Differential Geometry

Winter Term 2024/2025

Lecturer: Dr. N. Tsakanikas
Assistant: L. E. Rösler

Differential Geometry II - Smooth Manifolds

January 18, 2025

 CONTENTS

1 Topological and Smooth Manifolds 1

1.1 Topological Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Smooth Maps 9
2.1 Smooth Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Partitions of Unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 The Tangent Bundle 19

3.1 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 The Differential of a Smooth Map . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Computations in Local Coordinates . . . . . . . . . . . . . . . . . . . . . . 25
3.4 The Tangent Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Velocity Vectors of Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Maps of Constant Rank 33

4.1 Immersions, Submersions, and Embeddings . . . . . . . . . . . . . . . . . . 33
4.2 Local Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 The Rank Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5 Submanifolds 45
5.1 Embedded Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Immersed Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 The Tangent Space to a Submanifold . . . . . . . . . . . . . . . . . . . . . 52

6 Vector Bundles 55
6.1 Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Sections of a Vector Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.3 Subbundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

7 Vector Fields and Flows 63

7.1 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.2 Integral Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

iii
iv CONTENTS

7.3 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

8 Differential Forms 79
8.1 Differential 1-Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.2 Differential k-Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

9 Manifolds with Boundary 95

9.1 Topological Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . 95
9.2 Smooth Manifolds with Boundary . . . . . . . . . . . . . . . . . . . . . . . 96

10 Orientations 99
10.1 Orientations of Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.2 Orientations of Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . 101

11 Integration on Manifolds 107

11.1 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
11.2 Integration of Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . 111

A The Real Projective Space 119

B Sard’s theorem and Whitney’s theorems 125

C Multilinear Algebra 127

1 The Dual of a Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . 127
2 Multilinear Maps and Tensors . . . . . . . . . . . . . . . . . . . . . . . . . 129
3 Symmetric Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4 Alternating Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

Bibliography 141

Index 143
CHAPTER 1
TOPOLOGICAL AND SMOOTH MANIFOLDS

1.1 Topological Manifolds

Definition 1.1. A topological manifold of dimension n (or topological n-manifold) is a
topological space M with the following properties:

• M is a Hausdorff space: for each pair of distinct points p, q ∈ M there are disjoint
open sets U, V ⊆ M such that p ∈ U and q ∈ V .

• M is second-countable: there is a countable basis for the topology of M .

• M is locally Euclidean of dimension n: each point of M has a neighborhood which

is homeomorphic to an open set of Rn ; that is, for each p ∈ M we can find

 an open subset U ⊆ M containing p,

 an open subset U
b ⊆ Rn , and
 a homeomorphism φ : U → U b.

Recall : Let X and Y be topological spaces. A continuous bijective map F : X → Y with

continuous inverse is called a homeomorphism. If there exists a homeomorphism from
X to Y , then we say that X and Y are homeomorphic.

Comments:

(1) Every topological manifold has, by definition, a specific, well-defined dimension. Thus,
we do not consider spaces of mixed dimension, such as the disjoint union of a plane
and a line, to be manifolds at all. It can be shown (using de Rham cohomology or
singular homology) that the dimension of a (non-empty) topological manifold is in fact a
topological invariant: A non-empty topological n-manifold cannot be homeomorphic to a
topological m-manifold unless n = m.

(2) The three conditions in Definition 1.1 ensure that topological manifolds behave in the
ways we expect from our experience with Euclidean spaces. For example, in a Hausdorff

1
2 Chapter 1. Topological and Smooth Manifolds

space, finite subsets are closed and limits of convergent sequences are unique. The mo-
tivation for second-countability is less evident, but stems from the existence of so-called
partitions of unity; see Section 2.2.

(3) There are also examples of topological spaces which are not topological manifolds;
see Exercise Sheet 1. For example:

 The line with two origins is locally Euclidean and second-countable, but not Haus-
dorff.
 A disjoint union of uncountably many copies of R is locally Euclidean and Haus-
dorff, but not second-countable.

Definition 1.2. Let M be a topological n-manifold. A coordinate chart on M is a pair

(U, φ), where U is an open subset of M and φ : U → U b is a homeomorphism from U to
an open subset Ub = φ(U ) ⊆ Rn . The set U is called a coordinate domain, or a coordinate
neighborhood of each of its points. If, in addition, φ(U ) is an open ball in Rn , then U is
called a coordinate ball ; if φ(U ) is an open cube in Rn , then U is called a coordinate cube.
1 2 n
The map φ called a (local) coordinate map
and its component functions (x , x , . . . , x ),
1 2 n
defined by φ(p) = x (p), x (p), . . . , x (p) , are called local coordinates on U .

Figure 1.1: A coordinate chart

By definition of a topological manifold, each point p ∈ M is contained in the domain

of some chart (U, φ). If φ(p) = 0, then we say that the chart is centered at p. If (U, φ)
is any chart whose domain contains p, it is easy to obtain a new chart centered at p by
subtracting the constant vector φ(p).

Example 1.3.

(0) The basic example of a topological n-manifold is Rn itself. It is Hausdorff, because it

is a metric space, and it is second-countable, because the collection of all open balls with
rational centers and rational radii is a countable basis for the topology.
Moreover, every open subset of a topological n-manifold is itself a topological n-
manifold (with the subspace topology), because the Hausdorff and second-countability
properties are inherited by subspaces.

(1) Graphs of continuous functions: Let U ⊆ Rn be an open subset and let f : U → Rk

be a continuous function. The graph of f is the subset

Γ(f ) := (x, y) ∈ Rn × Rk | x ∈ U, y = f (x) ⊆ Rn × Rk



with the subspace topology. Let π1 : Rn × Rk → Rn be the projection onto the first factor
and let φ : Γ(f ) → U be the restriction of π1 to Γ(f ):

φ(x, y) = x, (x, y) ∈ Γ(f ).

 Section 1.1. Topological Manifolds 3

Since φ is the restriction of a continuous map, it is continuous; and

it is homeomorphism,
−1
because it has a continuous inverse given by φ (x) = x, f (x) . Thus, Γ(f ) is a topo-

logical manifold of dimension n. In fact, Γ(f ) is homeomorphic to U itself, and Γ(f ), φ
is a global coordinate chart, called graph coordinates.
The same observation applies to any subset of Rn+k defined by setting any k of the
coordinates (not necessarily the last k) equal to some continuous function of the other n,
which are restricted to lie in an open subset of Rn . (This observation will be used in (2)
below for k = 1.)
(2) Spheres: For each n ∈ N, the unit n-sphere is the subset
Sn := x ∈ Rn+1 |x| = 1 ⊆ Rn+1 .


It is Hausdorff and second-countable, because it is a subspace of Rn+1 . To show that it is

locally Euclidean, for each i ∈ {1, . . . , n + 1} consider the sets
Ui+ := (x1 , . . . , xn+1 ) ∈ Rn+1 | xi > 0


and
Ui− := (x1 , . . . , xn+1 ) ∈ Rn+1 | xi < 0 .


Consider also the (open) unit ball of dimension n

Bn := x ∈ Rn |x| < 1


and the continuous function

p
f : Bn → R, u 7→ 1 − |u|2 .
Then for each i ∈ {1, . . . , n + 1} it is easy to check that Ui+ ∩ Sn is the graph of the
function
xi = f (x1 , . . . , xbi , . . . , xn+1 )
↰
omitted

and that Ui− n

∩ S is the graph of the function

xi = −f (x1 , . . . , xbi , . . . , xn+1 ) .

↰
omitted

Thus, each subset Ui± ∩ Sn is locally Euclidean of dimension n, see (1) above, and the
maps
φ± ± n
i : Ui ∩ S → B
n

(x1 , . . . , xn+1 ) 7→ (x1 , . . . , xbi , . . . , xn+1 )

are graph coordinates for Sn . Since each point of Sn is in the domain of at least one of
these 2n + 2 charts (see Figure 1.2), we conclude that Sn is a topological n-manifold.
(3) Projective spaces: see Appendix A.
We will encounter many more examples of topological manifolds later in the course
and in the exercise sheets as well.
4 Chapter 1. Topological and Smooth Manifolds

Figure 1.2: Charts for Sn

1.2 Smooth Manifolds

⇝ Topological manifolds are:

 suitable for the study of topological properties (e.g. compactness, connected-

ness, etc)
 not suitable for doing calculus: being “differentiable” is not an invariant un-
der homeomorphisms (in other words, it is not a topological property). For
instance, the map
√ √

φ : R2 → R2 , φ(u, v) 7→ 3 u, 3 v

is a homeomorphism, the map

f : R2 → R, f (x, y) 7→ x
√
is differentiable, but the composite map (f ◦ φ)(u, v) = 3
u is not differentiable
at (0, 0).

⇝ To make sense of derivatives of maps between manifolds, we need to introduce a new

kind of manifold. It will be a topological manifold with some extra structure (in
addition to its topology), which will allow us to decide which maps are “smooth”.

⇝ Plausible definition of “smoothness” of a function on M : f : M → R smooth if

f ◦ φ−1 : U
b ⊆ Rn → R smooth (in the usual sense), which makes sense if it does not
depend on the choice of coordinate chart (U, φ). To guarantee this independence,
we will restrict our attention to “smooth charts”.

Definition 1.4. Let M be a topological manifold. If (U, φ) and (V, ψ) are two charts
such that U ∩ V ̸= ∅, then the composite map ψ ◦ φ−1 : φ(U ∩ V ) → ψ(U ∩ V ) is called
the transition map from φ to ψ (see Figure 1.3) and is clearly a homeomorphism. Two
charts (U, φ) and (V, ψ) are said to be smoothly compatible if either U ∩ V = ∅ or the
transition map ψ ◦ φ−1 is diffeomorphism (i.e., smooth and bijective with smooth inverse).
Since φ(U ∩ V ) and ψ(U ∩ V ) are open subsets of Rn , smoothness of this map is to be
interpreted in the ordinary sense of having continuous partial derivatives of all orders
(C ∞ ).
An atlas for M is a collection of charts whose domains cover M . An atlas A is called
a smooth atlas if any two charts in A are smoothly compatible. Finally, a smooth atlas A
on M is called maximal (or complete) if it is not properly contained in any larger smooth
atlas. This just means that any chart which is smoothly compatible with every chart in
A is already in A.
Remark 1.5. To show that an atlas is smooth, we need only verify that each transition
map ψ ◦ φ−1 is smooth whenever (U, φ) and (V, ψ) are charts in A; once we have proved
this, it follows that ψ ◦ φ−1 is a diffeomorphism because its inverse φ ◦ ψ −1 = (ψ ◦ φ−1 )−1
is one of the transition maps we have already shown to be smooth. Alternatively, given
 Section 1.2. Smooth Manifolds 5

two particular charts (U, φ) and (V, ψ), it is often easier to show that they are smoothly
compatible by verifying that ψ ◦ φ−1 is smooth with non-singular Jacobian at each point
of its domain, since then the Inverse Function Theorem = [Lee13, Theorem C.34] implies
that ψ ◦ φ−1 is a diffeomorphism, see [Lee13, Corollary C.36].

M
V

U
ψ
φ

ψ ◦ φ−1
ψ(V )
φ(U )

Rn Rn

Figure 1.3: A transition map

Comment: Our plan (see Chapter 2) is to define a “smooth structure” on M by giving

it a smooth atlas, and to define a function f : M → R to be smooth if and only if f ◦ φ−1
is smooth in the sense of ordinary calculus for each coordinate chart (U, φ) in this atlas.
There is one minor technical problem with this approach (which led to the definition
of a maximal smooth atlas): in general, there will be many possible atlantes that give
the “same” smooth structure, in that they all determine the same collection of smooth
functions on M . For example, consider the following pair of smooth atlantes on Rn :

A1 := (Rn , IdRn )},



A2 := (B1 (x), IdB1 (x) ) | x ∈ Rn }.



Although these are different smooth atlantes, clearly a function f : Rn → R is smooth

with respect to either atlas if and only if it is smooth in the sense of ordinary calculus.

We can now define the main concept of this chapter.

Definition 1.6. Let M be a topological manifold. A smooth structure on M is a maximal

smooth atlas. A smooth manifold is a pair (M, A), where M is a topological manifold
and A is a smooth structure on M .

Remark 1.7. A smooth structure is an additional piece of data that must be added to
a topological manifold before we are entitled to talk about a “smooth manifold”. Note
that a given topological manifold may have many smooth structures (in fact, if it has one,
then it has infinitely many, see [Lee13, Problem 1.6]), but it may also admit no smooth
structures at all.
6 Chapter 1. Topological and Smooth Manifolds

It is generally not convenient to define a smooth structure by explicitly describing a

maximal smooth atlas, because such an atlas contains very many charts. The next result
shows that we need only specify some smooth atlas.

Proposition 1.8. Let M be a topological manifold.

(a) Every smooth atlas A for M is contained in a unique maximal smooth atlas, called
the smooth structure determined by A.

(b) Two smooth atlantes for M determine the same smooth structure if and only if their
union is a smooth atlas.

Proof.
(a) Given a smooth atlas A on M , set

A := (U, φ) chart for M | ∀(V, ψ) ∈ A : (U, φ) and (V, ψ) are smoothly compatible .


By definition of a smooth atlas we have A ⊆ A. Now, let A′ be a smooth atlas on M such

that A ⊆ A′ and take (U ′ , φ′ ) ∈ A′ . Since it holds that A ⊆ A′ , we infer that (U ′ , φ′ ) is
smoothly compatible with every chart (U, φ) ∈ A (by virtue of A′ being a smooth atlas).
Hence, (U ′ , φ′ ) ∈ A, which implies that A = A′ ; in particular, A is a smooth atlas on M .
As A′ was arbitrary, we also conclude that A is maximal.
It remains to show that A the unique maximal smooth atlas containing A. So let
A′ be a maximal smooth atlas containing A. In particular, any chart in A′ is smoothly
compatible with any chart in A, and thus A′ ⊆ A. By maximality, we conclude that
A′ = A, and thus we obtain the uniqueness.
(b) Assume first that two smooth atlantes A1 and A2 for M determine the same smooth
structure, that is, both A1 and A2 are contained in the same (unique) maximal smooth
atlas for M . Then every chart (U, φ) ∈ A1 is smoothly compatible with every chart
(V, ψ) ∈ A2 , so the union A1 ∪ A2 is a smooth atlas for M .
Conversely, assume that the union A1 ∪ A2 of two smooth atlantes A1 and A2 for M
is also a smooth atlas for M . Then every chart (U, φ) ∈ A1 is smoothly compatible with
every chart (V, ψ) ∈ A2 . If Ā1 (resp. Ā2 ) is the smooth structure on M determined by
A1 (resp. A2 ), then by the construction in (a) we infer that A1 ⊆ Ā2 and A2 ⊆ Ā1 , and
hence Ā1 = Ā2 due to the uniqueness in (a).

For example, if a topological manifold M can be covered by a single chart, then

the smooth compatibility condition is trivially satisfied, so any such chart determines
automatically a smooth structure on M ; see Example 1.3(1) and Example 1.10(1).

Definition 1.9. Let (M, A) be a smooth manifold. Any chart (U, φ) contained in the
maximal smooth atlas A is called a smooth coordinate chart. The corresponding coor-
dinate map φ is called a smooth coordinate map, and its domain U is called a smooth
coordinate domain, or smooth coordinate neighborhood of each of its points. A smooth
coordinate ball is a smooth coordinate domain whose image under a smooth coordinate
map is a ball in Euclidean space. A smooth coordinate cube is defined similarly.
 Section 1.2. Smooth Manifolds 7

Here is how one usually thinks about (smooth) coordinate charts on a smooth manifold.
Once we choose a (smooth) coordinate chart (U, φ) on M n , the (smooth) coordinate map
φ: U → U b ⊆ Rn can be thought of as giving a temporary identification between U and U b.
Using this identification, while we work in this chart, we can think of U simultaneously
as an open subset of M and as an open subset of Rn . Under this identification, we can
represent a point p ∈ M by its coordinates (x1 , . . . , xn ) = φ(p), and think of this n-
tuple as being the point p. We typically express this by saying “(x1 , . . . , xn ) is the (local)
coordinate representation for p” or “p = (x1 , . . . , xn ) in local coordinates”.
Example 1.10.
(0) For each n ∈ N the Euclidean
 n space Rn is a smooth n-manifold with smooth structure
determined by the atlas (R , IdRn ) . We call this the standard smooth structure on Rn
and the resulting coordinate map the standard coordinates on Rn . (Unless we explicitly
say otherwise, we always use this smooth structure on Rn .) With respect to this smooth
structure, the smooth coordinate charts for Rn are exactly those charts (U, φ) such that
φ is diffeomorphism (in the usual sense) from U ⊆ Rn to another open set U b ⊆ Rn .

(1) Graphs of smooth functions: If U ⊆ Rn is an open set and if f : U → Rk is a

smooth function, then by Example 1.3(1) the graph Γ(f ) of f is a topological n-manifold
in the subspace topology. Since Γ(f ) is covered by the single graph coordinate chart
φ : Γ(f ) → U , we can put
a canonical smooth structure on Γ(f ) by declaring the graph
coordinate chart Γ(f ), φ to be a smooth chart.
(2) Spheres: The unit n-sphere Sn ⊆ Rn+1 is a topological n-manifold according to
Example 1.3(2). We put a smooth structure on Sn as follows. For each i ∈ {1, . . . , n + 1}
consider the graph coordinate charts (Ui± ∩ Sn , φ± i ). For any i ̸= j and any choice of ±
signs, the transition maps φi ◦ (φj ) and φi ◦ (φ∓
± ± −1 ±
j )
−1
are easily computed. For example,
when i < j, we get:
+ −1
p
φ+ (u , . . . , un ) = φ+

1 1
1 − |u|2 , . . . , un


i ◦ (φj ) i u ,...,
↓
j-th
p
= u1 , . . . , ubi , . . . , 1 − |u|2 , . . . , un ,


↓ ↓
i-th j-th

and similar formulas hold in the other cases. When i = j, the domains of φ+ and φ−i are
 ± in ± n+1
disjoint, so there is nothing to check. Thus, the collection of charts (Ui ∩ S , φi ) i=1 is
a smooth atlas, so it defines a smooth structure on Sn , which we call its standard smooth
structure.
(3) Projective spaces: see Appendix A.
(4) Open submanifolds: If U is any open subset of Rn , then U is a topological n-manifold,
and the single chart (U, IdU ) determines a smooth structure on U .
More generally, let M be a smooth n-manifold and let U ⊆ M be an open subset.
Define an atlas on U by

AU = smooth charts (V, φ) for M such that V ⊆ U .


8 Chapter 1. Topological and Smooth Manifolds

Every point p ∈ U is contained in the domain of some chart (W, φ) for M . If we set
V = W ∩ U , then (V, φ|V ) is a chart in AU whose domain contains p. Therefore, U is
covered by the domains of the charts in AU , and it is easy to verify that AU is a smooth
atlas for U . In conclusion, any open subset of M is itself a smooth n-manifold in a natural
way. Endowed with this smooth structure, we call any open subset an open submanifold
of M .

We will encounter many more examples of smooth manifolds later in the course and
in the exercise sheets as well.
In the examples we have seen so far, we constructed a smooth manifold structure in
two stages: we started with a topological space and checked that it was a topological
manifold, and then we specified a smooth structure (by means of a smooth atlas due to
Proposition 1.8(a)). The following lemma shows how, given a set with suitable “charts”
that overlap smoothly, we can use these charts to define both a topology and a smooth
structure on the set.

Lemma 1.11 (Smooth manifold chart lemma). Let M be a set. Suppose that we are
given a collection {Uα } of subsets of M together with maps φα : Uα → Rn such that the
following properties are satisfied:

(i) For each α, φα is a bijection between Uα and an open subset φα (Uα ) ⊆ Rn .

(ii) For each α and β, the sets φα (Uα ∩ Uβ ) and φβ (Uα ∩ Uβ ) are open in Rn .

(iii) Whenever Uα ∩ Uβ ̸= ∅, the map

φβ ◦ φ−1
α : φα (Uα ∩ Uβ ) → φβ (Uα ∩ Uβ )

is smooth.

(iv) Countably many of the sets Uα cover M .

(v) Whenever p, q ∈ M with p ̸= q, either there exists some Uα containing both p and q
or there exist disjoint sets Uα and Uβ with p ∈ Uα and q ∈ Uβ .

Then M has a unique smooth manifold structure such that each (Uα , φα ) is a smooth
chart.

Proof. For the details of the proof we refer to [Lee13, Lemma 1.35]. The basic idea is to
define a topology on M by taking all sets of the form φ−1 n
α (V ), where V ⊆ R is open, as
a basis.
CHAPTER 2
SMOOTH MAPS

The main reason for introducing smooth structures was to enable us to define smooth
functions on manifolds and smooth maps between manifolds. In Section 2.1 we carry out
this project. In Section 2.2 we introduce a powerful tool for blending together locally
defined smooth objects, called partitions of unity. They are used throughout smooth
manifold theory for building global smooth objects out of local ones. At the end of this
chapter we will give the first applications of partitions of unity.

2.1 Smooth Maps

Definition 2.1. Let M be a smooth n-manifold and let f : M → Rk be a function, where
k, n ∈ N. We say that f is smooth if for every point p ∈ M there exists a smooth chart
(U, φ) for M such that p ∈ U and the composite function f ◦ φ−1 is smooth on the open
subset Ub = φ(U ) ⊆ Rn .

Figure 2.1: Definition of smooth functions

Remark 2.2. Let M be a smooth manifold. The set C ∞ (M ) of all smooth real-valued
functions on M is an R-vector space: sums and constant multiples of smooth functions are
smooth. Note that C ∞ (M ) is infinite-dimensional, see Exercise 2.21. Moreover, pointwise
multiplication turns C ∞ (M ) into a commutative ring and an associative and commutative
R-algebra.
Definition 2.3. Let M be a topological manifold. Given a function f : M → Rk and
a chart (U, φ) for M , the function fb = f ◦ φ−1 : φ(U ) → Rk is called the coordinate
representation of f .
Let M be a smooth manifold and let f : M → Rk be a function on M . By definition,
f is smooth if and only if its coordinate representation is smooth in some smooth chart
around each point. According to [Exercise Sheet 3, Exercise 3], smooth functions have
smooth coordinate representations in every smooth chart; that is, f ◦ φ−1 : φ(U ) → Rk is
smooth for every smooth chart (U, φ) for M .

