
.
3) There exists a vector bundle η such that the Whitney sum ξ ⊕η is trivial.
Proof. (1) =⇒ (2) follows from Corollary 4.16. (2) =⇒ (1) It is an exercise to check that γ
m
n
is
of ﬁnite 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 ξ
∼
= f
∗
γ
m
n
is of ﬁnite type.
(2) =⇒ (3). Let (γ
m
n
)
∗
be the vector bundle given by
E((γ
m
n
)
∗
) = ¦(V, v) ∈ G
n,m
(F) F
m
[ v ⊥ V ¦
with canonical projection E((γ
m
n
)
∗
) → G
n,m
(F). Then γ
m
n
⊕ (γ
m
n
)
∗
is an mdimensional trivial
Fvector bundle. It follows that
f
∗
(γ
m
n
⊕(γ
m
n
)
∗
) = f
∗
(γ
m
n
) ⊕f
∗
((γ
m
n
)
∗
)
is trivial. Let η = f
∗
((γ
m
n
)
∗
). Then ξ ⊕η is trivial.
(3) =⇒ (2). The composite
E(ξ)
⊂ 
E(ξ ⊕η) = B F
m

F
m
is a Gauss map into ﬁnite dimensional vector space, where m = dim(ξ ⊕ η). By Proposition 4.12,
there is a map f : B → G
n,m
such that ξ
∼
= f
∗
γ
m
n
.
Corollary 4.18. Let ξ be a Fvector bundle over a compact (Hausdorﬀ) space B. Then there is a
Fvector bundle η such that ξ ⊕η is trivial.
In the view of (stable) Ktheory, the Whitney sum is an operation on vector bundles over a (ﬁxed)
basespace, where the trivial bundles (of diﬀerent 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 ξ
with property that ξ ⊕ η is trivial for some η means that ξ is invertible. Those who are interested
in algebra can push notions in algebra to vector bundles by doing constructions ﬁbrewisely. 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 ﬁxed space B, or even more general category whose objects
are diagrams over spaces. In the terminology of ﬁbre 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), 593610.
[2] Lawrence Conlon, Diﬀerentiable 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 504537, (1963).
[5] Serge Lang, Diﬀerential Manifolds, AddisonWesley Publishing Company, Inc. [1972].
[6] Victor Guillemin and Alan Pollack, Diﬀerential Topology, Prentice Hall, Inc, Englewood Cliﬀs, 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 SpringerVerlag [1966].
[9] J. Milnor, Construction of universal bundles I, Ann. Math. 63 (1956), 272284.
[10] J. Milnor, Construction of universal bundles II, Ann. Math. 63 (1956), 430436.
[11] Norman Steenrod, The Topology of Fibre Bundles, Princeton University Press, Princeton, New Jersey, Ninth
Printing, [1974].
2
JIE WU
1. Tangent Spaces, Vector Fields in Rn and the Inverse Mapping Theorem 1.1. Tangent Space to a Level Surface. Let γ be a curve in Rn : γ : t → (γ 1 (t), γ 2 (t), . . . , γ n (t)). (A curve can be described as a vectorvalued function. Converse a vectorvalued function gives a curve in Rn .) The tangent line at the point γ(t0 ) is given with the direction dγ (t0 ) = dt dγ 1 dγ n (t0 ), . . . , (t0 ) . dt t
(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 (x1 , x2 , . . . , xn ) = c of a diﬀerentiable function f , where xi refers to ith coordinate. The gradient vector of f at a point P = (x1 (P ), x2 (P ), . . . , xn (P )) is ∂f ∂f , . . . , n ). ∂x1 ∂x Given a vector u = (u1 , . . . , un ), the directional derivative is f =( ∂f 1 ∂f u + · · · + n un . ∂x1 ∂x The tangent space at the point P on the level surface f (x1 , . . . , xn ) = 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 ∂f (P )(x1 − x1 (P )) + · · · + n (P )(xn − xn (P )) = 0. ∂x1 ∂x 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 γ is a curve through P (when t = t0 ) that lies in the level surface f (x1 , . . . , xn ) = c, that is Du f = f ·u= f (γ 1 (t), γ 2 (t), . . . , γ n (t)) = c. By taking derivatives on both sides, ∂f ∂f (P )(γ 1 ) (t0 ) + · · · + n (P )(γ n ) (t0 ) = 0 ∂x1 ∂x and so the tangent line of γ is really normal (orthogonal) to f . When γ 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 Rn . Now consider the tangent space of Rn . According to the ideas in the previous subsection, ﬁrst we assume a given point P ∈ Rn . 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 ndimensional space. But we can look at in a little detail. Let γ be a smooth curve through P . We may assume that γ(0) = P . Let ω be another smooth curve with ω(0) = P . γ is called to be equivalent to ω if the directives γ (0) = ω (0). The tangent space of Rn at P , denoted by TP (Rn ), is then the set of equivalence class of all smooth curves through P .
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS
3
Let T (Rn ) =
P ∈Rn P ∈S
TP (Rn ), called the tangent bundle of Rn . If S is a region of Rn , let T (S) =
TP (S), called the tangent bundle of S. Note. Each TP (Rn ) is an ndimensional vector space, but T (S) is not a vector space. In other words, T (S) is obtained by attaching a vector space TP (Rn ) to each point P in S. Also S is assumed to be a region of Rn , otherwise the tangent space of S (for instance S is a level surface) could be a proper subspace of TP (Rn ). If γ is a smooth curve from P to Q in Rn , then the tangent space TP (Rn ) moves along γ to TQ (Rn ). The direction for this moving is given γ (t), which introduces the following important concept. Deﬁnition 1.1. A vector ﬁeld V on a region S of Rn 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 ﬁelds and let f : S → R be a smooth function. Then V + W : P → v(P ) + w(P ) and f V : P → f (P )v(P ) give (pointwise) addition and scalar multiplication structure on vector ﬁelds. 1.3. Operator Representations of Vector Fields. Let J be an open interval containing 0 and let γ : J → Rn be a smooth curve with γ(0) = P . Let f = f (x1 , . . . , xn ) be a smooth function deﬁned on a neighborhood of P . Assume that the range of γ is contained in the domain of f . By applying the chain rule to the composite T = f ◦ γ : J → R, Dγ (f ) := Proposition 1.2. Dγ (af + bg) = aDγ (f ) + bDγ (g), where a, b are constant. dT = dt
n
i=1
dγ i (t) ∂f dt ∂xi
xi =γ i (t)
Dγ (f g) = Dγ (f )g + f Dγ (g). Let C ∞ (Rn ) denote the set of smooth functions on Rn . An operation D on C ∞ (Rn ) is called a derivation if D maps C ∞ (Rn ) to C ∞ (Rn ) and satisﬁes the conditions D(af + bg) = aD(f ) + bD(g), Example: For 1 ≤ i ≤ n, ∂i : f → is a derivation. Proposition 1.3. Let D be any derivation on C ∞ (Rn ). Given any point P in Rn . Then there exist real numbers a1 , a2 , . . . , an ∈ R such that
n
where a, b are constant.
D(f g) = D(f )g + f D(g). ∂f ∂xi
D(f )(P ) =
i=1 ∞ n i
ai ∂i (f )(P )
for any f ∈ C (R ), where a depends on D and P but is independent on f .
n D(f (x)) = D(f (x) − f (P )) = i=1 n D(xi − xi (P ))gi (x) + (xi − xi (P ))D(gi (x)) = i=1 i D(xi )gi (x) + (xi − xi (P ))D(gi (x)) because D(f (P )) = D(x (P )) = 0. From this proposition. . D(1) = D(1 · 1) = D(1) · 1 + 1 · D(1) and so D(1) = 0. . . By applying D to the above equations. ∂2 . Let ai = D(xi )(P ) which only depends on D and P . Deﬁne 1 gi (x) = 0 ∂f (t(x − P ) + P )dt. . The operation v i (x)∂i is given as follows: for any f ∈ C ∞ (Rn ). Since gi (P ) = 0 1 ∂f (t(P − P ) + P )dt = ∂xi n ∂f ∂f (P )dt = (P ) = ∂i (f )(P ). i ∂x ∂xi D(f )(P ) = i=1 ai ∂i (f )(P ). . i=1 Since D is a derivation. n Dv (f )(P ) = i=1 v i (P ) · ∂i (f )(P ) for any P . xn ). n n D(f )(P ) = i=1 D(xi )(P )gi (P ) + 0 = i=1 1 0 ai gi (P ). v 2 (P ). . a derivation n Dv = i=1 ∞ n v i (P ) · ∂i on C (R ) is called an operator representation of the vector ﬁeld P → v(P ). v n (P )). ∂n }. the tangent spaces T (Rn ) admits a basis {∂1 . which is the conclusion. . . . It follows that D(c) = 0 for any constant c. . From this new view. ∂xi 1 Then f (x) − f (P ) = 0 1 n d f (t(x − P ) + P )dt dt = 0 n i=1 ∂f (t(x − P ) + P ) · (xi − xi (P ))dt ∂xi 1 = i=1 (xi − xi (P )) 0 ∂f (t(x − P ) + P )dt = ∂xi n (xi − xi (P ))gi (x). . . By evaluating at P . Write x for (x1 .4 JIE WU Proof. Note. we can give a new way to looking at vector ﬁelds: Given a vector ﬁelds P → v(P ) = (v 1 (P ). .
a2 ) = (a1 . x2 (P ). y(t)) is x (t) = v 1 (s(t)) = x(t) y (t) = v 2 (s(t)) = −y(t) with initial conditions (x(0). . Let n = 2 and let V : P → v(P ) = (v 1 (P ). . . . . s (0) = v(P ) Thus the statement follows from the fundamental theorem of ﬁrst order ODE. Example 1. a2 ) and (x (0). Integral Curves. Given a point P = (a1 . . a2 ). .LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 5 1. sn (t)) dt ········· n ds (t) = v n (s1 (t). .6. Given a point P = (a1 . deﬁned for suitable δ.4. . Let n = 2 and let V : P → v(P ) = (v 1 (P ). y(0)) = (a1 . . Let V : x → v(x) be a (smooth) vector ﬁeld on an neighborhood U of P . s2 (0). . . a2 ) and (x (0). Thus the solution is s(t) = (a1 et . such that s (t) = v(s(t)) for −δ < t < . . . . a2 et ). . . −a2 ). Theorem 1. s2 (t). a2 e−t ). . Then there exists an integral curve to V through P . Example 1. . Thus the solution is s(t) = (a1 et . The requirement for a curve s(t) = (s1 (t). . y) = y. y) = x and v 2 (x. . Proof. sn (t)) ds2 (t) = v 2 (s1 (t). sn (0)) = (x1 (P ). sn (t)) to be an integral curve is: 1 (t) dsdt = v 1 (s1 (t). where v 1 (x. . ) → U . . y) = x and v 2 (x. .5. a2 ). > 0. a2 ) = (a1 . y (0)) = v(a1 . The proof is given by assuming the fundamental existence and uniqueness theorem for systems of ﬁrst order diﬀerential equations. (0) dt dt = (v 1 (P ). . . y(t)) is x (t) = v 1 (s(t)) = x(t) y (t) = v 2 (s(t)) = y(t) with initial conditions (x(0). s2 (t). s2 (t). . where v 1 (x. Any two such curves agree on their common domain. the equation for the integral curve s(t) = (x(t). . v 2 (P )). v 2 (P )). . An integral curve to V is a smooth curve s : (−δ. xn (P )) ds1 dsn (0). Let V : x → v(x) be a (smooth) vector ﬁeld on an neighborhood U of P . a2 ). . the equation for the integral curve s(t) = (x(t). . y(0)) = (a1 . y) = −y.4. y (0)) = v(a1 . v n (P )). . . sn (t)) dt with the initial conditions s(0) = P (s1 (0).
