DIFFERENTIAL TOPOLOGY LECTURES

WENYUAN LI

1. Lecture 1: Introduction and Preliminaries

Our course is on differentiable manifolds or smooth manifolds, which are topological
spaces with sufficiently nice properties that allow one to do calculus.
There are many situations where we would like to do calculus not on Rd but on more
general spaces. For instance, sometimes we need to do differentiation and integrations on
curves and surfaces in many circumstances, and sometimes we need to do differentiation
and integrations on spaces given by cutting and pasting.
We elaborate more on the second case. For instance, in real analysis, we may want to
do calculus on spaces like the circle S 1 = [0, 1]/{0 ∼ 1}, without putting it into Euclidean
spaces as {(x, y) | x2 + y 2 = 1} ⊂ R2 all the time. In complex analysis, we may also want
to do calculus on spaces like the domain of the n-th root function f (z) = z 1/n . Note that
this function is not well-defined on C. However, it is well defined as a continuous (in fact,
smooth) function on the topological space
e n = {(z, k) | z ∈ C, k ∈ N, 1 ≤ k ≤ n}/{(0, i) ∼ (0, j)}
C
by f (reiθ , n) = r1/n eiθ/n . The latter case motivated B. Riemann to consider the notion
which is now known as Riemann surfaces in 1851, which later lead him to introduce the
first preliminary definition of manifolds.
In order to do calculus, we would like to assume that the spaces locally just look like
standard Euclidean spaces. Recall that a topological space is Hausdorff if for any two
distinct points x and y, there exist neighbourhoods Ux and Uy that are also disjoint.
Definition 1.1. A locally Euclidean space M of dimension d is a Hausdorff topological
space M for which each point has a neighborhood homeomorphic to an open subset of the
Euclidean space Rd .
Example 1.1. (1) Any open subset U ⊆ Rd is locally Euclidean of dimension d. (2) The
circle S 1 = [0, 1]/{0 ∼ 1}, which is homeomorphic to {(x, y) | x2 + y 2 = 1} ⊂ R2 , is locally
Euclidean of dimension 1. (3) The cone C+ 2 = {(x, y, z) | z 2 = x2 + y 2 , z ≥ 0} ⊂ R3 is

locally Euclidean of dimension 2. (4) The sphere

S d = {(x1 , x2 , . . . , xd+1 ) | x21 + x22 + · · · + x2d+1 = 1} ⊂ Rd+1
is locally Euclidean of dimension d. Let n = (0, . . . , 0, 1) and s = (0, . . . , 0, −1) be the north
and south poles and pn and ps be the stereographic projections from n and s:
pn : S d \ {n} → Rd , ps : S d \ {s} → Rd .
This gives the local Euclidean structure at any point.
Example 1.2. (1) The double-sided cone C± 2 = {(x, y, z) | z 2 = x2 + y 2 } ⊂ R3 is not locally

Euclidean. This can be proved by observing that C± 2 \{(0, 0, 0)} is disconnected. (2) The one-
1 1
point union of two circles S ∧ S = [0, 2]/{0 ∼ 1 ∼ 2} is not locally Euclidean. This can
be proved by observing that S 1 ∧ S 1 \{0} is has three connected components. (3) The half
1
2 WENYUAN LI

space R2+ = {(x1 , x2 ) | x2 ≥ 0} is not locally Euclidean. This can be proved by observing
that R2+ \{(0, 0)} is simply connected.
The following example explains when the condition of being a Hausdorff space fails.
One reason we would like to exclude this type of examples is that, on these (non-Hausdorff)
spaces, we may not be able to find continuous functions that have different values at different
points. One of the most effective ways to distinguish points is to look at functions with
different values on such points. This is only possible when we consider Hausdorff spaces.
Example 1.3. Consider the disjoint union of two real lines R ⊔ R. We label a real number
x in the first R-component by x1st and a real number x in the second R-component by x2nd .
The fat real line Rfat = R ⊔ R/{x1st ∼ x2nd | x ̸= 0} is not locally Euclidean because it is
not Hausdorff. In fact, any neighbourhood of 01st and 02nd have non-empty intersections.
Local homeomorphisms with Euclidean spaces allow us to consider coordinates on the
spaces using coordinates on the Euclidean spaces. Coordinates in Euclidean spaces are
simply functions Rd → R. Therefore, on locally Euclidean spaces, we also define coordinates
as some special functions.
Definition 1.2. Let M be a locally Euclidean space. If x is a homeomorphism of a connected
open set U ⊆ M onto an open subset of Rd , then φ is called a coordinate map. Let πi :
Rd → R be the standard coordinates. The functions xi = πi ◦ x are called the coordinate
functions, and the pair (U, x) (also denoted by (U, x1 , ..., xd )) is called a coordinate system.
Note that the coordinate function is not always defined on the whole space M . Instead,
it is only defined on the open subset U .
Example 1.4. Consider the circle S 1 = [0, 1]/{0 ∼ 1}. Then the open subset (0, 1) ⊂ S 1 is
homeomorphic to an open subset in R. The coordinate function is given by φ : (0, 1) → R.
e : S 1 → R.
It is an exercise that this cannot be extended to a continuous function φ

2. Lecture 2: Differentiable Manifolds

On locally Euclidean spaces, we now introduce differentiable structures or smooth struc-
tures. These will allow us to do calculus on such spaces. The only situation where we know
how to characterize differentiability is on Rd . Therefore, we will define smooth structure
by requiring certain maps between Euclidean spaces to be smooth. The natural source of
such maps come from coordinate systems. The following definition was first introduced by
H. Weyl.
Definition 2.1. A C k -differentiable structure (0 ≤ k ≤ ∞ or ω, the latter means analytic
functions) on a locally Euclidean space M is a collection of coordinate systems {(Uα , xα ) |
α ∈ A} satisfying the following properties:
S
(1) α∈A Uα = M ;
(2) (xα |Uα ∩Uβ ) ◦ (xβ |Uα ∩Uβ )−1 are C k -maps for all α, β ∈ A;
(3) The collection is maximal with respect to (2); that is, if (U, x) is a coordinate system
such that (x|U ∩Uα ) ◦ (xα |U ∩Uα )−1 and (xα |U ∩Uα ) ◦ (x|U ∩Uα )−1 are C k -maps for all
α ∈ A, then (U, x) is also in the collection.
When k = ∞ or ω, we also call this a smooth or analytic structure on M .
A d-dimensional C k -differentiable manifold is a pair (M, F ) consisting of a d-dimensional,
second countable, locally Euclidean space M together with a C k -differentiable structure F .
When k = ∞ or ω, we also call this a smooth or analytic manifold.
 DIFFERENTIAL TOPOLOGY LECTURES 3

Remark 2.1. We will not be able to prove this now, but it is true that any C 1 -manifold
admits a C k -structure for k ≥ 1 or k = ω (by considering a subcollection of coordinate
systems). However, it is not true that any C 0 -manifold admits a C 1 -structure.
We recall that a space is second countable if it has a countable topological basis (where
a topological basis is a collection of open subsets such that all open subsets are forms by
unions of them). Later (on Friday) we will explain why we require a manifold to be second
countable and Hausdorff.
In the definition of differentiable structures, the last property is to make the defini-
tion canonical: whenever two differentiable structures F = {(Uα , xα ) | α ∈ A} and
G = {(Vβ , yβ ) | β ∈ B} are compatible in the sense that all transition maps xα ◦ yβ−1
α are differentiable, then F = G by property (3).
and yβ ◦ x−1
If F0 = {(Uα , xα ) | α ∈ A} is any collection of coordinate systems satisfying properties
(1) and (2), then we can always find unique differentiable structure F containing F0 by
F = {(U, x) | φ ◦ x−1
α , xα ◦ x
−1
are C k for all α ∈ A}.
Definition 2.2. Let M and N be smooth manifolds. We say that f : M → R is a smooth
function on M (denoted by C ∞ (M ) or C ∞ (M, R)) if (f |Uα ) ◦ x−1 α is smooth for each co-
ordinate map xα : Uα → Rd . A continuous map f : M → N is a smooth map (denoted by
C ∞ (M, N )) if (yβ |f (Uα )∩Vβ ) ◦ (f |Uα ∩f −1 (Vβ ) ) ◦ (xα |Uα ∩f −1 (Vβ ) )−1 is a smooth function for all
coordinate maps xα : Uα → Rd and yβ : Vβ → Rc .
Lemma 2.1. Let M and N be smooth manifolds. A continuous map f : M → N is a
smooth map if and only if for any smooth function g on N , the composition g ◦ f is a
smooth function on M .
Now we introduce some examples of smooth manifolds. The details are left as exercises.
Example 2.2. (1) Any open subset in Rd is a manifold. (2) The general linear group
GL(n, R) = {A ∈ Mn×n (R) | det(A) ̸= 0} is a manifold. (3) The product of any two
manifolds M and N is still a manifold. (4) The standard sphere
S d = {(x1 , x2 , . . . , xd+1 ) | x21 + x22 + · · · + x2d+1 = 1} ⊂ Rd+1
is a manifold. Let n = (0, . . . , 0, 1) and s = (0, . . . , 0, −1) be the north and south poles and
pn and ps be the stereographic projections from n and s:
pn : S d \ {n} → Rd , ps : S d \ {s} → Rd .
Then the maximal collection that contains {(S d \ {n}, pn ), (S d \ {s}, ps )} gives a smooth
structure on S d . (4) The projective space
RPd = {[z0 , z1 , . . . , zd ] ∈ Rd+1 | z02 + z12 + · · · + zd2 ̸= 0}/{[z0 , z1 , . . . , zd ] ∼ [λz0 , λz1 , . . . , λzd ]}
is a manifold. One can consider local coordinate systems (Uα , xα ) for 0 ≤ α ≤ d by
Uα = {[z0 , z1 , . . . , zd ] ∈ Rd+1 | zα ̸= 0}, xα ([z0 , z1 , . . . , zd ]) = (z0 /zα , . . . , zd /zα ).