9
10 Chapter 2. Smooth Maps

Definition 2.4. Let F : M → N be a map between smooth manifolds. We say that F is

a smooth map if for every p ∈ M there exist smooth charts (U, φ) containing p and (V, ψ)
containing F (p) such that F (U ) ⊆ V and the composite map ψ ◦ F ◦ φ−1 : φ(U ) → ψ(V )
is smooth.

Figure 2.2: Definition of smooth maps

Observe that Definition 2.1 is a special case of Definition 2.4 by taking N = V = Rk

and ψ = IdRk .
The first important observation about our definition of smooth maps is that, as one
might expect, smoothness implies continuity.

Proposition 2.5. Every smooth map is continuous.

Proof. Let F : M → N be a map between smooth manifolds. Fix p ∈ M . Since F is

smooth, there are smooth charts (U, φ) containing p and (V, ψ) containing F (p) such that
F (U ) ⊆ V and ψ ◦ F ◦ φ−1 : φ(U ) → ψ(V ) is smooth, and hence continuous. Since
F (U ) ⊆ V and the maps φ : U → φ(U ) and ψ : V → ψ(V ) are homeomorphisms, the map

F U
= ψ −1 ◦ (ψ ◦ F ◦ φ−1 ) ◦ φ : U → V

is continuous as a composition of continuous maps. Therefore, F is continuous in a

neighborhood of each point, and thus continuous on M .

Comment: The requirement that

∀p ∈ M ∃ (U, φ) ∋ p ∃ (V, ψ) ∋ F (p) such that F (U ) ⊆ V

in the definition of smoothness is included precisely so that smoothness automatically

implies continuity. [Lee13, Problem 2.1] illustrates what can go wrong if this requirement
is omitted.

Definition 2.6. Let F : M → N be a map between topological manifolds. If (U, φ) and

(V, ψ) are charts for M and N , respectively, then we call Fb = ψ ◦ F ◦ φ−1 the coordinate

representation of F with respect to the given coordinates. It maps the set φ U ∩ F −1 (V )
to ψ(V ).

Remark 2.7. If F : M → N is a smooth map between smooth manifolds, then the

coordinate representation of F with respect to every pair of smooth charts for M and N
is smooth, see [Exercise Sheet 3, Exercise 3].

Proposition 2.8 (Equivalent characterizations of smoothness). Let M and N be smooth

manifolds and let F : M → N be a map. Then F is smooth if and only if either of the
following conditions is satisfied:

(a) For every p ∈ M there exist smooth charts (U, φ) containing p and (V, ψ) containing
F (p) such that U ∩F
−1 (V ) is open in M and the composite map ψ◦F ◦φ−1 is smooth
from φ U ∩ F −1 (V ) to ψ(V ).
 Section 2.1. Smooth Maps 11

 
(b) F is continuous and there exist smooth atlases (Uα , φα ) and (Vβ , ψβ ) for M
and N , respectively, such that for each α and β, ψβ ◦ F ◦ φ−1
α is a smooth map from
−1


φα Uα ∩ F (Vβ ) to ψβ (Vβ ).

Proof. See [Exercise Sheet 3, Exercise 1].

Proposition 2.9 (Smoothness is a local property). Let M and N be smooth manifolds

and let F : M → N be a map. The following statements hold:

(a) If every point p ∈ M has a neighborhood U such that F |U is smooth, then F is

smooth.

(b) If F is smooth, then its restriction to every open subset of M is smooth.

Proof. See [Exercise Sheet 3, Exercise 2].

The next result is essentially just a restatement of the previous proposition, but it
gives a highly useful way of constructing smooth maps.

Lemma 2.10 (Gluing lemma for smooth maps). Let M and N be smooth manifolds and
let {Uα }α∈A be an open cover of M . Suppose that for each α ∈ A we are given a smooth
map Fα : Uα → N such that the maps agree on overlaps: Fα |Uα ∩Uβ = Fβ |Uα ∩Uβ for all
α, β ∈ A. Then there exists a unique smooth map F : M → N such that F |Uα = Fα for
each α ∈ A.

Proposition 2.11. Let M , N and P be a smooth manifolds.

(a) Every constant map c : M → N is smooth.

(b) The identity map IdM of M is smooth.

(c) If U ⊆ M is an open submanifold, then the inclusion map ι : U ,→ M is smooth.

(d) If F : M → N and G : N → P are smooth, then so is G ◦ F : M → P .

Proof. See [Exercise Sheet 3, Exercise 3].

Comment: We now have enough information in order to produce a number of interesting

examples of smooth maps. In spite of the apparent complexity of the definition, it is
usually not hard to prove that a particular map is smooth. There are basically only three
common ways to do so:

(1) Write the map in smooth local coordinates and recognize its component functions
as compositions of smooth elementary functions.

(2) Exhibit the map as a composition of maps that are known to be smooth.

(3) Use some special-purpose theorem that applies to the particular case under consid-
eration.

We give below an example of a smooth map utilizing the first method above, and we will
encounter many more examples of smooth maps in the exercise sheets.
12 Chapter 2. Smooth Maps

Example 2.12. Consider the unit n-sphere Sn ⊆ Rn+1 with its standard smooth struc-
ture, see Example 1.10(2). The inclusion map ι : Sn ,→ Rn+1 is continuous (inclusion map
of topological spaces). It is a smooth map, because its coordinate representation with
respect to any of the graph coordinates of Example 1.3(2) is

ι(u1 , . . . , un ) = ι ◦ (φ± −1

1
b i ) (u , . . . , un )
p
= u1 , . . . , ui−1 , ± 1 − |u|2 , ui , . . . , un ,


↓
i-th

which is smooth on its domain (the set where |u|2 < 1).
Definition 2.13. Let M and N be smooth manifolds. A diffeomorphism from M to N
is a smooth bijective map M → N that has smooth inverse. We say that M and N are
diffeomorphic if there exists a diffeomorphism between them.
Example 2.14.
(1) Consider the maps
x
F : Bn → Rn , x 7→ p
1 − |x|2
and
y
G : Rn → Bn , y 7→ p .
1 + |y|2
These maps are smooth, and it is straightforward to check that they are inverses to each
other. Thus, they are both diffeomorphisms, so Bn is diffeomorphic to Rn .
(2) If M is any smooth manifold and if (U, φ) is any smooth coordinate chart on M , then
the coordinate map φ : U → φ(U ) ⊆ Rn is a diffeomorphism. Indeed, it has an identity
map as a coordinate representation.
Proposition 2.15 (Properties of diffeomorphisms).
(a) Every composition of diffeomorphisms is a diffeomorphism.
(b) Every finite product of diffeomorphisms between smooth manifolds is a diffeomor-
phism.
(c) Every diffeomorphism is a homeomorphism and an open map.
(d) The restriction of a diffeomorphism to an open submanifold is a diffeomorphism
onto its image.
(e) “Diffeomorphic” is an equivalence relation on the class of all smooth functions.
Proof. Exercise! (See also Proposition 4.9.)
Just as two topological spaces are considered to be “the same” if they are homeo-
morphic, two smooth manifolds are essentially indistinguishable if they are diffeomorphic.
The central concern of smooth manifold theory is the study of properties of smooth man-
ifolds that they preserved by diffeomorphisms. The dimension is one such property (cf.
p. 1):
 Section 2.2. Partitions of Unity 13

Theorem 2.16 (Diffeomorphism invariance of the dimension). A non-empty smooth m-

manifold cannot be diffeomorphic to a non-empty smooth n-manifold unless m = n.

Proof. Let M be a non-empty smooth m-manifold, let N be a non-empty smooth n-

manifold, and assume that there exists a diffeomorphism F : M → N . Choose any point
p ∈ M and consider smooth charts (U, φ) for M containing p and (V, ψ) for N containing
F (p). Then (the restriction of the coordinate representation) Fb = ψ ◦ F ◦ φ−1 is a diffeo-
morphism from an open subset of Rm to an open subset of Rn . It is now a consequence
of the chain rule that m = n, see [Lee13, Proposition C.4].

2.2 Partitions of Unity

We briefly discuss here partitions of unity, which are tools for “blending together” local
smooth objects into global ones without necessarily assuming that they agree on overlaps
(cf. Lemma 2.10). They are indispensable in smooth manifold theory, and we will soon see
some first applications of partitions of unity. For further information we refer to [Lee13,
Chapter 2, Partitions of Unity].

Definition 2.17. Let M be a topological space and let f : M → Rk be a function. The

support of f is defined as

supp f = {p ∈ M | f (p) ̸= 0}.

Moreover,

 if supp f is contained in some open subset U ⊆ M , then we say that f is supported

in U ;

 if supp f is a compact set (e.g., if M is a compact space), then we say that f is

compactly supported .

Definition 2.18. Let M be a topological space and let X = (Xα )α∈A be an open cover
of M , indexed by a set A. A partition of unity subordinate to X is an indexed family
(ψα )α∈A of continuous functions ψα : M → R with the following properties:

(i) 0 ≤ ψα (x) ≤ 1, ∀ α ∈ A, ∀ x ∈ M .

(ii) supp ψα ⊆ Xα , ∀ α ∈ A.

(iii) The family of supports supp ψα α∈A is locally finite, i.e., every point p ∈ M has a
neighborhood Wp such that Wp ∩ supp ψα = ∅ for all but a finite number of α ∈ A.
X
(iv) ψα (x) = 1, ∀ x ∈ M .
α∈A

If now M is a smooth manifold, then a smooth partition of unity is one for which each
of the functions ψα is smooth.

Observe that, due to the local finiteness condition (iii), the sum in (iv) has only finitely
many non-zero terms in a neighborhood of each point, so there is no issue of convergence.
14 Chapter 2. Smooth Maps

Theorem 2.19 (Existence of smooth partitions of unity). Let M be a smooth manifold

and let X = (Xα )α∈A be an open cover of M . Then there exists a smooth partition of
unity subordinate to X.

Proof. For a detailed proof of the statement we refer to [Lee13, Theorem 2.23], see also
[Lee09, Theorem 1.73].

Comment: The hypothesis that M is second-countable is used implicitly in the proof of

Theorem 2.19 via the following characterization: If X is a locally Euclidean Hausdorff
space, then X is second-countable if and only if it is paracompact and has countably many
connected components, see [Lee13, Problem 1.5]. In particular, every topological manifold
is paracompact, see [Lee13, Theorem 1.15].

We finally present three interesting applications of partitions of unity.

1 Existence of smooth bump functions:

If M is a topological space, A ⊆ M is a closed subset and U ⊆ M is an open subset
such that A ⊆ U , a continuous function ψ : M → R is called a bump function for A
supported in U if

• 0 ≤ ψ(x) ≤ 1, ∀ x ∈ M ,

• ψ ≡ 1 on A, and

• supp ψ ⊆ U .

In other words, a bump function is a continuous real-valued function that is equal to 1

on a specified set and zero outside a specified neighborhood of that set.

Proposition 2.20. Let M be a smooth manifold. For every closed subset A ⊆ M and any
open subset U ⊆ M containing A, there exists a smooth bump function for A supported
in U .

Proof. Set U0 := U and U1 := M \ A, and let {ψ0 , ψ1 } be a smooth partition of unity

subordinate
P1 to the open cover {U0 , U1 } of M . Since ψ1 ≡ 0 on A, and therefore ψ0 =
i=0 ψi ≡ 1 on A, the function ψ0 has the required properties.

Exercise 2.21: If M is a smooth manifold of dimension n ≥ 1, then the vector space

C ∞ (M ) is infinite-dimensional.
[Hint: Show that if f1 , . . . , fk are elements of C ∞ (M ) with non-empty disjoint supports,
then they are linearly independent.]

Solution: Assume first that there is a countable collection F of smooth functions on

M with non-empty disjoint supports. Pick an integer k ≥ 1. We will show that any k
elements f1 , . . . , fk ∈ F are linearly independent. To this end, write

λ1 f1 + . . . + λk fk = 0 (2.1)

for some λi ∈ R. For each i ∈ {1, . . . , k}, pick x ∈ supp(fi ) such that fi (x) ̸= 0, and
note that fj (x) = 0 for every j ∈ {1, . . . , k} \ {i} by assumption. By evaluating (2.1)
 Section 2.2. Partitions of Unity 15

at the chosen point x, we obtain λi fi (x) = 0, which implies λi = 0. This shows that
f1 , . . . , fk ∈ F are linearly independent, as claimed.
We will now show that there exists a countable collection of smooth functions on M
with non-empty disjoint supports, which in turn implies that the R-vector space C ∞ (M )
is infinite-dimensional, as desired. Fix a point p ∈ M and consider a smooth coordinate
chart (U, φ) containing p. In view of [Exercise Sheet 1, Exercise 1] and by further shrinking
U , we may assume that U is a smooth coordinate cube, i.e.,

φ(U ) = (0, 1) × . . . × (0, 1) ⊆ Rn .

For each integer i ≥ 1, consider the open subset

 1 1
Bi := (0, 1) × . . . × (0, 1) × , ⊆ φ(U )
i+1 i

and pick any non-empty closed subset Ai of Bi . Since φ : U → φ(U ) is a homeomorphism,

by Proposition 2.20 for every i ≥ 1 there exists a smooth bump function fi ∈ C ∞ (M ) for
φ−1 (Ai ) supported in φ−1 (Bi ). Since Bi ∩ Bj = ∅ whenever i ̸= j, we also have

supp(fi ) ∩ supp(fj ) = ∅ for i ̸= j.

Therefore, the family (fi )∞

i=1 is a countable collection of smooth functions on M with
non-empty disjoint supports. This completes the proof of the above assertion.

2 Extension lemma for smooth functions:

In view of Lemma 2.10 it is often possible to construct smooth maps by “gluing
together” maps defined on open subsets. However, one cannot expect to “glue together”
smooth maps defined on closed subsets and obtain a smooth result. For example, the two
functions

f+ : [0, +∞) → R, x 7→ +x,

f− : (−∞, 0] → R, x →7 −x,

are both smooth and agree at the point 0 where they overlap, but the continuous function
f : R → R, x 7→ |x| that they define is clearly not smooth at the origin. Our second
application of partitions of unity is an important result concerning the possibility of
extending smooth functions from closed sets.
Let M and N be smooth manifolds and let A ⊆ M be an arbitrary subset. We say
that a map F : A → N is smooth on A if it admits a smooth extension in a neighborhood
of each point; namely, if for every p ∈ A there exists an open subset W ⊆ M containing
p and a smooth map Fe : W → N whose restriction to W ∩ A agrees with F .

Lemma 2.22 (Extension lemma for smooth functions). Let M be a smooth manifold, let
A ⊆ M be a closed subset, and let f : A → Rk be a smooth function. For any open subset
U ⊆ M containing A, there exists a smooth function fe: M → Rk such that fe|A = f and
supp fe ⊆ U .
16 Chapter 2. Smooth Maps

Proof. For each p ∈ A choose an open neighborhood Wp of p and a smooth function

fep : M → Rk such that
fep |Wp ∩A = f. (2.2)
Replacing Wp by Wp ∩ U , we may assume that Wp ⊆ U . Observe that the family of sets
{Wp }p∈A ∪ (M \ A) is an open cover of M . Let {ψp }p∈A ∪ {ψ0 } be a smooth partition of
unity subordinate to this cover, with supp ψp ⊆ Wp and supp ψ0 ⊆ M \ A.
For each p ∈ A, the product ψp fep is smooth on Wp , and has a smooth extension to
all of M if we interpret it to be zero on M \ supp ψp . (The extended function is smooth
because the two definitions agree on the open subset Wp \ supp ψp where they overlap.)
Thus, we can define the function
X
fe: M → Rk , x 7→ ψp (x) fep (x).
p∈A

Since the collection of supports {supp ψp }p∈A is locally finite, the sum actually has only
finitely many zero terms in a neighborhood of any point of M , and therefore defines a
smooth function. If x ∈ A, then ψ0 (x) = 0 by construction and fep (x) = f (x) for each p
such that ψp (x) ̸= 0 by (2.2), so
X  X 
fe(x) = ψp (x) fep (x) = ψ0 (x) + ψp (x) f (x) = f (x).
p∈A p∈A

Thus, fe is indeed an extension of f . Finally, we have

[ [
supp fe ⊆ supp ψp = supp ψα ⊆ U,
p∈A p∈A

where the equality in the middle is a property of locally finite collections, see [Lee13,
Lemma 1.13].
Comments:
(1) The conclusion of the extension lemma can be false if A is not closed; see [Lee13,
Exercise 2.27].
(2) The assumption in the extension lemma that the codomain is Rk , and not some other
manifold, is necessary: for other codomains, extensions can fail to exist for topological
reasons.
3 Closed subsets as level sets: The next result tells us that every closed subset of
a smooth manifold can be expressed as a level set of some smooth real-valued function.
This remarkable fact will not be used anywhere in these notes, so we omit its proof and
we refer to [Lee13, Theorem 2.29] for the details.
Theorem 2.23. Let M be a smooth manifold. If K is a closed subset of M , then there
exists a smooth non-negative function f : M → R such that f −1 (0) = K.
Exercise 2.24: Let A and B be disjoint closed subsets of a smooth manifold M . Show
that there exists f ∈ C ∞ (M ) such that 0 ≤ f (x) ≤ 1 for all x ∈ M , f −1 (0) = A and
f −1 (1) = B.
 Section 2.2. Partitions of Unity 17

Solution: By Theorem 2.23 there exist non-negative smooth functions fA and fB on M

such that
fA−1 (0) = A and fB−1 (0) = B. (2.3)
Consider now the function
fA (x)
f : M → R, x 7→
fA (x) + fB (x)

and observe that it is well-defined (that is, fA (x) + fB (x) ̸= 0 for all x ∈ M ) due to (2.3)
and since A ∩ B = ∅. Moreover, f is smooth as a quotient of smooth functions, and it
satisfies
0 ≤ f (x) ≤ 1 for all x ∈ M,
since fA and fB are non-negative. Finally, it follows from (2.3) that

f −1 (0) = A and f −1 (1) = B.

Hence, f ∈ C ∞ (M ) has the desired properties.

CHAPTER 3
THE TANGENT BUNDLE

3.1 Tangent Vectors

Given a point a ∈ Rn , we define the geometric tangent space to Rn at a to be the set

Rna := {a} × Rn = (a, v) v ∈ Rn .



A geometric tangent vector in Rn is an element of Rna for some a ∈ Rn . We abbreviate

(a, v) as va or v|a , and we think of va as the vector v with initial point at a.

Figure 3.1: Geometric tangent space

The set Rna is an R-vector space under the natural operations

va + wa := (v + w)a ,
λ va := (λ v)a .

The vectors ei |a , 1 ≤ i ≤ n, (where ei denotes the i-th standard basis vector of Rn ) are a
basis of Rna . In fact, Rna is essentially the same as Rn itself; the only reason why we add
the index a is so that the geometric tangent spaces Rna and Rnb at distinct points a and b
are disjoint sets.
Geometric tangent vectors provide a means of taking directional derivatives of func-
tions. For example, any geometric tangent vector v ∈ Rna yields a map

Dv a : C ∞ (Rn ) → R
d
f 7→ Dv a f = Dv f (a) = f (a + tv),
dt t=0

the directional derivative of f in the direction v at a. This operation is R-linear and

satisfies the product rule:

Dv a (f g) = f (a) Dv a g + g(a) Dv a f.

19
20 Chapter 3. The Tangent Bundle

If va = v i ei |a in terms of the standard basis, then by the chain rule Dv a f can be written
more concretely as
∂f
Dv a f = v i i (a). (3.1)
∂x
In particular, if va = ej |a , then
∂f
Dej a f = (a). (3.2)
∂xj
With this construction in mind, we make the following definition.
Definition 3.1. Given a ∈ Rn , a map w : C ∞ (Rn ) → R is called a derivation at a if it is
R-linear and satisfies the product rule:

w(f g) = f (a) w(g) + g(a) w(f ).

We denote by Ta Rn the set of all derivations of C ∞ (Rn ) at a. Clearly, Ta Rn is an

R-vector space under the natural operations

(w1 + w2 )(f ) := w1 f + w2 f,
(λw)f := λ wf.

The most important fact about Ta Rn is that it is finite-dimensional; in fact, it is

naturally isomorphic to the geometric tangent space Rna that we defined above. The proof
will be based on the following lemma.
Lemma 3.2 (Properties of Derivations). Let a ∈ Rn , w ∈ Ta Rn and f, g ∈ C ∞ (Rn ).
(a) If f is constant, then wf = 0.

(b) If f (p) = g(p) = 0, then w(f g) = 0.

Proof.
(a) Consider the constant function f1 ≡ 1 ∈ C ∞ (Rn ). By the product rule we obtain

*1 *1
wf1 = w(f1 · f1 ) = 
f1
(a) wf1 + 
f1
(a) wf1 = 2wf1 ,
 

which implies that wf1 = 0. Now, since f ≡ c is constant, by linearity we obtain

wf = w(cf1 ) = c wf1 = 0.

(b) Follows immediately from the product rule.

Proposition 3.3. Let a ∈ Rn .
(a) For each geometric tangent vector va ∈ Rna , the map

Dv a : C ∞ (Rn ) → R
d
f 7→ Dv a f = Dv f (a) = f (a + tv)
dt t=0

(directional derivative of f in the direction v at a) is a derivation of C ∞ (Rn ) at a.

 Section 3.1. Tangent Vectors 21

(b) The map

Φ : Rna → Ta Rn
v 7→ Dv a

is an R-linear isomorphism.

(c) The n derivations

∂ ∂
,...,
∂x1 a ∂xn a

defined by
∂ ∂f
i
f := (a) , 1 ≤ i ≤ n,
∂x a ∂xi
form a basis of Ta Rn , and thus

dimR Ta Rn = n.

Proof.
(a) Easy to check (using calculus).
(b) Linearity: For every f ∈ C ∞ (Rn ) we have

Φ(λ1 v1 + λ2 v2 )(f ) = Dλ1 v1 +λ2 v2 a (f )

d

= f a + t(λ1 v1 + λ2 v2 )
dt t=0

= Df (a) · (λ1 v1 + λ2 v2 )
d d
= λ1 f (a + tv1 ) + λ2 f (a + tv2 )
dt t=0 dt t=0

= λ1 Φ(v1 )(f ) + λ2 Φ(v2 )(f )

= λ1 Φ(v1 ) + λ2 Φ(v2 ) (f ),

which shows the R-linearity of Φ.