. 0 = lim F (x1 (ks ). xn ) with values z ∈ (z0 − 0 0 c. xn . . x2 . . . . . . xn . . . . We may assume that ∂F (x1 . . There is 0 0 a unique z. . 0 0 2) g has continuous partial derivatives with F i (x1 . . x2 . xn ). Thus for ﬁxed 0 0 0 (x1 . . x2 . . (x1 . xn ) ∈ Nδ (x1 . . a neighborhood Nδ (x0 . xn ). . xn ). . . xn . there exists a small δ > 0 such that F (x1 . x2 . such that 1 n F (x1 . Fixed (x1 . .5. . . . zks ) 1 1 s→∞ = F ( lim x1 (ks ). . . xn . . . . . . . . xn . . . xn ) is continuous. . . . . . . x2 .6 JIE WU 1. xn . . . . . . . . xn ). z0 ) > 0. . x2 . . . . . . . xn ) ∈ Nδ (x1 . . xn ) where z = g(x1 . 0 0 0 0 0 0 0 0 0 Step 2. . . xn . . . . . F is strictly increasing on z in this neighborhood. . z0 − c) < 0 with (x1 . . xn ) and 0 0 0 F (x1 . . . . . . z0 − c < z < z0 + c. x2 . z0 + c) > 0 F (x1 . . . . . . . 0 0 0 0 ∂z 0 0 1 Then there is a neighborhood N (z0 ) ⊆ R. . xn ). 0 0 Proof. xn (ks ). 0 0 3) If F is smooth on D. . x2 . xn . . . xn . By the continuity of F . x2 . (x1 . z0 + c) > 0 0 0 0 0 0 0 with (x1 . . xn ). . xn (k)) be any sequence in Nδ (x1 . . xn . . . . z) ∂g 1 (x . . . . xn . . Step 4. . . z0 + c) ∈ N (x1 . . . . x2 . z0 − c). . . x . . . . . . xn . xn . . x2 . Let (x1 . xn ) is smooth on Nδ (x1 . z0 + c) ∈ N (x1 . z) for all (x1 . Theorem 1. . xn ) ⊆ Rn . F is continuous and strictly increasing on z. . . . . . xn . lim xn (ks ). then z = g(x1 . Let A be any 1 1 1 0 0 1 subsequential limit of {zk = g(x1 (k). . x2 . x2 . . Let D be an open region in Rn+1 and let F be a function welldeﬁned on D with continuous partial derivatives. . . . . . . Let 1 1 0 0 (x1 (k). . . Then. It follows that there exists c > 0 such that F (x1 .and InverseMapping Theorems. xn . . that is A = lim zks . x2 . z0 − c) < 0 F (x1 . .7. xn ) = − x 1 ∂xi Fz (x . . 0 0 Step 3. . . . xn . . . Step 1. . . . z0 ) in which Fz is continuous and positive. . . . . z0 + c). . . . xn ). . . . x2 . . . xn ) ∈ Nδ (x1 . . . . z0 ) = 0. x2 . . . z0 ) 0 0 0 for (x . . z0 ) = 0 (x . there 0 0 0 ∂z exists a neighborhood N (x1 . Implicit. . by the continuity of 1 1 s→∞ F. xn ) for (x1 . xn . This deﬁnes a function z = g(x1 . . . xn )) = 0 for all (x1 . . . Since Fz is continuous. . . . . . xn (k))}. xn ) ∈ Nδ (x1 . z) = 0. . . xn . Prove that z = g(x1 . . . . lim zks ) 1 1 s→∞ s→∞ s→∞ . . Let (x1 . . . z0 ). . . . . and a unique 0 function z = g(x1 . . . . xn ) with values z ∈ N (z0 ) such 0 0 that 1) z0 = g(x1 . . z0 − c). g(x1 . x ) ∈ Nδ (x1 . z0 ) be a point in D where 0 0 0 ∂F 1 2 F (x1 . . . . . . x2 . . . xn ) ∈ Nδ (x1 . . . . . . . . . . . . . . xn ) converging to (x1 . . xn ). . . xn ) deﬁned for (x1 . . . xn . . . xn ) ∈ Nδ (x1 . .
Let h be small enough. = − lim Theorem 1. . u2 . u0 . . . . N n (un ). xi−1 . . z ) ˜ ˜ . . xm ). . xn ) − z k = lim = lim h→0 h→0 h ∂xi h Fx Fxi (x1 . xn ). Then there are neighborhoods Nδ (x1 . . . Let z + k = g(x1 . . x2 . . . . . . . h→0 Fz (x1 . . xi . . . then all higher derivatives of g are continuous and so g is also smooth. . .LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 7 = F (x1 . . . . x2 . . . xi . . . . x ) deﬁned for (x1 . that is F (x1 . z )k by the mean value theorem (Consider the function φ(t) = F (x1 . . . . xi + th. . . . . A = g(x1 .). . . . xn . z) ˜ ˜ ˜ ˜ = Fxi (x1 . . . . . . . . . u1 . Let D be an open region in Rm+n and let F1 . . . . . . z )h + Fz (x1 . . and z is between ˜ ˜ z and z + k. xi+1 . un ∈ N n (un ) such that 0 0 0 0 . . . z ) ˜ ˜ Fz where z → z as h → 0 because g is continuous (and so k → 0 as h → 0). Fn ) = det ∂(u1 . . un ) = 0 0 0 0 0 0 0 and the Jacobian J= ∂(F1 . . ˜ Step 6. un ) ∂Fi ∂uj =0 at the point (x1 . un ). . . xn . . x . . . xi + h. . x2 . . . . . . . . ∂z Step 5. . . . xi + h. z + tk) for 0 ≤ t ≤ 1. xi + h. . u2 . . . xn ). . . gxi is continuous for each i. . .=− i. . .8 (Implicit Function Theorem). . . xi−1 . . . xn . . . . and unique functions 0 0 1 u = g1 (x1 . . . . . . . xm ) n · · · · ·1· · ·2· · · · m u = gn (x . . . u1 . . u1 . . . . . Since Fz is not zero in this small neighborhood. . z + k) − F (x1 . . xn . xn . . . . xi . . un ) = 0 F2 (x0 . N 1 (u1 ). xn . Then 0 = F (x1 . xi + h. . . F2 . xm . . . . . . . . xn . . . . . . . Then φ(1) − φ(0) = φ (ξ)(1 − 0). un ) 0 0 0 0 0 0 be a point in D where 1 2 m 1 2 n 0 0 0 0 0 0 F1 (x1 . u0 . . . . . . .. Now ∂g g(x1 . . . 0 0 0 0 0 0 0 0 0 N 2 (u2 ). Compute the partial derivatives ∂xi . . Let (x1 . . xn ) and 1 1 1 1 so g is continuous. . xi . . . F2 . xn . x2 . . xn . . 1 1 By the unique solution of the equation. . . . . . u2 . If F is smooth. x2 . xm . xm ) ∈ Nδ (x1 . where xi is between xi and xi + h. . u0 ) = 0 ············ Fn (x1 . . x0 . . . . . xm . xi+1 . . . x2 . . xm ) 2 u = g2 (x1 . . xm ) with values u1 ∈ N 1 (u1 ). . x0 . . . . . . . . . z + k) = 0 with z0 − c < z + k < z0 + c. xm . . . . . Thus {zk } converges g(x1 . u2 . . u1 . . Fn be functions welldeﬁned on D with continuous partial derivatives. A). u2 . . . .
. . . . . . . . . . x2 . . then each ui = gi (x1 . . . . x2 . . . . . j ≤ n − 1. . . if ∂un (P ) = 0. xm ). xn . ∗ ∂Fn ∂un 0 . . Fn ) (x . . 0 0 3) If each Fi is smooth on D. 0 0 2) Each gi has continuous partial derivatives with ∂gi 1 1 ∂(F1 . . . Consider G1 = F1 (x1 . . . . . . . . . 0 0 0 0 0 0 we may assume that ∂Fn (P ) = 0. ∂un ∂Fi (The entries in the last column can not be all 0 and so. . . . ∂uj ∂un ∂uj Let B= Then ∂Fi ∂uj ·B = ∂Gi ∂uj n−1. . . gi (x1 . . gn ) ············ Gn−1 = Fn−1 (x1 . there is a solution un = gn (x1 . . . u1 . u2 . we can interchange Fi and Fn . .) Since the matrix ∂Fi ∂uj is invertible at the point P = (x1 . . Assume that the statement holds for n − 1 with n > 1. . . . . . . xm ) and 0 0 0 0 Fi (x1 . xm . u1 . xm ) ∈ Nδ (x1 . xm . u2 . . . . . gn ) G2 = F2 (x1 . . . xm ) is smooth on Nδ (x1 . . . . 0 0 Sketch of Proof. un−1 . .n−1 ∂Gi ∂Fi ∂Fi ∂gn = + · ∂uj ∂uj ∂un ∂uj 1 0 0 ··· 0 ∂gn ∂u1 0 1 0 ··· 0 ∂gn ∂u2 0 0 1 ··· 0 ∂gn ∂u3 ··· ··· ··· ··· ··· ··· 0 0 0 ··· 1 ∂gn ∂un−1 0 0 0 ··· 0 1 . un ) (because the determinant is not zero). . xm ). xm ). . un ) ∂x J ∂(u for all (x1 . (We already prove that the statement holds for n = 1. . u1 . . . uj+1 . . un−1 ) to the last equation. . . . . . The proof is given by induction on n.8 JIE WU 1) ui = gi (x1 . where ∂Fn ∂Fn ∂gn + · = 0. xj . . xm . . gn ). xm ) where ui = gi (x1 . . . . uj−1 . . xm . . . . . xm )) = 0 for all 1 ≤ i ≤ n and all (x1 . xm . . . . x2 . . un−1 . . . . . u1 . .) From the previous theorem. un−1 . . . . u1 . Then for 1 ≤ i. . . . . xm ) ∈ Nδ (x1 . . xm ) = − · j 1 .
. . . . .. . A subset N ⊆ X with x ∈ N is called a neighborhood of x if there is an open set U with x ∈ U ⊆ N . . then the closed ball D (x) and the open ball B (x) are neighborhoods of x. . . Proof. . . . xn ) =0 ∂(u1 . . .1. 0 0 Then there are neighborhood Nδ (x1 . . . . . . then each fi−1 is smooth. . 0 0 0 0 Moreover if each fi is smooth. . . un ) ∂un ∂(u1 . . . The assertion follows from the Implicit Function Theorem. . there are solutions ui = gi (x1 .1.Gn−1 ) ∂(u1 . un ).. . . . Deﬁnition 2. J= Thus ∂(G1 . . . Let D be an open region in Rn . by induction. A subset C is said to closed if X \ C is open. Topological and Differentiable Manifolds.9 (Inverse Mapping Theorem). . . . The members U ∈ U are called the open sets. xn ) is welldeﬁned and has continuous partial derivatives on Nδ (x1 . un−1 ) = 0 at P and. . .LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 9 By taking the determinant. . For example. . . iii) the union of any number of members of U is in U.. . . . un ) 2 x = f2 (u1 . . un ) such that 0 0 0 0 1 −1 1 n u = f1 (x . Let Fi = fi (u1 . . A topology U for X is a collection of subsets of X satisfying i) ∅ and X are in U. Topological Spaces. Let X be a topological space. . . . The set X with U is called a topological space. Theorem 1. . . Let X be a set. xm ) for 1 ≤ i ≤ n − 1. . . . xn ) with values in N (u1 . . 2.. . . un ) be functions deﬁned on D with continuous partial derivatives. . Immersions. . . . . Let (u1 . ii) the intersection of two members of U is in U. . .. Deﬁnition 2. un ) − xi . un ) ∈ D satisfy xi = 0 0 0 fi (u1 . xn ) and N (u1 . . . . A function f : X → Y between two topological spaces is said to be continuous if for every open set U of Y the preimage f −1 (U ) is open in X. . . . . . .un−1 ) ∂(F1 . . un ) at (u1 . xn ) ······ n −1 u = fn (x1 .. . x ) 2 −1 u = f2 (x1 . . . un ) and the Jacobian 0 0 ∂(x1 . . . . un ). un ) ······ n x = fn (u1 . .. if X is a metric space. Submersions and Submanifolds 2. Diffeomorphisms. . Fn ) ∂Fn ∂(G1 . . . Let 1 x = f1 (u1 . ∂(u1 . . . . Gn−1 ) · = . .2..