3. Lecture 3: Partitions of Unity

We mentioned that (one of) the most effective ways to distinguish points is to look at
functions with different values on such points. We prove that on manifolds such functions
can indeed be constructed. Second countability and Hausdorff will play a major
S role.
Recall that a collection {Uα | α ∈ A} of subsets of M is a cover if M = α∈A Uα . It is
an open cover if each Uα is open. A subcollection which still covers M is called a subcover.
A refinement {Vβ | β ∈ B} of the cover {Uα | α ∈ A} is a cover such that for each α ∈ A
there is an β ∈ B such that Vβ ⊆ Uα . A collection {Uα | α ∈ A} of subsets of M is locally
4 WENYUAN LI
S
finite if whenever x ∈ α∈A Uα there exists a neighborhood Wx of x such that Uα ∩ Wx is
non-empty for only finitely many α ∈ A. A topological space is paracompact if every open
cover has an open locally finite refinement.
Recall that a space is locally compact if each point has at least one compact neighborhood.
It is a simple exercise to show that any locally Euclidean space is locally compact.
Lemma 3.1. Let M be a topological space which is locally compact, Hausdorff, and second
countable (manifolds, for example). Then X is paracompact. In fact, each open cover has
a countable, locally finite refinement consisting of open sets with compact closures.
Proof. We decompose M into a countable union of compact subsets and then use (local)
compactness to find a (locally) finite refinement. The Hausdorff property is used to show
that a compact set is closed.
Given a countable basis of M , since M is local compact, for each point there exists an
open subset in the basis that are contained in a compact set. Since M is Hausdorff, we
know that such open subsets have compact closures. Consider the countable subcollection
{Ui }i∈N of all theSopen subsets with compact closures. This collection {Ui }i∈N stillSforms a
basis. Let Gi = 1≤j≤i Uj . Then Gi has compact closures, Gi ⊆ Gi+1 and M = i∈N Gi .
This implies that Gi+1 \ Gi are compact subsets.
Now, for any open cover {Vα | α ∈ A}, we can consider the compact subsets {Gi+1 \ Gi |
i ∈ N} of M . For each i ∈ N, there exists a finite subcover. We choose the subcover
inductively and this gives a locally finite refinement. □
Lemma 3.2. Let M be a topological space which is paracompact and Hausdorff. Then M
is normal: for any disjoint closed subsets Z1 and Z2 , there exist open neighbourhoods Ui of
Zi such that U1 and U2 are also disjoint.
Then, using the paracompact and Hausdorff property, we can construct smooth functions
that distinguish points. In fact, we can construct partitions of unity:
Definition 3.1. A partition of unity on M is a collection {φα | α ∈ A} of smooth non-
negative functions on M such that
P collection of supports {supp φα | α ∈ A} is locally finite.
(1) The
(2) α∈A φα (x) = 1 for all x ∈ M .
A partition of unity {φα | α ∈ A} is subordinate to the cover {Uβ | β ∈ B} if for any β ∈ B,
there exists some α ∈ A such that supp φα ⊂ Uβ .
We remark that for a partition of unity to be subordinate to a cover, one open subset Uα
may contain the support of many functions φβ .
Theorem 3.3 (Existence of Partitions of Unity). Let M be a differentiable manifold and
{Uα | α ∈ A} an open cover of M . Then there exists a countable partition of unity with
compact supports {φi | i ∈ N} subordinate to the cover {Uα | α ∈ A}.
Lemma 3.4. There exists a non-negative smooth function f on Rd which equals 1 on
[−1, 1]d and 0 on the complement of (−2, 2)d .
Proof. We define f (x1 , . . . , xd ) = h(x1 )h(x2 ) . . . h(xd ), where
(
u(t) e−1/t , t > 0,
h(t) = g(2 + t)g(2 − t), g(t) = , u(t) =
u(t) + u(1 − t) 0, t ≤ 0.
This satisfies the condition. □
 DIFFERENTIAL TOPOLOGY LECTURES 5

Proof of Theorem. Consider a collection {(Vx , φx ) | x ∈ M } of coordinate systems that is

a refinement of {Uα | α ∈ A}. For each (Vx , φx ), by the lemma above, we can consider
a function ρ that is compactly supported in φx (Vx ) and equal to 1 on an open subset
Bx ⊂ φx (Vx ). Then the function fx = f ◦ φx is compactly supported in Vx and equal to 1
on an open subset Wx ⊂ Vx (where Wx = φ−1 x (Bx )). Now the collection {Wx | x ∈ M } forms
an open cover that refines {Uα | α ∈ A}. Since M is second countable and paracompact,
we can choose a countable locally finite cover {Wi | i ∈ N} that also refines {Uα | α ∈ A}.
We consider
fi
ρi = P .
j∈N fj
Then {ρi | i ∈ N} forms a partition of unity subordinate to {Uα | α ∈ A}. □
Remark 3.1. We remark that the above proof also shows that any second countable para-
compact Hausdorff space admits a continuous partition of unity.
Corollary 3.5. Let M be a differentiable manifold, Z be a closed subset and U be an open
neighbourhood of Z. Then there exists a smooth function φ on M such that
(1) 0 ≤ φ(x) ≤ 1;
(2) φ(x) = 1 for x ∈ Z;
(3) φ(x) = 0 for x ∈
/ U.
Such functions are called bump functions or cut-off functions.

4. Lecture 4: Tangent Vectors and Differentials

Our goal this week is to define derivatives and differentials locally on a smooth manifold.
More precisely, we will define tangent spaces and cotangent spaces at a point. Tangent vec-
tors will give directional derivatives at a point, and cotangent vectors will give differentials
of functions at a point.
The difficulty of defining the tangent and cotangent spaces is that manifolds are only
topological spaces and there is no vector space structures. However, one natural vector
space that arises from a manifold is the vector space of functions C ∞ (M ). We will start
from that.
We will first try to define cotangent spaces. This is because cotangent vectors are differ-
entials of functions at a point, and differentials, roughly speaking, are simply the first-order
approximations of functions.
Definition 4.1. Let M be a smooth manifold and p be a point on M . Consider the set
of smooth functions defined on some (open) neighbourhood of p, namely, the set of pairs
(U, f ) such that U is an open neighbourhood of p and f is in C ∞ (U ). Define an equivalence
relation such that (U, f ) ∼ (V, g) if there exists an open neighbourhood W ⊂ U ∩ V of x
such that f |W = g|W . Define
Cp∞ = {(U, f ) | U open neighbourhood of p, f ∈ C ∞ (U )}/ ∼
to be the vector space of germs of smooth functions at p ∈ M .
Use partitions of unity, one can prove that germs of smooth functions have a simpler
description. (The reason to make the more complicated definitions is that germs make
sense even when we do not have partitions of unity. For instance, one can define germs of
analytic functions in the above way.)
Lemma 4.1. Let M be a smooth manifold and p be a point on M . Define an equivalence
relation on C ∞ (M ) such that f ∼ g if there exists an open neighbourhood W of p such that
f |W = g|W . Then Cp∞ = C ∞ (M )/ ∼ .
6 WENYUAN LI

Differentials of functions the space of functions that vanish at the point modulo second
and higher order terms. Recall the following simple fact: if two smooth functions f1 and
f2 vanish at a point x, then f1 f2 vanish at x up to the second order (by the Leibniz rule).
This gives an alternative description of functions that vanish up to the second order.
Definition 4.2. Let M be a smooth manifold and Cp∞ is the space of germs of smooth
functions at p. Define mp ⊂ Cp∞ to be the subspace of germs of functions f ∈ Cp∞ such
that f (p) = 0 and m2p ⊂ Cx∞ to be subspace spanned by germs of functions of the form
f1 f2 ∈ Cx∞ such that f1 (p) = f2 (p) = 0. Then the cotangent space at p ∈ M is
Tp∗ M = mp /m2p .
Theorem 4.2. dim Tp∗ M = dim M .
Recall that Newton–Leibniz rule implies that for a C 1 -differentiable function on Rd and
x = (x1 , . . . , xd ) ∈ Rd , we have
d Z 1
X ∂f
f (x) = f (0) + xi (1 − t) (tx)dt.
0 ∂xi
i=1

Doing it once more implies that for a C 2 -differentiable

function on Rd and x = (x1 , . . . , xd ) ∈
d
R , we have Taylor’s formula with integral remainders
d d Z 1
X ∂f X ∂2f
f (x) = f (0) + xi (0) + xi xj (1 − t) (tx)dt.
∂xi 0 ∂xi ∂xj
i=1 i,j=1

Proof of the Theorem. Consider a local coordinate system (U, x) around p such that x(p) =
0 in Rd . Then for any germ of function f ∈ mp that vanishes at p, we can consider the
function f ◦ x−1 and write
d d
∂(f ◦ x−1 ) ∂ 2 (f ◦ x−1 )
X X Z 1
f (q) = xi (q) (0) + xi (q)xj (q) (1 − t) (tx(q))dt.
∂xi 0 ∂xi ∂xj
i=1 i,j=1

Since the last term is a linear combination of functions that vanish at p, we have in mp /m2p
that
d
X ∂(f ◦ x−1 )
f (q) ≡ xi (q) (p).
∂xi
i=1
This implies that {x1 , . . . , xd } spans mp /m2p . Now, we can also show that they are linear
independent. This completes the proof. □
Definition 4.3. Let M be a smooth manifold and p be a point in M . We write the image
of a germ of smooth functions f in mp /m2p as the differential dfp . In particular, for a
coordinate system (U, x) around p. We write the image of x1 , . . . , xd in mp /m2p as the
differentials (dx1 )p , . . . , (dxd )p .
Next, we try to define the tangent space. They are, roughly speaking, directional deriva-
tives at a point. Derivatives are operators that act on smooth functions, and we can define
them in the following axiomatic way.
Definition 4.4. A tangent vector v at the point p ∈ M is a linear derivation on Cp∞ . That
is, for all f1 , f2 ∈ Cp∞ and λ1 , λ2 ∈ R, the map v : Cp∞ → R satisfies
(1) v(λ1 f1 + λ2 f2 ) = λ1 v(f1 ) + λ2 v(f2 );
(2) v(f1 f2 ) = v(f1 )f2 (p) + f1 (p)v(f2 ).
The tangent space Tp M at p ∈ M is the vector space of tangent vectors at p ∈ M .
 DIFFERENTIAL TOPOLOGY LECTURES 7

Lemma 4.3. Tp M is naturally isomorphic to the linear dual of Tp∗ M (that is, all linear
functions on Tp∗ M ).
Proof. Using Leibniz rule, we know that for any f ∈ m2p and tangent vector v : Cp∞ → R,
we have v(f ) = 0. Therefore, a tangent vector induces a linear function mp /m2p → R.
Conversely, for any linear function ℓ : mp /m2p → R, we can define a tangent vector by
vℓ (f ) = ℓ(f − f (p)). It satisfies Leibniz rule because
vℓ (f1 f2 ) = ℓ(f1 f2 − f1 (p)f2 (p)) = ℓ((f1 − f1 (p))f2 (p) + f1 (f2 − f2 (p)))
= ℓ((f1 − f1 (p))f2 (p)) + ℓ(f1 (p)(f2 − f2 (p))) = vℓ (f1 )f2 (p) + f1 (p)vℓ (f2 ).
Here, the second last equality is because (f1 − f1 (p))(f2 − f2 (p)) ∈ m2p , which implies that
ℓ(f1 (p)(f2 − f2 (p))) = ℓ(f1 (f2 − f2 (p))).
This shows that there is a natural bijection. □
Definition 4.5. Let M be a smooth manifold and p be a point in M . For a local coor-
dinate system (U, x) around p, we define the dual basis of (dx1 )p , . . . , (dxd )p ∈ Tp∗ M to be
(∂/∂x1 )p , . . . , (∂/∂xd )p ∈ Tp M . That means
∂
∂xi

dxi = = δij .
∂xj p ∂xj p

5. Lecture 5: Pull Back and Push Forward

We can explain why we call the linear derivatives tangent vectors in a geometric way.
Intuitively, tangent vectors should be the direction vectors of smooth curves, in other words,
they are first order approximations of smooth curves. We have seen that one can define
first order term using equivalence relations. Here, we do the same thing.
The difficulty is to measure equality up to the first order. This means we need to consider
smooth functions and take the first derivatives. This leads to the following definition:
Definition 5.1. Let M be a smooth manifold and p be a point in M . Consider the set of
smooth curves defines on some neighbourhood of p, namely, the set of pairs (U, γ) where U
is an open neighbourhood of x and γ : (−1, 1) → U is a smooth map such that γ(0) = p.
Define the equivalence relation such that (U, γ) ∼ (V, δ) if there exists an open neighbourhood
W ⊂ U ∩ V of x such that for any smooth function f ∈ C ∞ (W ),
d  d 
(f ◦ γ(t)) = (f ◦ δ(t)) .
dt t=0 dt t=0
Define Tp′ M = {(U, γ) | U an open neighbourhood of x, γ ∈ C ∞ ((−1, 1), U ), γ(0) = p}.
It is not clear that this definition gives a vector space, but it turns out that it does give
the same tangent space as we have in the previous lecture.
Lemma 5.1. There is a natural bijection between Tp′ M and Tp M .
Now we explain how to compute derivatives and differentials of smooth functions on a
manifold.
Remark 5.1. (1) For a manifold M and a coordinate system (U, φ) around p, a tan-
gent/cotangent vector is of the form
d d
X ∂
X
v= vi , α= αi (dxi )p .
∂xi p
i=1 i=1
8 WENYUAN LI