Injectivity: Suppose that Φ(va ) = Dv |a = 0 is the zero derivation. Writing va = v i ei |a
in terms of standard basis and considering the j-th coordinate function xj : Rn → R,
thought of as a smooth function on Rn , we obtain:

(3.1) ∂
0 = Dv a xj === v i (xj ) = vj ,
∂xi x=a

∂xj ∂xi
where the last equality follows because ∂xi
= 0 for i ̸= j, and ∂xi
= 1. Hence, va = 0 ∈ Rna .
Surjectivity: Let w ∈ Ta Rn . Set v := v i ei |a ∈ Rna , where v i = w(xi ) ∈ R. We will
show that w = Φ(v) = Dv |a , To this end, let f ∈ C ∞ (Rn ). By Taylor’s theorem [Lee13,
22 Chapter 3. The Tangent Bundle

Theorem C.15] we can write

n
X ∂f
f (x) = f (a) + i
(a) (xi − ai ) +
i=1
∂x
n 1
∂ 2f
X Z
i i j j


+ (x − a ) (x − a ) (1 − t) a + t(x − a) dt .
i,j=1 0 ∂xi ∂xj
| {z }

Note that each term in the last sum above is a product of two smooth functions of x that
vanish at x = a: one is (xi − ai ) and the other is (xj − aj ) · (integral). (The integral is a
smooth function of x by iterative application of [Lee13, Theorem C.14].) By Lemma 3.2(b)
the derivation w annihilates this entire sum. Thus, thanks to the R-linearity of w, we
obtain
n  

 0 X ∂f i i
(a) (x − a )
:
wf =  wf (a) +
 w i
i=1
∂x
n i
X ∂f *v
i *
0
i
= (a) w(x ) −  )
w(a
  

i=1
∂xi 

n
X ∂f
= vi (a) = Dv |a f.
i=1
∂xi

∂
(c) By (3.2) we know that ∂xi
= Dei |a . Hence, (c) follows immediately from (b).

Definition 3.4. Let M be a smooth manifold and let p ∈ M . A map v : C ∞ (M ) → R is

called a derivation at p if it is R-linear and satisfies the product rule:

v(f g) = f (p) v(g) + g(p) v(f ), ∀ f, g ∈ C ∞ (M ).

We denote by Tp M the set of all derivations of C ∞ (M ) at p. Clearly, Tp M is an R-vector

space, called the tangent space to M at p ∈ M . An element of Tp M is called a tangent
vector at p.

Lemma 3.5. Let M be a smooth manifold, p ∈ M , v ∈ Tp M and f, g ∈ C ∞ (M ).

(a) If f is constant, then vf = 0.

(b) If f (p) = g(p) = 0, then v(f g) = 0.

Proof. Exercise! (cf. Lemma 3.2)

With the motivation of geometric tangent vectors in Rn in mind, we visualize tangent

vectors to M as “arrows” that are tangent to M and whose base points are attached to
M at the given point. For alternative descriptions of tangent vectors to M , see [Exercise
Sheet 4, Exercise 4], Section 3.5 and [Exercise Sheet 4, Exercise 5].
 Section 3.2. The Differential of a Smooth Map 23

v
p Tp M

Figure 3.2: Tangent vector to a manifold at a point

3.2 The Differential of a Smooth Map

Definition 3.6. If F : M → N is a smooth map, then for each p ∈ M we define a map

dFp : Tp M → TF (p) N,

called the differential (or tangent map) of F at p, as follows. Given v ∈ Tp M , we let

dFp (v) be the derivation at F (p) that acts on f ∈ C ∞ (N ) by

dFp (v)(f ) = v(f ◦ F ) .

The operator dFp (v) : C ∞ (N ) → R is a derivation at F (p). Indeed, it is R-linear, since v

is so, and it satisfies the product rule:



dFp (v)(f g) = v (f g) ◦ F = v (f ◦ F )(g ◦ F )
= (f ◦ F )(p) v(g ◦ F ) + (g ◦ F )(p) v(f ◦ F )



= f F (p) dFp (v)(g) + g F (p) dFp (v)(f ).

Proposition 3.7 (Properties of differentials). Let F : M → N and G : N → P be smooth

maps and let p ∈ M .
(a) dFp : Tp M → TF (p) N is an R-linear map.
(b) d(G ◦ F ) = dGF (p) ◦ dFp : Tp M → T(G◦F )(p) P .
(c) d(IdM )p = IdTp M : Tp M → Tp M .
(d) If F is a diffeomorphism, then dFp : Tp M → TF (p) N is an isomorphism, and it
holds that (dFp )−1 = d(F −1 )F (p) .
Proof. See [Exercise Sheet 4, Exercise 1].
Our first important application of the differential will be to use coordinate charts to
relate the tangent space to a point on a smooth manifold with the Euclidean tangent
space. But there is an important technical issue that we must address first. While the
tangent space is defined in terms of smooth functions on the whole manifold, coordinate
charts are in general defined only on open subsets. The key point, expressed in the next
proposition, is that tangent vectors act locally.
Proposition 3.8. Let M be a smooth manifold, p ∈ M and v ∈ Tp M . If f, g ∈ C ∞ (M )
agree on some neighborhood of p, then vf = vg.
24 Chapter 3. The Tangent Bundle

Proof. Set h := f − g and observe that h is a smooth function on M that vanishes in

a neighborhood U of p. By Proposition 2.20 there exists a smooth bump function ψ for
supp h supported in M \ {p} (open subset of M which contains supp h, since h(x) = 0
for all x ∈ U ). Since ψ ≡ 1 where h is non-zero, the product ψh is identically equal to h.
Since h(p) = ψ(p) = 0, by Lemma 3.5(b) we obtain v(h) = v(ψh) = 0, so v(f ) = v(g) by
linearity.
Using Proposition 3.8, we can identify the tangent space to an open submanifold with
the tangent space to the whole manifold.
Proposition 3.9 (The tangent space to an open submanifold). Let M be a smooth man-
ifold, let U ⊆ M be an open subset and let ι : U ,→ M be the inclusion map. For every
p ∈ U , the differential dιp : Tp U → Tp M is an isomorphism.
Proof. Recall that U is a smooth manifold by Example 1.10(4) and that ι is a smooth
map by Proposition 2.11(c). Fix now p ∈ U , consider the differential dιp : Tp U → Tp M ,
and let V be a neighborhood of p such that V ⊆ U .
Injectivity: Let v ∈ Tp U such that dιp (v) = 0 ∈ Tp M . If f ∈ C ∞ (U ) is arbitrary,
then by Lemma 2.22 there exists fe ∈ C ∞ (M ) such that fe|V = f . Since then f and fe|U
are smooth functions on U that agree in a neighborhood of p, Proposition 3.8 implies


 
vf = v fe U = v fe ◦ ι = dιp (v) fe = 0.

Hence, v = 0 ∈ Tp U , so dιp is injective.

Surjectivity: Let w ∈ Tp M . Define

v : C ∞ (U ) → R, f 7→ wfe,

where fe is any smooth function on M that agrees with f on V , see Lemma 2.22. By
Proposition 3.8, vf is independent of the choice of fe, so v is well-defined, and it is easy
to check that it is a derivation of C ∞ (U ) at p. For any g ∈ C ∞ (M ) we have

dιp (v)(g) = v(g ◦ ι) = w(gg

◦ ι) = wg,

where the last equality follows from the fact that g ◦ ι, gg

◦ ι and g all agree on V . Hence,
dιp (v) = w, and thus dιp is surjective.
Given an open subset U ⊆ M , the isomorphism dιp from Tp U to Tp M is canonically
defined, independent of any choices. From now on we identify Tp U with Tp M for any
p ∈ U . This identification just amounts to the observation that dιp (v) is the same
derivation as v, though of as acting on functions on the bigger manifold M instead of
on functions on U . Since the action of a derivation on a function depends only on the
values of the function in an arbitrarily small neighborhood (see Proposition 3.8), this is
a harmless identification. In particular, this means that any tangent vector v ∈ Tp M
can be unambiguously applied to the functions defined only in a neighborhood of p, not
necessarily on all of M .
Proposition 3.10. If M is a smooth n-manifold, then for each p ∈ M , the tangent space
Tp M is an n-dimensional R-vector space.
 Section 3.3. Computations in Local Coordinates 25

Proof. Fix p ∈ M and let (U, φ) be a smooth coordinate chart containing p. Since
φ: U → U b ⊆ Rn is a diffeomorphism by Example 2.14(2), dφp : Tp U → Tφ(p) U b is an
∼
isomorphism by Proposition 3.7(d). Since Proposition 3.9 guarantees that Tp U = Tp M
b∼
and Tφ(p) U = Tφ(p) Rn , it follows from Proposition 3.3(c) that

dim Tp M = dim Tφ(p) Rn = n.

3.3 Computations in Local Coordinates

Next, we will show how to do computations with tangent vectors and differentials in local
coordinates.
Let M be a smooth manifold and let (U, φ) be a smooth coordinate chart on M . Then
φ is a diffeomorphism from U to an open subset Ub ⊆ Rn . By Propositions 3.7(d) and 3.9
n
we deduce that dφp : Tp M → Tφ(p) R is an R-linear isomorphism (for each p ∈ U ). By
Proposition 3.3(c) the derivations

∂ ∂
,...,
∂x1 φ(p) ∂xn φ(p)

form a basis of Tφ(p) Rn . Therefore, the preimages of these vectors under the isomorphism
dφp , denoted by ∂x∂ i p , form a basis of Tp M . These vectors are characterized by
   
∂ −1 ∂ 3.7(d) −1 ∂
= (dφp ) === d(φ )φ(p) . (3.3)
∂xi p ∂xi φ(p) ∂xi φ(p)

Unwinding the definitions, we see that ∂

∂xi p
acts on a function f ∈ C ∞ (U ) by

∂ ∂ ∂ fb
i
f= (f ◦ φ−1 ) = (b
p),
∂x p ∂xi φ(p) ∂xi

where fb := f ◦ φ−1 is the coordinate representation of f and pb = (p1 , . . . , pn ) = φ(p)

is the coordinate representation of p. In other words, ∂x∂ i p is the derivation at p that
takes the i-th partial derivative of (the coordinate representation of) f at (the coordinate
representation of) p. The vectors ∂x∂ i p are called the coordinate vectors at p associated
with the given coordinate system. In the special case of standard coordinates on Rn , the
vectors ∂x∂ i p are literally the partial derivative operators
To summarize, if M is a smooth n-manifold and if p ∈ M , then
Tp M is an n-
dimensional R-vector space, ∂ and for any smooth coordinate chart U, (xi ) containing p,
n
the coordinate vectors ∂xi p i=1 form a basis for Tp M . Thus, a tangent vector v ∈ Tp M
can be written uniquely as a linear combination
∂
v = vi .
∂xi p

The ordered basis ∂x∂ i p is called a coordinate basis for Tp M and the numbers (v i ) are

called the components of v with respect to the coordinate basis. If v is known, then its
26 Chapter 3. The Tangent Bundle

components can be easily computed from its action on the coordinate functions. For each
j ∈ {1, . . . , n}, the components of v are given by v j = v(xj ) (where we think of xj as a
smooth real-valued function on U ), because

j
 
j i ∂ j i ∂x
v(x ) = v (x ) = v (p) = v j .
∂xi p ∂xi

We now explore how differentials look in coordinates. We begin by considering the case
of a smooth map F : U ⊆ Rn → V ⊆ Rm between open subsets of Euclidean spaces. For
any p ∈ U we will determine the matrix of dFp : Tp Rn → TF (p) Rm in terms of the standard
coordinate bases. Denoting by (x1 , . . . , xn ) (respectively (y 1 , . . . , y m )) the coordinates in
the domain (respectively codomain), we use the chain rule to compute the action of dFp
on a typical basis vector as follows:

∂F j
 
∂ ∂ ∂f
dFp f= (f ◦ F ) = F (p) (p)
∂xi p ∂xi p ∂y j ∂xi
 j 
∂F ∂
= (p) j f.
∂xi ∂y F (p)

Thus,
∂F j
 
∂ ∂
dFp = (p) . (3.4)
∂xi p ∂xi ∂y j F (p)

In other words, the matrix of dFp in terms of the coordinate bases is

 
∂F 1 ∂F 1
 ∂x1 (p) · · · ∂xn
(p) 
.. ..

..

. . . ,
 

 
 ∂F m ∂F m 
(p) · · · (p)
∂x1 ∂xn

that is, the Jacobian matrix of F at p, which is the matrix representation of the total
derivative DF (p) : Rn → Rm . Therefore, in this case, dFp : Tp Rn → TF (p) Rm corresponds
to the total derivative DF (p) : Rn → Rm , under the usual identification of Euclidean
spaces with their tangent spaces.
We now consider the more general case of a smooth map F : M → N between two
smooth manifolds. Choosing smooth coordinate charts (U, φ) for M containing p and
(V, ψ) for N containing F (p), we obtain the coordinate representation

Fb = ψ ◦ F ◦ φ−1 : φ U ∩ F −1 (V ) → ψ(V )

see Figure 3.3, and we also denote by pb = φ(p) the coordinate representation of p. By the
computation above, dFbpb is represented with respect to the standard coordinates bases by
 Section 3.3. Computations in Local Coordinates 27

the Jacobian matrix of Fb at pb. Using the fact that F ◦ φ−1 = ψ −1 ◦ Fb, we compute
    
∂ dfn −1 ∂
dFp == dFp d(φ )pb
∂xi p ∂xi pb
 
Prop. −1 ∂
=== d(F ◦ φ )pb
3.7(b) | {z } ∂xi pb
ψ −1 ◦Fb
  
Prop. −1 ∂
=== d(ψ )Fb(bp) dFbpb
3.7(b) ∂xi pb
 bj 
(3.4) −1 ∂F ∂
== d(ψ )Fb(bp) (b
p) j
∂xi ∂y pb

dfn and ∂ Fbj ∂

===== i
(b
p) j . (3.5)
linearity ∂x ∂y F (p)

Thus, dFp is represented in coordinate bases by the Jacobian matrix of (the coordinate
representation of) F . In fact, the definition of the differential was cooked up precisely in
order to give a coordinate-independent meaning to the Jacobian matrix.

Figure 3.3: The differential in coordinates

Finally, suppose that U, φ = (xi ) and V, ψ = (e xi ) are two smooth charts on M
and that p ∈ U ∩ V . Any
tangent
vector at p can be represented with respect to either
coordinates basis ∂x∂ i p or ∂e∂xi p . How are the two representations related?

Figure 3.4: Change of coordinates

In this situation it is customary to write the transition map ψ ◦ φ−1 : φ(U ∩ V ) →

ψ(U ∩ V ) in the following notation:

ψ ◦ φ−1 (x) = x
e1 (x), . . . , x
en (x) .

Here we are indulging in a typical abuse of notation: in the expression x ei (x) we think
ei as a coordinate function (whose domain is an open subset of M , identified with an
of x
open subset of Rn ), but we think of x as representing a point (in this case, in φ(U ∩ V )).
By (3.4) we have

xj
 
−1 ∂ ∂e
∂
d(ψ ◦ φ )φ(p) = φ(p) .
∂xi φ(p) ∂xi xj
∂e ψ(p)
28 Chapter 3. The Tangent Bundle

Using the definition of coordinate vectors, we obtain

 
∂ (3.3) −1 ∂
== d(φ )φ(p)
∂xi p ∂xi φ(p)
 
Prop. −1 −1 ∂
=== d(ψ )ψ(p) · d(ψ ◦ φ )φ(p)
3.7(b) ∂xi φ(p)

xj
 
−1 ∂e
∂
= d(ψ )ψ(p) φ(p)
∂xi xj
∂e ψ(p)

(3.3) xj
∂e
∂
===== i
φ(p) . (3.6)
linearity ∂x |{z} ∂e xj p
= pb

(This formula looks exactly the same as the chain rule for partial derivatives in Rn .)
Applying this to the components of a vector
∂ ∂
v = vi = vej ,
∂xi p xj
∂e p

we find that the components of v transform by the rule

xj
∂e
vej = p) v i .
(b (3.7)
∂xi

3.4 The Tangent Bundle

Definition 3.11. Let M be a smooth manifold. The tangent bundle of M is denoted by
TM and is defined as the disjoint union of the tangent spaces at all points of M :
G
TM = Tp M.
p∈M

We usually write an element of this disjoint union as an ordered pair (p, v) with p ∈ M
and v ∈ Tp M ; we sometimes also write vp for (p, v). The tangent bundle comes equipped
with a natural projection map π : TM → M , which sends each vector in Tp M to the
point p at which is tangent: (p, v) 7→ p.
For example, when M = Rn , using Proposition 3.3, we see that the tangent bundle of
n
R can be canonically identified with the disjoint union of its geometric tangent spaces,
which in turn is just the Cartesian product of Rn with itself:

Tp Rn ∼
G G G
T(Rn ) = = Rnp = {p} × Rn = Rn × Rn .
p∈Rn p∈Rn p∈Rn

An element of this Cartesian product can be thought of as representing either the geo-
metric tangent vector vp or the derivation Dv |p defined in Proposition 3.3(a). In general,
however, the tangent bundle of a smooth manifold cannot be identified in a natural way
with a Cartesian product, because there is no canonical way to identify tangent spaces at
distinct points with each other.
 Section 3.4. The Tangent Bundle 29

The tangent bundle of a smooth manifold can be thought of simply as a disjoint union
of vector spaces, but it is much more than that. The next proposition shows that the
tangent bundle of a smooth manifold can be considered as a smooth manifold in its own
right. For its proof we will use Lemma 1.11.

Proposition 3.12. For any smooth n-manifold M , the tangent bundle TM has a natural
topology and smooth structure that make it into a smooth (2n)-manifold. With respect to
this structure, the projection π : TM → M is smooth.

Proof. We begin by defining the maps that will become our smooth charts. Given any
smooth chart (U, φ) for M , observe that π −1 (U ) is the set of all tangent vectors to M at
all points of U . Denote by (x1 , . . . , xn ) the coordinate functions of φ, and define a map

e : π −1 (U ) → R2n ,
φ
 
i ∂ 1 n 1 n


φ
e v = x (p), . . . , x (p), v , . . . , v . (3.8)
∂xi p

Its image is the set φ(U ) × Rn , which is an open subset of R2n . It is a bijection onto its
image, because its inverse can be explicitly written as

∂
e−1 x1 , . . . , xn , v 1 , . . . , v n = v i i


φ .
∂x φ−1 (x)

Now, suppose that we are given two smooth charts (U, φ) and (V, ψ) for M , and
consider the corresponding “charts” π −1 (U ), φ e and π −1 (V ), ψe for TM . The sets

e π −1 (U ) ∩ π −1 (V ) = φ(U ∩ V ) × Rn


φ

and
ψe π −1 (U ) ∩ π −1 (V ) = ψ(U ∩ V ) × Rn

are open in R2n , and the transition map

e−1 : φ(U ∩ V ) × Rn → ψ(U ∩ V ) × Rn

ψe ◦ φ

can be written explicitly as

 
−1

1 n 1 n

i ∂
ψe ◦ φ x ,...,x ,v ,...,v = ψe v
∂xi φ−1 (x)
e

xj  ∂
 
(3.7) ∂e
== ψ v i i
e
∂x ∂e xj φ−1 (x)
 x1
∂e xn 
∂e
= xe1 , . . . , x
en , i v i , . . . , i v i ,
∂x ∂x
which is clearly smooth.
Choosing a countable cover {Ui } of M by smooth coordinate domains, we obtain a
countable cover of TM by coordinate domains {π −1 (Ui )} satisfying conditions (i)-(iv) of
Lemma 1.11. To check the Hausdorff condition (v), just note that any two points in the
30 Chapter 3. The Tangent Bundle

same fiber of π lie in one chart, while if (p, v) and (q, w) lie in different fibers, there exist
disjoint smooth coordinate domains U and V for M such that p ∈ U and q ∈ V , and then
π −1 (U ) and π −1 (V ) are disjoint coordinate neighborhoods containing (p, v) and (q, w),
respectively. This completes the proof of the first part of the statement.

π : TM → M is smooth, note that with respect
Finally, to check that to charts (U, φ)
for M and π −1 (U ), φ e for TM , its coordinate representation φ◦π ◦ φ e−1 is π
b(x, v) = x.

The coordinates (xi , v i ) given by (3.8) are called natural coordinates on TM .

Proposition 3.13. If M is a smooth n-manifold which can be covered by a single smooth

chart, then its tangent bundle TM is diffeomorphic to M × Rn .

Proof. If (U, φ) is a global smooth chart for M , then φ is, in particular, a diffeomor-
phism from U = M to an open subset U b ⊆ Rn , see Example 2.14(2). The proof of
Proposition 3.12 showed that the natural coordinate chart φe is a bijection from TM to
n
U × R , and the smooth structure on TM is defined essentially by declaring φ
b e to be
diffeomorphism.

Comment: In general, the tangent bundle is not globally diffeomorphic (or ever homeo-
morphic) to a product of the manifold with Rn .

Let F : M → N be a smooth map. By putting together the differentials of F at all

points of M , we obtain a globally defined map between tangent bundles, called the global
differential and denoted by dF : TM → TN . This is just the map whose restriction to
each tangent space Tp M ⊆ TM is dFp .
One important feature of the smooth structure we have defined on the tangent bundle is
that it makes the differential of a smooth map into a smooth map between tangent bundles;
namely, if F : M → N is a smooth map, then its global differential dF : TM → TN is
also a smooth map, see [Exercise Sheet 5, Exercise 4].

Proposition 3.14 (Properties of the global differential). Let F : M → N and G : N → P

be smooth maps. The following statements hold:

(a) d(G ◦ F ) = dG ◦ dF : TM → TP .

(b) d(IdM ) = IdTM : TM → TM .

(c) If F is a diffeomorphism, then dF : TM → TN is also a diffeomorphism, and it

holds that (dF )−1 = d(F −1 ).

Proof. See [Exercise Sheet 5, Exercise 4].