The manifolds are the objects that we are going to discuss in this course. Thus a homeomorphism between X and Y is a bijective between the points and the open sets of X and Y . given any two points. which states that any simply connected e 3dimensional (topological) manifold is homeomorphic to the 3sphere S 3 . However the topological classiﬁcation of 3dimensional manifolds (we will learn manifolds later. + Manifold is one of models that we can do calculus ‘locally’. A very general question in topology is how to classify topological spaces under homeomorphisms. that is f sends open sets to open sets. A 2dimensional manifold is called a surface. an nmanifold is locally Rn . Deﬁnition 2. that is. Dn is an example of ‘manifolds with boundary’.3. Let X and Y be topological spaces. g : Y → X such that f ◦ g = idY and g ◦ f = idX . For example. By the deﬁnition. any loop can be deformed to be the constant loop in X). Equivalently f is a homeomorphism if and only if 1) f is a bijective. The collection {(U (x).) is quite open. A space means a Hausdorﬀ space. By the deﬁnition of manifold. (Note. you just think that M is a subspace of RN for a large N . = By the deﬁnition. Roughly speaking. Sometimes M is denoted as M n for mentioning the dimension of M . xn ) ∈ Rn xn ≥ 0}. The objects traditionally called ‘surfaces in 3space’ can be made into manifolds in a standard way.10 JIE WU A continuous function from a topological space to a topological space is often simply called a map. we know (from complex analysis and others) that any simple closed loop is homeomorphic to the unit circle S 1 . The famous Poicar´ conjecture is related to this problem. By means of calculus. 2. the closed ndisk Dn is not an nmanifold because it has the ‘boundary’ S n−1 . If you are not familiar with topological spaces. we need local coordinate systems. Rn and the nsphere S n is an nmanifold. The topological classiﬁcation of (twodimensional) surfaces is known as well. A Hausdorﬀ space M is called a (topological) nmanifold if each point of M has a neighborhood homeomorphic to an open set in Rn . Roughly speaking topological classiﬁcation of curves is known. a function f : X → Y is a homeomorphism if and only if i) f is a bijective. The compact surfaces have been classiﬁed as spheres or projective planes with various numbers of handles attached. 2) f is continuous and 3) f is an open map. ii) f is continuous and iii) f −1 is also continuous.) For example. We say that X and Y are homeomorphic if there exist continuous functions f : X → Y. there is a continuous path joining them) and (2) the fundamental group π1 (X) is trivial (roughly speaking.2. A space X is called simply connected if (1) X is pathconnected (that is. φx )x ∈ M } has + the property that 1) {U (x)x ∈ M } is an open cover and 2) φx is a homeomorphism from U (x) . · · · . there is a an open neighborhood U (x) of x and a homeomorphism φx from U (x) onto an open set in Rn . Let x ∈ M . We give the deﬁnition of manifold with boundary as follows. Topological Manifolds. A Hausdorﬀ space M is called an nmanifold with boundary (n ≥ 1) if each point in M has a neighborhood homeomorphic to an open set in the half space Rn = {(x1 . a topological spaces where any two points has disjoint neighborhoods. We write X ∼ Y and say that f and g are homeomorphisms between X and Y .
φα ) such that x ∈ Uα and φα (x) ∈ Rn−1 = {x ∈ Rn xn = 0}. For any x ∈ M there exists α such that x ∈ Uα and so Uα is an open neighborhood of x that is homeomorphic to an open set in Rn . For example the boundary of Dn is S n−1 . They are called local coordinates on the manifold.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 11 onto an open set in Rn . If P denotes a point of U . Proposition 2. φn . φx )x ∈ M } is somewhat too large and we may like less local coordinate systems. Proposition 2. . Suppose that M is a manifold. then U ∩ Rn−1 is also open because: Since U + is open in Rn . Diﬀerentiable Manifolds. . ∂M is called the boundary of M . φα )α ∈ J} be an atlas for M . A Hausdorﬀ space M is a manifold (with boundary) if and only if M has an atlas.3. Deﬁnition 2. The key point here is that if U is open in Rn . xn (P ) or simply x1 . φ) such that 1) U is an open set in M and 2) φ is a homeomorphism from U onto an open set in Rn . . φα Uα ∩∂M α ∈ J } can be made into an atlas for ∂M . Thus M is a manifold. φα )α ∈ J} such that . Let M be a nmanifold with boundary. . Proof.6. . . Then Clearly {(Uα ∩ ∂M. Then the collection {(U (x). Now if x ∈ U ∩ Rn−1 . . Proof. The map + φ : U → Rn + can be given by n coordinate functions φ1 . x2 . The + + collection {(U (x). . An atlas for M means a collection of charts {(Uα . there + + is an open disk E (x) ⊆ V and so E (x) ∩ Rn−1 ⊆ V ∩ Rn−1 = U ∩ Rn−1 is an open (n − 1)dimensional disk in Rn−1 centered at x. Let {(Uα . these functions are often written as x1 (P ). A chart of M is a pair (U. Note.5. This can be done as follows. . . Let M be a space. Let J ⊆ J be the set of indices such that Uα ∩ ∂M = ∅ if α ∈ J . 2. .4. A Hausdorﬀ space M is called a diﬀerential manifold of class C k (with boundary) if there is an atlas of M {(Uα . x2 (P ). The subspace φx (Ux ) in Rn plays a role as a local coordinate system. Conversely suppose that M has an atlas. φα )α ∈ J} such that {Uα α ∈ J} is an open cover of M . . φx )x ∈ M } is an atlas. + We deﬁne a subset ∂M as follows: x ∈ ∂M if there is a chart (Uα . xn . Then ∂M is an (n − 1)manifold without boundary. there is an open subset V of Rn such that U = V ∩ Rn .
ηγ β ∈ J }. . Since f is smooth with respect with the atlases {(Uα . (Note. φα )α ∈ I} and {(Vβ . {(Uδ . θδ α ∈ I }.12 JIE WU For any α. In this case we can + not talk diﬀerentiability of φα unless Uα is regarded as a subspace of a (large dimensional) Euclidian space. A diﬀerential structure of class C k on M is an equivalence class of diﬀerential atlases of class C k on M . N then it is smooth with respect to equivalent atlases Proof. then M is a diﬀerentiable manifold of + class C k . Note: A general manifold is also called topological manifold. If M has an atlas {(Uα . A map f : M → N is called smooth if for some smooth atlases {(Uα . the composites φα ◦ φ−1 : φβ (Uα ∩ Uβ ) → Rn + β is diﬀerentiable of class C k . we only assume that M is a (Hausdorﬀ) topological space and so φα is only an identiﬁcation of an abstract Uα with an open subset of Rn . θδ α ∈ I }. Deﬁnition 2. φα )α ∈ J} such that each φα : Uα → Rn is diﬀerentiable of class C k . Thus a diﬀerential manifold of class C k means a manifold with a diﬀerential structure of class C k . φβ β ∈ J}. ηγ β ∈ J } for M . φα α ∈ J} is called a diﬀerential atlas of class C k on M . φβ β ∈ J} ∪ {(Vγ . {(Vγ . {(Vβ . If f : M → N is smooth with respect to atlases {(Uα . φα α ∈ I}.7. Kervaire and Milnor [4] have shown that the topological sphere S 7 has 28 distinct oriented smooth structures. φβ β ∈ J} {(Vγ . φα α ∈ I} ∪ {(Uδ . φα α ∈ I}. Assume that M is a subspace of RN with N >> 0. The atlas {(Uα . In our deﬁnition (the usual deﬁnition of diﬀerentiable manifolds using charts). φα α ∈ I} for M and {(Vβ . ηγ β ∈ J } by look at the local coordinate systems. A smooth manifold means a diﬀerential manifold of class C ∞ . f is smooth with respect to the smooth atlases {(Uα .) Two diﬀerential atlases of class C k {(Uα . let M and N be smooth manifolds (with boundary) of dimensions m and n respectively. Thus f is smooth with respect to the atlases {(Uδ . ψβ )β ∈ J} is again a diﬀerential atlas of class C k (this is an equivalence relation). {(Vβ . {(Vβ .8. This is the deﬁnition of diﬀerentiable (smooth) manifolds in [6] as in the beginning they already assume that M is a subspace of RN with N large. θδ α ∈ I }. Proposition 2. ψβ )β ∈ J} for N the functions ψβ ◦ f ◦ φ−1 φα (f −1 (Vβ )∩Uα ) : φα (f −1 (Vβ ) ∩ Uα ) → Rn α + are of class C ∞ . ψβ )β ∈ J} are called equivalent if {(Uα . β ∈ J. φα )α ∈ I} ∪ {(Vβ .
Let T be an open region of Rm and let f = (f1 . By the deﬁnition. Let (U. . Let S be an open region of Rn . denoted by (U. Now (U.9. w). If ∂M = ∅. v) is called equivalent to (V. Tangent Space. Then f induces a linear transformation T f : TP (S) → Tf (P ) (T ) given by ∂fi ∂xj 1 v1 v ∂1 (f1 ) + v 2 ∂2 (f1 ) + · · · + v n ∂n (f1 ) v2 v 1 ∂1 (f2 ) + v 2 ∂2 (f2 ) + · · · + v n ∂n (f2 ) · = ··· ············ m×n v 1 ∂1 (fm ) + v 2 ∂2 (fm ) + · · · + v n ∂n (fm ) v n n×1 . Let M be a smooth nmanifold. possibly with boundary. A smooth map f : M → N is called a diﬀeomorphism if f is onetoone and onto. . T f (v) = namely T f is obtained by taking directional derivatives of (f1 . A subset X is called a properly embedded submanifold of dimension k ≤ n if X is a closed in M and.4. 2. (U. the collection {(U ∩ X. φ) and (V. v) ∼ (V. X is called simply a submanifold of M . if T (ψ ◦ φ−1 )(v) = w. φ. The point is that there are possibly many charts around P . v) and (V. In the above deﬁnition. fm ) : S → T be a smooth map. So we need to deﬁne an equivalence relation in TP such that. by dropping the requirement that X is a closed subset but keeping the requirement on local charts. φ. Now we are going to deﬁne the tangent space to a (diﬀerentiable) manifold M at a point P as follows: First we consider the set TP = {(U. + where Rk ⊆ Rn is the standard inclusion. That is (U. ψ. φ) is a chart v ∈ T (φ(P ))(φ(U ))}. w). .10. Deﬁnition 2. for each P ∈ X.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 13 Thus the deﬁnition of smooth maps between two smooth manifolds is independent of choice of atlas. there exists a chart (U. .ψ(U ∩ V ) is diﬀeomorphism and so it induces an isomorphism of vector spaces T (ψ ◦ φ−1 ) : Tφ(P ) (φ(U ∩ V )) . Deﬁnition 2. Recall that. w) be two elements in TP . . for P ∈ S.Tψ(P ) (ψ(U ∩ V )). φ. . ψ) are two charts with P ∈ U and P ∈ V . the tangent space TP (S) is just the ndimensional vector space by putting the origin at P . TP modulo these relations is only one copy of ndimensional vector space which is also independent on the choice of charts. φ. . . ψ ◦ φ−1 : φ(U ∩ V ) . and if the inverse f −1 : N → M is also smooth. v)  P ∈ U. φ) about P in M such that φ(U ∩ X) = φ(U ) ∩ Rk . + + Note. ψ. Each chart creates an ndimension vector space. fm ) along vector v for any v ∈ TP (S). ψ. φU ∩X )} is an atlas for making X to a smooth kmanifold with boundary ∂X = X ∩ ∂M . .