(2) When M = Rd and x is the standard coordinate system, the differential of a smooth
function f ∈ C ∞ (Rd ) at p ∈ Rd is simply
d
X ∂f
dfp = (p)(dxi )p .
∂xi
i=1

More generally, for M with a coordinate system (U, x) around p, the differential of f ∈
C ∞ (M ) at p ∈ M is simply
d
X ∂(f ◦ x−1 )
dfp = (x(p))(dxi )p .
∂xi
i=1

(3) For two coordinate systems (U, x) and (V, y) around p, we have the coordinate change
formula that
d d
∂ X ∂xi ∂ X ∂yj
= , dyj = dxi .
∂yj ∂yj ∂xi ∂xi
i=1 i=1

Here the matrix (∂yj /∂xi )ni,j=1 is the Jacobian matrix of the smooth function y ◦ x−1 .
Given a smooth map, it induces maps between tangent spaces and cotangent spaces.
This is a generalization of change of coordinates in Euclidean spaces.
Definition 5.2. Let φ : M → N be a smooth map and p ∈ M . Then the pull back map of
φ is a linear map φ∗ : Tφ(p)
∗ N → T ∗ M defined by
p

φ∗ (df )φ(p) = d(f ◦ φ)p .

The push forward map or differential of φ is a linear map φ∗ : Tp M → Tφ(p) N (also denoted
by dφ) defined by
φ∗ v(f ) = v(f ◦ φ).
In particular, for φ : M → R, we have φ∗ v = dφ(v).
Remark 5.2. (1) For coordinate systems (U, x) and (V, y) around p and φ(p), we can write
φj = yj ◦ φ for the j-th component of φ and get
d
∂
∂
X ∂φj ∂
φ∗ = dφ = .
∂xi ∂xi ∂xi ∂yj
j=1

Here the matrix (∂φj /∂xi )i,j is the Jacobian matrix of the smooth map φ.
(2) For a smooth curve γ : (−1, 1) → M , considering the tangent vector d/dt on R with
respect to the standard coordinate system, we write
d
d

γ∗ = dγ = γ ′ (t).
dt dt
One can show that this is the tangent vector defined at the beginning of the lecture as
d
d
γ∗ (f ) = (f ◦ γ).
dt dt
(3) For two maps φ : M → N and ψ : N → P , we have chain rules that (ψ ◦φ)∗ = ψ∗ ◦φ∗
or d(ψ ◦ φ) = dψ ◦ dφ, and (ψ ◦ φ)∗ = φ∗ ◦ ψ ∗ . The latter follows from the definition of pull
back maps and compositions of functions.
 DIFFERENTIAL TOPOLOGY LECTURES 9

6. Lecture 6: Submanifolds and Inverse Function Theorem

We define submanifolds and give more examples of manifolds and submanifolds.
Definition 6.1. Let φ : M → N be a smooth map.
(1) φ is an immersion if dφp is injective for all p ∈ M .
(2) φ is a submersion if dφp is surjective for all p ∈ M .
(3) φ is an injective immersion if it is an immersion and injective.
(4) φ is an embedding if it is an immersion and homeomorphism onto the image.
(5) φ is a diffeomorphism if it is an embedding and surjective.
Example 6.1. A curve γ : R → M is an immersion if γ ′ (t) = γ∗ (d/dt) ̸= 0. When
γ(s) ̸= γ(t) whenever s ̸= t, the curve is an injective immersion. The curve γ : R → T 2 is
defined by γ(t) = (t, αt) for some α ∈/ Q is an injective immersion but not an embedding.
We recall the inverse function theorem in Euclidean spaces. Let U ⊂ Rd be an open
subset containing 0 and φ : U → Rd be a smooth map such that φ(0) = 0. Suppose at
0 ∈ U , the Jacobian
 ∂φ 
i
J(φ)0 =
∂xj 1≤i,j≤d
is invertible. Then the inverse function theorem states that there exists a smaller open
subset V ⊂ U an inverse function ψ : φ(V ) → V such that
φ ◦ ψ = id, ψ ◦ φ = id.
The outline of the proof is as follows:
First, using the contraction mapping principle, We show that the linear map J(φ)0 is
injective implies that φ is also injective in a neighbourhood of 0. Since J(φ)0 is injective,
we may assume that |J(φ)0 (p) − J(φ)0 (q)| ≥ C|p − q|. For ϵ(p) = φ(p) − J(φ)0 (p), there is
a small neighbourhood of 0 so that
C
|p − q|.
|ϵ(p) − ϵ(q)| ≤
2
Then we have the inequality which shows that φ is injective
C
|p − q| ≤|φ(p) − φ(q)|.
2
Using the contraction mapping principle, we can show that φ is surjective on a neigh-
bourhood of φ(0) because F : p 7→ q − ϵ(p) is a contraction mapping and therefore has a
fixed point. Then, we prove that the inverse ψ of φ is continuously differentiable using the
definition.
This theorem can be generalized to smooth manifolds.
Theorem 6.1 (Inverse Function Theorem). Let φ : M → N be a smooth map and
φ∗ : Tp M → Tφ(p) N is an isomorphism for some p ∈ M . Then there exists an open
neighbourhood U of p such that φ : U → φ(U ) is a diffeomorphism.
Proof. Consider local coordinate systems (U, x) and (V, y) around p and φ(p). Under the
basis dx1 , . . . , dxd and dy1 , . . . , dyd , the pull back map φ∗ : Tφ(p)
∗ N → T ∗ M is represented
p
by the matrix
 ∂φ 
i
.
∂xj 1≤i,j≤d
Since φ∗ is an isomorphism, we know that φ∗ is also an isomorphism and the matrix is
invertible. This is now reduced to the case of Euclidean spaces. □
10 WENYUAN LI

7. Lecture 7: Inverse Function Theorem and Coordinates

The inverse function theorem provides an effective way to construct local coordinate
systems.
Definition 7.1. A set y1 , . . . , yd of smooth functions defined on some neighborhood of p ∈ M
is called an independent set at p if dy1 , . . . , dyd are linear independent in Tp∗ M .
Corollary 7.1. Let M be a d-dimensional manifold and y1 , . . . , yd be independent functions
at p. Then y1 , . . . , yd forms a coordinate system in a neighbourhood U of p.
Proof. Consider the smooth map y : M → Rd . Apply the inverse function theorem, we
know that y is a diffeomorphism in a neighbourhood U of p. Consider any coordinate
system x : V → Rd . Then (y|U ∩V ) ◦ (x|U ∩V )−1 and (x|U ∩V ) ◦ (y|U ∩V )−1 are smooth maps.
This implies that (U, y) is also a coordinate system. □
Corollary 7.2. Let M be a d-dimensional manifold and x1 , . . . , xl be independent functions
at p. Then x1 , . . . , xl form part of a coordinate system in a neighbourhood U of p.
Proof. Consider a local coordinate system (V, y) around p. Then dx1 , . . . , dxl , dy1 , . . . , dyd
span Tp∗ M . Since dx1 , . . . , dxl are linear independent, we can choose a basis of Tp∗ M given
by dx1 , . . . , dxl , dyi1 , . . . , dyid−l . Then apply the previous corollary. □
Corollary 7.3. Let φ : M → N be a submersion at p and x1 , . . . , xd be a coordinate system
around φ(p). Then x1 ◦ φ, . . . , xd ◦ φ form part of a coordinate system in a neighbourhood
U of p.
Proof. Since the push forward φ∗ is surjective, we know that the pull back φ∗ is injective.
∗ N , we conclude that d(x ◦ φ), . . . , d(x ◦ φ) are
Since dx1 , . . . , dxd form a basis in Tφ(p) 1 d
linear independent in Tp∗ M . Then apply the previous corollary. □
Corollary 7.4. Let φ : M → N be an immersion at p and x1 , . . . , xd be a coordinate
system around φ(p). Then a subcollection of x1 ◦ φ, . . . , xd ◦ φ forms a coordinate system
in a neighbourhood U of p.
Finally, we show that all immersions have standard local forms. Namely, they are slices
in local coordinate systems:
{p ∈ U | xi (p) = ci , l + 1 ≤ i ≤ d}.
Proposition 7.5. Let φ : M → N be an immersion. Then for any p ∈ M , there exists
a coordinate system (V, x) centered at φ(p) and a neighbourhood U of p such that φ|U is
injective and φ(U ) is a slice in (V, x).
Proof. Consider a coordinate system (W, y) centered at φ(p). By the corollary, a subcollec-
tion of y1 ◦ φ, . . . , yd ◦ φ forms a coordinate system in a neighbourhood U of p. We denote
that by z1 , . . . , zl . Then on the open neighbourhood V = W ∩ y −1 (Rl × φ(U ))), we can
consider the projection π : Rd → Rl and define
xi = yi , 1 ≤ i ≤ l; xi = yi − yi ◦ φ ◦ z −1 ◦ π ◦ y, l + 1 ≤ i ≤ d.
Suppose the linear homomorphism dφ : Tp M ,→ Tφ(p) N is given by dyi = dzi for 1 ≤ i ≤ l
and dyi = lj=1 aij dzi for l + 1 ≤ i ≤ d. One can show that this defines a coordinate system
P
by computing the differentials:
l
X
dxi = dyi , 1 ≤ i ≤ l, dxi = dyi − aij dyj , l + 1 ≤ i ≤ d.
j=1

Moreover, for q ∈ φ(N ), we always have xi (q) = yi (q) − yi ◦ φ ◦ z −1 ◦ π ◦ y(q) = 0. This

completes the proof. □
 DIFFERENTIAL TOPOLOGY LECTURES 11

We remark that the above proposition is local in M but not local in N . Namely, we need
to fix an open set U in M and then find an open set V in N so that φ(U ) ⊂ V . It is not
true that we can directly find an open set V in N so that φ(M ) ∩ V is a slice, unless φ is
an embedding.