3.5 Velocity Vectors of Curves

Recall that a continuous (parametrized) curve in a topological manifold M is a continuous
map γ : J → M , where J ⊆ R is an interval. Our definition of tangent spaces leads to a
natural interpretation of velocity vectors.

Definition 3.15. Let M be a smooth manifold.

 Section 3.5. Velocity Vectors of Curves 31

(a) A smooth (parametrized) curve in M is a smooth map γ : J → M , where J ⊆ R is

an interval.

(b) Given a smooth curve γ : J → M in M and an instant t0 ∈ J, the velocity of γ at

t0 is defined to be the tangent vector
 
′ d
γ (t0 ) := dγ ∈ Tγ(t0 ) M,
dt t=t0

where d/dt|t0 is the standard coordinate basis vector in Tt0 R. Other common nota-
tions for the velocity vector are:
dγ
γ̇(t0 ) and (t0 )
dt

Figure 3.5: Velocity vector of a curve

Assume that M , γ and t0 are as in Definition 3.15. The tangent vector γ ′ (t0 ) acts on
functions f ∈ C ∞ (M ) by
 
′ d d
γ (t0 )f = dγ f= (f ◦ γ) = (f ◦ γ)′ (t0 ).
dt t=t0 dt t=t0

In other words, γ ′ (t0 ) is the derivation at γ(t0 ) obtained by taking the derivative of a
function along γ. (If t0 is an endpoint of the interval J ⊆ R, this still holds, provided that
we interpret the derivative with respect to t as a one-sided derivative, or equivalently as
the derivative of any smooth extension of f ◦ γ to an open subset of R.)
Now, let (U, φ) be a smooth chart for M with coordinate functions (xi ). If γ(t0 ) ∈ U ,
then we can write the coordinate representation of γ as

b(t) = γ 1 (t), . . . , γ n (t) ,

γ

at least for t ∈ J sufficiently close to t0 ∈ J, and then the coordinate formula for the
differential (3.5) yields
dγ i ∂
γ ′ (t0 ) = (t0 ) i .
dt ∂x γ(t0 )
This means that γ ′ (t0 ) is given by essentially the same formula as it would be in Euclidean
space: it is the tangent vector whose components in a coordinate basis are the derivatives
of the component functions of γ.
The next proposition shows that every tangent vector on a manifold is the velocity
vector of some curve. This gives a different and somewhat more geometric way to think
about the tangent bundle: it is just the set of all velocity vectors of smooth curves in M .
Proposition 3.16 (Tangent vectors as velocity vectors of smooth curves). Let M be
a smooth manifold. If p ∈ M , then for any v ∈ Tp M there exists a smooth curve
γ : (−ε, ε) → M such that γ(0) = p and γ ′ (0) = v.
Proof. See [Exercise Sheet 4, Exercise 5].
32 Chapter 3. The Tangent Bundle

Proposition 3.17 (The velocity of a composite curve). If F : M → N is a smooth map

and if γ : J → M is a smooth curve, then for any t0 ∈ J, the velocity at t = t0 of the
composite curve F ◦ γ : J → N is given by

(F ◦ γ)′ (t0 ) = dF γ ′ (t0 ) .

Proof. See [Exercise Sheet 4, Exercise 5].

Corollary 3.18 (Computing the differential using a velocity vector). If F : M → N is a

smooth map, p ∈ M and v ∈ Tp M , then

dFp (v) = (F ◦ γ)′ (0)

for any smooth curve γ : J → M such that 0 ∈ J, γ(0) = p and γ ′ (0) = v.

Proof. See [Exercise Sheet 4, Exercise 5].

CHAPTER 4
MAPS OF CONSTANT RANK

Because the differential of a smooth map is supposed to represent the “best linear ap-
proximation” to the map near a given point, we can learn a great deal about a map by
studying linear-algebraic properties of its differential. The most essential property of the
differential is its rank (the dimension of its image). In this chapter we study the ways
in which geometric properties of smooth maps can be detected from their differentials.
The maps for which differentials give good local models turn out to be the ones whose
differentials have constant rank.

4.1 Immersions, Submersions, and Embeddings

Definition 4.1. Given a smooth map F : M → N and a point p ∈ M , the rank of
F at p is defined to be the rank of the linear map dFp : Tp M → TF (p) N ; it is the
rank of the Jacobian matrix of F in any smooth chart, or the dimension of the image
Im(dFp ) ⊆ TF (p) N . If F has the same rank r at every point, then we say that it has
constant rank and we write rk F = r.
Note that the rank of F at each point is bounded above by min{dim M, dim N }. If
the rank of dFp is equal to this upper bound, then we say that F has full rank at p. If F
has full rank everywhere, then we say that F has full rank.
Definition 4.2. A smooth map F : M → N is called
(a) a smooth immersion if its differential is injective at each point or, equivalently, if
rk F = dim M ;

(b) a smooth submersion if its differential is surjective at each point or, equivalently, if
rk F = dim N ; and

(c) a smooth embedding if it is a smooth immersion that is also topological embedding,

i.e., a homeomorphism onto its image F (M ) ⊆ N in the subspace topology.
Smooth immersions and embeddings are essential ingredients in the theory of subman-
ifolds (see Chapter 5), while smooth submersions play a role in smooth manifold theory
closely analogous to the role played by quotient maps in topology (see Subsection 4.3.1

33
34 Chapter 4. Maps of Constant Rank

and [Lee13, Chapter 4, Smooth Covering Maps]). We will see that smooth immersions
and submersions behave locally like injective and surjective linear maps, respectively (see
Theorem 4.11).

Comment: A smooth embedding is a map that is both a topological embedding and

a smooth immersion, not just a topological embedding that happens to be smooth; see
Example 4.5(1).

Lemma 4.3. Let F : M → N be a smooth map. If dFp is injective (respectively surjective)

for some p ∈ M , then p has a neighborhood U such that F |U is an immersion (respectively
submersion).

Proof. If we choose any smooth coordinates for M near p and for N near F (p), either
hypothesis means that the Jacobian matrix of F in coordinates has full rank at p ∈ M .
By [Exercise Sheet 2, Exercise 3] we know that the set of n × m matrices of full rank is
an open subset of M(n × m, R) (where m = dim M and n = dim N ), so by continuity, the
Jacobian of F (in coordinates) has full rank in some neighborhood of p ∈ M .

Example 4.4.

(1) If γ : J → M is a smooth curve in a smooth manifold M , then γ is an immersion if

and only if γ ′ (t) ̸= 0 for all t ∈ J.

(2) If M is a smooth manifold and its tangent bundle TM is given the smooth manifold
structure described in Proposition 3.12, then the projection π : TM → M is a smooth
submersion. Indeed, we showed that with respect to any smooth local coordinates (xi ) on
an open subset U ⊆ M and the corresponding natural coordinates (xi , v i ) on π −1 (U ) ⊆
TM , the coordinate representation of π is π
b(x, v) = x, and thus


Jπb = Iddim M O ,

which has rank rk Jπb = dim M .

(3) If M is a smooth manifold and U ⊆ M is an open subset, then the inclusion map
U ,→ M is a smooth embedding, see Proposition 3.9.

We will encounter more examples of smooth immersions, smooth embeddings and

smooth submersions later in the course and in the exercise sheets as well.
To understand more fully what it means for a map to be a smooth embedding, it
is useful to bear in mind some examples of injective smooth maps that are not smooth
embeddings. The next three examples illustrate three rather different ways in which this
can happen.

Example 4.5.

(1) The map

γ : R → R2 , t 7→ (t3 , 0)
is a smooth map and a topological embedding, but it is not a smooth embedding, because
γ ′ (0) = 0, see Example 4.4(1).
 Section 4.1. Immersions, Submersions, and Embeddings 35

(2) The figure-eight curve: Consider the smooth curve

β : (−π, π) → R2 , t 7→ sin(2t), sin t .

Its image is a set that looks like a figure-eight in the plane, sometimes called a lemniscate,
see Figure 4.1. It is the locus of points (x, y) ∈ R2 such that x2 = 4y 2 (1 − y 2 ), as one can
easily check.

−1 0 1

−1

Figure 4.1: Lemniscate

Note that β is injective, since

β(t1 ) = β(t2 ) =⇒ t1 = t2 ,
and β ′ (t) ̸= 0 for all t ∈ (−π, π), since
2
∥β ′ (t)∥2 = = 4 cos2 (2t) + cos2 t ̸= 0.


2 cos(2t), cos t
Hence, β is an injective smooth immersion, but it is not a topological embedding, because
its image is compact in the subspace topology, while its domain is not.
(3) A dense curve on the torus: Let T2 = S1 ×S1 ⊆ C2 denote the torus, and let α ∈ R\Q.
The map
γ : R → T2 , t 7→ e2πi t , e2πi αt

is a smooth immersion, because γ ′ (t) never vanishes. It is also injective, because

γ(t1 ) = γ(t2 ) =⇒ t1 − t2 , αt1 − αt2 ∈ Z =⇒ t1 = t2 .
However, γ is not a topological embedding. Indeed, using Dirichlet’s approximation theo-
rem [Lee13, Lemma 4.21], one can show that γ(0) is a limit point of γ(Z) = {γ(n) | n ∈ Z},
while Z has no limit point in R. Finally, it can also be shown that γ(R) is dense in T2 ,
see [Lee13, Problem 4.4].
Recall : Let F : X → Y be a (not necessarily continuous) map between topological
spaces. We say that F is

(a) an open map if for every open subset U of X, the image F (U ) is an open subset
of Y ;

(b) a closed map if for every closed subset C of X, the image F (C) is a closed subset
of Y ;

(c) a proper map if for every compact subset K ⊆ Y , the preimage F −1 (K) is a
compact subset of X.
36 Chapter 4. Maps of Constant Rank

The following proposition gives a few simple sufficient criteria for an injective immer-
sion to be an embedding.

Proposition 4.6. Let F : M → N be an injective smooth immersion. If any of the

following holds, then F is a smooth embedding.

(a) F is an open map or a closed map.

(b) F is a proper map.

(c) M is compact.

(d) dim M = dim N .

Proof. We first prove the following three claims, which will be then used crucially in the
proof of the statement.
 Claim 1 : Let F : X → Y be a continuous map between topological spaces that is either
open or closed. If F is injective, then it is a topological embedding.
 Proof : Assume that F is open and injective. Then F : X → F (X) is bijective, so
F −1 : F (X) → F exists. If U ⊆ X is open, then (F −1 )−1 (U ) = F (U ) is open in Y by
hypothesis, and therefore also open in F (X) by definition of the subspace topology on
F (X). Hence, F −1 is continuous, so that F is a topological embedding.
The proof of the assertion is similar when F is closed and injective.
 Claim 2 (Closed map lemma): Let X be a compact space, let Y be a Hausdorff space,
and let F : X → Y be a continuous map. Then F is a closed map.
 Proof : Let K ⊆ X be a closed subset. Since X is compact, K is also compact, and
since F is continuous, F (K) is also compact. Since Y is Hausdorff, F (K) ⊆ Y is a closed
subset. Thus, F is a closed map.
 Claim 3 : Let X be a topological space and let Y be a locally compact Hausdorff space.
Then every proper continuous map F : X → Y is closed.
 Proof : Let K ⊆ X be a closed subset. To show that F (K) ⊆ Y is closed, we will
show that its complement is open. Let y ∈ Y \ F (K). Since Y is locally compact, y
has an open neighborhood V with compact closure V , and since F is proper, F −1 ( V ) is
compact. Set E := K ∩ F −1 ( V ) and note that E is a compact set. Since F is continuous,
F (E) is also compact, and since
Y is Hausdorff, F (E) is a closed subset of Y . Set
U := V \ F (E) = V ∩ Y \ F (E) and observe that U is open neighborhood of y, which
is disjoint from F (K). Hence, Y \ F (E) is open, which implies that F (K) is closed.
We are now ready to prove the statement.
(a) By assumption and by Claim 1, F is a topological embedding. Since it is also a smooth
immersion by assumption, we conclude that F is a smooth embedding.
(b) By assumption and by Claim 3, F is a closed map, so it is a smooth embedding by
(a).
(c) By assumption and by Claim 2, F is a closed map, so it is a smooth embedding by
(a).
 Section 4.2. Local Diffeomorphisms 37

(d) By assumption and by Proposition 4.10(b), F is a local diffeomorphism (see Def-

inition 4.7), and thus an open map by Proposition 4.9(c). Therefore, F is a smooth
embedding by (a).
Comment: There exist smooth embeddings which are neither open nor closed maps, see
[Lee13, Exercise 4.24].

4.2 Local Diffeomorphisms

Definition 4.7.
(a) Let X and Y be topological spaces. A map F : X → Y is called a local homeomor-
phism if every point p ∈ X has a neighborhood U such that F (U ) is open in Y and
F |U : U → F (U ) is a homeomorphism.

(b) Let M and N be smooth manifolds. A map F : M → N is called a local diffeomor-

phism if every point p ∈ M has a neighborhood U such that F (U ) is open in N and
F |U : U → F (U ) is a diffeomorphism.
The next theorem is the key to the most important properties of local diffeomorphisms.
Theorem 4.8 (Inverse function theorem for manifolds). Let F : M → N be a smooth
map. If p ∈ M is a point such that the differential dFp of F at p is invertible, then there
exist connected neighborhoods U0 of p in M and V0 of F (p) in N such that F |U0 : U0 → V0
is a diffeomorphism.
Proof. See [Exercise Sheet 6, Exercise 4].
Proposition 4.9 (Elementary properties of local diffeomorphisms).
(a) Every composition of local diffeomorphisms is a local diffeomorphism.

(b) Every finite product of local diffeomorphisms between smooth manifolds is a local
diffeomorphism.

(c) Every local diffeomorphism is a local homeomorphism and an open map.

(d) The restriction of a local diffeomorphism to an open submanifold is a local diffeo-

morphism.

(e) Every diffeomorphism is a local diffeomorphism.

(f) Every bijective local diffeomorphism is a diffeomorphism.

(g) A map between smooth manifolds is a local diffeomorphism if and only if in a neigh-
borhood of each point of its domain, it has a coordinate representation that is a local
diffeomorphism.
Proof. Exercise! (See also Proposition 2.15.)
Proposition 4.10. Let M and N be smooth manifolds and let F : M → N be a map.
The following statements hold:
38 Chapter 4. Maps of Constant Rank

(a) F is a local diffeomorphism if and only if it is both a smooth immersion and a smooth
submersion.

(b) If dim M = dim N and if F is either a smooth immersion or a smooth submersion,

then it is a local diffeomorphism.

Proof. See [Exercise Sheet 6, Exercise 5].

4.3 The Rank Theorem

The most important fact about maps of constant rank is the following consequence of the
inverse function theorem (see Theorem 4.8), which says that a smooth map of constant
rank can be placed locally into a particularly simple canonical form by a change of coor-
dinates. (This is a non-linear version of the canonical form theorem for linear maps; see
[Lee13, Theorem B.20].)

Theorem 4.11 (Rank theorem). Let M and N be smooth manifolds of dimension m and
n, respectively, and let F : M → N be a smooth map of constant rank r. For each p ∈ M
there exist smooth charts (U, φ) for M centered at p and (V, ψ) for N centered at F (p)
such that F (U ) ⊆ V , in which F has a coordinate representation of the form

Fb(x1 , . . . , xr , xr+1 , . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0).

In particular, if F is a smooth submersion (so that r = n), then this becomes

Fb(x1 , . . . , xn , xn+1 , . . . , xm ) = (x1 , . . . , xn ),

while if F is a smooth immersion (so that r = m), then this becomes

Fb(x1 , . . . , xm ) = (x1 , . . . , xm , 0, . . . , 0).

Proof. Since the theorem is local, after choosing smooth coordinates we can replace M
and N by open subsets U ⊆ Rm and V ⊆ Rn . The fact that DF (p) has rank r implies
that its matrix has some r × r submatrix with non-zero determinant. By
reordering
i
the coordinates, we may assume that it is the upper left submatrix ∂F ∂xj
(p) for i, j ∈
{1, . . . , r}. We relabel the standard coordinates as

(x, y) = (x1 , . . . , xr , y 1 , . . . , y m−r ) in Rm

and
(v, w) = (v 1 , . . . , v r , w1 , . . . , wn−r ) in Rn .
By initial translation of the coordinates, without loss of generality we may assume that
p = (0, 0) and F (p) = (0, 0). If we write


F (x, y) = Q(x, y), R(x, y)

for some smooth maps Q : U → Rr and R : U → Rn−r , then our hypothesis is that the
∂Qi


matrix ∂xj is non-singular at (0, 0).
 Section 4.3. The Rank Theorem 39

Define the function

φ : U → Rm , φ(x, y) = Q(x, y), y ,

and observe that its total derivative at (0, 0) is

 i 
∂Q ∂Qi
(0, 0) (0, 0)
Dφ(0, 0) =  ∂xj ∂y j ,

i
O δj

which is non-singular by virtue of the hypothesis. Therefore, by the inverse function

theorem [Lee13, Theorem C.34], there are connected neighborhoods U0 of (0, 0) and U e0 of
φ(0, 0) = (0, 0) such that φ|U0 : U0 → U e0 is a diffeomorphism. By shrinking U0 and U
e0 if
1
necessary, we may assume that U0 ∋ (0, 0) is an open cube. Writing the inverse map as
e

φ−1 (x, y) = A(x, y), B(x, y)

for some smooth functions A : Ue0 → Rr and B : Ue0 → Rn−r , we see that

 
(x, y) = φ ◦ φ−1 (x, y) = φ A(x, y), B(x, y) = Q A(x, y), B(x, y) , B(x, y) .

Comparing y components shows that B(x, y) = y, and therefore φ−1 has the form
φ−1 (x, y) = A(x, y), y . Comparing


now x components and taking this into account
also shows that Q A(x, y), y = x, and therefore F ◦ φ−1 has the form

F ◦ φ−1 (x, y) = x, R(x,

e y) ,


where the function Re: Ue0 → Rn−r is defined by R(x, e y) = A(x, y), y . The Jacobian
matrix of F ◦ φ−1 at an arbitrary point (x, y) ∈ U e0 is

δji
 
O
D(F ◦ φ−1 )(x, y) =  ∂ R .
 
ei ei
∂R
(x, y) (x, y)
∂xj ∂y j
Since composing with a diffeomorphism does not change the rank of a map, the above
matrix has rank r everywhere in U e0 . The first r columns are obviously linearly indepen-
ei
dent, so the rank can be r only if ∂∂yRj vanish identically on U e0 , which implies that Re is
actually independent of (y 1 , . . . , y m−r ). (This is one reason why we arranged for U e0 to be
a cube.) Thus, if we let S(x) = R(x, e 0), then we have

F ◦ φ−1 (x, y) = x, S(x) .

(4.1)
To complete the proof, we need to define an appropriate smooth chart in some neigh-
borhood of F (p) = (0, 0) ∈ V . Consider the open subset

V0 = (v, w) ∈ V | (v, 0) ∈ U
e0 ⊆ V
1
A closed rectangle in Rk is a set of the form [a1 , b1 ] × . . . × [ak , bk ], whereas an open rectangle in Rk is
a set of the form (a1 , b1 ) × . . . × (ak , bk ), for real numbers ai < bi . A (closed or open) rectangle is called
a (closed or open) cube if all of its side lengths bi − ai are equal.
40 Chapter 4. Maps of Constant Rank

and note that V0 is a neighborhood of (0, 0). Since
Ue0 ∋ (0, 0) = φ(0, 0) is a cube and
−1 −1
F ◦ φ has the form (4.1), it follows that F ◦ φ (U e0 ) ⊆ V0 (because (v, w) ∈ U
e0 ⇒
−1



F ◦ φ (v, w) = v, S(v) ∈ V by construction and (v, 0) ∈ U e0 by the form of U
e0 ), and
hence F (U0 ) ⊆ V0 . Define the function

ψ : V0 → Rn , ψ(v, w) = v, w − S(v) .

This is an open map and a diffeomorphism onto its image, because its inverse is given
−1


explicitly by ψ (s, t) = s, t + S(s) . Thus, (V0 , ψ) is a smooth chart. It follows now
immediately from (4.1) that

ψ ◦ F ◦ φ−1 (x, y) = ψ x, S(x) = x, S(x) − S(x) = (x, 0) ,

which was to be proved.

The next corollary can be viewed as a more invariant statement of the rank theorem.
It says that maps of constant rank are precisely the ones whose local behavior is the same
as that of their differentials.

Corollary 4.12. Let F : M → N be a smooth map. Assume that M is connected. Then

the following are equivalent:

(a) For each p ∈ M there exists smooth charts containing p and F (p) in which the
coordinate representation of F is linear.

(b) F has constant rank.

Proof.
(b) ⇒ (a): Follows immediately from the rank theorem.
(a) ⇒ (b): Since every linear map has constant rank, it follows that the rank of F is
constant in a neighborhood of each point, and thus by connectedness it is constant on all
of M .
The rank theorem is a purely local statement. However, it has the following powerful
global consequence.

Theorem 4.13 (Global rank theorem). Let F : M → N be a smooth map of constant

rank.

(a) If F is surjective, then it is a smooth submersion.

(b) If F is injective, then it is a smooth immersion.

(c) If F is bijective, then it is a diffeomorphism.

Proof. Assume that dim M = m, dim N = n and rk F = r.

(a) See the proof of [Lee13, Theorem 4.14(a)] for the details.
(b) Assume that F is not a smooth immersion, so that r < m. By the rank theorem, for
each p ∈ M we can choose smooth charts around p and F (p) in which F has the coordinate
 Section 4.3. The Rank Theorem 41

representation Fb(x1 , . . . , xr , xr+1 , . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0). Thus, Fb(0, . . . , 0, ε) =

(0, . . . , 0, 0) for any 0 < ε ≪ 1, which shows that F is not injective, a contradiction.
(c) We have the following implications:

F : bijective ====⇒ F : injective & surjective

(a)
====⇒ F : smooth immersion & smooth submersion
(b)
4.10(a)
====⇒ F : local diffeomorphism
4.9(f )
====⇒ F : diffeomorphism.
F :bijective

4.3.1 Applications of the Rank Theorem

1 Applications to Smooth Immersions:

Theorem 4.14 (Local embedding theorem). Let F : M → N be a smooth map. Then F

is a smooth immersion if and only if every point in M has a neighborhood U such that
F |U : U → N is a smooth embedding.