Then use Diagram (1) to check that T f is independent on choices of charts. ψ) is another chart with P ∈ V . Then prove that this operation gives a welldeﬁned vector space structure on TP . (V ˜ ˜ ˜ ˜ φ ◦ ψ −1 : ψ(U ∩ V ) → φ(U ∩ V ) φ ◦ ψ −1 : ψ(U ∩ V ) → φ(U ∩ V ) and deﬁne ˜ ˜ ˜ a(V.Tf (P ) (N ) such that the diagram TP (M ) Tf Tφ∼ Tφ(P ) (φ(U )) = T (ψ ◦ f ◦ φ−1 ) ? ? Tψ Tf (P ) (N ) ∼ Tψ(f (P )) (ψ(V )) = commutes for any chart (U. φ. aT (φ ◦ ψ −1 )(x) + bT (φ ◦ ψ −1 )(y)). ψ. x) + b(V . then there is a commutative diagram Tφ TP (M ) ∼ Tφ(P ) (φ(U ∩ V )) = (1) TP (M ) T (ψ ◦ φ−1 ) ? Tψ . x). y) = (U. Prove that TP (M ) is an ndimensional vector space. . φ) and deﬁned a(U. If (V. φ. av + bw). Exercise 2. independent on the equivalence relation. ψ) with f (P ) ∈ V . there is a unique isomorphism Tφ : TP (M ) → Tφ(P ) (φ(U )). Exercise 2. ψ) with f (P ) ∈ V .] The tangent space TP (M ). φ. where M and N need not to have the same dimension. v) as representatives for its equivalence class. can be described as follows: given any chart (U. Prove that there is a unique linear transformation T f : TP (M ) . ψ. φ) with P ∈ U . φ) with P ∈ U and any chart (V. ∼ = where T (ψ ◦ φ−1 ) is the linear isomorphism induced by the Jacobian matrix of the diﬀerentiable map ψ ◦ φ−1 : φ(U ∩ V ) → ψ(U ∩ V ). w) := (U. v) + b(U. that is.2. φ. ˜ ˜ .14 JIE WU Deﬁne TP (M ) to be the quotient TP (M ) = TP / ∼ . [First ﬁx a choice of (U. Let M be a diﬀerentiable nmanifold and let P be any point in M . consider the map Now given any (V. φ. φ) with P ∈ U and (V. [Hint: Fixed a chart (U.Tψ(P ) (ψ(U ∩ V )). y) ∈ TP . ψ. as a vector space. by choosing (U. Let f : M → N be a smooth map. the linear transformation T f is uniquely deﬁned by the above diagram.1. ψ.
V ψ(f (P )) = 0 ψ canonical coordinate inclusion ? . Suppose that f : M m → N n is immersion at P .T0 (ψ(V )) . Immersions. . φ) about P and (V. We may assume that φ(P ) = 0 and ψ(f (P )) = 0. T g : T0 (φ(U )) is an injective linear transformation and so rank(T g) = m ? Rn .Rn Proof.5. (Otherwise replacing φ and ψ by φ − φ(P ) and ψ − ψ(f (P )).ψ(V ) φ(U ) ∩ ∩ ? Rm By the assumption. respectively.11 (Local Immersion Theorem).LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 15 2.) Consider the commutative diagram f U U φ . Then there exist charts (U.V ψ ? g = ψ ◦ f ◦ φ−1 ? . . ψ) about f (P ) such that the diagram f U U φ(P ) = 0 φ ? Rm ⊂ commutes. A smooth map f : M → N is called immersion at P if the linear transformation T f : TP (M ) → Tf (P ) (M ) is injective. Theorem 2.
. xm . . . . . . xn ) = g i (x1 . . . . . Deﬁne a function h = (h1 .Rn B(n−m)×m where B is taken from (m + 1)st row to nth row in the matrix (2).16 JIE WU at the origin. xm ) for 1 ≤ i ≤ m and hi (x1 . . . h is an diﬀeomorphism in a small neighborhood ˜ ˜ of the origin. xm . . . . . xm+1 . we may assume that the ﬁrst m rows forms an invertible matrix Am×m at the origin. It follows that there exist open neighborhoods U ⊆ U of P and V ⊆ V of f (P ) such . h2 . . xm ) + xi for m + 1 ≤ i ≤ n. . The matric for T g is ∂g 1 ∂x1 2 ∂g ∂x1 ··· m ∂g ∂x1 ··· n ∂g ∂x1 ∂g 1 ∂x2 ∂g 2 ∂x2 ··· ∂g m ∂x2 ··· ∂g n ∂x2 ∂g 1 ∂xm ∂g 2 ∂xm ··· . By the Inverse Mapping Theorem. . . . . hn ) : φ(U ) × Rn−m by setting hi (x1 . xm+1 . Then Jacobian matrix of h is Am×m 0m×(n−m) In−m . . . . . . xn ) = g i (x1 . . Thus the Jacobian of h is not zero at the origin. ∂g m ∂xm ··· ∂g n ∂xm ··· ··· ··· ··· ··· ··· (2) By changing basis of Rn (corresponding to change the rows). .
then there is an open subset V of N such that V ∩ f (M ) = f (U ). . ψf (U ) )} gives an atlas for f (M ) such that f (M ) is a submanifold of M .11.V ψV ˜ ? ˜ ψ(V ) ∼ h−1 = ˜ φ(U ) × 0 ∩ ⊂ ? ˜ . The charts {(f (U ). Submersions.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 17 that the following diagram commutes ˜ U φU ˜ ? ˜ φ(U ) g = ψ ◦ f ◦ φ−1 f U ˜ ˜ . f (U ) is an open subset of f (M ). For any point P in M .Rn . Then f (M ) is a smooth submanifold of M and f : M → f (M ) is a diﬀeomorphism. Let f : M → N be a smooth map. 2) f is onetoone and 3) f : M → f (M ) is a homeomorphism.) An injective immersion satisfying condition 3 is called an embedding. Suppose that 1) f is immersion at every point P ∈ M . namely an injective immersion need not give a diffeomorphism with its image. Theorem 2. where U2 is a small neighborhood of the origin in Rn−m .6. Now f : M → f (M ) is a diﬀeomorphism because it is locally diﬀeomorphism and the inverse exists. Condition 3 is important in this theorem. By Condition 3.φ(U ) × U2 ∩ ? Rm = Rm × 0 ⊂ ? . In Condition 3. we need that if U is an open subset of M . we can choose the charts as in Theorem 2. Note. 2.12. A smooth map f : M → N is called submersion at P if the linear transformation T f : TP (M ) → Tf (P ) (M ) is surjective. (Construct an example for this. Proof.
the composite f −1 (V ) f f −1 (V ) f U . Let Z be a submanifold of N . Proof. Proof. ψV1 ) is a chart of Z about f (P ). If a smooth map f : M → N is transversal to a submanifold Z ⊆ N . since Z is a submanifold. Then there exist charts (U.15. Then f −1 (Q) is a submanifold of M with dim f −1 (Q) = dim M − dim N . Moreover the codimension of f −1 (Z) in M equals to the codimension of Z in N . Suppose that Q is regular. Let f : M → N be a smooth map and let Q ∈ N such that f −1 (Q) is not empty. ψ) of N about f (P ) such that V = V1 × V2 with V1 = V ∩ Z and (V1 . for any P ∈ f −1 (Q). Then the transversal condition is TP (M ) + TP (Z) = TP (N ) for any P ∈ M ∩ Z. Suppose that f : M m → N n is submersion at P .Rn .13 (Local Submersion Theorem). φf −1 (Q) : f −1 (Q) ∩ U gives a chart about P .V proj. . ψ) about f (P ) such that the diagram U φ(P ) = 0 φ ? Rm ⊂ commutes. φ) about P and (V.14 (Preimage Theorem). Theorem 2.18 JIE WU Theorem 2. For a smooth map of manifolds f : M → N .V2 is submersion. A smooth map f : M → N is said to be transversal to Z if Im(T f : TP (M ) → Tf (P ) (N )) + Tf (P ) (Z) = Tf (P ) (N ) for every x ∈ f −1 (Z). ? . then f −1 (Z) is a submanifold of M . By the assumption. With respect to the assertion about the codimensions. Consider the special case that both M and Z are submanifolds of N . a point Q ∈ N is called regular if T f : TP (M ) → TQ (N ) is surjective for every P ∈ f −1 (Q). Theorem 2. Given P ∈ f −1 (Z). codim(f −1 (Z)) = dim V2 = codim(Z).V ψ(f (P )) = 0 ψ canonical coordinate proj. 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 . . By the Preimage Theorem. the preimage of Q. there is a chart (V. From the above theorem.Rm−n .
and so Gk. vk )T = (w1 .n given by the quotient topology. . Given an kplane X and w ∈ Rn . The open Stiefel manifold is the space of ktuples of linearly independent vectors in Rn : ˜ Vk. .n is considered as the subspace of k × n matrixes M (k. .n is an open = kn ˜ kn subset of M (k. .n → R is continuous and Gk. . Open Stiefel Manifolds and Grassmann Manifolds. . . GLk (R) consists of k × k nonsingular matrixes. First we prove that Gk.n . . .n be the quotient by sending ktuples of linearly independent vectors to the kplanes spanned by k vectors. So we assume that k < n. If k = n.n to any space ˜ is continuous if and only if f ◦ π from Vk. . . there is an open neighborhood UX of α such that UX ∼ Rk(n−k) . The Grassmann manifold Gk.n . . . Now we check that Gk. vk . vk )T  vi ∈ Rn . The intersection of two transversal submanifolds of N is again a submanifold.n is an open submanifold of R . n) = R . . write v1 . Let ˜ π : Vk. k) = Rk . .n by the action of GLk (R). .n . Then Gk. Examples of Manifolds 3. vk ) ∈ V two ktuples (v1 . vk for the kplane spanned by v1 . . then Gn.n is only one point. . we obtain the continuous map ρw : Gk. for any X in Gk. .n is the set of kdimensional subspaces of Rn .n . we can choose a w such that ρw (X) = ρw (Y ). Let V1 and V2 be disjoint open subsets of R such that ρw (X) ∈ V1 and ρw (Y ) ∈ V2 . . . U is an open set of Gk. . that is. n) ∼ Rkn . = . .n = {(v1 . . .R ˜ because ρw ◦ π : Vk. Let GLk (R) be the space of general linear groups on Rk . namely.1.n is given by quotient topology of π.n is the quotient of Vk. Moreover codim(M ∩ Z) = codim(M ) + codim(Z) in N . . . all kplanes through the origin.n that separate X and Y . . that is. wk )T spanned the same kplane if and only if each of them is basis for the common plane. . . . . ˜ ˜ where Vk. . This gives the identiﬁcation rule for the Grassmann manifold Gk. Let {e1 . . {v1 . .n → Gk.n . (Here we use the property of quotient topology that any function f from the quotient space Gk. which is an 2 ˜ open subset of M (k. The topology in Gk.) Given any two distinct points X and Y in Gk.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 19 Corollary 2. wk )T . Vk. Observe that For (v1 . then k ρw (X) = w · w − j=1 (w · ej )2 .n is Hausdorﬀ. if and only if there is nonsingular k × k matrix P such that P (v1 . Then ρ−1 (V1 ) and ρ−1 (V2 ) are two open w w subset of Gk. .n .n is manifold of dimension k(n − k) by showing that.n is Hausdorﬀ. let ρw be the square of the Euclidian distance from w to X. .n to that space is continuous.16. . Since Vk. . T ˜k. vk )T and (w1 . Fixing any w ∈ Rn .n ˜ if and only if π −1 (U ) is open in Vk. ek } be the orthogonal basis for X. . 3. vk } linearly independent}.