8. Lecture 8: Implicit Function Theorem

Many interesting subspaces in Euclidean spaces are cut out by certain (system of) equa-
tions. In this lecture, we provide a general criterion on when equations of smooth functions
cut out a submanifold. This will provide a large number of examples.
Definition 8.1. Let φ : M → N be a smooth map. Then q ∈ N is a regular value of φ if
for any p ∈ φ−1 (q), dφp is a submersion.
Theorem 8.1 (Implicit Function Theorem). Let φ : M → N be a smooth map and q is a
regular value of φ. Then φ−1 (q) has a (unique) manifold structure such that the inclusion
φ−1 (q) ,→ M defines an embedded submanifold whose dimension is dim M − dim N .
Proof. We prove that in the relative topology, there is a smooth structure on f −1 (q) that
makes f −1 (q) ⊂ M a submanifold of dimension dim M − dim N . For this it is sufficient
to prove that if p ∈ f −1 (q), then there exists a coordinate system on a neighborhood U of
p ∈ M of which f −1 (q) ∩ U is a slice of that dimension and the transition maps are smooth.
Let x1 , . . . , xd be a coordinate system centered at q ∈ N . Since dφp : Tp M → Tq N is
surjective, it follows from the corollary yesterday that y1 = x1 ◦ φ, . . . , yd = xd ◦ φ form part
of a coordinate system of p ∈ M . Complete the coordinate system to y1 , . . . , yd , yd+1 , . . . , yl .
Then φ−1 (q) ∩ U is give by the slice
y1 = 0, . . . , yd = 0,
and we can take yd+1 , . . . , yl to be a coordinate system on φ−1 (q) ∩ U . For smoothness of
the transition maps, we leave the details to the readers. □
Theorem 8.2 (Constant Rank Theorem). Let φ : M → N be a smooth map such that the
rank of dφp is constant for all p ∈ M . Show that for each p ∈ M and φ(p) ∈ N , there exists
local coordinate systems (U, x) around p and (V, y) around φ(p) such that
y ◦ φ ◦ x−1 (x1 , . . . , xk , xk+1 , . . . , xd ) = (x1 , . . . , xk , 0, . . . , 0).
Example 8.1. Consider the function r : Rd+1 → R, r(x1 , . . . , xd+1 ) = x21 + · · · + x2d+1 .
Then we know
(dr)(x1 ,...,xd+1 ) = 2x1 dx1 + · · · + 2xd+1 dxd+1 .
Therefore, we know that dr ̸= 0 whenever (x1 , . . . , xd+1 ) ̸= 0. Hence r−1 (c) is a d-
dimensional submanifold. The tangent space at (x1 , . . . , xd+1 ) is
D d+1
X E
d+1
(v1 , . . . , vd+1 ) ∈ R | vi xi = 0 .
i=1

Example 8.2. Consider the smooth map F : Md×d (R) → Md×d (R), F (A) = AAT − Id . We
show that F −1 (0) is a smooth submanifold of dimension d(d − 1)/2. This is usually called
the orthogonal group O(d).
Write down all the components of the map F ((aij )1≤i,j≤d ) = ( dk=1 aik ajk − δij )1≤i,j≤d .
P
First, note that the image of F lands in the space of all symmetric matrices Sd×d (R). Then
we can compute the differential
d
X d
X 
(dF )(aij )1≤i,j≤d = aik dajk + ajk daik .
1≤i,j≤d
k=1 k=1
12 WENYUAN LI

Consider the differential at A = Id . It is

d
X d
X 
(dF )(δij )1≤i,j≤d = δik dajk + δjk daik = (daji + daij )1≤i,j≤d .
1≤i,j≤d
k=1 k=1
It follows that (dF )Id is a surjection from TId Md×d (R) = Md×d (R) to TO Sd×d (R) = Sd×d (R),
and the tangent space at Id is the set of all skew symmetric matrices
D E
(vij )1≤i,j≤d | (vji + vij )1≤i,j≤d = O .

Finally, for any other P ∈ F −1 (O), consider the smooth map

φP : Md×d (R) → Md×d (R), A 7→ P A.
It has a smooth inverse φP −1 and moreover φP (F −1 (O)) = F −1 (O). This implies that
(dF )P is also surjective at any P ∈ F −1 (O). Therefore, O is a regular value of F and this
finishes the proof.

9. Lecture 9: Tangent and Cotangent Bundles

Rather than describing the tangent and cotangent vectors at a single point on the mani-
fold, we would often like to describe how the tangent and cotangent vectors smoothly change
when the point moves around in the manifold. This requires us to put some topology on
the space of all (co)tangent vectors.
Definition 9.1. Let M be a smooth manifold. Then the tangent bundle T M and cotangent
bundle T ∗ M of M are defined as
[ [
TM = Tp M, T ∗ M = Tp∗ M.
p∈M p∈M

The topology on T M is generated by the sets {(p, v) ∈ T M | p ∈ U, dx(v) ∈ Ω}, where (U, x)
is a coordinate system on M and Ω is an open subset in Rd . The differentiable structure on
T M is generated by the coordinate systems (U × Rd , x × dx), where (U, x) is a coordinate
system on M and Ω is an open subset in Rd .
It may seem like the tangent bundle is just the space M × Rd . However, this is rarely the
case. In fact, the local coordinate maps secretly encode the twistings of the tangent spaces.
For example, consider the coordinate systems on S 2 given by S 2 \ {n} and S 2 \ {s}. Then
on the overlap, one can see that the tangent spaces are identified in a nontrivial way. In
fact, we can give the following alternative description:
F
Proposition 9.1. Let M be a smooth manifold. Consider the topological space α∈A Uα ×
Rd . Define the equivalence relation such that
(p, vα ) ∼ (p, vβ ) if p ∈ Uα ∩ Uβ , dxα (v) = vα , dxβ (v) = vβ for some v ∈ Tp M.
Then α∈A Uα × Rd /{(p, vα ) ∼ (p, vβ ), vα = dxα ◦ dx−1
F ∼
β (vβ )} = T M .

Example 9.1. In the situation when M = Rd , the proposition shows that we have
T Rd = Rd × Rd .
For a smooth map φ : M → N , the push forward or the differential
dφ : T M → T N, (p, vp ) 7→ (φ(p), dφp (vp ))
is a smooth map. Moreover, if φ : M → N is an embedding, we can show that dφ : T M →
T N is also an embedding. However, the pull back map does NOT define a map between
cotangent bundles. We will see how to resolve the issue on Friday.
 DIFFERENTIAL TOPOLOGY LECTURES 13

Example 9.2. Let S 2 = {(x1 , x2 , x3 ) | x21 + x22 + x23 = 1} ⊂ R3 . We have seen how
to compute the tangent space of S 2 . Namely, setting r(x1 , x2 , x3 ) = x21 + x22 + x23 , at
(x1 , x2 , x3 ) ∈ S 2 , we have
T(x1 ,x2 ,x3 ) S 2 = {v ∈ R3 | dr(v) = 0} = {(v1 , v2 , v3 ) | x1 v1 + x2 v2 + x3 v3 = 0}.
Then we know that T S 2 ⊂ T R3 ∼
= R3 × R3 is defined as the following subset:
T S 2 = {(x1 , x2 , x3 ; v1 , v2 , v3 ) | x21 + x22 + x23 = 1, x1 v1 + x2 v2 + x3 v3 = 0}.
We have seen that the tangent bundle and cotangent bundle are smooth manifolds.
However, we also note that the transition maps dxα ◦ dx−1 β are not only smooth maps,
d
they are linear maps along the R -directions. In the next lecture, we will explain the extra
structure that can be put on a special type of manifolds.

10. Lecture 10: Vector Bundles

In this lecture, we introduce the notion of vector bundles, which are not only manifolds
but have some linear vector space directions.
Definition 10.1. Let M be a topological space. Then a rank-k vector bundle over M is a
triple (π, E, M ) where E is a topological space and π : E → M is a continuous map such
that
(1) for each p ∈ M , the fiber Ep = π −1 (p) is a k-dimensional vector space;
(2) for each p ∈ M , there is a neighbourhood U and a homeomorphism Φ : π −1 (U ) →
U × Rk such that πU ◦ Φ = π, and Φq : π −1 (q) → Rk is a linear isomorphism.
The pair (U, Φ) is called a vector bundle coordinate system or chart.
Definition 10.2. Let M be a smooth manifold and (π, E, M ) be a rank-k vector bundle.
Then a C k -differentiable vector bundle structure/atlas is a collection F of vector bundle
charts (Uα , Φα ) such that
S
(1) α∈A Uα = M ;
(2) Φα ◦ Φ−1 k
β is a C -differentiable map for any α, β ∈ A;
(3) F is maximal so that, for any vector bundle chart (U, Φ), if Φ ◦ Φ−1
α and Φα ◦ Φ
−1

are C -differentiable for any α ∈ A, then (U, Φ) ∈ F .

k

In this case, (π, E, M ) is called a smooth vector bundle of rank-k.

Example 10.1. (1) The natural projection π : M × Rk → M defines a vector bundle on M .
This is usually called the trivial vector bundle. (2) The first nontrivial example of a vector
bundle is the Mobiüs bundle, which is a rank-1 vector bundle over the circle π : E → S 1 .
Let us discuss the vector bundle structure more carefully. The condition that πUα ◦ Φα =
π = πUβ ◦ Φβ implies that the map Φα ◦ Φ−1β can be written as the form

Φα ◦ Φ−1
β (p, v) = (p, gαβ (p)v),

where gij (p) ∈ GL(k, R) is a linear isomorphism. They are called the transition maps.
Lemma 10.1. Let (π, E, M ) be a rank-k (smooth) vector bundle on M and (Uα , Φα ),
(Uβ , Φβ ) are local vector bundle coordinate charts. Then
Φα ◦ Φ−1
β (p, v) = (p, gαβ (p)v),

and gαβ : Uα ∩ Uβ → GL(k, R) is a (smooth) map.

Using transition maps, we can reconstruct the vector bundle.
14 WENYUAN LI

Lemma 10.2. Let M be a manifold and {Uα | α ∈ A} be an open cover. Let {gαβ | α, β ∈
A} be a collection of smooth maps Uα ∩ Uβ → GL(k, R) such that
gαβ (p)gβγ (p) = gαγ (p), for any p ∈ Uα ∩ Uβ ∩ Uγ .
Then there exists a unique rank-k vector bundle (π, E, M ) such that {gαβ | α, β ∈ A} are
the transition maps.
Proof. We can define the vector bundle by considering
G
E= Uα × Rk /{(p, vα ) ∼ (p, vβ ) if p ∈ Uα ∩ Uβ , vα = gαβ (vβ )}.
α∈A

One can check that π : E → M, (p, v) → p is a well defined (smooth) map and the collection
id : Uα × Rk → Uα × Rk defines a vector bundle structure. □

For vector bundles on a given topological space, there are many constructions that can
be performed. It is an exercise to prove that these are vector bundles.
Definition 10.3. Let (π1 , E1 , M ) and (π2 , E2 , M ) be vector bundles over M . Then a vector
bundle map is a map φ : E1 → E2 such that π1 = π2 ◦φ and φp : E1p → E2p are linear maps
between vector spaces. It is called a monomorphism if φp are injective, an epimorphism if
φp are surjective, and an isomorphism if φp are isomorphic.
Let (π, E, M ) be a vector bundle. Then a subbundle is a vector bundle (π0 , E0 , M ) where
E0 ⊆ E and π0 = π|E0 . In other words suppose the transition maps of (π, E, M ) is {gαβ |
α, β ∈ A}, then the transition maps are
{g0αβ | α, β ∈ A, i ◦ g0αβ = gαβ ◦ i}.
Then the quotient bundle (π, E/E0 , M ) is defined by fibers (E/E0 )p = Ep /E0p and transition
maps
{g αβ | α, β ∈ A}
where g αβ : Rk /Rl → Rk /Rl is the quotient of gαβ : Rk → Rk .
Let (π1 , E1 , M ) and (π2 , E2 , M ) be vector bundles over M . Then the direct sum (or
Whitney sum) is the vector bundle (π1 ⊕ π2 , E1 ⊕ E2 , M ) with fibers (E1 ⊕ E2 )p ∼
= E1p ⊕ E2p
and transition maps
{g1αβ ⊕ g2αβ | α, β ∈ A}.
The internal hom is the vector bundle (πHom , Hom(E1 , E2 ), M ) with fibers Hom(E1 , E2 )p ∼
=
Hom(E1p , E2p ) and transition maps
{g2αβ (−)g1αβ | α, β ∈ A}.
In particular, the dual of (π, E, M ) is the vector bundle (π ∗ , E ∗ , M ) with fibers Ep∗ ∼
= (Ep )∗
∗
and transition maps {gβα | α, β ∈ A}.
Using partitions of unity, many properties on vector spaces can be translated to vector
bundles using the local-to-global argument.
Theorem 10.3. Let M be a manifold and (π, E, M ) is a vector bundle and (π0 , E0 , M ) is
a subbundle. Then there is an isomorphism of vector bundles E ∼
= E0 ⊕ E/E0 .
Proof. Consider a vector space chart (Uα , Φα ) of (π, E, M ) and a vector bundle chart
(Uα , Ψα ) of (π0 , E0 , M ). Then we have homeomorphisms Φα : π −1 (Uα ) → Uα × Rk and
Ψα : π0−1 (Uα ) → Uα × Rl . Their composition have the form
Φα ◦ Ψ−1 l k
α : Uα × R → Uα × R , (p, v0 ) 7→ (p, gα (v0 ))
 DIFFERENTIAL TOPOLOGY LECTURES 15