Proof. If every point in M has a neighborhood on which F is a smooth embedding, then

F has full rank everywhere, so it is a smooth immersion.
Conversely, assume that F is a smooth immersion, and let p ∈ M . We first claim that
p has a neighborhood on which F is injective. Indeed, by the rank theorem there is an
open neighborhood U1 of p on which F has a coordinate representation of the form

Fb(x1 , . . . , xm ) = (x1 , . . . , xm , 0, . . . , 0),

and thus F |U1 is injective. Now, consider a precompact neighborhood U of p such that
U ⊆ U1 . The restriction of F to U is an injective continuous map with compact domain
and Hausdorff codomain, so it is a topological embedding according to Claims 1 and 2
from the proof of Proposition 4.6. Since any restriction of a topological embedding is again
a topological embedding, F |U is both a topological embedding and a smooth immersion,
so it is a smooth embedding.

2 Applications to Smooth Submersions:

Recall : Let π : M → N be a continuous map between topological spaces. A section

of π is a continuous right inverse for π, i.e., a continuous map σ : N → M such that
π ◦ σ = IdN . A local section of π is a continuous map σ : U → M defined on some open
subset U ⊆ N and satisfying the analogous relation π ◦ σ = IdU .
Many of the important properties of smooth submersions follow from the fact that
they admit an abundance of smooth local sections, which we prove below.

Theorem 4.15 (Local section theorem). Let π : M → N be a smooth map. Then π is

a smooth submersion if and only if every point of M is in the image of a smooth local
section of π.
42 Chapter 4. Maps of Constant Rank

Proof. Set m = dim M and n = dim N .

“⇒”: Fix p ∈ M and set q = π(p) ∈ N . By the rank theorem we can choose smooth
coordinates (x1 , . . . , xm ) centered at p and (y 1 , . . . , y n ) centered at q in which π has the
coordinate representation

π(x1 , . . . , xn , xn+1 , . . . , xm ) = (x1 , . . . , xn ) .

If 0 < ε ≪ 1, then the coordinate cube

Cε := x |xi | < ε, 1 ≤ i ≤ m


is a neighborhood of p whose image under π is the cube

Cε′ := y |y i | < ε, 1 ≤ i ≤ n .


The map σ : Cε′ → Cε whose coordinate representation is

σ(x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0)

is a smooth local section of π satisfying σ(q) = p, see Figure 4.2.

“⇐”: Given p ∈ M , let σ : U → M be a smooth local section of π such that σ(q) = p,
where q = π(σ(q)) = π(p) ∈ N . The equation π ◦ σ = IdU implies that dπp ◦ dσq = IdTq N
by Proposition 3.7(b), which in turn implies that dπp is surjective. Since p ∈ M was
arbitrary, we conclude that π is a smooth submersion.

Figure 4.2: Local section of a smooth submersion

Recall : If X is a topological space, Y is a set, and π : X → Y is a surjective map, then

the quotient topology on Y determined by π is defined by declaring a subset V ⊆ Y to
be open if π −1 (V ) is open in X. If X and Y are topological spaces, a map π : X → Y is
called a quotient map if it is surjective and continuous and Y has the quotient topology
determined by π.
Proposition 4.16. Let π : M → N be a smooth submersion. Then π is an open map.
Moreover, if it is surjective, then it is a quotient map.
Proof. The second assertion follows from the first one, because a surjective, open and
continuous map is a quotient map by [Lee13, Exercise A.29 and Theorem A.38].
It remains to prove that π is an open map. To this end, let W be an open subset of
M and let q ∈ π(W ). For any p ∈ W such that π(p) = q, by Theorem 4.15 there is a
neighborhood U of q on which there exists a smooth local section σ : U → M
of π with
σ(q) = p. For each y ∈ σ −1 (W ), the fact that σ(y) ∈ W implies that y = π σ(y) ∈ π(W ).
Thus, σ −1 (W ) is an open neighborhood of q contained in π(W ), which implies that π(W )
is open.
The next three theorems provide important tools that are frequently used when
studying submersions and demonstrate that surjective smooth submersions play a role
in smooth manifold theory analogous to the quotient maps in topology.
 Section 4.3. The Rank Theorem 43

Theorem 4.17 (Characteristic property of surjective smooth submersions). Let π : M →

N be a surjective smooth submersion. For any smooth manifold P , a map F : N → P is
smooth if and only if F ◦ π : M → P is smooth.

M
F ◦π
π

N F
P

Proof. See [Exercise Sheet 7, Exercise 3].

⇝ [Exercise Sheet 7, Exercise 4] explains the sense in which the above property is
“characteristic”.

⇝ [Exercise Sheet 7, Exercise 5] shows that the converse of the Theorem 4.17 is false.

Theorem 4.18 (Pushing smoothly to the quotient). Let π : M → N be a surjective

smooth submersion. If P is a smooth manifold and if F : M → P is a smooth map that is
constant on the fibers of π, then there exists a unique smooth map Fe : N → P such that
Fe ◦ π = F .
M
F
π

N P
Fe

Proof. See [Exercise Sheet 7, Exercise 6].

Theorem 4.19 (Uniqueness of smooth quotients). Let π1 : M → N1 and π2 : M → N2 be

surjective smooth submersions that are constant on each other’s fibers. Then there exists
a unique diffeomorphism F : N1 → N2 such that F ◦ π1 = π2 .

M
π1 π2

N1 F
N2

Proof. See [Exercise Sheet 7, Exercise 7].

CHAPTER 5
SUBMANIFOLDS

Many familiar manifolds appear naturally as subsets of other manifolds. We have already
seen that open subsets of smooth manifolds can be viewed as smooth manifolds in their
own right. However, there are many interesting examples beyond the open ones. In this
chapter we explore smooth submanifolds, which are smooth manifolds that are subsets of
other smooth manifolds.

5.1 Embedded Submanifolds

Definition 5.1. Let M be a smooth manifold. An embedded submanifold of M is a subset
S ⊆ M that is a topological manifold in the subspace topology, endowed with a smooth
structure with respect to which the inclusion map S ,→ M is a smooth embedding.
If S is an embedded submanifold of M , then the difference dim M − dim S is called the
codimension of S in M , and the containing manifold M is called the ambient manifold
for S.

For instance, an embedded submanifold of codimension 1 is called an embedded hy-

persurface. The empty set ∅ is an embedded submanifold of any dimension. The easiest
embedded submanifolds to understand are those of codimension 0, as the following result
demonstrates.

Proposition 5.2 (Open submanifolds). Let M be a smooth manifold. The embedded

submanifolds of codimension 0 in M are exactly the open submanifolds.

Proof. If U ⊆ M is an open submanifold, then we have already seen that U is a smooth

manifold of dim U = dim M (Example 1.10(4)) and that the inclusion map ι : U ,→ M is
a smooth embedding (Example 4.4(3)). Therefore, U ⊆ M is an embedded manifold of
codimension 0.
Conversely, if U ⊆ M is an embedded submanifold of codimension 0, then the inclusion
ι : U ,→ M is a smooth embedding. Since dim U = dim M , it is actually a local diffeomor-
phism by Proposition 4.10(b), and thus an open map by Proposition 4.9(c). Therefore,
U is an open subset of M .

45
46 Chapter 5. Submanifolds

Proposition 5.3 (Images of embeddings as submanifolds). Let F : N → M be a smooth

embedding and set S := F (N ). With the subspace topology, S is a topological manifold,
and it has a unique smooth structure making it into an embedded submanifold of M with
the property that F is a diffeomorphism onto its image.

Proof. If we give S the subspace topology that it inherits from M , then the assumption
that F is an embedding means that F can be considered as a homeomorphism from N
onto S, and thus S is a topological manifold. We now give S a smooth structure by
taking the smooth charts to be those of the form F (U ), φ ◦ F −1 , where (U, φ) is a

smooth chart for N . Note that the smooth compatibility of these charts follows from the
smooth compatibility of the corresponding charts for N . With this smooth structure on
S, the map F is a diffeomorphism onto its image (essentially by definition), and this is
obviously the only smooth structure with this property. The inclusion map ι : S ,→ M is
equal to the composition of a diffeomorphism followed by a smooth embedding
F −1 F
S N M,
so it is a smooth embedding itself by [Exercise Sheet 6, Exercise 1(a)(iii)].
Since every embedded submanifold is the image of a smooth embedding (namely its
own inclusion map), Proposition 5.3 shows that embedded submanifolds are exactly the
images of smooth embeddings of smooth manifolds.

Example 5.4 (Graphs as embedded submanifolds). Let M be a smooth m-manifold and

let N be a smooth n-manifold. Let U ⊆ M be an open subset and let f : U → N be a
smooth map. Then the graph of f ,

Γ(f ) := (x, y) ∈ M × N | x ∈ U, y = f (x) ,

is an embedded m-dimensional submanifold of M ×N diffeomorphic to U . Indeed, consider

the map

γf : U → M × N, x 7→ x, f (x) .
It is a smooth map by [Exercise Sheet 3, Exercise 4(b)] whose image is Γ(f ). Since the
projection πM : M × N → M satisfies


πM ◦ γf (x) = x = IdU (x) for all x ∈ U,

the composition d(πM )(x,f (x)) ◦ d(γf )x is the identity on Tx U ∼

= Tx M for each x ∈ U .
Thus, d(γf )x is injective, so γf is a smooth immersion. It is also a homeomorphism onto
its image, since πM |Γ(f ) is a continuous inverse for it. Therefore, Γ(f ) is an embedded
submanifold of M × N diffeomorphic to U by Proposition 5.3.

An embedded submanifold S ⊆ M is said to be properly embedded if the inclusion

S ,→ M is a proper map. It will be shown in [Exercise Sheet 8, Exercise 1(b)] that
an embedded submanifold S ⊆ M is properly embedded if and only if S is a closed
subset of M . Consequently, every compact embedded submanifold is properly embedded,
since compact subspaces of Hausdorff spaces are closed. We refer to [Exercise Sheet 8,
Exercises 1(d), 2 and 3 ] and [Exercise Sheet 9, Exercise 6 ] for further examples of properly
embedded submanifolds.
 Section 5.1. Embedded Submanifolds 47

5.1.1 Slice Charts for Embedded Submanifolds

Definition 5.5.

(a) Given an open subset U ⊆ Rn and an integer k ∈ {0, . . . , n}, a k-dimensional slice of
U (or simply a k-slice) is any subset of the form

S = (x1 , . . . , xk , xk+1 , . . . , xn ) ∈ U xk+1 = ck+1 , . . . , xn = cn



for some constants ck+1 , . . . , cn ∈ R (often taken to be zero). (When k = 0, we have

S = {point} ⊆ U , while when k = n, then S = U .)
Note that every k-slice is homeomorphic to an open subset of Rk . (Sometimes it is
convenient to consider slices defined by setting some subset of the coordinates other than
the last ones equal to constants.)

(b) Let M be a smooth manifold and let (U, φ) be a smooth chart for M . If S is a subset
of U such that φ(S) is a k-slice of φ(U ) ⊆ Rn , then we say that S is a k-slice of U .

(c) Given a smooth manifold M , a subset S ⊆ M and an integer k ∈ N, we say that

S satisfies the local k-slice condition if each point of S is contained in the domain of a
smooth chart (U, φ) for M such that S ∩ U is a single k-slice in U . Any such chart is
called slice chart for S in M , and the corresponding coordinates (x1 , . . . , xn ) are called
slice coordinates.

Theorem 5.6 (Local slice criterion for embedded submanifolds). Let M be a smooth
n-manifold. If S is an embedded k-dimensional submanifold of M , then S satisfies the
local k-slice condition. Conversely, if S ⊆ M is a subset that satisfies the local k-slice
condition, then with the subspace topology, S is a topological manifold of dimension k, and
it has a smooth structure making it into a k-dimensional embedded submanifold of M .

Proof.
“⇒”: Fix p ∈ S. Since the inclusion map ι : S ,→ M is in particular a smooth immersion,
by the rank theorem there are smooth charts (U, φ) for S (in its given smooth manifold
structure) and (V, ψ) for M , both centered at p, in which the inclusion map ιU : U ,→ V
has the coordinate representation

(x1 , . . . , xk ) 7→ (x1 , . . . , xk , 0, . . . , 0).

Now, choose 0 < ε ≪ 0 so that both U and V contain coordinate balls U0 ⊆ U and V0 ⊆ V
of radius ε > 0 centered at p. It follows that U0 ∼ = ι(U0 ) is exactly a single slice in V0
(using the above local description). Since S ⊆ M has the subspace topology and since U0
is open in S, there is an open subset W ⊆ M such that U0 = W ∩S. Setting V1 := W ∩V0 ,
we obtain a smooth chart (V1 , ψ|V1 ) for M containing p such that V1 ∩ S = U0 ∩ V0 = U0 ,
which is a single slice of V1 (as U0 is a single slice of V0 ).
“⇐”: With the subspace topology, S is Hausdorff and second-countable, because both
properties are inherited by subspaces. To show that S is locally Euclidean, we construct
an atlas. The basic idea of the construction is that if (x1 , . . . , xn ) are slice coordinates for
S in M , then we can use (x1 , . . . , xk ) as local coordinates for S.
48 Chapter 5. Submanifolds

Figure 5.1: A chart for a subset satisfying the k-slice condition

Let π : Rn → Rk be the projection onto the first k-coordinates. Let (U, φ) be a slice chart
for S in M , and define

V = U ∩ S, Vb = (π ◦ φ)(V ), ψ = (π ◦ φ)|V : V → Vb .

By definition of slice charts, φ(V ) is the intersection of φ(U ) with a certain k-slice A ⊆ Rn
defined by setting xk+1 = ck+1 , . . . , xn = cn , and therefore φ(V ) is open in A. Since
π|A : A → Rk is a diffeomorphism, it follows that Vb is open in Rk . Moreover, ψ is a
homeomorphism, because it has a continuous inverse given by (φ−1 ◦ j)|Vb , where

j : Rk → Rn , j(x1 , . . . , xk ) = (x1 , . . . , xk , ck+1 , . . . , cn ).

Thus, S is a topological manifold.

We now check that the charts constructed above are smoothly compatible. Let (U, φ)
and (U ′ , φ′ ) be two slice charts for S in M and let (V, ψ) and (V ′ , ψ ′ ) be the corresponding
charts for S. The transition map is given by

ψ ′ ◦ ψ −1 = π ◦ φ′ ◦ φ−1 ◦ j,

which is smooth by Proposition 2.11(d) as a composite of four smooth maps. Hence, the
atlas we have constructed is actually a smooth atlas (see Remark 1.5), and it defines a
smooth structure on S by Proposition 1.8(a). In terms of a slice chart (U, φ) for S in M
and the corresponding chart (V, ψ) for S, the inclusion map ι : S ,→ M has a coordinate
representation of the form

(x1 , . . . , xk ) 7→ (x1 , . . . , xk , ck+1 , . . . , cn ),

which is a smooth immersion. Since the inclusion map is also a topological embedding,
we are done.

Notice that the local slice condition for S ⊆ M is a condition on the subset S only;
it does not presuppose any particular topology or smooth structure on S. According to
[Exercise Sheet 9, Exercise 1], the smooth manifold structure constructed in Theorem 5.6
is the unique one in which S can be considered as a submanifold, so a subset satisfying
the local slice condition is an embedded submanifold in only one way.

5.1.2 Level Sets

Let Φ : M → N be a map between sets. Recall that if c ∈ N , then Φ−1 (c) is called a level
set of Φ. In the special case when N = Rk and c = 0, the level set Φ−1 (c) is usually called
the zero set of Φ.

Definition 5.7. Let Φ : M → N be a smooth map.

• A point p ∈ M is called regular point of Φ if dΦp : Tp M → TΦ(p) N is surjective;

otherwise, we say that p is a critical point of Φ.
 Section 5.1. Embedded Submanifolds 49

• A point c ∈ N is called a regular value of Φ if every point of the level set Φ−1 (c) is
a regular point; otherwise, we say that c is a critical value of Φ. (In particular, if
Φ−1 (c) = ∅, then c is a regular value.)

• A level set Φ−1 (c) is called a regular level set if c is a regular value of Φ.

Remark 5.8. Let Φ : M → N be a smooth map.

(1) If dim M < dim N , then every point of M is critical point of Φ.

(2) Every point of M is regular if and only if Φ is a smooth submersion.

(3) By Lemma 4.3, the set of regular points of Φ is an open subset of M (but may well
be empty).

Consider the three smooth functions

Θ : R2 → R, (x, y) 7→ x2 − y,
Φ : R2 → R, (x, y) 7→ x2 − y 2 ,
Ψ : R2 → R, (x, y) 7→ x2 − y 3 .

Although the zero set Θ−1 (0) of Θ is an embedded submanifold of R2 , it will be shown in
[Exercise Sheet 8, Exercise 3(b)] and [Exercise Sheet 9, Exercise 5(c)], respectively, that
neither the zero set Φ−1 (0) of Φ nor the zero set Ψ−1 (0) of Ψ is an embedded submanifold
of R2 . Hence, it is fairly easy to find level sets of smooth functions that are not smooth
submanifolds. In fact, without further assumptions on the smooth function, the situation
is about as bad as could be imagined; namely, according to Theorem 2.23, every closed
subset of M can be expressed as the zero set of a smooth non-negative real-valued function.
However, using the rank theorem, we can prove the following result:

Theorem 5.9 (Constant-rank level set theorem). Let Φ : M → N be a smooth map of

constant rank r. Each level set of Φ is a properly embedded submanifold of codimension r
in M .
In particular, if Φ is a smooth submersion, then each level set of Φ is a properly
embedded submanifold of M of codimension r = dim N .

Proof. Set m = dim M , n = dim N and k = m − r. Pick c ∈ N and set S = Φ−1 (c). By
the rank theorem, for each p ∈ S there are smooth charts (U, φ) centered at p and (V, ψ)
centered at c = Φ(p) in which Φ has a coordinate representation of the form

e 1 , . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0).
Φ(x

Therefore, S ∩ U = Φ−1 (c) ∩ U is the slice

 1
(x , . . . , xr , xr+1 , . . . , xm ) ∈ U | x1 = . . . = xr = 0 .

In conclusion, S satisfies the local (k = m − r)-slice condition, so it is an embedded

submanifold of dimension k by Theorem 5.6. It is also closed in M by continuity of Φ, so
it is actually properly embedded in M by [Exercise Sheet 8, Exercise 1(b)].
50 Chapter 5. Submanifolds

Corollary 5.10 (Regular level set theorem). Every regular level set of a smooth map
between smooth manifolds is a properly embedded submanifold of the domain whose codi-
mension is equal to the dimension of the codomain.
Proof. Let Φ : M → N be a smooth map and let c ∈ N be a regular value of Φ. By
Remark 5.8(3) the set

U = p ∈ M | rk(dΦp ) = dim N ⊆ M

is open in M , and contains Φ−1 (c) by assumption. Thus, Φ|U : U → N is a smooth

submersion, so Φ−1 (c) is an embedded submanifold of U by Theorem 5.9. It follows now
from Proposition 5.2 and [Exercise Sheet 6, Exercise 1] that

Φ−1 (c) ,→ U ,→ M

is a smooth embedding (as a composite of smooth embeddings), so Φ−1 (c) is an embedded

submanifold of M . It is also closed in M by continuity of Φ, so it is actually properly
embedded in M by [Exercise Sheet 8, Exercise 1(b)].
Not all embedded submanifolds can be expressed as level sets of smooth submersions.
However, the next proposition shows that every embedded submanifold is at least locally
of this form.
Proposition 5.11. Let S be a subset of a smooth m-manifold M . Then S is an embedded
k-submanifold of M if and only if every point of S has a neighborhood U in M such that
U ∩ S is a level set of a smooth submersion.
Proof. See [Exercise Sheet 8, Exercise 4].
If S ⊆ M is an embedded submanifold, then a smooth map Φ : M → N such that S
is a regular level set of Φ is called a defining map for S. In the special case N = Rm−k
it is usually called a defining function for S. For several examples, see [Exercise Sheet 8 ]
and [Exercise Sheet 9 ]. More generally, if U ⊆ M is an open subset and Φ : U → N is a
smooth map such that S ∩U is a regular level set of Φ, then Φ is called a local defining map
(or local defining function) for S. Proposition 5.11 says that every embedded submanifold
admits a local defining map in a neighborhood of each of its points.

Figure 5.2: An embedded submanifold is locally a level set

5.2 Immersed Submanifolds

Definition 5.12. Let M be a smooth manifold. An immersed submanifold of M is a
subset S ⊆ M endowed with a topology (not necessarily the subspace topology) with
respect to which it is a topological manifold, and with a smooth structure with respect to
which the inclusion map S ,→ M is (an injective) smooth immersion. The codimension
of S in M is defined as dim M − dim S.
Observe that every embedded submanifold is an immersed submanifold, but the con-
verse fails in general; see [Exercise Sheet 9, Exercise 5(b)] for a counterexample.
 Section 5.2. Immersed Submanifolds 51

Proposition 5.13 (Images of immersions as submanifolds). Let F : N → M be an in-

jective smooth immersion. Set S := F (N ). Then S has a unique topology and smooth
structure such that it is an immersed submanifold of M and such that F : N → S is a
diffeomorphism onto its image.

Proof. The proof is very similar to that of Proposition 5.3, except that now we also have
to define the topology on S.
We give S a topology by declaring a subset U ⊆ S to be open if and only if F −1 (U ) ⊆ N
is open, and then we give it
a smooth structure by taking the smooth charts to be those
of the form F (U ), φ ◦ F −1 , where (U, φ) is a smooth chart for N . (As in the proof
of Proposition 5.3, the smooth compatibility of these charts follows from the smooth
compatibility of the corresponding charts for N .) With this topology and smooth structure
on S, the map F is a diffeomorphism onto its image, and these are the only topology and
smooth structure on S with this property. The inclusion map ι : S ,→ M can be written
as the composition
F −1 F
S N M,

where the first map is a diffeomorphism and the second map is a smooth immersion, so ι
is itself a smooth immersion by [Exercise Sheet 6, Exercise 1(a)(ii)].