There exists a nonsingular n × n matrix Q such that (v1 . 0)Q. . n − k). It follows that there is a nonsingular k × k matrix P such that ˜ ˜ (Ik . A)Q φ−1 X (P. . AZ )QQ−1 . A)Q ˜ A → (Ik . P AZ )T = (P T11 + P AZ T21 . Fixing Q. vk )T = (Ik . . . Bk.20 JIE WU Let X ∈ Gk. respectively. (w1 . n − k) = Rk(n−k) and so Gk. . For checking that Gk. AZ )Q ⇔ (Ik .UX . Let UX = π(EX ) ⊆ Gk. . ˜ Then EX is an open subset of Vk. . 0)Q.n with X ∈ UX . BZ )Q = P (Ik . T22 T11 T12 . vk )T = (Ik . . wk )T = (Ik . . n − k)} ⊆ Vk. let X and Y ∈ Gk.n be spanned by (v1 .n−k )Q  det(Pk ) = 0. T11 + AZ T21 is invertible). . UX is homeomorphic to M (k. . n − k) proj.n is a smooth manifold. deﬁne ˜ Xα = {(Pk . . BZ )Q for unique A. . Consider the maps: M (k. . then ˜ Z = (Ik . . . UX is open in Gk.UY A → (Ik . B ∈ M (k.n . There exists nonsingular n × n matrixes Q and Q such that ˜ (v1 .n−k ∈ M (k. . AZ )Q = (Ik . φ−1 X φ−1 Y . ? M (k. wk )T . From the commutative diagram GLk (R) × M (k. .n . vk )T ˜ and (w1 . P A)Q . .EX ∼ = π ? UX . vk )T . . BZ ) = (P. Bk. A)Q . n − k) M (k. . P T12 + P AZ T22 ) Ik = P (T11 + AZ T21 ) BZ = P (T12 + AZ T22 ). n − k) A → (Ik . . . BZ ) = P (Ik . . n − k) If Z ∈ UX ∩ UY . A) → (P.n be spanned by (v1 .n . It follows that Z ∈ UX ∩ UY if and only if det(T11 + AZ T21 ) = 0 (that is. . Since π −1 (UX ) = EX ˜ is open in Vk.n . 0)Q. where Ik is the unit k × kmatrix.n is a (topological) manifold. Let ˜ T = QQ−1 = T21 Then (Ik .
By the identiﬁcation of M (k. a smooth manifold 2 plus a topological group such that the multiplication and inverse are smooth. Thus Vk. which is smooth.n ⊆ M (k.n is a smooth submanifold of M (k.2. n) of dimension (k + 1)k k(2n − k − 1) = . n) and Tf (A) (S(k)) = S(k)). As a special case. n)) → Tf (A) (S(k)) is onto and so Ik is a regular value of f . n) AAT =Ik 1 1 1 1 C=C T AAT C T + CAAT = = = = C T + C = = = C. Thus Gk.n = O(n) the orthogonal group. denoted by Vk. The Stiefel manifold. Then S(k) ∼ R 2 is a smooth manifold of = dimension. Deﬁne T fA (B) = lim B= Then T fA (B) = AB T + BAT = 1 CA ∈ TA (M (k. One can check that 2 Vn−1. f (A + sB) − f (A) s→0 s (A + sB)(A + sB)T − AAT = lim s→0 s T AA + sAB T + sBAT + s2 BB T − AAT = lim = AB T + BAT .) When k = 1.n is the space of lines (through the origin) of Rn . (Note. then Vn. the composite φX (UX ∩ UY ) is given by A → (T11 + AT21 ) (T12 + AT22 ) . is deﬁned to be the set of k orthog˜ onal unit vectors in Rn with topology given as a subspace of Vk.n is a smooth manifold. We prove that Vk. T fA : TA (M (k. O(n) is a Lie group. n).n = {A ∈ M (k. n)). From the above. n) → S(k) A → AAT .n = S n−1 which is manifold of dimension n − 1. When k = n − 1. then V1. G1. For any A ∈ M (k.n . s→0 s We check that T fA : TA (M (k. Thus. 2 2 Special Cases: When k = n. 2 −1 φ−1 X . Vk. n)  A · AT = Ik }. O(n) is a (smooth) manifold of dimension n(n−1) . n) and S(k) with Euclidian spaces. === == 2 2 2 2 Thus T f : TA (M (k. (k+1)k Let S(k) be the space of symmetric matrixes. that is.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 21 From the above. which is also called projective space denoted by RPn−1 . n). n)) → Tf (A) (S(k)) is given by setting T fA (B) is the directional derivative along B for any B ∈ TA (M (k. n)) → Tf (A) (S(k)) is surjective for any A ∈ f −1 (Ik ). n) by using Preimage Theorem. then Vn−1. Consider the map f : M (k. 3. n)) = M (k. by Preimage Theorem. RPn−1 is a manifold of dimension n − 1. Stiefel Manifold.M (k. n)). From the above. TA (M (k.n = f −1 (Ik ) is a smooth submanifold of M (k.n is a manifold of dimension (n−1)n . Let A ∈ f −1 (Ik ) and let C ∈ Tf (A) (S(k)).UX ∩ UY φY . namely.n ∼ SO(n) = kn − .
where p : E → B is a (continuous) map. a ﬁbre bundle p : E → B is a “locally trivial” bundle with a “ﬁxed ﬁbre” F . Fibre Bundles.Vk. This gives the diagram Vk. y) → x. etc) is used to denote a bundle. Roughly speaking. there exists an open neighborhood U of x such that p−1 (U ) is a trivial . λ. . .n /O(k) = O(n)/(O(k) × O(n − k)).n is a submanifold.n is compact. Vk. φ) : ξ → ξ is a pair of (continuous) maps φ : E(ξ) → E(ξ ) and ¯ : B(ξ) → B(ξ ) such that the diagram φ E(ξ) p(ξ) ? B(ξ) ¯ φφE(ξ ) p(ξ ) ? B(ξ ) commutes.n is given by k unit vectors (v1 . p−1 (b) is called the ﬁbre of the bundle over b ∈ B.n  is onto and so the Grassmann manifold Gk.1. The trivial bundle is the projection of the Cartesian product: p : B × F → B. B). Vk. By the construction. In general case. The space B is called the base space.n is a closed subspace of the kfold Cartesian product of S n−1 because Vk. 4. . a bundle can be thought as a union of ﬁbres f −1 (b) for b ∈ B parameterized by B and glued together by the topology of the space E.n = O(n)/O(n − k).22 JIE WU the subgroup of O(n) with determinant 1.n = Vk. Usually a Greek letter ( ξ. (x. then E(ξ) denotes the total space of ξ.Gk. For each b ∈ B. the space E is called the total space. η. vk )T in Rn that are solutions to vi · vj = 0 for i = j. n) submersion atk S(k) T A → AA smooth ? ? Gk. Moreover the above composite is a smooth map because π is smooth and Vk.n Note.n by the action of O(k). Intuitively. p. As a space. This follows from that Vk. More precisely.n ⊂ ˜ . Fibre Bundles and Vector Bundles 4. and B(ξ) denotes the base space of ξ. and the map p is called the projection of the bundle. A bundle means a triple (E. ¯ A morphism of bundles (φ.n is the quotient of Vk.n ⊂ submanifold I . The composite Vk.n is also compact. This gives identiﬁcations: Gk. Gk. and the fact that any closed subspace of compact Hausdorﬀ space is compact. for any x ∈ B.M (k.n π . ζ. .
An atlas is a collection of charts {(Uα . we can use “chart” to describe ﬁbre bundles. Fixing any x. in other words. F ) is given by compactopen topology brieﬂy reviewed as follows: Let X and Y be topological spaces. y) is a homeomorphism with inverse given by gβα (x.(Uα ∩ Uβ ) × F . Proof. Similar to manifolds. F ). F ) such that the transition functions gαβ are continuous. p(φ(x . gαβ (x. By choosing α = γ. This gives a transition function gαβ : Uα ∩ Uβ . φ) for a bundle p : E → B is (1) an open set U of B and (2) a homeomorphism φ : U × F → p−1 (U ) such that p(φ(x .LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 23 bundle. Y ) denote the set of all continuous maps from X to Y . y)) = x for any x ∈ U and y ∈ F . Prove that the transition functions {gαβ } satisfy the following equation (3) (4) .p−1 (Uα ∩ Uβ ) . Y )  f (K) ⊆ U }. −). F ) is the group of all homeomorphisms from F to F . y) = (x. y → gαβ (x. Then the collection {(U (x). y)) = x for any x ∈ U and y ∈ F .−1 x p (U ) p ? ? U ===== U ==== commutes. where Homeo(F.1. φα )} such that {Uα } is an open covering of B. Thus p : E → B is a ﬁbre bundle. Conversely suppose that p : E → B has an atlas. Let ξ be a ﬁbre bundle with ﬁbre F and an atlas {(Uα . gαα (x) ◦ gαα (x) = gαα (x) and so φβ φ−1 α . For any x ∈ B there exists α such that x ∈ Uα and so Uα is an open neighborhood of x with the property that pp−1 (U ) : p−1 (Uα ) → Uα is a trivial bundle. The composite φ−1 ◦ φβ : (Uα ∩ Uβ ) × F α has the property that φ−1 ◦ φβ (x. A chart (U. Suppose that p : E → B is a ﬁbre bundle. y)) α for any x ∈ Uα ∩ Uβ and y ∈ F .Homeo(F. Let Map(X.1. gαβ (x) ◦ gβγ (x) = gαγ (x) gαα (x) = x x ∈ Uα ∩ Uβ ∩ Uγ . Exercise 4. φx )x ∈ B} is an atlas. A bundle p : E → B is a ﬁbre bundle with ﬁbre F if and only if it has an atlas. gαβ (x. Given any compact set K of X and any open set U of Y . The topology on Homeo(F. x ∈ Uα By choosing α = β = γ. We need to introduce a topology on Homeo(F. Consider the continuous map gαβ : Uαβ × F → F . let WK. Proposition 4. −) : F → F . that is. gαβ (x) ◦ gβα (x) = gαα (x) = x and so (5) gβα (x) = gαβ (x)−1 x ∈ Uα ∩ Uβ .U = {f ∈ Map(X. there is a homeomorphism φU : p−1 (U ) → U × F such that the diagram U ×F π1 φ. φα )}.