where gα (p) is an injective linear map Rl → Rk . Using implicit function theorem, we can
define a projection map hα (p) : Rk → Rl such that hα (p)gα (p) = id, which gives a projection
map fα (p) : Ep → E0p . Then we use a partition of unity on M and define
 X 
f : E → E0 , f (p, v) = p, φα (p)fα (p)(v) .
α∈A
∼
This together with the natural quotient map defines the isomorphism E −
→ E0 ⊕ E/E0 . □
Remark 10.2. The continuous version of the result also holds for all paracompact Hausdorff
spaces.
Given two topological spaces M and N and a map φ : M → N , from a vector bundle on
N we can naturally define a vector bundle on M :
Definition 10.4. Let M and N be topological spaces or manifolds and φ : M → N a
continuous or smooth map. Given a vector bundle (π, E, N ) on N , the pull-back vector
bundle on M is defined as (πφ , φ∗ E, M ) where
φ∗ E = {(p, q, v) | φ(p) = π(q, v) = q}, πφ (p, q, v) = p.
Using partition of unity, we can prove the following strong result, that a homotopy of
vector bundle gives an isomorphism of vector bundle.
Theorem 10.4. Let M be a manifold, I be an interval and (πI , EI , M × I) be a vector
bundle on M × I. Consider the vector bundle (πt , Et , M ) where Et = π −1 (M × {t}) and
πt = π|Et . Then (πI , EI , M × I) is isomorphic to the pull-back bundle of (π0 , E0 , M ). In
particular, (π0 , E1 , M ) is isomorphic to (π1 , E1 , M ).
Remark 10.3. The continuous version of the result also holds for all paracompact Hausdorff
spaces.
Corollary 10.5. Let M be a manifold, f, g : M → N be smoothly homotopic maps, and
(π, E, N ) a vector bundle over N . Then (f ∗ π, f ∗ E, M ) is isomorphic to (g ∗ π, g ∗ E, N ).
Finally, we can resolve the issue that the pull back map of cotangent vectors do not define
a map of cotangent bundles. Instead, the pull back map defines a vector bundle map
f ∗ : f ∗ T ∗ N → T ∗ M, (p, φ(p), αφ(p) ) 7→ (p, φ∗ αp ).

11. Lecture 11: Sections on Vector Bundles

The motivation we have at the beginning is to characterize what it means to have a
smooth family of tangent vectors, and more generally, what it means to have a smooth
family of vectors in a vector bundle.
Definition 11.1. Let (π, E, M ) be a (smooth) fiber bundle over M . A section is a (smooth)
map s : M → E such that π ◦ s = idM . The zero section is the section s : M → E such that
s(p) = 0 ∈ Ep for any p ∈ M . The space of sections is denoted by C ∞ (M, E).
Lemma 11.1. Let (π, E, M ) be a (smooth) fiber bundle over M . Then s : M → E is a
(smooth) section if and only if for any local coordinate system (Uα , xα , Φα ),
Φα ◦ s(p) = (p, sα (p))
such that sα : Uα → Rk is a (smooth) map.
Proposition 11.2. Let φ : M → N be a smooth map and (π, E, N ) be a smooth vector
bundle over N . Then there is a map between the space of smooth sections C ∞ (N, E) →
C ∞ (M, φ∗ E).
16 WENYUAN LI

Using partition of unity, we can construct sections on any vector bundles from a given
vector at a point, just like we construct smooth functions.
Proposition 11.3. Let (π, E, M ) be a smooth fiber bundle over M . Then for any vp ∈ Ep ,
there exists a section s ∈ C ∞ (M, E) such that s(p) = (p, vp ).

12. Lecture 12: Vector Fields

In the last week, we started with the question of how to characterize a smooth family of
(co)tangent vectors on a manifold, and ended by introducing vector bundles and sections,
which assign each point on a manifold with a vector.
Now, we can define a vector field as a section in the tangent bundle.
Definition 12.1. Let M be a smooth manifold. Then a smooth (resp. continuous) vector
field X on a subset U ⊂ M is a smooth (resp. continuous) section in the tangent bundle
X ∈ C ∞ (U, T M |U ) (resp. X ∈ C 0 (U, T M |U )).
Proposition 12.1. Let M be a smooth manifold and X be a vector field. Then X is smooth
if and only if either of the conditions hold:
(1) for any local coordinate system (U, x),
d
X ∂
X|U (p) = ai (p) ,
∂xi
i=1

and ai (p) ∈ C ∞ (U );
(2) for any open set U ⊂ M and smooth function f ∈ C ∞ (U ), Xf = df (X) ∈ C ∞ (U ).
For two vector fields, we can define the Lie bracket between them:
Definition 12.2. Let M be a smooth manifold and X and Y be smooth vector fields. Then
the Lie bracket [X, Y ] is a section in (T ∗ M )∗ defined by
[X, Y ]p (f ) = Xp (Y f ) − Yp (Xf ).
Proposition 12.2. Let M be a smooth manifold and X, Y and Z be smooth vector fields.
(1) [X, Y ] is a smooth vector field;
(2) (anti-symmetry) [X, Y ] = −[Y, X];
(3) (Jacobi identity) [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0;
(4) for f, g ∈ C ∞ (M ), [f X, gY ] = f g[X, Y ] + f (Xg)Y − g(Y f )X.
Remark 12.1. On a local coordinate system (U, x), suppose the vector fields are
d d
X ∂ X ∂
X|U (p) = ai (p) , Y |U (p) = bi (p) ,
∂xi ∂xi
i=1 i=1

then their Lie bracket is given by the vector field

d 
X ∂bj ∂aj  ∂
[X, Y ]|U (p) = ai (p) (p) − bi (p) (p) .
∂xi ∂xi ∂xj
i,j=1

Finally, we remark that push forwards of vector fields under smooth maps are not nec-
essarily well defined, since a smooth map M → N may send different points on M to the
same point on N .
 DIFFERENTIAL TOPOLOGY LECTURES 17

13. Lecture 13: Intergral Curves and Flows

Given a vector field on a manifold, a point can move along the direction of the vector
field and trace will form a curve. This is called the integral curve of the vector field.
Definition 13.1. Let X be a smooth vector field on a manifold M . Then a smooth curve
γ : (a, b) → M is called an integral curve of X if for any t ∈ (a, b),
d
γ ′ (t) = dγ = X(γ(t)).
dt t
Consider a local coordinate system (U, x), we can write down the differential equation
for the integral curve of the vector field X = di=1 ai (p)∂/∂xi :
P

dγi
(t) = ai (γ(t)).
dt
Since the coefficients ai (p) are smooth, the ordinary differential equation has a unique
solution. In fact, let U ⊂ Rn be open subsets, a : U → Rn be Lipschitz continuous. Then
for any p ∈ U , there exists ϵ > 0 such that
dγi
(t) = ai (γ(t)), γ(0) = p,
dt
has a unique solution. This can be proved by considering the fixed point of the following
mapping on Banach spaces T : C 0 ((−ϵ, ϵ), Rn ) → C 0 ((−ϵ, ϵ), Rn ):
Z t
(T γ)(t) = p + X(γ(s))ds.
0
This is a contraction mapping for sufficiently small ϵ > 0 because
Z t
∥T γ − T σ∥ = sup |X(γ(s)) − X(σ(s))|ds ≤ Cϵ∥γ − σ∥.
t∈(−ϵ,ϵ) 0

Then the contraction mapping principle ensures that the existence of a solution. The
uniqueness of the solution follows from the computation that
|γ ′ (t) − σ ′ (t)| = |X(γ(t)) − X(σ(t))| ≤ C|γ(t) − σ(t)|, |γ(t) − σ(t)| ≤ eCt |γ(0) − σ(0)|.
Therefore, we have the following global result on integral curves:
Theorem 13.1. Let M be a smooth manifold and X be a smooth vector field on M . Then
for any p ∈ M , there exists −∞ ≤ a(p) < b(p) ≤ +∞ and a smooth maximal integral curve
γp : (a(p), b(p)) → M
of the vector field X such that for any integral curve σ : (c, d) → M with σ(0) = p and
γp′ (t) = X(σ(t)), (c, d) ⊂ (a(p), b(p)) and σ = γ|(c,d) .
Moreover, since the coefficients ai (p) are smooth, the unique solution to the ordinary
differential equation depends smoothly on the initial condition p ∈ M (the proof is much
harder and we will omit that). Therefore, we have the following global result on integral
flows:
Theorem 13.2. Let M be a smooth manifold and X be a smooth vector field on M . For
t ∈ R, we can define a smooth map φX
t called the integral flow on the domain
Dt = {p ∈ M | a(p) < t, b(p) > t}
such that φX
t (p) = γp (t) and
(1) for any p ∈ M , there is a neighbourhood U and ϵ > 0 such that U ⊂ D−ϵ ∩ Dϵ ;
(2) φX t : Dt → D−t and φ−t : D−t → Dt are inverse diffeomorphisms;
X
18 WENYUAN LI

(3) φX X X X X
s ◦ φt = φs+t on the domain of φs ◦ φt .

Corollary 13.3. Let M be a compact smooth manifold (without boundary) and X be a

smooth vector field on M . For t ∈ R, we can define a smooth diffeomorphism
φX
t : M → M, φX
t (p) = γp (t)

such that φX X X
s ◦ φt = φs+t .