Since every immersed submanifold is the image of an injective smooth immersion

(namely its own inclusion map), Proposition 5.13 shows that immersed submanifolds are
exactly the images of injective smooth immersions of smooth manifolds.

Example 5.14. The figure-eight curve (lemniscate) from Example 4.5(2) is the image of
the injective smooth immersion

β : (−π, π) → R2 , t 7→ sin(2t), sin t

(which is not an embedding), so it is an immersed submanifold of R2 when given an

appropriate topology and smooth structure. As a smooth manifold, it is diffeomorphic
2
to R. It is not an embedded submanifold
of R , because it does not have the subspace
topology. In fact, the image set β (−π, π) cannot be made into an embedded submanifold
of R2 even if we are allowed to change its topology and smooth structure, see Exercises
1, 2 and 5(a) from [Exercise Sheet 9 ].

Remark 5.15. In general, smooth (immersed) submanifolds can be closed without being
embedded (as is, for example, the figure-eight curve from Example 5.14) or embedded
without being closed (as is, for example, the open unit ball Bn in Rn ).

The following observation is sometimes useful when thinking about the topology of an
immersed submanifold.

Comment: Let M be a smooth manifold and let S be an immersed submanifold of M .

Then every subset of S that is open in the subspace topology is also open in its given
submanifold topology (that is, the submanifold topology on an immersed submanifold is
finer than the subspace topology); and the converse is true if and only if S is embedded.
52 Chapter 5. Submanifolds

Given a smooth submanifold that it is only known to be immersed, it is often useful

to have simple criteria that guarantee that it is embedded. The next proposition gives
several such criteria.
Proposition 5.16. Let M be a smooth manifold and let S be an immersed submanifold
of M . If any of the following conditions holds, then S is embedded.
(a) The inclusion map ι : S ,→ M is proper.
(b) S is compact.
(c) codimM S = 0.
Proof. Since S is an immersed submanifold of M , the inclusion map ι : S ,→ M is an
injective smooth immersion. If any of the above conditions holds, then Proposition 4.6
implies that ι is a smooth embedding; in particular, ι(S) = S is endowed with the
subspace topology inherited from M . Therefore, in any of these three cases, S is an
embedded submanifold of M .
Even though many immersed submanifolds are not embedded, such as the one from
Example 5.14, the next result shows that the local structure of an immersed submanifold
is the same as that of an embedded one.
Proposition 5.17 (Immersed submanifolds are locally embedded). If M is a smooth
manifold and if S is an immersed submanifold of M , then for each p ∈ S there exists a
neighborhood U of p in S that is an embedded submanifold of M .
Proof. By assumption, ι : S ,→ M is a smooth immersion. By Theorem 4.14 every p ∈ S
has a neighborhood U in S such that ι|U : U ,→ M is a smooth embedding, so Proposi-
tion 5.3 yields the assertion.

5.3 The Tangent Space to a Submanifold

We discuss here the tangent space to submanifolds. If S is a submanifold of Rn , we
intuitively think of the tangent space Tp S at a point p ∈ S as a subspace of the tangent
space Tp Rn . Similarly, the tangent space to a smooth submanifold of an abstract smooth
manifold can be viewed as a subspace of the tangent space to the ambient manifold, once
we make appropriate identifications.
Let M be a smooth manifold and let S be an immersed or embedded submanifold of M .
Since the inclusion map ι : S ,→ M is (at least) a smooth immersion, at each point p ∈ S
we have an injective linear map dιp : Tp S ,→ Tp M . In terms of derivations, this injection
works in the following way: for any vector v ∈ Tp S, the image vector ve = dιp (v) ∈ Tp M
acts on smooth functions on M by


vef = dιp (v)(f ) = v(f ◦ ι) = v f S .
We usually identify Tp S with its image dιp (Tp S) under dιp , thereby thinking of Tp S as a
certain linear subspace of Tp M . This identification makes sense regardless of whether S
is immersed or embedded.
There are several alternative ways of characterizing Tp S as a subspace of Tp M . The
first one is the most general; it is just a straightforward generalization of Proposition 3.16.
 Section 5.3. The Tangent Space to a Submanifold 53

Proposition 5.18. Let M be a smooth manifold, let S ⊆ M be an immersed or embedded

submanifold, and let p ∈ S. A vector v ∈ Tp M is in Tp S if and only if there exists a
smooth curve γ : J → M whose image is contained in S, and which is also smooth as a
map into S, such that 0 ∈ J, γ(0) = p and γ ′ (0) = v.
Proof. See [Exercise Sheet 9, Exercise 3].
The next proposition gives a useful way to characterize Tp S in the embedded case.
However, according to [Lee13, Problem 5.20], this does not work in the non-embedded
case; see the Remark after the solution of [Exercise Sheet 9, Exercise 3(b)] for a coun-
terexample (relying on Example 4.5(2)).
Proposition 5.19. Let M be a smooth manifold, let S ⊆ M be an embedded submanifold
of M and let p ∈ S. As a subspace of Tp M , the tangent space Tp S is characterized by

Tp S = v ∈ Tp M | vf = 0 whenever f ∈ C ∞ (M ) with f |S = 0 .


Proof. Pick v ∈ Tp S ⊆ Tp M . Then v = dιp (w) for some w ∈ Tp S, where ι : S
,→ M is

the inclusion map. If f ∈ C ∞ (M ) with f |S = 0, then vf = dιp (w)(f ) = w f |S = 0.
Conversely, if v ∈ Tp M satisfies vf = 0 whenever f vanishes on S, then we have to
show that v = dιp (w) for some w ∈ Tp S. Let (x1 , . . . , xn ) be slice coordinates for S in
some neighborhood U of p, so that

U ∩ S = (x1 , . . . , xn ) ∈ U | xk+1 = . . . = xn = 0 ,


and (x1 , . . . , xk ) are coordinates for U ∩ S. Since the inclusion map ι : U ∩ S ,→ M has
the coordinate representation

ι(x1 , . . . , xk ) = (x1 , . . . , xk , 0, . . . , 0)

in these coordinates (see the proof of Theorem 5.6), it follows that Tp S ∼

= dιp (Tp S) is
exactly the subspace of Tp M spanned by

∂ ∂
,..., .
∂x1 p ∂xk p

If we write the coordinate representation of v ∈ Tp M as

n
X ∂
v= vi ,
i=1
∂xi p

then v ∈ Tp S if and only if v j = 0 for all j > k.

Let φ be a smooth bump function supported in U that is equal to 1 in a neighborhood
of p. Choose an index j > k and consider the function f (x) = φ(x) xj , extended to be
zero on M \ supp φ. Then f vanishes identically on S, so
n
X
i∂(φ(x) xj ) product rule
0 = vf = v i
(p) ======= v j .
i=1
∂x +properties

Thus, v ∈ Tp S, as desired.
54 Chapter 5. Submanifolds

Finally, if an embedded submanifold is characterized by a defining map, then this map

gives a concise characterization of its tangent space at each point; see [Exercise Sheet 9,
Exercise 4] and recall also Proposition 5.11.

Exercise 5.20: Let S ⊆ M be a level set of a smooth map Φ : M → N of constant rank.

Show that Tp S = ker dΦp for every p ∈ S.
Given a smooth manifold M and a subset S of M , it is important to bear in mind that
there are two very different questions one can ask. The simplest question is whether S is
an embedded manifold. Since embedded submanifolds are exactly those subsets satisfying
the local slice condition, this is simply a question about the subset S itself: either it is
an embedded submanifold or it is not, and if so, then the topology and smooth structure
making it into an embedded submanifold are uniquely determined according to [Exercise
Sheet 9, Exercise 1].
A more subtle question is whether S can be an immersed submanifold. In this case,
neither the topology nor the smooth structure is known in advance, so one needs to ask
whether there exist any topology and smooth structure on S making it into an immersed
submanifold. This question is not always straightforward to answer, and it can be es-
pecially tricky to prove that S is not an immersed submanifold. A typical approach is
to assume that it is, and then use one or more of the following phenomena to derive a
contradiction:

• At each p ∈ S, the tangent space Tp S is a subspace of Tp M , with the same dimension

at each point.

• Each vector tangent to S is the velocity vector of some smooth curve in S.

• Each vector tangent to S annihilates every smooth function that is constant on S.

Here is an example of how this can be done; another one is given in [Exercise Sheet 9,
Exercise 5(c)].

Example 5.21. Consider the subset

S = (x, y) ∈ R2 | y = |x| ⊆ R2 .


It is easy to check that S \ {(0, 0)} is an embedded 1-dimensional submanifold of R2 , so if

S itself is an immersed submanifold at all, then it must be 1-dimensional. Suppose there
were some smooth manifold structure on S making it into an immersed submanifold. Then
T(0,0) S would be a 1-dimensional subspace of T(0,0) R2 , so by Proposition 5.18 there would
be a smooth curve γ : (−ε, ε) → R2 whose
image is in S, and that satisfies γ(0) = (0, 0)
and γ ′ (0) ̸= 0. Writing γ(t) = x(t), y(t) , we see that y(t) takes a global minimum at
t = 0, so y ′ (0) = 0. On the other hand, since every point (x, y) ∈ S satisfies x2 = y 2 , we
have x2 (t) = y 2 (t) for all t ∈ (−ε, ε). Differentiating twice and setting t = 0, we conclude
that 2x′ (0) = 2y ′ (0) = 0, which is a contradiction. Thus, there is no such smooth manifold
structure on S.
CHAPTER 6
VECTOR BUNDLES

In Section 3.4 we saw that the tangent bundle of a smooth manifold has a natural structure
as a smooth manifold in its own right. The natural coordinates we constructed on TM
make it look locally like the Cartesian product of an open subset of M n with Rn . This
kind of structure arises quite frequently – a collection of vector spaces, one for each point
in M , glued together in a way that looks locally like the Cartesian product of an open
subset of M n with Rn , but globally may be “twisted”. Such structures are called vector
bundles and will be briefly discussed in this chapter.
There is a deep and extensive body of theory about vector bundles on manifolds,
which we will not touch in this course. We introduce them primarily in order to have
a convenient language for talking about the tangent bundle and structures like it; see
Chapter 7 and Chapter 8.

6.1 Vector Bundles

Definition 6.1. Let M be a topological space. A (real) vector bundle of rank k over M
is a topological space E together with a continuous surjective map π : E → M satisfying
the following conditions:

(i) For each p ∈ M , the fiber Ep = π −1 (p) over p is endowed with the structure of a
k-dimensional R-vector space.

(ii) For each p ∈ M , there exists a neighborhood U of p in M and a homeomorphism

Φ : π −1 (U ) → U × Rk (called a local trivialization of E over U ), satisfying the
following conditions (see Figure 6.1):

 πU ◦ Φ = π, where πU : U × Rk → U is the projection.

 For each q ∈ U , the restriction of Φ to Eq is an R-vector space isomorphism
from Eq to {q} × Rk ∼= Rk .

The space E is called the total space of the bundle, M is called its base, and π is called
its projection.

55
56 Chapter 6. Vector Bundles

If M and E are smooth manifolds, π is a smooth map, and the local trivializations
can be chosen to be diffeomorphisms, then E is called a smooth vector bundle over M .
In this case, any local trivialization that is a diffeomorphism onto its image is called a
smooth local trivialization.

Figure 6.1: A local trivialization of a vector bundle

Definition 6.2. If there exists a local trivialization of E over all of M , called a global
trivialization of E, then E is called a trivial bundle. In this case, E itself is homeomorphic
to the product space M × Rk .
If E → M is a smooth vector bundle that admits a smooth global trivialization, then
we say that E is smoothly trivial . In this case, E is diffeomorphic to M × Rk , not just
homeomorphic (as in previous case).

Example 6.3.

(1) Product bundles: Given any topological space M , the product space E = M × Rk
with the map π = πM : M × Rk → M as its projection is a rank-k vector bundle over
M . Any such bundle, called a product bundle, is clearly trivial (with the identity map
Φ = IdE : M × Rk → M × Rk as a global trivialization). If M is a smooth manifold, then
the (smooth) product bundle M × Rk is smoothly trivial.

(2) The Möbius bundle: see [Lee13, Example 10.3].

Proposition 6.4 (The tangent bundle as a vector bundle). Let M be a smooth n-manifold
and let TM be its tangent bundle. With its standard projection map π : TM → M ,
its natural vector space structure on each fiber, and the topology and smooth structure
constructed in Proposition 3.12, π : TM → M is a smooth vector bundle of rank n over
M.

Proof. Given any smooth chart (U, φ) for M with coordinate functions (xi ), define a map

∂
Φ : π −1 (U ) → U × Rn , v i 7→ p, (v 1 , . . . , v n ) .


∂xi p

This is linear on the fibers and satisfies πU ◦ Φ = π. The composite map

Φ ϕ ◦ IdRn
π −1 (U ) U × Rn φ(U ) × Rn ,

is equal to the coordinate map φe : π −1 (U ) → φ(U ) × Rn constructed in Proposition 3.12.

Since both φ e and φ × IdRn are diffeomorphisms, so is Φ. Therefore, Φ satisfies all the
conditions for a smooth local trivialization.

Any bundle that is not trivial requires more than one local trivialization. The next
lemma shows that the composition of two smooth local trivializations has a simple form
where they overlap.
 Section 6.1. Vector Bundles 57

Lemma 6.5. Let π : E → M be a smooth vector bundle of rank k over M . Suppose that

Φ : π −1 (U ) → U × Rk and Ψ : π −1 (V ) → V × Rk

are two smooth local trivializations of E with U ∩ V ̸= ∅. Then there exists a smooth map

τ : U ∩ V → GL(k, R),

called the transition function between the smooth local trivializations Φ and Ψ, such that
the composition
Φ ◦ Ψ−1 : (U ∩ V ) × Rk → (U ∩ V ) × Rk
has the form
Φ ◦ Ψ−1 (p, v) = p, τ (p) · v .

Proof. Note that the following diagram commutes:

Ψ Φ
(U ∩ V ) × Rk π −1 (U ∩ V ) (U ∩ V ) × Rk

π
π1 π1 =πU ∩V

U ∩V
and thus π1 ◦ (Φ ◦ Ψ−1 ) = π1 , which means that

Φ ◦ Ψ−1 (p, v) = p, σ(p, v)

for some smooth map σ : (U ∩ V ) × Rk → Rk .

k −1 k
(p, v) ∈ (U ∩ V ) × R
⇝ (Φ ◦ Ψ )(p, v) = (q, w) ∈ (U ∩ V ) × R ⇝ q = p and
w = w(p, v) =: σ(p, v) .
Moreover, for each fixed p ∈ U ∩ V , the map v 7→ σ(p, v) is an invertible linear map (since
both Φ|Ep and Ψ|Ep are R-linear isomorphisms), so there is an invertible k × k matrix
τ (p) such that σ(p, v) = τ (p) · v. It remains to show that the map τ : U ∩ V → GL(k, R)
is smooth; this is [Exercise Sheet 10, Exercise 1(b)].
Vector bundles are often most easily described by giving a collection of vector spaces,
one for each point of the base manifold. In order to make such a set into a vector bundle,
we would first have to construct a manifold topology and a smooth structure on the
disjoint union of all the vector spaces, and then construct the local trivializations and
show that they have the requisite properties. The next lemma provides a shortcut (cf.
Lemma 1.11) by showing that it is sufficient to construct the local trivializations, as long
as they overlap with smooth transition functions. (See [Exercise Sheet 10, Exercise 2] for
a stronger form of the result.)
Lemma 6.6 (Vector bundle chart lemma). Let M be a smooth manifold. Suppose that
F p ∈ M we are given an R-vector space Ep of some fixed dimension k. Set
for each
E := p∈M Ep and consider the map π : E → M, v ∈ Ep 7→ p ∈ M . Suppose furthermore
that we are given the following data:
(i) an open cover {Uα }α∈A of M ,
58 Chapter 6. Vector Bundles

(ii) for each α ∈ A, a bijective map Φα : π −1 (Uα ) → Uα × Rk whose restriction to each

Ep is an R-vector space isomorphism from Ep to {p} × Rk ∼ = Rk ,
(iii) for each α, β ∈ A with Uα ∩ Uβ ̸= ∅, a smooth map ταβ : Uα ∩ Uβ → GL(k, R) such
that the composition

Φα ◦ Φ−1 k
β : (Uα ∩ Uβ ) × R → (Uα ∩ Uβ ) × R
k

has the form

Φα ◦ Φ−1



β (p, v) = p, ταβ (p, v) .

Then E has a unique topology and smooth structure making it into a smooth manifold
and a smooth vector bundle of rank k over M , with π as projection and (Uα , Φα ) α∈A as
smooth local trivializations.
Proof. For the details of the proof, which relies essentially on Lemma 1.11, we refer to
[Lee13, Lemma 10.6].
Here are some examples showing how the vector bundle chart lemma can be used to
construct new vector bundles from old ones.
Example 6.7 (Whitney sums). Let M be a smooth manifold. Let π ′ : E ′ → M and
π ′′ : E ′′ → M be two smooth vector bundles of ranks k ′ and k ′′ , respectively, over M . We
will construct a new smooth vector bundle π : E → M of rank k ′ + k ′′ over M , denoted
by E ′ ⊕ E ′′ and called the Whitney sum of E ′ and E ′′ , whose fiber over each p ∈ M is
the direct sum Ep := Ep′ ⊕ Ep′′ , which is a (k ′ + k ′′ )-dimensional R-vector space. For each
p ∈ M choose a small enough neighborhood U of p so that there exist local trivializations
(U, Φ′ ) of E ′ and (U, Φ′′ ) of E ′′ , and define the map
′ ′′
Φ : π −1 (U ) → U × Rk +k , Φ(v ′ , v ′′ ) := π ′ (v ′ ), πRk′ ◦ Φ′ (v ′ ), πRk′′ ◦ Φ′′ (v ′′ ) .

e ′ and U e ′′ .



Suppose that we are given another such pair of local trivializations U e, Φ e, Φ
Let τ ′ : U ∩ U e → GL(k ′ , R) and τ ′′ : U ∩ U e → GL(k ′′ , R) be the corresponding transition
functions. Then the transition function for E ′ ⊕ E ′′ has the form
e ◦ Φ−1 p, (v ′ , v ′′ ) = p, τ (p)(v ′ , v ′′ ) ,



Φ

where τ (p) := τ ′ (p) ⊕ τ ′′ (p) ∈ GL(k ′ + k ′′ , R) is the block diagonal matrix

 ′ 
τ (p) O
.
O τ ′′ (p)

Since this depends smoothly on p, it follows from Lemma 6.6 that E ′ ⊕ E ′′ is a smooth
vector bundle over M .
Example 6.8 (Restriction of a vector bundle). Let π : E → M be a rank-k vector bundle
S let S ⊆ M be any subset. We define the restriction of E to S to be
and
−1
the set E|S :=
E
p∈S p , with the projection E| S → S obtained by restricting π. If Φ : π (U ) → U × Rk
is a local trivialization of E over U ⊆ M , it restricts to a bijective map from (π|S )−1 (U ∩S)
to (U ∩ S) × Rk , and it is easy to check that these form local trivializations for a vector
bundle structure on E|S .
 Section 6.2. Sections of a Vector Bundle 59

• If E is a smooth vector bundle over M and S ⊆ M is an embedded submanifold,

then it follows easily from Lemma 6.6 that E|S is a smooth vector bundle over M , taking
also [Exercise Sheet 8, Exercise 5(a)] into account.

• If E is a smooth vector bundle over M , but S ⊆ M is merely immersed, then we

give E|S a topology and a smooth structure making it into a smooth rank-k vector bundle
over S as follows: For each p ∈ S, choose a neighborhood U of p in M over which there
is a smooth local trivialization Φ of E, and a neighborhood V of p in S that is embedded
in M and contained in U (see Proposition 5.17). Then the restriction of Φ to π −1 (V ) is
a bijection from π −1 (V ) to V × Rk , and we can apply Lemma 6.6 to these bijections to
yield the desired structure.
In particular, if S ⊆ M is a smooth (immersed or embedded) submanifold, then TM |S is
called the ambient tangent bundle over S.

6.2 Sections of a Vector Bundle

Definition 6.9. Let π : E → M be a vector bundle. A local section of E is a continuous
map σ : U → E defined on some open subset U ⊆ M and satisfying π ◦ σ = IdU (see
Figure 6.2). This means that σ(p) ∈ Ep for every p ∈ U . A global section of E is a section
of E defined on all of M , i.e., a continuous map σ : M → E such that π ◦ σ = IdM .
A rough (local or global) section of E over an open subset U ⊆ M is defined to be a
(not necessarily continuous) map σ : U → E such that π ◦ σ = IdU . (Note that a local
section of E over U is the same as a global section of the restricted bundle E|U .)
The zero section of E is the global section ζ : M → E of E defined by ζ(p) = 0 ∈ Ep
for each p ∈ M . Note that ζ is continuous, and if E → M is a smooth vector bundle,
then ζ is smooth; see [Exercise Sheet 10, Exercise 3(a)].
If M is a smooth manifold and if E is a smooth vector bundle over M , then a smooth
(local or global) section of E is one that is a smooth map from its domain to E.

σ(U )
π
( σ
)
U M

Figure 6.2: A local section of a vector bundle

If E → M is a smooth vector bundle, then the set of all smooth global sections of E
is an R-vector space under pointwise addition and scalar multiplication:

(c1 σ1 + c2 σ2 )(p) := c1 σ1 (p) + c2 σ2 (p).

60 Chapter 6. Vector Bundles

This vector space is usually denoted by Γ(E) (but for particular smooth vector bundles we
often introduce specialized notation for their spaces of global sections) and it is infinite-
dimensional, see [Exercise Sheet 10, Exercise 3] and Exercise 2.21. Moreover, smooth
sections of E → M can be multiplied by smooth real-valued functions: If f ∈ C ∞ (M )
and σ ∈ Γ(E), then we obtain a new smooth section f σ ∈ Γ(E) defined by

(f σ)(p) := f (p) σ(p). (6.1)

∈
R Ep

⇝ The various claims made above are proved in [Exercise Sheet 10, Exercise 3(b)].

⇝ The global sections of a product bundle are discussed in [Exercise Sheet 10, Exercise
3(c)].