because g0 is a homeomorphism. f ) → f ◦ g Homeo(F.U ) is open in Uα ∩ Uβ . F ) be the set of all continuous maps from F to F with compact open topology.U and y ∈ V . then f ◦ g ∈ WK. We need to show that there is a neighborhood V is x0 such that gαβ (V ) ⊆ WK.F K. F ) × Homeo(F.U ) . . Since U is open and gαβ : (Uα ∩ Uβ ) × F → F is continuous. . Thus WK.U . This implies that ¯ Homeo(F. If F is regular and locally compact.U .U . Note that F U is compact. Then Proposition 4. Then Homeo(F. N (yn )} of K. F ) is continuous.U are neighborhoods of g and f whose composition product lies in WK. F ) is a topological monoid. an open set in Map(X.2. then the inverse map f → f −1 Homeo(F. Map(F. If Homeo(F. Since {N (y)  y ∈ K} is an open cover n of the compact set K. Proposition 4. then. from the above arguments. −1 −1 Proof. F ) → Homeo(F. we show that gαβ (WK. F ) × F → F is ¯ continuous.U . F K is open and g0 ∈ WF If g ∈ WF U.4. V (yj ).3.U . Let x0 ∈ Uα ∩ Uβ such that gαβ (x0 ) ∈ W(K. . F ) is a subset of Map(F. y) → f (y) F K⊇F g0 (U ) = g0 (F U) U.V and f ∈ WV . If g ∈ WV . F ) .F K . Since F is regular and locally compact. F ) has the compactopen topology. Suppose that f ◦ g ∈ WK. namely compactopen topology only fails in the continuity of g −1 . Let V = j=1 −1 V × K ⊆ gαβ (U ) and so gαβ (V ) ⊆ WK.U and hence the result. It follows that . A modiﬁcation on compactopen topology eliminates this defect [1].U . If F is compact Hausdorﬀ. F ) with subspace topology.Homeo(F. −1 Proof. then the transition functions gαβ : Uα ∩ Uβ → Homeo(F.U . For each y ∈ K. F ) × F are continuous. then Homeo(F. then g(y) ∈ U and so the evaluation map Homeo(F. Note.U . F ) Homeo(F. and the latter is open. there is an open set V such that ¯ g(K) ⊆ V ⊆ V ⊆ f −1 (U ) ¯ and the closure V is compact.Homeo(F. then the composition and evaluation maps (g. If g ∈ WK. there is a ﬁnite cover {N (y1 ). there is a neighborhood V of y0 such that V is compact and y0 ∈ V ⊆ V ⊆ g0 (U ). Proof. . F ) are continuous. Then f (g(K)) ⊆ U . that is. −1 Let U be an open set of F and let f0 (y0 ) ∈ U or y0 ∈ f0 (U ). F ) is continuous. or g(K) ⊆ f −1 (U ). . g −1 ∈ WK.U . g −1 (U ) is an −1 open set of (Uα ∩ Uβ ) × F with x0 × K ⊆ gαβ (U ). Proposition 4. Since F is regular an locally −1 ¯ ¯ compact. If F regular and locally compact. F ) .V and ¯ WV . Y ) is an arbitrary union of a ﬁnite intersection of subsets with the form WK. Given WK.U .F (f. Then g0 (K) ⊆ U or K ⊆ g0 (U ). there exist open neighborhoods −1 V (y) of x and N (y) of y such that V (x) × N (y) ⊆ gαβ (U ). F ) × Homeo(F. or gαβ (V × K) ⊆ U . Suppose that g0 ∈ WK.24 JIE WU Then the compactopen topology is generated by WK.
More precisely. x x ¯ We need to ﬁnd the third condition such that φx is a homeomorphism. (1). This property depends on the topology structure of X and X/G. Let G be a topological group and let X be a space.2. the local crosssection means ¯ the preimages can be chosen “continuously” in a neighborhood. GSpaces and Principal GBundles. (x. Proof. A continuous map f : X → Y is called a Gmap if f (x · g) = f (x) · g for any x ∈ X and g ∈ G. we call X a (right) Gspace. and so π(U ) is open by quotient topology. Our second condition is that the projection π : X → X/G has local crosssections.sx (¯) · g y ¯ ¯ y for any y ∈ U (¯). π −1 (π(U )) = g∈G U ·g is open because it is a union of open sets. choose x ∈ X such that π(x) = x. x · g)  x ∈ X. For having constant ﬁbre G. Given any point x ∈ X/G. 2) The projection π : X → X/G is an open map. Proposition 4. Let X be a (right) free Gspace with local crosssections to π : X → X/G. A right Gaction on X means a(continuous) map µ : X × G → X. x Exercise 4. we need to assume that the Gaction on X is free.5. This is equivalent to the property that =⇒ g=h for any x ∈ X. Thus π is an open map. with quotient topology. g) . Let X/G be the set of Gorbits xG.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 25 4. x where Hx = {g ∈ G  x · g = x}. . x ∈ X. Let U (¯) be a neighborhood of x with a (continuous) crossesection sx : U (¯) → X. In this case we call X a free Gspace. In this case.) Assume that X is a (right) free Gspace with local crosssections to π : X → X/G. Let ¯ X ∗ = {(x. g ∈ G} ⊆ X × X. x ¯ x ¯ Deﬁne φx : U (¯) × G . (2) If U is an open set of X. we can always choose a preimage of π. there is always a local crosssection from the base space to the total space. namely x·g =x x·g =x·h =⇒ g=1 for any x ∈ X. Then the continuous map φx : U (¯) × G → π −1 (U (¯)) is onetoone and onto. x ¯ ¯ x (Note. The inverse is given by x → x · g −1 .2. 1) For ﬁxing any g ∈ G. Then ¯ ¯ π −1 (¯) = {x · g  g ∈ G} = G/Hx . We are going to ﬁnd some conditions such that π : X → X/G has canonical ﬁbre bundle structure with ﬁbre G. there is an open neighborhood U (¯) ¯ x with a continuous map sx : U (¯) → X such that π ◦ sx = idU (¯) . Since a ﬁbre bundle is locally trivial (locally Cartesian product). Let X be a Gspace. Let x be any ¯ point in X/G. Let X and Y be (right) Gspaces.π −1 (U (¯)) x x (¯. for any x ∈ X/G. For every point x. g) → x · g such that x · 1 = x and (x · g) · h = x · (gh). the map x → x · g is a homeomorphism.
U (¯) × G x z → (π(z). ¯ .6. ¯ (2) =⇒ (3) is obvious. there is a unique g ∈ G such that x = x · g. If X is a free Gspace. h)) = g −1 h y y is continuous.π −1 (Uα ) . τ (sx (π(z)). then translation function is unique because. x ) ∈ X ∗ . (1) =⇒ (2). y y Proof. 2) For any x ∈ X/G. g). the map φx : U (¯) × G → π −1 (U (¯)) is a homeomorphism.26 JIE WU A function τ : X∗ such that x · τ (x. τ (sx (π(z)). Then the following statements are equivalent each other: 1) The translation function τ : X ∗ → G is continuous. τ (x. Since τ (Uα × G)((¯. that is φα is a homeomorphism of Gspaces. z)). ∗ −1 . (Note. for any (x. Consider the (continuous) map θ : π −1 (U (¯)) x Then θ ◦ φx (¯. and so. τ (sx (¯).G satisfy the condition φα (¯.) Proposition 4.G τ (Uα × G) τ (X) ? ? ===== G = = = = = = G. φα } of X/G such that the homeomorphisms φα : Uα × G . the translation function restricted to π −1 (Uα ) τ (X) = τ (Uα × G) ◦ ((φα )∗ ) is continuous for each α and so τ (X) is continuous. gh) = φα (¯. g) · h. g) = θ(sx (¯) · g) = (¯. ¯ x x ¯ 3) There is an atlas {(Uα . x ) = g. Consider the commutative diagram (Uα × G)∗ φ∗ α . x ) = x for all (x. (¯. Let X be a (right) free Gspace with local crosssections to π : X → X/G. x ) ∈ X is called a translation function. z)) = sx (π(z)) · τ (sx (π(z)). g). It suﬃces to show that the restriction τ (X) : X ∗ ∩ (π −1 (Uα ) × π −1 (Uα )) = π −1 (Uα ) is continuous. (3) =⇒ (1). Note that the translation function is unique for free Gspaces. by deﬁnition. sx (¯) · g)) = (¯.−1 ∼ π (Uα ) = ∗ ∗ . y y ¯ y ¯ y ¯ y ¯ y φx ◦ θ(z) = φx (π(z). ¯ ¯ ¯ ¯ ¯ Thus φx is a homeomorphism. z) = z.
3. then ﬁbre bundle is trivial. In other words. θ is assume to be a monomorphism (equivalently. which is continuous. F ).LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 27 Now we give the deﬁnition. that is. 4. We are going to construct a principal Gbundle π : E G → B. Note. φα )} be an atlas and let gαβ : Uα ∩ Uβ . the composite G×F θ×idF . if we can choose G to be trivial (that is.Homeo(F. the Gaction on F is eﬀective. Also the topological group G plays an important role for ﬁbre bundles. Let {(Uα . F ) × F evaluation .Homeo(F. Let Γ be a topological group and let G be a closed subgroup.6 holds. if y · g = y for all y ∈ F . g) → ag for a ∈ Γ and g ∈ G is free. F ) be the transition functions. The Associated Principal GBundles of Fibre Bundles.).) 3) The transition functions gαβ : Uα ∩ Uβ . A principal Gbundle is a free Gspace X such that π : X → X/G has local crosssections and one of the (equivalent) conditions in Proposition 4. by choosing diﬀerent topological groups G. 2) There exists an atlas of ξ such that the transition functions gαβ lift to G via θ.G. We come back to look at ﬁbre bundles ξ given by p : E → B with ﬁbre F . φα )}.F is continuous. there is commutative diagram gαβ Uα ∩ Uβ . gαβ )} for the set of transition functions to the atlas {(Uα .Homeo(F. Then prove that the total space E = F ×G E G and p : E → B can be obtained canonically from π : E G → B. We will see that the bundle group G for ndimensional vector bundles can be chosen as the general linear group GLn (R). A topological group G is called a group of the bundle ξ if 1) There is a group homomorphism θ : G . then g = 1. Example. p.Homeo(F. we may get diﬀerent properties for the ﬁbre bundle ξ. that is. gαβ lifts to the trivial group). (where we use the same notation gαβ . that is. Thus Γ → Γ/G is principal Gbundle if and only if it has local crosssections. F ) 6 θ Uα ∩ Uβ gαβ . b) = a−1 b.7]. The . ¯ We write ξ = {(Uα . Then the action of G on Γ given by (a. In Steenrod’s deﬁnition [11. For instance. all ﬁbre bundles can obtained through principal Gbundles through this way.G are continuous. Then translation function is given by τ (a. 4) The Gaction on F via θ is continuous. Namely.
φα )}. The map φα is a homeomorphism because. α} = {b. α). gαβ (b)(gh). If n = 2m. g. g. The right Gaction on E G is deﬁned by {b. then φα is continuous and satisﬁes π ◦ φα (b. φα )} such that ξ is the set of transition functions to this atlas. one can consider the structure on the tangent bundles. Let ¯ E= α Uα × G × α. β) → (b. β) = (b.B ¯ and an atlas {(Uα . α} · h = {b. g. an oriented manifold means its tangent bundle is oriented. β} = θβ (b. 1 · g. α) in E . gαβ (b)g) = {b. g) = {b. g. g) = b and φα (b. The proof is given by construction. α}. gh. β) ∼ (b. gαβ (b)g. gh. g. for ﬁxing α. gαβ (b)g ) induces a map π −1 (Uα ) → Uα × G which the inverse of φα . g . Moreover. (gαβ (b)g)h. ¯ ¯ that is E is the disjoint union of Uα × G. g . . The vector bundle admits (almost) complex structure if and only if the transition functions can left to GLm (C). Let E G = E/ ∼ with quotient topology and G G let {b. α). g. Thus the {(Uα . β) ∼ (b. Deﬁne φα : Uα × G → π −1 (Uα ) by setting φα (b.7. gαβ )} is the set of transition function to the atlas {(Uα . Now deﬁne a relation on E by (b. Deﬁne π : E → B by π{b.) ¯ Proposition 4. the map (Uα ∩ Uβ ) × G × β β . g) = {b. φα (b. g) for b ∈ Uα ∩ Uβ and g ∈ G. If ξ is the set of transition functions for the space B and topological group G. g. α} = {b. then GLm (C) ⊆ GL2m (R). β) = (b. α} for the class of (b. α} · g for b ∈ Uα and g ∈ G. then (b . g = gαβ (b)g . 1. This is welldeﬁned (and so continuous) because if (b . α) ∼ (b .28 JIE WU vector bundle is orientable if and only if the transition functions can left to the subgroup of GLn (R) consisting of n × n matrices whose determinant is positive. β) ⇐⇒ b = b . Proof. then there is a principal Gbundle ξ G given by π : EG . ¯ This is an equivalence relation by Equations (3)(5). (For manifolds. α}. g. For instance. g . α} = b.Uα × G (b. g h. then π is clearly welldeﬁned (and so continuous).
LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 29 Let X be a right Gspace and let Y be a left Gspace. For any x ∈ X/G. y) πX X → x π→x ¯ X/G factors through X ×G Y . then the projection p : X ×G Y → X/G has the constant ﬁbre Y . gy) with quotient topology. φα ) for π : X → X/G. Observe that the action of Homeo(F. Now let p : E → B is a ﬁbre bundle with ﬁbre F and bundle group G. choose ¯ x ∈ π −1 (¯) ⊆ X. Thus G acts by left on F via θ : G → Homeo(F. The product over G is deﬁned by X ×G Y = X × Y /(xg. y) ∼ (x. Let p : X ×G Y → X/G be the resulting map. x x where Hx = {g ∈ G  xg = x}.X/G p : X ×G Y is a ﬁbre bundle with ﬁbre Y . Thus if X is a free right Gspace. there is a commutative diagram φα Uα × Y = = (Uα × G) ×G Y ∼ π −1 (Uα ) ×G Y = = p−1 (Uα ) = = = πU α πU α p p ? ? ? ? Uα = = = = = = Uα = = = = = = = Uα = = = = = Uα ===== ====== ===== and hence the result. Then ﬁbre bundle . then x p−1 (¯) = π −1 (¯) ×G Y = x × Y /Hx .8. denoted by ξ[Y ]. Proof. F ) on F is a left action because (f ◦ g)(x) = f (g(x)). F ).X/G p : X ×G Y is called induced ﬁbre bundle of ξ. Proposition 4. Then . Since the homeomorphism φα : Uα × G → π −1 (Uα ) is a Gmap. Let ξ be a (right) principal Gbundle given by π : X → X/G. A bundle morphism E(ξ) p(ξ) ? B(ξ) ¯ φφE(ξ ) p(ξ ) ? B(ξ ) . Consider a chart (Uα . Let π : X → X/G be a (right) principal Gbundle and let Y be any left Gspace. Note that the composite X ×Y (x. Let Y be any left Gspace.
gαβ (b. Rough introduction to this theory is as follows: There exists a universal Gbundle ωG as π : EG → BG. (There are few assumptions on the topology on B such as B is paracompact. Let ξ be a ﬁbre bundle given by p : E → B with ﬁbre F and bundle group G. gαβ (b)(y)) ((Uα ∩ Uβ ) × G × α) ×G F ((b. We write φα for φα in the proof of Proposition 4.7 according to a set of transitions functions to ξ Then ξ G [F ] ∼ ξ. y) → (b. β). Let ξ G be the principal Gbundle constructed in Proposition 4. y) (Uα ∩ Uβ ) × F ∼ φβ = (b. Given any principal Gbundle ξ over B. gαβ (b)g .) In this case. y) → ((b. it suﬃces to classify the principal Gbundles over B. g . This theorem tells that any ﬁbre bundle with a bundle group G is an induced ﬁbre bundle of a principal Gbundle.9. this means that (φ−1 . α). φα )} be an atlas for ξ. The assertion follows. y) → (b. (φ)−1 ) ∼ξ. is isomorphic to . Consider the map θα given by the composite: ˜ φα φα × idF π −1 (Uα ) ×G F ∼ (Uα × G × α)G × F = = Uα × F ∼ p−1 (Uα ). = ˜ Proof. there exists a (continuous) map f : B → BG such that ξ. for classifying ﬁbre bundles over a ﬁxed base space B. Thus. = = = From the commutative diagram ((Uα ∩ Uβ ) × G × β) ×G F (b. the map θα induces a bundle map E G ×G F θE(ξ) ? ? === B(ξ) = = = B(ξ).7. we write ξ = Theorem 4. The latter is actually done by the homotopy classes from B to the classifying space BG of G.) The theory for classifying ﬁbre bundles is also called (unstable) Ktheory. Let {(Uα . and θ is a local homeomorphic by restricting each chart. y)) . g . as a principal Gbundle. g . This is an bundle isomorphism because θ is onetoone and onto.30 JIE WU ¯ ¯ is call an isomorphism if both φ and φ are homeomorphisms. which is one of important applications of homotopy theory to geometry.(Uα ∩ Uβ ) × F ∼ φα = ? ? ===================== p−1 (Uα ∩ Uβ ) = = = = = = = = = = = = = = = = = = = = = = p−1 (Uα ∩ Uβ ). are continuous. (Note.
Fn . all nFplanes through the origin.m (F) × Fm  x ∈ V }. the total spaces of all vector bundles are just given by E(ξ GLn (F) ) ×GLn (F) Fn . y) → x . there is a Moreover.Homeo(Fn .p−1 (b) . ∞ F.4. complex or quaternion numbers. 10].) 4. From the composite (Uα ∩ Uβ ) × Fn the transition functions gαβ : Uα ∩ Uβ have that property that. that is.) Let j=1 . Seminar Topic: The classiﬁcation of principal Gbundles and ﬁbre bundles. the set of homotopy classes [B. In other words. In other words. The Construction of Gauss Maps. for continuous maps f. g : B → BG. It follows that the bundle group for a vector bundle can be chosen as the general linear group GLn (F). 0) = f (x) and F (x. (If m = ∞. Vector Bundles. F∞ = m E(γn ) = {(V. Proposition 4.(Uα ∩ Uβ ) × Fn . BG] is onetoone correspondent to the set of isomorphic classes of principal Gbundles over G. The Grassmann manifold Gn. we have the following. C or Hthe real. for each x ∈ Uαβ . b ∈ B. that is. is a linear isomorphism. Let ξ be an ndimensional Fvector space over B.9. 4. Let ξ be a vector bundle.EG π ? f . Then there exists a principal GLn (F)bundle ξ GLn (F) over B such that ξ ∼ ξ GLn (F) [Fn ]. An ndimensional Fvector bundle is a ﬁbre bundle ξ given by p : E → B with ﬁbre Fn and an atlas {(Uα . f ωG = continuous map (called homotopy) F : B ×[0.m (F) is the set of ndimensional Fsubspaces of Fm . x) ∈ Gn. Fn ) φβ . ∗ ∼ g ∗ ωG if and only if f g. pp. gαβ (x) : Fn . Conversely.10. y) ∈ B × EG  f (x) = π(y)} (x.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 31 the pullback bundle f ∗ ωG given by E(f ∗ ωG ) = {(x.p−1 (Uα ∩ Uβ ) φ−1 α .? B BG. ξ GLn (F) [Fn ] is an ndimensional Fvector bundle over B. φα )} in which each ﬁbre p−1 (b). Let F denote R. has the structure of vector space over F such that each homeomorphism φα : Uα × Fn → p−1 (Uα ) has the property that φα {b}×Fn : {b} × F n is a vector space isomorphism for each b ∈ Uα . for any principal GLn (F)= bundle over B.5. (References: for instance [8. 1) = g(x). By Theorem 4. with the topology described as in the topic on the examples of Manifolds.4858] Or [9. 1] → BG such that F (x.
1) If there is a vector bundle morphism E(ξ) p(ξ) ? B(ξ) um E(γn ) m p(γn ) ? f . Since g restricted to each ﬁbre is a monomorphism. e1 ). For each b ∈ B(ξ). g(b. e1 ).11. (1) is obvious.m (F) m γn . The map m q : E(γn ) → Fm (V. e1 ). . Deﬁne the functions f : B(ξ) → Gn. Let ξ be an ndimensional Fvector bundle. . en )) is continuous. . g(b.m (F) m u : E(ξ) → E(γn ) f (b) = g(p(ξ)−1 (b)). . . x) → x is a Gauss map. H) = Sp(n).m (F) ×O(n. Let {e1 . g(b. the vectors {g(b. . g(b.m (F) is a principal O(n. where Vn. en } be the standard Fbases for Fn . then q ◦ u : E(ξ) → Fm is a Gauss map.] A Gauss map of an ndimensional Fvector bundle in Fm (n ≤ m ≤ ∞) is a (continuous) map g : E(ξ) → Fm such that g restricted to each ﬁbre is a linear monomorphism. R) = O(n).m (F) b → (g(b. For checking the continuity of f and u. (V. Thus h: B ˜ .Gn. . e2 ). where O(n.32 JIE WU Exercise 4. then there is a vector bundle morphism (u.3. 2) If there is a Gauss map g : E(ξ) → Fm . u(z) = (f (p(z)). The functions f and u are welldeﬁned. O(n. . . Show that m p : E(γn ) → Gn. . .Vn. en )) ∈ Vn. e2 ). check that Vn. denoted by [Hint: By reading the topic on the examples of manifolds. e2 ). e1 ). e2 ). . Proposition 4. en )) . . x) → V is an ndimensional Fvector bundle. F). g(p(ξ)−1 (b)) is an ndimensional Fsubspace of Fm and so a point in Gn. g(z)). g : B(ξ) × Fn → Fm restricted to each ﬁbre is a linear monomorphism. . .m (F) for each b. . g(b. en )} are linearly independent and so ˜ (g(b. (2). .F) Fn . one can look at a local coordinate of ξ and so we may assume that ξ is a trivial bundle. Then check that E(γn ) = Vn. . Then the map h : B .m (F) → Gn. Proof. .m (F) that is an isomorphism when restricted to any ﬁbre of ξ.m (F). Example. C) = m U (n) and O(n. g(b.m (F) is the open Stiefel manifold over F. . g(b. namely. f ) : ξ → m γn such that qu = g. g(b.Fm × · · · × Fm b → (g(b.
Proposition 4. u ˜ m . From Part (2) of Proposition 4.11.m (F) quotient .LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 33 is continuous and so the composite f: B h ˜ .Vn. =⇒ Assume that ξ has a Gauss map g.Gn.Gn. there is commutative diagram E(ξ) p(ξ) ? ? B(ξ) = = = B(ξ).12. there is a commutative diagram um E(γn ) E(ξ) p(ξ) ? B(ξ) m p(γn ) ? f .E(ξ) E(f ∗ ξ) = {(x. This ﬁnishes the proof.m (F) × Fm is continuous.m (F). === where u restricted to each ﬁbre is a linear isomorphism because both vectorbundle has the same ˜ dimension and the Gauss map g restricted to each ﬁbre is a monomorphism. It follows that . Let ξ be a vector bundle and let f : X → B(ξ) be a (continuous) map. Proof.m (F). There exists a Gauss map g : E(ξ) → Fm (n ≤ m ≤ ∞) if and only if ξ ∼ f ∗ (γ m ) = n over B(ξ) for some map f : B(ξ) → Gn. The function u is continuous because the composite u m .m (F)  is continuous. ⇐= is obvious.E(γn ) E(ξ) ∩ ∆ ? ? (f ◦ p(ξ)) ×g E(ξ) × E(ξ) Gn. Then the induced vector f ∗ ξ is the pullback .E(f ∗ γn ) p m Since E(f ∗ γn ) is deﬁned to be the pullback. y) ∈ X × E(ξ) f (x) = p(y)} p ? X f? B(ξ).
we need some preliminary results for bundles over paracompact spaces. Lemma 4. Let {(Uα .) 3) The equation λα (x) = 1 α∈J for all x ∈ X. First.13. by the deﬁnition of partition of unity. φα  α ∈ J} be an atlas for ξ. 1] such that 1) The support supp(λα ) ⊆ Uα for each α. (In other words. let S(b) = {α ∈ J  λα (b) > 0}. We are going to ﬁnd another atlas with countable charts. where the summation is welldeﬁned for each given x because there are only ﬁnitely many nonzeros. Let ξ be a ﬁbre bundle over a paracompact space B. (For further information on paracompact spaces. Theorem 4. then there exists a partition of unity subordinate to U. there is a neighborhood Wx of x such that λλα Wx ≡ 0 for only ﬁnitely many indices α ∈ J. Vβ ⊆ Uα for some α. Let Vα = λ−1 (0. We give the following wellknown theorem without proof. V is called a reﬁnement of U if for each β. one can see [3. Let U = {Uα  α ∈ J} be an open cover of a space X. 2) for each x ∈ X. By Theorem 4. subordinate to U. We are going to construct a Gauss map for each vector bundle over a paracompact space. Moreover u is a homeomorphism by considering a ˜ local coordinate. For each b ∈ B. by the deﬁnition of partition of unity.13.14. Then. A Hausdorﬀ space X is called paracompact if it is regular and if every open cover of X admits a locally ﬁnite reﬁnement. α ¯ Then. where the support supp(λα ) = {x ∈ X  λα (x) = 0} is the closure of the subset of X on which λα = 0.1720]. the supports of λα ’s are locally ﬁnite. 1] = {b ∈ B  λα (b) > 0}. If X is a paracompact space and U = {Uα } is an open cover of X. Proof.34 JIE WU m ˜ u : E(ξ) → E(f ∗ γn ) is onetoone and onto. pp. S(b) is a ﬁnite subset of J. Let U = {Uα } and V = {Vβ } be two open covers of X. A partition of unity. One may read a proof in [2. there is a partition of unity {λα  α ∈ J} subordinate to {Uα  α ∈ J}. Vα ⊆ Vα ⊆ Uα . Now for each ﬁnite subset S of J deﬁne W (S) = {b ∈ B λα (b) > λβ (b) for each α ∈ S and β ∈ S} . is a collection {λα  α ∈ J} of continuous functions λα : X → [0. Then ξ admits an atlas with countable charts. 162169]. A family of C = {Cα  J} of subsets of a space X is called locally ﬁnite if each x ∈ X admits a neighborhood Wx such that Wx ∩ Cα = ∅ for only ﬁnitely many indices α ∈ J.