Remark 13.1. In general, a vector field X on M is complete if Dt = M for any t ∈ M . The

corollary shows that all vector fields on compact manifolds (without boundary) are complete.
As a corollary, one can show that a non-zero vector field always generate a local coordinate
function.
Proposition 13.4. Let M be a smooth manifold and X be a vector field such that X(p) ̸= 0.
Then there exists a local coordinate system (U, x) around p such that x(p) = 0 and
∂
X|U = .
∂x1
Proof. First, by corollaries of the inverse function theorem, we can find a local coordinate
system (V, y) around p such that Xp = (∂/∂y1 )p . Now consider the smooth map
σ(t, a2 , . . . , ad ) = y ◦ φt ◦ y −1 (0, a2 , . . . , ad ).
Then dσ0 ̸= 0 and by inverse function theorem, there is a small neighbourhood U of p such
that σ is a diffeomorphism. Then x = σ −1 satisfies the condition. □

14. Lecture 14: Distributions and Frobenius Theorem

We have seen that a vector field generates integral curves. A natural generalization one
can consider is to use vector subspaces to generate submanifolds.
Definition 14.1. Let M be a d-dimensional smooth manifold. Then an l-dimensional
distribution D is a choice of l-dimensional tangent subspaces Dp ⊂ Tp M for all p ∈ M . D
is smooth if for each p ∈ M there is a neighborhood U and l smooth vector fields X1 , . . . , Xl
that span Dq for each q ∈ U .
Definition 14.2. We say that a vector field X belongs to (or lies in) the distribution D
(denoted by X ∈ D) if Xp ∈ Dp for all p ∈ M . A distribution D on M is incolutive if for
any X, Y ∈ D, [X, Y ] ∈ D.
We say that a (injectively immersed) submanifold φ : N ,→ M is an integral manifold of
a distribution D if dφ(Tq N ) = Dφ(q) for all q ∈ N .
Example 14.1. Consider T 2 = R2 /Z2 and consider the vector field X = ∂/∂x1 + α∂/∂x2
for α ∈
/ Q, which defines a 1-dimensional distribution. Then the curves {(x1 , x2 ) | x1 ≡
αx2 + c} are integral manifolds of the distribution. They are only injectively immersed.
The following proposition explains a necessary condition for a distribution to have an
integral manifold and justifies the definition of involutive distributions.
Proposition 14.1. Let D be a smooth distribution on M such that for each point of M
there is an integral manifold D passing through. Then D is involutive.
Proof. Consider an integral manifold φ : N ,→ M of D with φ(q) = p. Consider two vector
∼
fields X, Y ∈ D, Since we have an isomorphism dφ : Tq N − → Dφ(q) , there exists X ′ , Y ′ on
N defined by
dφ ◦ X ′ = X ◦ φ, dφ ◦ Y ′ = Y ◦ φ : N → T M.
 DIFFERENTIAL TOPOLOGY LECTURES 19

One can check that X ′ , Y ′ are smooth vector fields and dφ ◦ [X ′ , Y ′ ] = [X, Y ] ◦ φ, as
(dφ ◦ [X ′ , Y ′ ])(f ) = [X ′ , Y ′ ](f ◦ φ) = X ′ Y ′ (f ◦ φ) − Y ′ X ′ (f ◦ φ)
= X ′ (dφ(Y ′ )(f )) − Y ′ (dφ(X ′ )(f )) = X ′ (Y (f ) ◦ φ) − Y ′ (X(f ) ◦ φ)
= dφ(X ′ )(Y (f )) − dφ(Y ′ )(X(f )) = XY (f ) − Y X(f ) = [X, Y ](f ),
which completes the proof. □
Theorem 14.2 (Frobenius Theorem). Let D be an l-dimensional smooth involutive distri-
bution on M . Then for any p ∈ M , there exists a coordinate system (U, x) around p such
that the slices
xi = ci , l+1≤i≤d
are integral manifolds of D, and if φ : N ,→ U is a connected integral manifold, then φ(N )
lies in one of the slices.
Proof. We prove by induction on the dimension. Suppose X1 , . . . , Xl span D on a neigh-
bourhood W in M . Consider a coordinate system (V, y) around p such that
X1 = ∂/∂y1 .
Then we try to consider the distribution D ′ spanned by
X2′ = X2 − X2 (y1 )X1 , . . . , Xl′ = Xl − Xl (y1 )X1 .
We claim that this is still involutive. Note that Xi′ (y1 ) = Xi (y1 ) − Xi (y1 )X1 (y1 ) = 0. This
means that X2′ , . . . , Xl′ lie in the slices y1 = c. Since D is involutive, we can write
l
X
[Xi′ , Xj′ ] = cijk Xk .
k=1

However, since X2′ , . . . , Xl′ lie in the slices y1 = c, so is [Xi′ , Xj′ ]. Therefore, we can write
l
X l
X
[Xi′ , Xj′ ] = cijk Xk = cijk Xk′ .
k=2 k=2

This shows that D′

is still involutive. By the induction hypothesis, we can find a coordinate
system (U, x) such that x1 = y1 , and the slices
xi = ci , l+1≤i≤d
are integral manifolds of D ′ . Finally, we notice that the equality Xi′ (xj ) = 0 for l+1 ≤ j ≤ d
will also imply that Xi (xj ) = 0 for l + 1 ≤ j ≤ d. This will complete the proof. □

Moreover, similar to the situation of integral curves, we can find a maximal integral
submanifold for an involutive distribution.
Definition 14.3. A maximal integral manifold φ : N ,→ M of a distribution D on a
manifold M is a connected integral manifold of D such that for any connected integral
manifold φ′ : N ′ ,→ M of D, if φ(N ) ⊂ φ′ (N ′ ), then φ(N ) = φ′ (N ′ ).
Theorem 14.3. Let D be a smooth involutive distribution on M . Then for any p ∈ M
there exists a unique maximal connected integral manifold of D passing through, and every
connected integral manifold of D through p is contained in the maximal one.
20 WENYUAN LI

15. Lecture 15: Tensor Algebra

We recall some basic notions in multilinear algebra. First, we recall the tensor product.
Definition 15.1. Let V and W be vector spaces over a field. Consider the free vector space
generated by elements (v, w) ∈ V × W
k
nX o
F (V, W ) = λi (vi , wi ) | (vi , wi ) ∈ V × W .
i=1

Let R(V, W ) ⊂ F (V, W ) be the subspace generated by the collections

(v1 + v2 , w) − (v1 , w) − (v2 , w),
(v, w1 + w2 ) − (v, w1 ) − (v, w2 ),
(av, w) − a(v, w), (v, aw) − a(v, w).
Then we define the tensor product to be V ⊗ W = F (V, W )/R(V, W ), and denote the
equivalence class of (v, w) by v ⊗ w.
Here are some basic properties of the tensor product. We will omit the proofs.
Proposition 15.1 (Universal property). Let V and W be vector spaces over a field. Let
φ denote the bilinear map V × W → V ⊗ W, (v, w) 7→ v ⊗ w. Then whenever U is a vector
space and l : V × W → U is a bilinear map, there exists a unique linear map ˜l : V ⊗ W → U
such that the following diagram commutes:
V ⊗O W
l̃
φ

l &/
V ×W U.
Proposition 15.2. Let V and W be vector spaces over a field.
(1) V ⊗ W is naturally isomorphic to W ⊗ V ;
(2) (V ⊗ W ) ⊗ U is naturally isomorphic to V ⊗ (W ⊗ U );
(3) dim V ⊗ W = (dim V )(dim W ), and a basis {v1 , . . . , vm } and {w1 , . . . , wn } for V
and W determines a basis {v1 ⊗ w1 , . . . , vm ⊗ wn } of V ⊗ W .
Then we define the tensor algebra, consisting of tensors of arbitrary lengths.
Definition 15.2. Let V be a vector space over a field. Then the tensor space of type (r, s)
is defined by Vr,s = V ⊗r ⊗(V ∗ )⊗s . The two-sided tensor algebra of V is the noncommutative
algebra
M
T± (V ) = Vr,s ,
r,s≥0
where the product is given by the tensor product (v, w) 7→ v ⊗ w. We define the (one-sided)
tensor algebra to be the noncommutative algebra
M
T (V ) = Vr,0 .
r≥0

We write T k (V ) = Vk,0 = V ⊗k .
Consider the natural pairing between a vector space V and its dual space V ∗ together
with the universal property, we also get a natural pairing between T ∗ (V ) and T ∗ (V ∗ ) as
follows:
 DIFFERENTIAL TOPOLOGY LECTURES 21

Lemma 15.3. Let V be a vector space over a field. Then there exists a natural non-
degenerate pairing between Vr,s and (V ∗ )s,r given by
(v1 ⊗ · · · ⊗ vr ⊗ v1∗ ⊗ · · · ⊗ vs∗ , u1 ⊗ · · · ⊗ us ⊗ u∗1 ⊗ · · · ⊗∗r ), u∗1 (v1 ) . . . u∗r (vr ) · v1∗ (u1 ) . . . vs∗ (us ).
In particular, when dim V < ∞, (Vr,s )∗ ≃ (V ∗ )s,r .
We can define the tensor bundles of a manifold and tensor products of vector bundles
over a manifold as follows:
Definition 15.3. Let (π1 , E1 , M ) and (π2 , E2 , M ) be vector bundles over a space M . Then
the tensor product is the vector bundle (π1 ⊗π2 , E1 ⊗E2 , M ) with fibers (E1 ⊗E2 )p ∼
= E1p ⊗E2p
and transition maps
{g1αβ ⊗ g2αβ | α, β ∈ A}.
Definition 15.4. Let M be a smooth manifold. Then the tensor bundle Tr,s M of M are
defined as [
Tr,s M = (Tp M )r,s .
p∈M
The topology on Tr,s M is generated by the sets {(p, v1 ⊗ vr ⊗ v1∗ ⊗ . . . , vs∗ ) ∈ T M | p ∈
U, x∗ (vi ) ∈ Ωi , (x−1 )∗ (vj∗ ) ∈ Ω∗j }, where (U, x) is a coordinate system on M and Ωi and Ω∗j
are open subsets in Rd . The differentiable structure on T M is generated by the coordinate
systems (U × Rd , x × x∗ ⊗ · · · ⊗ x∗ ⊗ (x−1 )∗ ⊗ · · · ⊗ (x−1 )∗ ), where (U, x) is a coordinate
system on M and Ω is an open subset in Rd . A tensor field on M of type (r, s) is a section
of the tensor bundle Tr,s M .
Remark 15.1. Under a coordinate system (U, x) on M , a tensor field of type (r, s) is of
the form
X ∂ ∂
α(p) = ai1 ...ir j1 ...js (x(p)) ⊗ ··· ⊗ ⊗ dxj1 ⊗ · · · ⊗ dxjs .
∂xi1 ∂xi1
i1 ,...,ir ,j1 ,...,js

16. Lecture 16: Exterior Algebra

In the tensor algebra of a vector space, we are particularly interested in symmetric
algebras and exterior algebras. We introduce their definitions now.
Definition 16.1. Consider the tensor algebra of a vector space T (V ). Define the subspace
J(V ) ⊂ T (V ) be the vector space generated by
v1 ⊗ · · · ⊗ (vi ⊗ vi+1 − vi+1 ⊗ vi ) ⊗ · · · ⊗ vr .
Then the symmetric algebra is S(V ) = Sym(V ) = T (V )/J(V ), where the equivalence class
of v1 ⊗ · · · ⊗ vr is denoted by v1 · · · vr , and the product is induced by the tensor product. We
write S k (V ) = Symk (V ) = T k (V )/J k (V ) where J k (V ) = J(V ) ∩ T k (V ).
Define the subspace I(V ) ⊂ T (V ) be the vector space generated by
v1 ⊗ · · · ⊗ (vi ⊗ vi ) ⊗ · · · ⊗ vr .
Then the exterior algebra is Λ(V ) = T (V )/I(V ), where the equivalence class of v1 ⊗ · · · ⊗ vr
is denoted by v1 ∧ · · · ∧ vr , and the product is induced by the tensor product. We write
Λk (V ) = T k (V )/I k (V ) where I k (V ) = I(V ) ∩ T k (V ).
Definition 16.2. A multilinear map h : V ⊗ · · · ⊗ V → R is called symmetric if for any
σ ∈ Sn ,
h(vσ(1) , . . . , vσ(n) ) = h(v1 , . . . , vn ).
It is called alternating if for any σ ∈ Sn ,
h(vσ(1) , . . . , vσ(n) ) = sgn(σ)h(v1 , . . . , vn ).
22 WENYUAN LI