Lemma 6.10 (Extension lemma for smooth vector bundles). Let π : E → M be a smooth
vector bundle. Let A ⊆ M be a closed subset and let σ : A → E be a section of E|A that
is smooth in the sense that σ extends to a smooth local section of E in a neighborhood of
each point. Then for each open subset U ⊆ M containing A, there exists a smooth global
e ∈ Γ(E) such that σ
section σ e = {p ∈ M | σ
e|A = σ and supp σ e(p) ̸= 0} ⊆ U .

Proof. Exercise! (Similar to the proof of Lemma 2.22).

⇝ For two applications of Lemma 6.10 we refer to [Exercise Sheet 10, Exercises 3(d)
and 4(c)].

Lemma 6.11 (Extension lemma for sections of restricted bundles). Let π : E → M

be a smooth vector bundle over a smooth manifold M and let S ⊆ M be an embedded
submanifold. For any smooth section σ of the restricted bundle E|S → M , show that there
exists a neighborhood U of S in M and a smooth section σ e|S . If
e of E|U such that σ = σ
E has positive rank, then show that every smooth section of E|S extends smoothly to all
of M if and only if S is a properly embedded.

Proof. Exercise! (Similar to the solution of [Exercise Sheet 8, Exercise 6]).

6.2.1 Local and Global Frames

Definition 6.12. Let E → M be a vector bundle. If U ⊆ M is an open subset, then
a k-tuple of local sections (σ
1 , . . . , σk ) of E over U is said to be linearly independent if
their values σ1 (p), . . . , σk (p) form a linearly independent k-tuple in Ep for each p ∈ U .
Similarly, they are said to span E if their values span Ep for each p ∈ U .
A local frame for E over U is an ordered k-tuple (σ1 , . . . , σ
k ) of linearly independent
local sections of E over U that span E; thus, σ1 (p), . . . , σk (p) is a basis for the fiber Ep
for each p ∈ U . It is called a global frame if U = M . We often denote a frame (σ1 , . . . , σk )
by (σi ).
If, moreover, E → M is a smooth vector bundle, then a local or global frame for E is
said to be smooth if each σi is a smooth section of E.
 Section 6.2. Sections of a Vector Bundle 61

Example 6.13 (Global frame for a product bundle). If E = M × Rk → M is a (smooth)

product bundle over a (smooth) manifold M , then the standard basis (e1 , . . . , ek ) for Rk
yields a (smooth) global frame eei for E, defined by

eei : M → E, p 7→ (p, ei ).

⇝ For the correspondence between smooth local frames and smooth local trivializations
see [Exercise Sheet 10, Exercise 5] (which also settles the question of the existence
of smooth local frames).

⇝ For the completion of smooth local frames for smooth vector bundles see [Exercise
Sheet 10, Exercise 4].

We conclude this section with the important observation that smoothness of sections
of smooth vector bundles can be characterized in terms of local frames.
Assume that (σi ) is a smooth local frame for a smooth vector bundle E → M over
some open subset U ⊆ M . If τ : M → E is a rough section, then the value of τ at an
arbitrary point p ∈ U can be written as

τ (p) = τ i (p) σi (p)

for some uniquely determined numbers τ 1 (p), . . . , τ k (p) . This clearly defines k-functions
τ i : U → R, called the component functions of τ with respect to the given local frame (σi ).

Proposition 6.14 (Local frame criterion for continuity/smoothness). Let π : E → M be

a continuous (respectively smooth) vector bundle and let τ : M → E be a rough section. If
(σi ) is a continuous (respectively smooth) local frame for E over an open subset U ⊆ M ,
then τ is continuous (respectively smooth) if and only if its component functions with
respect to (σi ) are continuous (respectively smooth).

Proof. We prove the statement in the smooth case; the other case can be treated similarly.
Let Φ : π −1 (U ) → U × Rk be the smooth local trivialization associated with the smooth
local frame (σi ), see [Exercise Sheet 10, Exercise 5(b)]. Since Φ is a diffeomorphism, τ is
smooth on U if and only if Φ ◦ τ is smooth on U . By the construction of Φ in [Exercise
Sheet 10, Exercise 5(b)] we know that



Φ ◦ τ (p) = Φ τ i (p)σi (p) = p, τ 1 (p), . . . , τ k (p) ,

where (τ i ) are the component functions of τ with respect to (σi ). Therefore, Φ ◦ τ is

smooth if and only if the component functions τ i are smooth according to [Exercise Sheet
3, Exercise 4(b)].

Note that Proposition 6.14 applies equally well to local sections, since a local section
of E over an open subset V ⊂ M is a global section of the restricted bundle E|V .
62 Chapter 6. Vector Bundles

6.3 Subbundles
Definition 6.15. Given a vector bundle πE : E → M , a subbundle of E is a vector bundle
πD : D → M , in which D is a topological subspace of E and πD is the restriction of πE
to D, such that for each p ∈ M , the subset Dp = D ∩ Ep is a linear subspace of Ep , and
the vector space structure on Dp is the one inherited from Ep .
If E → M is a smooth vector bundle, then a subbundle of E is called a smooth
subbundle if it is a smooth vector bundle and an embedded submanifold of E.

Note that the condition that D be a vector bundle over M implies that all of the fibers
Dp must be non-empty and have the same dimension.

Figure 6.3: A subbundle of a vector bundle

Thefollowing lemma gives a convenient condition for checking that a union of sub-
spaces Dp ⊆ Ep | p ∈ M is a smooth subbundle.

Lemma 6.16 (Local frame criterion for subbundles). Let π : E → M be a smooth vector
bundle of rank k. Suppose that S for each p ∈ M we are given an m-dimensional linear
subspace Dp ⊆ Ep . Then D = p∈M Dp ⊆ E is a smooth subbundle of E if and only if the
following condition is satisfied: “Each point of M has a neighborhood U on which there
exist smooth local sections σ1 , . . . , σm : U → E with the property that σ1 (q), . . . , σm (q)
form a basis for Dq at each q ∈ U .”

Proof. If D → M is a smooth subbundle of E → M , then by definition each p ∈ M has a

neighborhood U over which there exists a smooth local trivialization of D, and [Exercise
Sheet 10, Exercise 5(a)] shows that there exists a smooth local frame for D over U , namely
a collection of smooth sections τ1 , . . . , τm : U → D whose images form a basis for Dq at
each point q ∈ U . The smooth sections of E that we seek are obtained by composing
with the inclusion map ι : D ,→ E; that is, σj = ι ◦ τj for j ∈ {1, . . . , m}.
For the details of the proof of the converse direction, which uses [Exercise Sheet 10,
Exercise 4(a)] and [Exercise Sheet 10, Exercise 5(b)], we refer to [Lee13, Lemma 10.32].

Example 6.17 (Subbundles).

(1) Let M be a smooth manifold and let S ⊆ M be an immersed k-submanifold. Then the
tangent bundle TS is a smooth rank-k subbundle of the ambient tangent bundle TM |S ;
see [Lee13, Problem 10-14].

(2) If M is a smooth manifold and if V is a nowhere-vanishing smooth vector field on M

(see Chapter 7), then the set D ⊆ TM whose fiber at each p ∈ M is the linear span of
Vp ∈ Tp M \ {0} is a smooth 1-dimensional subbundle of TM .

(3) Let E → M be any smoothly trivial vector bundle of rank k and let (E1 , . . . , Ek )
be a smooth global frame
for E. If m ∈ {0, . . . , k}, then the subset D ⊆ E defined by
Dp = span E1 |p , . . . , Em |p for each p ∈ M is a smooth rank-m subbundle of E.
CHAPTER 7
VECTOR FIELDS AND FLOWS

7.1 Vector Fields

Definition 7.1. A rough (resp. continuous, smooth) vector field on a smooth manifold M
is a rough (resp. continuous, smooth) global section of the tangent bundle π : TM → M .
More concretely, a vector field is a map X : M → TM , usually written p 7→ Xp , with
the property that π ◦ X = IdM or, equivalently, Xp ∈ Tp M for each p ∈ M . The support
of X is defined as the closure of the set {p ∈ M | Xp ̸= 0}. In particular, we say that X
is compactly supported if its support is a compact set.
If U ⊆ M is open, then the fact that Tp U is naturally identified with Tp M for each
p ∈ U (see Proposition 3.9) allows us to identify TU with the open subset π −1 (U ) ⊆ TM .
Therefore, a vector field on U can be thought of either as a map U → TU or as a map
U → TM . If X is a vector field on M , then its restriction X|U is a vector field on U ,
which is smooth if X itself smooth.
A (continuous) vector field on an open subset U ⊆ Rn is simply a continuous map
U → Rn , which can be visualized as attaching an “arrow” to each point of U . Similarly,
we think of a (continuous) vector field on an open subset U of a smooth manifold M as an
arrow attached to each point of M , chosen to be tangent to M and to vary continuously
from point to point (see Figure 7.1).

Figure 7.1: A vector field

⇝ The set X(M ) of all smooth (global) vector fields on a smooth manifold M is an
infinite-dimensional R-vector space and a module over the ring C ∞ (M ): this is a
special case of [Exercise Sheet 10, Exercise 3(b)].

⇝ Extension lemma for vector fields: this is a special case of Lemma 6.10; see also
[Exercise Sheet 10, Exercise 3(d)] for an application (any tangent vector at a point
can be extended to a smooth vector field on the entire manifold).

⇝ Local/global frame for M = local/global frame for TM , see Definition 6.12.

63
64 Chapter 7. Vector Fields and Flows

⇝ Completion of smooth local frames for M : this is a special case of [Exercise Sheet
10, Exercise 4].


Let M and X be as above. If U, (xi ) is a smooth coordinate chart for M , then we
can write the value of X at any point p ∈ U in terms of the coordinate basis vectors:

∂
Xp = X i (p) .
∂xi p

This defines n functions X i : U → R, called the component functions of X in the given

chart.

Proposition 7.2 (Smoothness criterion for vector fields). Let M
be a smooth manifold
i
and let X : M → TM be a rough vector field on M . If U, (x ) is a smooth coordinate
chart for M , then the restriction of X to U is smooth if and only if its components
functions with respect to this chart are smooth.

Proof. If U, φ = (xi ) is a smooth chart for M , then π −1 (U ), φ

e = (xi , v i ) are the natural
coordinates on TM (see Proposition 3.12), and the coordinate representation of X with
respect to these charts is
!
∂
e X i φ−1 (x)
b 1 , . . . , xn ) = φ


X(x
∂xi φ−1 (x)


1 n 1 −1 n −1


= x , . . . , x , X φ (x) , . . . , X φ (x) .

Therefore, X is smooth on U if and only if its component functions X i , i ∈ {1, . . . , n},

are smooth on U .

Example 7.3.

(1) If U, (xi ) is any smooth chart on M , then the assignment p 7→ ∂x∂ i p determines a

vector field on U , called the i-th coordinate vector field and denoted by ∂x∂ i . It is smooth
by Proposition 7.2, because its component functions are constant.
In particular, the coordinate vector fields form a smooth local frame ∂x∂ i for TM ,

called a coordinate frame. Note that every point of M is in the domain of such a local
frame.

(2) The Euler vector field V on Rn whose value at a point x = (x1 , . . . , xn ) ∈ Rn is

∂ ∂
V x = x1 + . . . + xn .
∂x1 x ∂xn x

It is discussed in [Exercise Sheet 12, Exercise 2].

We will encounter many more examples of vector fields (especially on Rn ) later in the
course and in the exercise sheets as well.
 Section 7.1. Vector Fields 65

7.1.1 Vector Fields as Derivations of C ∞ (M )

An essential property of vector fields is that they define operators on the space of smooth
real-valued functions. If X ∈ X(M ) and f ∈ C ∞ (U ), where U ⊆ M is open, then we
obtain a new function
Xf : U → R, p 7→ (Xf )(p) := Xp f.
(Do not confuse the notations f X and Xf : the former is a smooth vector field on U
obtained by multiplying X by f , that is, (f X)(p) = f (p)Xp , while the latter is the real-
valued function on U obtained by applying the vector field X to the smooth function
f .) Since the action of a tangent vector on a function is determined by the values of the
function in any arbitrary small neighborhood (see Proposition 3.8), it follows that Xf is
locally determined. In particular, for any open subset V ⊆ U , we have


(Xf )|V = X f |V .

This construction yields another useful smoothness criterion for vector fields.

Proposition 7.4 (Smoothness criterion for vector fields). Let M be a smooth manifold
and let X : M → TM be a rough vector field on M . The following are equivalent:

(a) X is smooth.

(b) For every f ∈ C ∞ (M ), the function Xf : M → R is smooth.

(c) For every open subset U ⊆ M and every f ∈ C ∞ (U ), the function Xf : U → R is

smooth.

Proof.


(a) ⇒ (b): Given p ∈ M , pick a smooth chart U, (xi ) for M containing p. For x ∈ U
we may write
 
i ∂ ∂ fb

f = X i (x) i x


Xf (x) = X (x) i b .
∂x x ∂x
Since the component functions X i of X are smooth on U by Proposition 7.2, it follows
that Xf is smooth on U . We conclude by [Exercise Sheet 3, Exercise 2(a)].
(b) ⇒ (c): Fix an open subset U ⊆ M and f ∈ C ∞ (U ). For any p ∈ U , let ψ be a
smooth bump function that is equal to 1 in a neighborhood of p and supported in U (see
Proposition 2.20), and define fe = ψf , extended to be zero on M \ supp ψ. Then X fe is
smooth by assumption, and equal to Xf in a neighborhood of p by construction (and by
the above discussion). We conclude by [Exercise Sheet 3, Exercise 2(a)].
(c) ⇒ (a): If (xi ) are smooth local coordinates on U ⊆ M , then we think of each coordinate
xi as a smooth function on U , and we have
∂x i
i
i
 ∂  i ∂x
j j = δj
X(x ) = X (x ) ==== Xi ,
∂xj
which is smooth by assumption. We conclude by Proposition 7.2 and [Exercise Sheet 3,
Exercise 2(a)].
66 Chapter 7. Vector Fields and Flows

One consequence of Proposition 7.4 is that a smooth vector field X ∈ X(M ) defines a
map
C ∞ (M ) → C ∞ (M ), f 7→ Xf,
which (as can be checked pointwise) is R-linear and satisfies the following product rule
for vector fields:
X(f g) = f Xg + g Xf ;
in other words, the map is a derivation of C ∞ (M ).
The next proposition shows that derivations of C ∞ (M ) can be identified with smooth
vector fields. Due to this result, we sometimes identify smooth vector fields on M with
derivations of C ∞ (M ), using the same letter for both the vector field (thought of as a
smooth map M → TM ) and the derivation (thought of as a linear map C ∞ (M ) →
C ∞ (M )).

Proposition 7.5. Let M be a smooth manifold. A map D : C ∞ (M ) → C ∞ (M ) is a

derivation if and only if it is of the form Df = Xf for some X ∈ X(M ).

Proof.
“⇒”: We just showed above that any smooth vector field induces a derivation of C ∞ (M ).
“⇐”: Let p ∈ M and consider the map

Xp : C ∞ (M ) → R, f 7→ (Df )(p).

Since D is R-linear, Xp is also R-linear, and since D is a derivation, we have

Xp (f g) = D(f g)(p) = f D(g) + g D(f ) (p)
= f (p) D(g)(p) + g(p) D(f )(p)
= f (p) Xp g + g(p) Xp f.

Hence, Xp is a derivation at p ∈ M , i.e., Xp ∈ Tp M (see Definition 3.4). We obtain thus

a rough vector field X : M → TM, p 7→ Xp , but since Xf = Df is smooth for every
f ∈ C ∞ (M ), X is actually smooth by Proposition 7.4, so we are done.

7.1.2 Vector Fields and Smooth Maps

If F : M → N is a smooth map and if X is a (rough) vector field on M , then for each
point p ∈ M we obtain a tangent vector dFp (Xp ) ∈ TF (p) N by applying the differential of
F at p to the tangent vector Xp ∈ Tp M . However, this does not define a (rough) vector
field on N in general. For example, if F is not surjective, there is no way to decide what
tangent vector to assign to a point q ∈ N \ F (M ), while if F is not injective, then for
some points of N there may be several different tangent vectors obtained by applying dF
to X at different points of M .
Let F : M → N be a smooth map and let X be a (rough) vector field on X. If there
exists a (rough) vector field Y on N such that dFp (Xp ) = YF (p) for each p ∈ M , then X
and Y are said to be F -related.
 Section 7.1. Vector Fields 67

Lemma 7.6. Let F : M → N be a smooth map. Let X ∈ X(M ) and Y ∈ X(N ). Then
X and Y are F -related if and only if for every smooth real-valued function f defined on
an open subset of N , we have

X(f ◦ F ) = (Y f ) ◦ F.

Proof. See [Exercise Sheet 11, Exercise 2(a)].

It is important to remember that for a given smooth map F : M → N and vector field
X ∈ X(M ), there may not be any vector field on N that is F -related to X. There is one
special case, however, in which there is always such a vector field, as the next proposition
shows.

Proposition 7.7. Let F : M → N be a diffeomorphism. For every smooth vector field

X on M there exists a unique smooth vector field Y on N that is F -related to X. The
smooth vector field Y is denoted by F∗ X and is called the pushforward of X by F .

Proof. See [Exercise Sheet 11, Exercise 2(c)].

7.1.3 Vector Fields and Submanifolds

If S ⊆ M is an immersed or embedded submanifold, a vector field X on M does not
necessarily restrict to a vector field on S, because Xp ∈ Tp M may not lie in the subspace
Tp S ⊆ Tp M at a point p ∈ S. Given a point p ∈ S, a vector field X on M is said to be
tangent to S at p if Xp ∈ Tp S ⊆ Tp M . It is called tangent to S if it is tangent to S at all
points of S (see Figure 7.2).

Figure 7.2: A vector field tangent to a submanifold

Proposition 7.8. Let M be a smooth manifold, S ⊆ M be an embedded submanifold,

and X ∈ X(M ). Then X is tangent to S if and only if (Xf )|S = 0 for every f ∈ C ∞ (M )
such that f |S ≡ 0.

Proof. The assertion is an immediate consequence of Proposition 5.19.

Proposition 7.9 (Restricting smooth vector fields to submanifolds). Let M be a smooth

manifold, let S be an immersed submanifold of M , and let ι : S ,→ M be the inclusion
map. The following statements hold:

(a) If Y ∈ X(M ) and if there is X ∈ X(S) that is ι-related to Y , then Y ∈ X(M ) is

tangent to S.

(b) If Y ∈ X(M ) is tangent to S, then there is a unique smooth vector field on S,

denoted by Y |S , which is ι-related to Y .

Proof. See [Exercise Sheet 11, Exercise 7(a)].

68 Chapter 7. Vector Fields and Flows

7.1.4 The Lie Bracket

We now introduce an important way of combining two smooth vector fields to obtain
another smooth vector field. Let M be a smooth manifold and let X, Y ∈ X(M ). Given
f ∈ C ∞ (M ), we can apply X to f to obtain Xf ∈ C ∞ (M ) (see Proposition 7.4) and we
can now apply Y to Xf to obtain Y (Xf ) ∈ C ∞ (M ). The operation f 7→ Y Xf does not
satisfy the product rule in general, and thus cannot be a vector field (see Proposition 7.5),
as the following example shows:
Example 7.10. Consider the smooth vector fields
∂ ∂
X= and Y = x
∂x ∂y
and the smooth functions

f (x, y) = x and g(x, y) = y

on R2 . We compute
∂(x2 )
 
∂(xy)
(XY )(f g) = X x = X(x2 ) = = 2x
∂y ∂x
and
 1
  0
 1
∂y



∂x



∂x



f XY g + gXY f = x X x
+ y X x
= x
= x,

∂y
∂y
∂x
so XY is not a derivation of C ∞ (R2 ).
However, we can also apply the same two vector fields in the opposite order, obtaining
a (usually different) smooth function Y Xf ∈ C ∞ (M ). Applying both of these operators
to f ∈ C ∞ (M ) and subtracting, we obtain the operator

[X, Y ] : C ∞ (M ) → C ∞ (M ), f 7→ XY f − Y Xf,

called the Lie bracket of X and Y . The key fact, following readily from Proposition 7.5,
is that this operator is a vector field.
Lemma 7.11. The Lie bracket of any pair of smooth vector fields on a smooth manifold
is a smooth vector field.
Proof. See [Exercise Sheet 11, Exercise 3].
We mention below the basic properties of the Lie bracket and we refer to Exercise
Sheet 11 for their proofs. The geometric interpretation of the Lie bracket will not be
covered in this course, but we refer to [Lee13, Chapter 9, Lie derivatives] for some details.
Proposition 7.12 (Coordinate formula for the Lie bracket). Let M be a smooth n-
manifold and let X, Y ∈ X(M ). Let
n n
X
i ∂ X ∂
X= X and Y = Yj
i=1
∂xi j=1
∂xj
 Section 7.1. Vector Fields 69

be the coordinate expressions for X and Y , respectively, in terms of some smooth local
coordinates (xi ) for M . Then the Lie bracket [X, Y ] has the following coordinate expres-
sion:
n X n  j j

i ∂Y i ∂X ∂
X
[X, Y ] = X i
−Y i j
.
j=1 i=1
∂x ∂x ∂x

Proof. See [Exercise Sheet 11, Exercise 4(a)]

Proposition 7.13 (Properties of the Lie bracket). Let M be a smooth manifold. The Lie
bracket satisfies the following identities for all X, Y, Z ∈ X(M ):

(a) Bilinearity: For all a, b ∈ R we have

[aX + bY, Z] = a[X, Z] + b[Y, Z],

[Z, aX + bY ] = a[Z, X] + b[Z, Y ].

(b) Antisymmetry:
[X, Y ] = −[Y, X].

(c) Jacobi identity:

     
X, [Y, Z] + Y, [Z, X] + Z, [X, Y ] = 0.

(d) For all f, g ∈ C ∞ (M ) we have

[f X, gY ] = f g[X, Y ] + (f Xg)Y − (gY f )X.

Proof. See [Exercise Sheet 11, Exercise 5].

Proposition 7.14 (Naturality of the Lie bracket). Let F : M → N be a smooth map.

Let X1 , X2 ∈ X(M ) and Y1 , Y2 ∈ X(N ) be smooth vector fields such that Xi is F -related
to Yi for i ∈ {1, 2}. Then [X1 , X2 ] is F -related to [Y1 , Y2 ].