(Then {Wm } induces an atlas for ξ. Otherwise there exists b ∈ W (S) ∩ W (S ). Proof. Fixing α ∈ S(b). . From the above.Ui × Fn projection p2 . λα (b) > λβ (b) because α ∈ S and β ∈ S. check that Wj = ∅ for j > k.15. β ∈ S and α ∈ S . where k is ﬁnite or inﬁnite. . φi )}1≤i≤k be an atlas of ξ with countable or ﬁnite charts. Let m = S(b). Wm is a disjoint union of W (S(b)). Then there exist α ∈ S S and β ∈ S S because S = S but S and S has the same number elements.} is an open cover of B. Now check that ξ restricted to Wm is trivial. and so the above (possibly inﬁnite) intersection of open sets restricted to Wb is only a ﬁnite intersection of open sets. [In this case. It suﬃces to check that ξ restricted to each W (S(b)) is trivial. By deﬁnition W (S). For each b ∈ B. . where S is the number of elements in S. In particular. then B admits an atlas of ﬁnite (at most k) charts. if ξ has an atlas of k charts. Since ξ restricted to Uα is trivial.] Theorem 4.Fn . . From the proof. Note. Let S and S be two subsets of J such that S = S and S = S  = m > 0. i where p2 ◦ φ−1 is the composite i p(ξ)−1 (Ui ) φ−1 i W (S(b)) . if for each b ∈ B there are at most k sets Uα with b ∈ Uα . and (2) ξ restricted to Wm is a trivial bundle for each m. then b ∈ Wm and so {Wm } is an open cover of B. for any x ∈ W (S(b)). On the other hand. deﬁne the map gi : E(ξ) → Fn as follows: gi restricted to p(ξ)−1 (Ui ) is given by gi (z) = λi (z)(p2 ◦ φ−1 (z)). λβ (b) > λα (b) because b ∈ W (S ). then λα (x) > λβ (x) for any β ∈ S.) To check {Wm } is an open cover. then ξ has a Gauss map g : E(ξ) → Fkn . We claim that W (S) ∩ W (S ) = ∅. S(b) is a ﬁnite set and b ∈ W (S(b)) because λβ (b) = 0 for β ∈ S(b) and λα (b) > 0 for α ∈ S(b). Now deﬁne Wm = b∈B S = m for each m ≥ 1. This ﬁnishes the proof. For each i. Then W (S) is open because for each b ∈ W (S). 1]. λα (x) > 0 for any x ∈ W (S(b)). Any ndimensional Fvector bundle ξ over a paracompact space B has a Gauss map g : E(ξ) → F∞ . by deﬁnition of partition of unity. Let {λi } be the partition of unity subordinate to {Ui }. note that each Wm is open.LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 35 = α∈S β∈S (λα − λβ )−1 (0. ξ restricted to W (S(b)) is trivial. We prove that (1) {Wm  m = 1. Let {(Ui . It follows that W (S(b)) ⊆ Vα ⊆ Uα . Moreover. there exists a neighborhood Wb of b such that there are ﬁnitely many supports intersect with Wb .
∞ (F). For instance. CP∞ ] = H 2 (B. then G1.m (F). This gives the following classiﬁcation theorem: Corollary 4.∞ (H)] = [B. G1. Some people even believe that it is impossible to compute the general homotopy groups πn (S 3 ).∞ (F)]. HP∞ ] is very hard problem even when B are spheres. the map g is a Gauss map. G1. However. Since each gi : E(ξ) → Fn is a monomorphism (actually isomorphism) on the ﬁbres of E(ξ) over b with λi (b) > 0. ξ ⊕ η is just the ﬁbrewise direct sum. 1] is contained in Ui .∞ (H) BSp(1) BS 3 HP∞ . From this. and so [B. Seminar Topic: Gauss Maps and the Classiﬁcation of Vector Bundles. m m Remarks: It can be proved that f ∗ γn ∼ g ∗ γn if and only if f g : B → Gn.∞ (C) BU (1) BS 1 CP∞ . and [B. if n = 1 and F = R. . RP ] = H 1 (B.) A vector bundle is called of ﬁnite type if it has an atlas with ﬁnite charts. If n = 1 and F = C. This a welldeﬁned (continuous) map because for each z. any real line bundles over a simply connected space is always trivial.E(ξ) × E(η) E(ξ ⊕ η) p(ξ) × p(η) ? B ? ∆ . i gi is a welldeﬁned (continuous) map. diagonal Intuitively.3133]. where BG is socalled ∞ the classifying space of the (topological) group G. In particular.B × B. the Whitney sum ξ ⊕ η is deﬁned to be the pullback: . by a lot of computations through many papers. Given two vector bundles ξ and η over B. HP∞ ] = πn−1 (S 3 ) that is only known for n less than 66 or so. Every vector bundle over a paracompact space B is isomorphic to an induced vector ∞ bundle f ∗ (γn ) for some map f : B → Gn. (References: for instance [8. and [B. Gn. then G1.m (F). Z). = one get that the set of isomorphism classes of ndimensional Fvector bundles over a paracompact space B is isomorphic to the set of homotopy classes [B.36 JIE WU and gi restricted to the outside of p(ξ)−1 (Ui ) is 0. which states that all complex line bundles are by the second integral cohomology. Moreover every vector bundle over a paracompact m space B with an atlas of ﬁnite charts is isomorphic to an induced vector bundle f ∗ (γn ) for some m and some map f : B → Gn. If B = S n .16. there is a neighborhood of z such that there are only ﬁnitely many gi are not identically zero on it. pp. If n = 1 and F = H. HP∞ ]. and since the images of gi are in complementary subspaces of Fkn . which states that all line bundles are by the ﬁrst cohomology with coeﬃcients in Z/2Z. Since the closure of λ−1 (0. Now deﬁne k k g : E(ξ) → i=1 Fn = Fkn g(z) = i=1 gi (z). the determination of [B.2629. then [B. Z/2).∞ (R) BO(1) BZ/2 RP∞ .
m (F) such that ξ is isomorphic to f ∗ γn . where O(n. = Corollary 4. m there is a map f : B → Gn. O(n.m (F) × Fm  v ⊥ V } m m m with canonical projection E((γn )∗ ) → Gn. the Whitney sum is an operation on vector bundles over a (ﬁxed) basespace.m (F). Let ξ be a Fvector bundle over a compact (Hausdorﬀ ) space B. the following statement are equivalent: 1) The bundle ξ is of ﬁnite type. It follows that m m m m f ∗ (γn ⊕ (γn )∗ ) = f ∗ (γn ) ⊕ f ∗ ((γn )∗ ) m is trivial.18. (1) =⇒ (2) follows from Corollary 4. people also study the category whose objects are just continuous maps f : E → B with ﬁxed space B. By Proposition 4. the Whitney sum plays as an addition (that is associative and commutative with 0). Then γn ⊕ (γn )∗ is an mdimensional trivial Fvector bundle. m 2) There exists a map f : B → Gn. In this sense. where m = dim(ξ ⊕ η). m Proof. F) × O(m − n. The composite E(ξ) ⊂ .E(ξ ⊕ η) = B × F m . F) is compact. Let (γn )∗ be the vector bundle given by m E((γn )∗ ) = {(V. C) = U (n) and O(n. Then there is a Fvector bundle η such that ξ ⊕ η is trivial. Let η = f ∗ ((γn )∗ ). where the trivial bundles (of diﬀerent dimensions) are all regarded as 0. a map f : E → B is called a bundle (without assuming locally trivial). v) ∈ Gn. F)/O(n. In the terminology of ﬁbre bundles. The bundle ξ with property that ξ ⊕ η is trivial for some η means that ξ is invertible. Then ξ ⊕ η is trivial.m such that ξ ∼ f ∗ γn . 3) There exists a vector bundle η such that the Whitney sum ξ ⊕ η is trivial. or even more general category whose objects are diagrams over spaces. = m (2) =⇒ (3). More general situation possibly is the sheaf theory (by removing the locally trivial condition) that is pretty useful in algebraic geometry. It follows m that ξ ∼ f ∗ γn is of ﬁnite type. In the view of (stable) Ktheory.Fm is a Gauss map into ﬁnite dimensional vector space. R) = O(n).16. In algebraic topology.12.17. (3) =⇒ (2).LECTURE NOTES ON DIFFERENTIABLE MANIFOLDS 37 Proposition 4. H) = Sp(n).m (F) = O(m. Those who are interested in algebra can push notions in algebra to vector bundles by doing constructions ﬁbrewisely. . For a vector bundle ξ over a paracompact space B. (2) =⇒ (1) It is an exercise to check that γn is of ﬁnite type by using the property that the Grassmann manifold Gn.
(1963). Tensor Algebras and Exterior Algebras 9. Topology.38 JIE WU 5. Princeton. 63 (1956). B. Boston. Edelen. 593610. The Topology of Fibre Bundles. Inc. [1974]. . Milnor. Groups of homotopy spheres I. Tangent Bundles and Vector Fields 6. Construction of universal bundles II. AddisonWesley Publishing Company. Stokes’ Theorem References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] R. 68 (1946). Orientation of Manifolds 8. Topologies for homeomorphism groups. Dale Husemoller. Ninth Printing. Graduate Texts in Mathematics SpringerVerlag [1966]. J. second edition. Dugundji. M. Applied Exterior Calculus. Math. [1972]. W. Englewood Cliﬀs. Math. Math. Construction of universal bundles I. 272284. Arens. 63 (1956). A. Diﬀerential Manifolds. Ann. DeRham Cohomology 10. Diﬀerentiable Manifolds. Lawrence Conlon. Kervaire and J. Inc. 430436. New Jersey [1974]. Diﬀerential Topology. Fibre Bundles. John Wiley & Sons. 77 504537. Ann. New Jersey. Amer. Milnor. [1985]. Jour. J. Integration on Manifolds 11. Milnor. Birkh auser Boston [2001]. Prentice Hall. Norman Steenrod. MA [1966]. Dominic G. J. Victor Guillemin and Alan Pollack. Cotangent Bundles and Tensor Fields 7. Serge Lang. Inc. Math. second Edition. Princeton University Press. Ann. Allyn and Bacon.