The symmetric algebra and the exterior algebra also satisfies the universal properties:
Proposition 16.1 (Universal Property). Let V be a vector space over a field.
(1) Let φs denote the bilinear map V × · · · × V → S k (V ), (v1 , v2 , . . . , vk ) 7→ v1 v2 . . . vk .
Then whenever U is a vector space and l : V ×· · ·×V → U is a symmetric multilinear
map, there exists a unique linear map ˜l : S k (V ) → U such that the following diagram
commutes:
S k (V )
O
l̃
φs

l '/
V × ··· × V U.
(2) Let φa denote the bilinear map V ×· · ·×V → Λk (V
), (v1 , v2 , . . . , vk ) 7→ v1 ∧v2 ∧· · ·∧
vk . Then whenever U is a vector space and l : V × · · · × V → U is an alternating
multilinear map, there exists a unique linear map ˜l : Λk (V ) → U such that the
following diagram commutes:
Λk (V )
O
l̃
φa

l '/
V × ··· × V U.
Proposition 16.2. Let V be a vector space over a field.
(1) If u ∈ Λk (V ) and v ∈ Λl (V ), then u ∧ v = (−1)kl v ∧ u;
n
(2) if dim V = n, then dim Λk (V ) = k , and a basis {v1 , . . . , vn } of V determines a

basis {vi1 ∧ vi2 ∧ · · · ∧ vik | i1 < i2 < · · · < ik } of Λk (V );

Pk n i+k
(3) if dim V = n, then dim S k (V ) =



i=0 i i , and a basis {v1 , . . . , vn } of V
determines a basis {vi1 vi2 . . . vik | i1 ≤ i2 ≤ · · · ≤ ik } of S k (V ) = Symk (V );
Using the natural pairing between a vector space V and its dual space V ∗ together with
the universal property, we also get a natural pairing between Λ∗ (V ) and Λ∗ (V ∗ ) as follows:
Lemma 16.3. Let V be a vector space over a field. Then there exists a natural non-
degenerate pairing between Λk (V ) and Λk (V ∗ ) given by
X
(v1 ∧ · · · ∧ vk , u∗1 ∧ · · · ∧ u∗k ), sgn(σ)u∗1 (vσ1 ) . . . u∗k (vσk ) = det(u∗i (vj )).
σ∈Sk

In particular, when dim V < ∞, Λk (V )∗ ≃ Λk (V ∗ ).

Definition 16.3. Let V be a vector space over a field. Then for any v ∈ V , the interior
multiplication by v is defined by
i(v) : Λ∗ (V ∗ ) → Λ∗ (V ∗ ), (i(v)u∗ , w) = (u∗ , v ∧ w).
Lemma 16.4. Let V be a vector space over a field. Then i(v) : Λ∗ (V ∗ ) → Λ∗ (V ∗ ) is an
antiderivation, that is,
i(v)(u∗1 ∧ u∗2 ) = i(v)u∗1 ∧ u∗2 + (−1)k u∗1 ∧ i(v)u∗2 , u∗1 ∈ Λk (V ∗ ), u∗2 ∈ Λl (V ∗ ).
We can define the symmetric and exterior tensor bundles of a manifold and symmetric
and exterior products of vector bundles over a manifold as follows:
Definition 16.4. Let (π, E, M ) be a vector bundle over a space M . Then the symmetric
tensor product is the vector bundle (S k (π), S k (E), M ) with fibers (S k (E))p ∼
= S k (Ep ) and
transition maps
k
{gαβ | α, β ∈ A}.
 DIFFERENTIAL TOPOLOGY LECTURES 23

the exterior tensor product is the vector bundle (Λk (π), Λk (E), M ) with fibers (Λk (E))p ∼
=
Λk (Ep ) and transition maps
∧k
{gαβ | α, β ∈ A}.
Definition 16.5. Let M be a smooth manifold. Then the symmetric and exterior tensor
bundle S k (T ∗ M ) and Λk (T ∗ M ) of M are defined as
[ [
S k (T ∗ M ) = S k (Tp∗ M ), Λk (T ∗ M ) = Λk (Tp∗ M ).
p∈M p∈M

The topology is generated by local coordinate systems on M .

Example 16.1. (1) Consider the symmetric tensor bundle S 2 (T ∗ M ). Then under a local
coordinate system, a section is of the form g(p) = di,j=1 gij (x(p))dxi dxj . Suppose g(p) :
P
Tp M ⊗ Tp M → R is positive definite. Then g is called a Riemannian metric on M .
(2) Consider the exterior tensorP bundle Λk (T ∗ M ). Then under a local coordinate system,
a section is of the form ω(p) = i1 <···<ik ωi1 ...ik (x(p))dxi1 ∧ · · · ∧ dxik . This is called a
differential form. When k = d, under a local coordinate system, a section is of the form
ω(p) = ω(x(p))dx1 ∧ · · · ∧ dxd . Suppose ω(p) ̸= 0. Then ω is called a volume form.

17. Lecture 17: Differential Forms

In the last lecture, we defined the exterior tensor product of vector bundles. In this
lecture, we will focus on the exterior tensor product of the cotangent bundles. Sections of
these vector bundles are called differential forms.
Definition 17.1. Let M be a smooth manifold. Then the space of differential k-forms
Ωk (M ) is the space of sections on the k-th exterior tensor product of the cotangent bundle:
Ωk (M ) = Γ(M, Λk (T ∗ M )).
Remark 17.1. Consider a coordinate system (U, x) in M , then a differential form ω can
be written as X
ω(p) = ωi1 ...ik (x(p))dxi1 ∧ · · · ∧ dxik .
i1 <···<ik
Given two differential forms ω and ω ′ ,
we can take their sum ω + ω ′ , their exterior product
′
ω ∧ ω , and given a function f , we can take the multiplication of a form by a function f ω.
Given a differential k-form ω and k vector fields X1 , . . . , Xk , using the pairing of exterior
tensor bundles
Λk (T ∗ M ) × Λk (T M ) → M × R,
we can define the pairing which gives a smooth function
ω(X1 , . . . , Xk )(p) = ωp (X1 (p), . . . , Xk (p)).
Conversely, we can show that any alternating C ∞ (M )-linear map on the space of vector
fields X(M ) can be defined this way:
Lemma 17.1. For any alternating C ∞ (M )-linear map l : X(M ) × · · · × X(M ) → C ∞ (M ),
there is a canonical k-form ωl ∈ Ωk (M ) so that ωl (X1 , . . . , Xk ) = l(X1 , . . . , Xk ).
Proof. First, we claim that the value l(X1 , . . . , Xk )(p) only depends on X1 (p), . . . , Xk (p).
For simplicity, we only consider C ∞ (M )-linear maps l : X(M ) → C ∞ (M ) and show that
l(X)(p) only depends on X(p). Using linearity, it suffices to show that l(X)(p) = 0 as long
as X(p) = 0. Consider a local coordinate system (U, x) and write X|U = di=1 ai (∂/∂xi )
P
24 WENYUAN LI

where ai (p) = 0. Consider a compact neighbourhood V ⊂ U of p and a bump function φ

on V ⊂ U . Then setting the globally defined vector fields Xi = φai (∂/∂xi ), we can write
d
X
X= Xi + (1 − φ)X.
i=1

Then, by the C ∞ (M )-linearity, we have

d
X
l(X)(p) = l(Xi )(p) + (1 − φ(p))l(X)(p) = 0.
i=1

Now, since l(X1 , . . . , Xk )(p) only depends on X1 (p), . . . , Xk (p), we can define the value of
k-form ωl at p ∈ M by setting (ωl )p (v1 , . . . , vk ) = l(X1 , . . . , Xk ) where X1 , . . . , Xk are any
smooth vector fields that satisfy X1 (p) = v1 , . . . , Xk (p) = vk . This completes the proof. □

For smooth functions, we have defined the notion of the differential, which is a linear map
d : Ω0 (M ) → Ω1 (M ). Now, we will extend the differential to a linear map on all differential
forms: d : Ω∗ (M ) → Ω∗+1 (M ):
Theorem 17.2. There exists a unique antiderivation d : Ω∗ (M ) → Ω∗+1 (M ) such that
(1) d2 = 0;
(2) d : Ω0 (M ) → Ω1 (M ) is the differential map;
(3) d(a1 ω1 + a2 ω2 )|p = a1 dω1 |p + a2 dω2 |p if ω1 , ω2 ∈ Ωk (M );
(4) d(ω1 ∧ ω2 )|p = dω1 |p ∧ ω2 |p + (−1)k ω1 |p ∧ dω2 |p if ω1 ∈ Ωk (M ), ω2 ∈ Ωl (M ).
Moreover, dωp = dωp′ if ω|U = ω ′ |U on some neighbourhood U .
Proof. First, we show the uniqueness. Let d be any exterior differential satisfying properties
(1)–(4). We prove that dωp = 0 if ω|U = 0 on some neighbourhood U . In fact, let φ be a
bump function on some neighbourhood U ′ ⊂ U of p. Then ω = (1 − φ)ω. We can compute
using (4) that
dωp = d((1 − φ)ω)p = d(1 − φ)p ∧ ωp + (1 − φ(p))dωp = 0.
Then,
Pwe can define the exterior differential d on local coordinate systems (U, x). Given
ω = i1 <···<ik ai1 ...ik dxi1 ∧ · · · ∧ dxik , by property (1)–(4), dω must be of the form
X
dω = dai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik .
i1 <···<ik

Using properties (1)–(4), we show that for any exterior differential d′ defined using a different
cooridinate system, we have
d′ ω = d′ (ai1 ...ik dxi1 ∧ · · · ∧ dxik )
Xk
= d′ ai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik + ai ...i dxi1 ∧ · · · ∧ d′ dxij ∧ · · · ∧ dxik
j=1 1 k
Xk
= dai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik + ai1 ...ik dxi1 ∧ · · · ∧ dd2 xij ∧ · · · ∧ dxik
j=1
= dai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik = dω.
Then, we show the existence. Consider a coordinate system (U, x), and define the exterior
differential of ω by
X
dω = dai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik .
i1 <···<ik
 DIFFERENTIAL TOPOLOGY LECTURES 25

Then we show that the dω we defined satisfies conditions (1)–(4). By the formula we write
down, (2)–(3) are obvious. (4) follows from the Leibniz rule:
d((ai1 ...ik dxi1 ∧ · · · ∧ dxik ) ∧ (bj1 ...jl dxj1 ∧ · · · ∧ dxjl ))
= d(ai1 ...ik bj1 ...jl dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl )
= (ai1 ...ik dbj1 ...jl + bj1 ...jl dai1 ...ik ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl ,
d(ai1 ...ik dxi1 ∧ · · · ∧ dxik ) ∧ (bj1 ...jl dxj1 ∧ · · · ∧ dxjl )
+ (−1)k (ai1 ...ik dxi1 ∧ · · · ∧ dxik ) ∧ d(bj1 ...jl dxj1 ∧ · · · ∧ dxjl )
= bj1 ...jl dai1 ...ik ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
+ ai1 ...ik dbj1 ...jl ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl .
Finally, (1) follows from the property that d2 f = 0:
d d X
d
X ∂f  X ∂2f
d2 f = d(df ) = d dxi = dxj ∧ dxi = 0.
∂xi ∂xi ∂xj
i=1 j=1 i=1
The fact that the definition is independent of coordinate changes follows from the unique-
ness. □
Just as vectors can pair with covectors, vector fields can pair with differential forms.
Definition 17.2. Let X be a smooth vector field on M , and let ω ∈ Ωk (M ). The interior
multiplication of ω by X is the form i(X)ω whose value at p ∈ M is the interior multiple
i(Xp )ωp .
Finally, we show that differential forms are functorial with respect to smooth maps.
Definition 17.3. Let φ : M → N be a smooth map. Then define φ∗ : Λ∗ Tφ(p)
∗ N → Λ∗ T ∗ M
p
to be the natural map induced by φ∗ : Tφ(p)
∗ N → T ∗ M . Let ω ∈ Ωk (M ). The pull-back of ω
p
by φ is a form φ∗ ω whose value at p ∈ M is the pull-back φ∗ ωp .
Proposition 17.3. Let φ : M → N be a smooth map. Then
(1) φ∗ : Ω∗ (N ) → Ω∗ (M ) is a ring/algebra homomorphism;
(2) φ∗ dω = d(φ∗ ω) for any ω ∈ Ω∗ (N );
(3) (φ∗ ω)p (X1p , . . . , Xkp ) = ωφ(p) (φ∗ X1p , . . . , φ∗ Xkp ).

18. Lecture 18: Lie Derivatives

Since the (co)tangent spaces at different points are not canonically identified, it is harder
to define derivatives of vector fields or differentials forms. Any such definition will require
a choice of the identification of (co)tangent spaces of nearby points. We now define the
derivatives of tensor fields and differential forms along a vector field called the Lie derivative,
where we identity nearby (co)tangent spaces using the integration flow. (Note that this is
different from the so-called covariant derivatives in Riemannian geometry.)
Definition 18.1. Let X and Y be vector fields on a manifold M and φtX : U → M be
integration flow of X in a neighbourhood U of p. Then the Lie derivative of the vector field
Y along X is
d  −t 
(LX Y )p = (φX )∗ YφtX (p) |t=0 .
dt
Let ω be a differential form on the manifold M . Then the Lie derivative of the differential
form ω along X is
d  t ∗ 
(LX ω)p = (φX ) ωφtX (p) |t=0 .
dt
26 WENYUAN LI

Roughly, the idea in the definition is to use the integration flow φtX to identity the
(co)tangent spaces at p and φtX (p), and then compare the (co)tangent vectors at these
points.
The following proposition shows that this is a reasonable definition, since it agrees with
the usual derivative on smooth functions:
Proposition 18.1. Let X be a vector field on M .
(1) Let f be a function on M . Then LX f = X(f );
(2) Let Y be a vector field on M . Then LX Y = [X, Y ];
(3) Let ω be a differential form on M . Then LX ω = di(X)ω + i(X)dω;
(4) Let ω be a differential form and X1 , . . . , Xk be vector fields on M . Then
k
X
LX (ω(X1 , . . . , Xk )) = (LX ω)(X1 , . . . , Xk ) + ω(X1 , . . . , LX Xi , . . . , Xk ).
i=1

Remark 18.1. (3) is also called Cartan’s magic formula.

Proof. First, we show that LX (Y (f )) = (LX Y )(f ) + Y (LX f ). This is because
(φtX )∗ (Y (f )) − Y (f )
LX (Y (f )) = lim
t→0 t
(φ )∗ Y ((φtX )∗ f ) − (φ−t
−t
X )∗ Y (f ) ((φ−t
X )∗ Y )(f ) − Y (f )
= lim X + lim
t→0 t t→0 t
Y ((φtX )∗ f ) − Y (f ) ((φ−tX ∗) Y )(f ) − Y (f )
= lim + lim
t→0 t t→0 t
= Y (LX f ) + (LX Y )(f ).
We can conclude that (4) holds in a similar way. Then, using (4), we can conclude that (2)
holds:
(LX Y )(f ) = LX (Y (f )) − Y (LX f ) = X(Y (f )) − Y (X(f )) = [X, Y ](f ).
Finally, we can show that (3) holds by the following local computation:
 Xd  Xd
LX ai dxi (Y ) = (LX ai )dxi (Y ) + ai (LX dxi )(Y )
i=1 i=1
Xd
= (Xai )(Y xi ) + ai X(Y (xi )) − ai [X, Y ](xi )
i=1
Xd
= (Xai )(Y xi ) − (Y ai )(Xxi ) + Y (ai X(xi ))
i=1
Xd
= dai (X)dxi (Y ) − dxi (X)dai (Y ) + d(ai dxi (X))(Y ).
i=1

Then we can conclude the proof by induction on the degree. □

Finally, using the above proposition, we can give the following coordinate free formula
for computing the exterior differential:
Proposition 18.2. Let ω be a differential form and X1 , X2 , . . . , Xk+1 be vector fields on
M . Then
Xk+1
dω(X1 , X2 , . . . , Xk+1 ) = (−1)i+1 Xi ω(X1 , . . . , X
bi , . . . , Xk+1 )
i=1
X
+ (−1)i+j ω([Xi , Xj ], X1 , . . . , X bi , . . . , X
bj , . . . , Xk+1 ).
i<j
 DIFFERENTIAL TOPOLOGY LECTURES 27

Proof. We prove by induction on the degree using Cartan’s formula. For a 1-form ω, we
can compute
dω(X, Y ) = (i(X)dω)(Y ) = (LX ω)(Y ) − (di(X)ω)(Y )
= LX (ω(Y )) − ω(LX Y ) − d(ω(X))(Y ) = Xω(Y ) − Y ω(X) − ω([X, Y ]).
Then we complete the proof by induction. □
Pd
Remark 18.2. (1) Under some local coordinate systems (U, x), let X = i=1 ai ∂/∂xi and
Y = di=1 bi ∂/∂xi , the Lie derivative on vector fields can be computed by
P

d 
∂bi
X ∂ai  ∂
LX Y = [X, Y ] = − bj
aj .
∂xj ∂xj ∂xi
i,j=1

(2) Under some local coordinate systems (U, x), let X = di=1 ai ∂/∂xi and ω = di=1 bi dxi ,
P P
the Lie derivative on 1-forms can be computed by
d 
X ∂bi ∂ai 
LX ω = aj + bj dxi .
∂xj ∂xj
i,j=1

19. Lecture 19: Differential Ideals

In the previous lecture, we have seen the relationship between the exterior differential
and the Lie bracket. In this lecture, we will review the Frobenius theorem for distributions,
which was stated in terms of Lie brackets of vector fields, and reformulate it in terms of
differntials of forms.
This was one of the original motivations that E. Cartan introduced the calculus of differ-
ential forms, and we will see why the antisymmetry of differntial forms play a crucial role
in the problem.
Definition 19.1. Let D be an l-dimensional smooth distribution on M . A k-form ω is said
to annihilate D if for each p ∈ M and any v1 , . . . , vk ∈ D(p),
ωp (v1 , . . . , vk ) = 0.
Define I (D) = {ω0 + · · · + ωd | ωi ∈ Ωi (M ) annihilates D}.
Definition 19.2. A collection ω1 , . . . , ωl of 1-forms on M is called independent if they form
an independent set in Tp∗ M for each p ∈ M .
We can show that the annihilator I (D) is an ideal of the ring Ω∗ (M ) and every distri-
bution corresponds to such an ideal and vice versa.
Proposition 19.1. Let D be an l-dimensional smooth distribution on M .
(1) I (D) is an ideal of Ω∗ (M ): for any ω ∈ I (D) and η ∈ Ω∗ (M ), ω ∧ η ∈ I (D);
(2) I (D) is locally generated by d − l independent 1-forms: ω ∈ I (D) iff for any
p ∈ M , there is a neighbourhood U and 1-forms ω1 , . . . , ωd−l on U such that ω|U is
generated by these 1-forms.
Conversely, let I be an ideal of Ω∗ (M ) locally generated by d−l independent 1-forms, there
exists an l-dimensional smooth distribution D on M such that I = I (D).
Proof. We only prove statement (2). Suppose D is locally generated by vector fields
X1 , . . . , Xl on an open subset U . Then by inverse function theorem, there exist vector
fields Xl+1 , . . . , Xd such that X1 , . . . , Xl , Xl+1 , . . . , Xd form a basis of tangent spaces at all
points in U . We take the dual 1-forms ω1 , . . . , ωd such that
ωi (Xj )|U = δij , 1 ≤ i, j ≤ d.
28 WENYUAN LI

For ω ∈ Ω∗ (M ), we write ω|U = i1 <···<ik ai1 ...ik ωi1 ∧ · · · ∧ ωik . Then if ω ∈ I (D), it
P

must be the case that ai1 ...ik = 0 whenever there exists ij ∈ {1, . . . , l}. Conversely, if ω|U is
generated by ω1 , . . . , ωd−l , then clearly ω ∈ I (D). This proves statement (2). □
In the previous weeks, we have seen that a distribution is integrable if and only if it is
involutive, namely, it is closed under Lie bracket. Using the exterior differential, we are
able to reformulate this condition in a more algebraic way:
Definition 19.3. A subspace I ⊂ Ω∗ (M ) is an ideal if for any ω ∈ I and η ∈ Ω∗ (M ),
ω ∧ η ∈ I . An ideal I ⊂ Ω∗ (M ) is a differential ideal if
d(I ) ⊂ I .
Theorem 19.2. A smooth distribution D on M is involutive if and only if I (D) is a
differential ideal.
Proof. Let D be an involutive distribution. Suppose D is locally generated by vector fields
X1 , . . . , Xl on an open set U . Complete the collection of vector fields to X1 , . . . , Xl , . . . , Xd
that form a basis of tangent spaces at all points in U . We take the dual 1-forms ω1 , . . . , ωd
such that
ωi (Xj )|U = δij , 1 ≤ i, j ≤ d.
Then I (D) is locally generated by the 1-forms ωl+1 , . . . , ωd . Since for l + 1 ≤ j ≤ d and
1 ≤ i, i′ ≤ l,
dωj (Xi , Xi′ ) = Xi ωj (Xi′ ) − Xi′ ωj (Xi ) − ω([Xi , Xi′ ]) = 0,
we know that dωj annihilates X1 , . . . , Xl .
Conversely, take the 1-forms ωl+1 , . . . , ωd that generate I (D). For l + 1 ≤ j ≤ d, if dωj
annihilates D, then
ω([Xi , Xi′ ]) = dωj (Xi , Xi′ ) − Xi ωj (Xi′ ) + Xi′ ωj (Xi ) = 0.
This implies that [Xi , Xi′ ] must be a linear combination of X1 , . . . , Xl . □
Finally, we explain the historical context in which Frobenius’ theorem was developed.
The classical Frobenius’ theorem was a theorem about when a differential
b11 dx1 + · · · + b1d dxd , . . . , bc1 dx1 + · · · + bcd dxd
can be written as a total differential of smooth functions α : U ⊂ Rd → V ⊂ Rc . Frobenius
proved that there exist smooth functions α that satisfies the equation
∂αi
= bij
∂xj
if and only if bij satisfy the following relation that
n
∂biβ ∂biγ X  ∂biβ ∂biγ 
− + bjγ − bjβ = 0.
∂xγ ∂xβ ∂yj ∂yj
j=1
Darboux realized that the following formal expression is in fact invariant under coordinate
changes
 ∂b ∂biγ 
iβ
− (dxβ dxγ − dxγ dxβ ).
∂xγ ∂xβ
Cartan finally developed the language of differential forms and exterior differentials under
which the above differential can now be written as simply
d(b11 dx1 + · · · + b1d dxd ), . . . , d(bc1 dx1 + · · · + bcd dxd ).

Department of Mathematics, University of Southern California.

Email address: [email protected]