Proof. See [Exercise Sheet 11, Exercise 6(a)].

Corollary 7.15 (Pushforwards of Lie brackets). Let F : M → N be a diffeomorphism.

For any X1 , X2 ∈ X(M ) we have F∗ [X1 , X2 ] = [F∗ X1 , F∗ X2 ].

Proof. See [Exercise Sheet 11, Exercise 6(b)].

Corollary 7.16 (Lie brackets of smooth vector fields tangent to submanifolds). Let M be
a smooth manifold and let S be an immersed submanifold of M . If Y1 and Y2 are smooth
vector fields on M that are tangent to S, then their Lie bracket [Y1 , Y2 ] is tangent to S as
well.

Proof. See [Exercise Sheet 11, Exercise 7(b)].

70 Chapter 7. Vector Fields and Flows

7.2 Integral Curves

Let M be a smooth manifold. If γ : J ⊆ R → M is a smooth curve, then for each t ∈ J the
velocity vector γ ′ (t) is an element of Tγ(t) M . We describe next a way to work backwards:
given a tangent vector at each point, we seek a curve whose velocity at each point is equal
to the given vector there.
Definition 7.17. Let M be a smooth manifold and let V be a vector field on M .
(a) An integral curve of V is a differentiable curve γ : J → M whose velocity at each
point is equal to the value of V at that point:

γ ′ (t) = Vγ(t) , ∀ t ∈ J .

If 0 ∈ J, then γ(0) ∈ M is called the starting point of γ.

(b) A maximal integral curve of V is one that cannot be extended to an integral curve
on any larger open interval.

Figure 7.3: An integral curve of a vector field

Finding integral curves of vector fields boils down to solving a system of ODEs in a
smooth chart. Suppose that V ∈ X(M ) and that γ : J ⊆ R → M is a smooth curve. On
a smooth coordinate domain U ⊆ M we can write γ in local coordinates as

γ(t) = γ 1 (t), . . . , γ n (t) .

Then the condition γ ′ (t) = Vγ(t) for γ to be an integral curve of V can be written as

∂
∂
γ̇ i (t) = V i γ(t) ,
∂xi γ(t) ∂xi γ(t)

which reduces to the following autonomous system of ODEs:

 1 1 1 n


γ̇ (t) = V γ (t), . . . , γ (t)

..
. . (7.1)

 n

n 1 n
γ̇ (t) = V γ (t), . . . , γ (t)

The fundamental fact about such systems is the following existence, uniqueness and
smoothness theorem. (This is the reason for the terminology “integral curves”, because
solving a system of ODEs is often referred to as “integrating” the system.)
Theorem 7.18 (Fundamental theorem for autonomous ODEs). Let V : U → Rn be a
smooth vector-valued function, where U ⊆ Rn is open. Consider the initial value problem

ẏ i (t) = V i y 1 (t), . . . , y n (t) , 1 ≤ i ≤ n

(1)

y i (t0 ) = ci , 1≤i≤n (2)

for arbitrary t0 ∈ R and c = (c1 , . . . , cn ) ∈ U .

 Section 7.2. Integral Curves 71

(a) Existence: For any t0 ∈ R and x0 ∈ U , there exists an open interval J0 ∋ t0 and an
open subset x0 ∈ U0 ⊆ U such that for each c ∈ U0 , there is a C 1 map y : Jo → U
that solves (1) - (2).

(b) Uniqueness: Any two differentiable solutions to (1) - (2) defined on intervals con-
taining t0 agree on their common domain.

(c) Smoothness: Let J0 and U0 be as in (a), and consider the map θ : J0 × U0 → U ,

(t, x) 7→ y(t), where y : J0 → U is the unique solution to (1) with initial condition
y(t0 ) = x. Then θ is smooth.

Proposition 7.19. Let V be a smooth vector field on a smooth manifold M . For each
point p ∈ M , there exists ε > 0 and a smooth curve γ : (−ε, ε) → M that is an integral
curve of V starting at p ∈ M .

Proof. Follows immediately from the existence and smoothness part of Theorem 7.18.

Comment: Given V and M as above, it is a consequence of the uniquness part of The-

orem 7.18 that for each p ∈ M there actually exists a unique, maximal integral curve of
V starting at p ∈ M ; see Theorem 7.26(a).

The next two lemmas show how affine reparametrizations affect integral curves.

Lemma 7.20 (Rescaling lemma). Let V be a smooth vector field on a smooth manifold
M , let J ⊆ R be an interval, and let γ : J → M be an integral curve of V . For any a ∈ R,
the curve
e : Je → M, t 7→ γ(at)
γ

is an integral curve of the vector field Ve := aV on M , where Je := {t ∈ R | at ∈ J}.

Proof. See [Exercise Sheet 12, Exercise 1(a)].

Lemma 7.21 (Translation lemma). Let V be a smooth vector field on a smooth manifold
M , let J ⊆ R be an interval, and let γ : J → M be an integral curve of V . For any b ∈ R,
the curve
b : Jb → M, t 7→ γ(t + b)
γ

is also an integral curve of V on M , where Jb := {t ∈ R | t + b ∈ J}.

Proof. See [Exercise Sheet 12, Exercise 1(b)].

Proposition 7.22 (Naturality of integral curves). Let F : M → N be a smooth map.

Then X ∈ X(M ) and Y ∈ X(N ) are F -related if and only if F takes integral curves of X
to integral curves of Y .

Proof. See [Exercise Sheet 11, Exercise 1(e)].

Example 7.23. Let (x, y) be the standard coordinates on R2 .

72 Chapter 7. Vector Fields and Flows

(1) Let
∂
V = ∈ X(R2 )
∂x
be the first coordinate vector field. Note that the integral curves of V are precisely the
straight lines parallel to the x-axis (see Figure 7.4a), with parametrization of the form
γ(t) = (a + t, b) for constants a, b ∈ R. Thus, there is a unique integral curve starting at
each point of the plane, and the images of different integral curves are either identical or
disjoint.

(2) Let
∂ ∂
W = −y +x ∈ X(R2 ).
∂x ∂y
To determine the integral curves of W we proceed as follows (see p. 70):
(
γ̇1 (t) = −γ2 (t)
γ(t) = γ 1 (t), γ 2 (t)


=⇒ γ̇(t) = Wγ(t) =⇒
γ̇2 (t) = γ1 (t)
(
γ̈1 (t)+γ1 (t)=0 γ1 (t) = a cos t − b sin t
========⇒

γ2 (t) = a sin t + b cos t = −γ̇1 (t)

for constants a, b ∈ R. Thus, each curve of the form

γ(t) = (a cos t − b sin t, a sin t + b cos t), t ∈ R,

is an integral curve of W . When (a, b) = (0, 0), this is the constant curve γ(t) = (0, 0);
otherwise, it is a circle traversed clockwise (see Figure 7.4b). Since γ(0) = (a, b), we
see again that there is a unique integral curve staring at each point (a, b) ∈ R2 , and the
images of the various integral curves are either identical or disjoint.

(a) Integral curves of V (b) Integral curves of W

 Section 7.3. Flows 73

7.3 Flows
Definition 7.24. Let M be a smooth manifold.

(a) A flow domain for M is an open subset D ⊆ R × M with the property that for each
p ∈ M , the set
D(p) := t ∈ R | (t, p) ∈ D


is an open interval containing 0 ∈ R.

(b) A (local) flow on M is a continuous map θ : D → M , where D ⊆ R × M is a flow

domain, which satisfies the following group laws:

• ∀ p ∈ M : θ(0, p) = p.
• ∀ s ∈ D(p) ∀ t ∈ D(θ(s,p)) such that s + t ∈ D(p) , we have


θ t, θ(s, p) = θ(t + s, p).

When D = R × M (and hence θ : R × M → M is a continuous left R-action on M )

we say that θ is a global flow on M (or a one-parameter group action).

(c) A maximal flow on M is a flow that admits no extension to a flow on a larger flow
domain.

Figure 7.5: A flow domain

Let θ : D → M be a flow on M .

• For each p ∈ M we define a map

θ(p) : D(p) → M, θ(p) (t) = θ(t, p).

• For each t ∈ R we define a set

Mt := p ∈ M | (t, p) ∈ D


and a map
θt : Mt → M, θt (p) = θ(t, p) = θ(p) (t) .

These maps satisfy

θt ◦ θs = θt+s and θ0 = IdM ,
so each θt is a homeomorphism, and if θ is smooth, then each θt is a diffeomorphism.

Note that
p ∈ Mt ⇐⇒ (t, p) ∈ D ⇐⇒ t ∈ D(p) .
74 Chapter 7. Vector Fields and Flows

Proposition 7.25. If θ : D → M is a smooth flow on M , then the infinitesimal generator

V of θ, defined as

′ d
V : M → TM, p 7→ Vp := θ(p) (0) = θ(p) (t) ,
dt t=0

is a smooth vector field on M , and each curve θ(p) is an integral curve of V starting at
p ∈ M.

Proof. If D = R × M , then this is shown in [Exercise Sheet 12, Exercise 4]. The proof of
the general case is essentially identical to the proof for global flows (after verifying that
all the expressions involved make sense).
The term “infinitesimal generator” comes from the following picture: in a smooth
chart, a good approximation to an integral curve can be obtained by composing many
small straight-line motions, with the direction and length of each motion determined by
the value of the vector field at the point arrived at in the previous step. Intuitively, one
can think of a flow as a sequence of infinitely many infinitesimally small linear steps.

Theorem 7.26 (Fundamental theorem on flows). Let V be a smooth vector field on

a smooth manifold M . There exists a unique smooth maximal flow θ : D → M whose
infinitesimal generator is V . This flow has the following properties:

(a) For each p ∈ M , the curve θ(p) : D(p) → M is the unique maximal integral curve of
V starting at p.

(b) If s ∈ D(p) , then D(θ(s,p)) is the interval

D(θ(s,p)) = D(p) − s = t − s | t ∈ D(p) .



(c) For each t ∈ R, the set Mt is open in M , and the map θt : Mt → M−t is a diffeo-
morphism with inverse θ−t .

Proof.
(a) Proposition 7.19 shows that there exists an integral curve of V starting at each point
p ∈ M . Suppose that γ and γ e are two integral curves of V defined on the same open
e(t0 ) for some t0 ∈ J. Consider the set
interval J such that γ(t0 ) = γ

S := t ∈ J | γ(t) = γ

e(t)

and observe that S ̸= ∅, because t0 ∈ S by hypothesis, and also that it is closed in J by

continuity. On the other hand, pick t1 ∈ S. Then in a smooth coordinate neighborhood
around the point p = γ(t1 ), γ and γ e are both solutions to the same ODE with the same
e(t1 ) = p. By the uniqueness part of Theorem 7.18, γ ≡ γ
initial condition γ(t1 ) = γ e on
an open interval containing t1 , which implies that S is open in J. Since J is connected,
we infer that S = J, which in turn shows that γ = γ e on all of J. Thus, any two integral
curves that agree at one point agree on their common domain.
For each p ∈ M , let D(p) be the union of all open intervals J ⊆ R containing 0 on
which an integral curve of V starting at p is defined. Define θ(p) : D(p) → M by letting
 Section 7.3. Flows 75

θ(p) (t) = γ(t), where γ is any integral curve starting at p and defined on an open interval
containing 0 and t. Since all such integral curves agree at t by the argument above, θ(p)
is well defined, and it is obviously the unique maximal integral curve of V starting at p.
Next, for the verification that the set

D := (t, p) ∈ R × M | t ∈ D(p)


is open (so that it is a flow domain) and that the map

θ : D → M, θ(t, p) := θ(p) (t)

satisfies the claimed properties, as well as for the proof of (b), we refer to [Lee13, Theorem
9.12], which makes heavy use of Theorem 7.18.
(c) The fact that Mt is open in M is an immediate consequence of the fact that D is open.
We have
(b)
p ∈ Mt =⇒ t ∈ D(p) =⇒ D(θ(t,p)) = D(p) − t
dfn
=⇒ −t ∈ D(θ(t,p)) =⇒ θt (p) ∈ M−t ,

which shows that θt maps Mt to M−t for any (fixed) t ∈ R. Moreover, the group laws
then show that θ−t ◦ θt is equal to the identity on Mt . Reversing the roles of t and −t
shows that θt ◦ θ−t is equal to the identity on M−t . This completes the proof of (c).

The flow whose existence and uniqueness are asserted in Theorem 7.26 is called the
flow generated by V , or just the flow of V .
The naturality of integral curves (see Proposition 7.22) translates into the following
naturality statement for flows.

Proposition 7.27 (Naturality of flows). Let F : M → N be a smooth map. Let X ∈

X(M ) and Y ∈ X(N ). Let θ be the flow of X and η be the flow of Y . If X and Y are
F -related, then for each t ∈ R it holds that F (Mt ) ⊆ Nt and ηt ◦ F = F ◦ θt on Mt :

F
Mt Nt
θt ηt

F
M−t N−t

Proof. See [Exercise Sheet 12, Exercise 5(a)].

Proposition 7.28 (Diffeomorphism invariance of flows). Let F : M → N be a diffeo-

morphism. If X ∈ X(M ) and if θ is the flow of X, then the flow of F∗ X ∈ X(N ) is
ηt = F ◦ θt ◦ F −1 , with domain Nt = F (Mt ) for each t ∈ R.

Proof. See [Exercise Sheet 12, Exercise 5(b)].

76 Chapter 7. Vector Fields and Flows

7.3.1 Complete Vector Fields

Example 7.29 (Global flows). The two smooth vector fields on the plane described in
Example 7.23 both had integral curves defined for all t ∈ R, so they generate global flows.
We can write them down explicitly:


(1) θV : R × R2 → R2 , t, (x, y) 7→ (x + t, y).
For each t ∈ R \ {0}, (θV )t translates the plane to the left (t < 0) or to the right (t > 0)
by a distance |t|.


(2) θW : R × R2 → R2 , t, (x, y) 7→ (x cos t − y sin t, x sin t + y cos t).
For each t ∈ R, (θW )t rotates the plane through an angle t about the origin.
There are also smooth vector fields whose integral curves are not defined for all t ∈ R.
Here are two such examples:
Example 7.30. Let (x, y) be the standard coordinates on R2 .
∂
(1) Consider M = R2 \ {(0, 0)} and V = ∈ X(M ).
∂x
The unique integral curve of V starting at (−1, 0) ∈ M is the curve γ(t) = (t − 1, 0),
cf. Example 7.23(1). However, it cannot be extended continuously past t = 1. This is
intuitively evident because of the “hole” in M at the origin.
(To prove it rigorously, suppose that γe is a continuous extension of γ past t = 1. Then
γ(t) → γe(1) ∈ R2 \ {(0, 0)} as t ↗ 1. But we may also consider γ as a map into
R2 by composing with the inclusion M ,→ R2 , and it is obvious from the formula that
γ(t) → (0, 0) as t ↗ 1. Since limits in R2 are unique, this is a contradiction.)
∂
(2) Consider M = R2 and W = x2 ∈ X(M ).
∂x
1


The unique integral curve of W starting at (1, 0) is γ(t) = 1−t , 0 . It cannot be extended
past t = 1, because its x-coordinate is unbounded as t ↗ 1.
Definition 7.31. A smooth vector field V on a smooth manifold M is called complete if
it generates a global flow or, equivalently, if each of its maximal integral curves is defined
for all t ∈ R.
It is not always easy to determine by looking at a vector field whether it is complete
or not. If one can solve the ODE explicitly to find all of the integral curves, and they
all exist for all time (as we did for the vector fields of Example 7.29), then the vector
field is complete. On the other hand, if one can find one single integral curve that cannot
be extended to all of R (as we did for the vector fields of Example 7.30), then it is not
complete. However, it is often impossible to solve the ODE explicitly, so it is useful to
have some general criteria for determining when a vector field is complete. The following
theorem provides such a criterion. For the details of its proof we refer to [Lee13, Lemma
9.15 and Theorem 9.16].
Theorem 7.32. Every compactly supported smooth vector field on a smooth manifold is
complete.
 Section 7.3. Flows 77

Corollary 7.33. Every smooth vector field on a compact smooth manifold is complete.

Exercise 7.34 (The escape lemma): Let M be a smooth manifold and let V be a smooth
vector field on M . Show that if γ : J → M is a maximal integral curve of V whose
domain
J has a finite least upper bound b ∈ R, then for any t0 ∈ J the image γ [t0 , b) of the
interval [t0 , b) under γ is not contained in any compact subset of M .
CHAPTER 8
DIFFERENTIAL FORMS

In this chapter we transfer the algebra of alternating tensors on a finite-dimensional

real vector space (see Appendix C) to smooth manifolds and begin to explore the basic
properties of differential forms. The heart of the chapter is the introduction of the most
important operation on differential forms, called the exterior derivative. It is one of
the very few differential operators that are naturally defined on every smooth manifold
without any arbitrary choices.

8.1 Differential 1-Forms

8.1.1 Covectors
Definition 8.1. Let M be a smooth manifold. For each p ∈ M we define the cotangent
space at p, denoted by T∗p M , to be the dual space of Tp M :

T∗p M := (Tp M )∗ .

Elements of T∗p M are called (tangent) covectors at p ∈ M .

Given smooth local
coordinates (xi ) on an open subset U ⊆ M , for each p ∈ U the

coordinate basis ∂x∂ i p for Tp M gives rise to a dual basis for T∗p M , which we denote
temporarily by λi |p . Any covector ω ∈ T∗p M can thus be written uniquely as

 
i ∂
ω = ωi λ p , where ωi = ω .
∂xi p

Given now another set of smooth local coordinates (exj ) whose domain contains p ∈ U ,
ej |p the basis for T∗ M dual to ∂ j . We can compute the components



denote by λ p ∂e
x p
∗
of the same covector ω ∈ Tp M with respect to the new coordinate system as follows.
According to (3.6), the coordinate vector fields transform as follows:

∂ xj
∂e ∂
= (p) (8.1)
∂xi p ∂xi xj
∂e p

79
80 Chapter 8. Differential Forms

Writing ω in both systems as

ω = ωi λi p
=ω ej ,
ej λ p

we can use (8.1) to compute ωi in terms of ωej :

 j
xj xj
    
∂ ∂e
x ∂ ∂e ∂ ∂e
ωi = ω = ω (p) = (p) ω = (p) ω
ej . (8.2)
∂xi p ∂xi xj p
∂e ∂xi xj p
∂e ∂xi

8.1.2 The Cotangent Bundle

Definition 8.2. Let M be a smooth manifold. The cotangent bundle of M is denoted by
T∗ M and is defined as the disjoint union
G
T∗ M = T∗p M.
p∈M

It has a natural projection map

π : T∗ M → M, ω ∈ T∗p M 7→ p.
As in Subsection 8.1.1, given any smooth
local coordinates (xi ) on an open
subset
U ⊆ M , for each p ∈ M we denote by λi |p the basis for T∗p M dual to ∂x∂ i p . This
defines n maps
λ1 , . . . , λ n : U → T ∗ M
(to be denoted differently soon), and λi is called the i-th coordinate covector field .
Proposition 8.3 (The cotangent bundle as vector bundle). Let M be a smooth n-
manifold. With its standard projection map and the natural vector space structure on
each fiber, the cotangent bundle T∗ M has a unique topology and smooth structure making
it into a smooth vector bundle of rank n over M for which all coordinate covector fields
are smooth local sections.
Proof. (Similar to the proof of Proposition 6.4.) Given any smooth chart (U, φ) for M
with coordinate functions (xi ), define a map
Φ : π −1 (U ) → U × Rn ,

ξi λi


p
7→ p, (ξ1 , . . . , ξn ) ,

where λi is the i-th coordinate covector field associated with (xi ). Suppose that (U e , φ)
e is
−1 e
another smooth chart for M with coordinate functions (e x ), and let Φ : π (U ) → U × Rn
j e e
−1
be defined analogously. On π (U ∩ U e ), it follows from (8.2) that
  j
xj

−1

1 n

∂e
x ∂e
Φ◦Φ e p, (ξ , . . . , ξ ) = p,
e e (p) ξj , . . . , n (p) ξj
e e .
∂x1 ∂x
xj
The GL(n, R)-valued function ∂e


∂x i is smooth, so it follows from the vector bundle chart
lemma (= Lemma 6.6) that T∗ M has a smooth structure making it into a smooth vector
bundle for which the maps Φ are smooth local trivializations. Uniqueness follows as in
the proof of [Exercise Sheet 10, Exercise 6].
 Section 8.1. Differential 1-Forms 81

As in the case of the tangent bundle (see the proof of Proposition 3.12), smooth local
coordinates for M yield smooth local coordinates for its cotangent bundle. If (xi ) are
smooth coordinates on an open subset U ⊆ M , then [Exercise Sheet 10, Exercise 5(d)]
shows that the map

π −1 (U ) → R2n , ξi λi p 7→ x1 (p), . . . , xn (p), ξ1 , . . . , ξn ,

is a smooth coordinate chart for T∗ M . We call (xi , ξi ) the natural coordinates for T∗ M
associated with (xi ).

8.1.3 Covector Fields

Definition 8.4. A rough (resp. continuous, smooth) local or global section of T∗ M is
called a rough (resp. continuous, smooth) covector field or a (differential) 1-form on the
smooth manifold M .

⇝ The set X∗ (M ) of all smooth (global) covector fields on a smooth manifold M is an

infinite-dimensional R-vector space and a module over the ring C ∞ (M ): this is a
special case of [Exercise Sheet 10, Exercise 3(b)].

⇝ Extension lemma for covector fields: this is a special case of Lemma 6.10; see also
[Exercise Sheet 10, Exercise 3(d)] for an application (any tangent covector at a point
can be extended to a smooth covector field on the entire manifold).

⇝ Local/global coframe for M = local/global frame for T∗ M , see Definition 6.12.

⇝ Completion of smooth local coframes for M : this is a special case of [Exercise Sheet
10, Exercise 4].

In any smooth local coordinates (xi ) on an open subset U ⊆ M , a (rough) covector

field ω can be written in terms of the coordinate covector fields (λi ) as ω = ωi λi for
n functions ωi : U → R, called the component functions of ω in the given chart and
characterized by  
∂
ωi (p) = ωp .
∂xi p
If ω is a (rough) covector field and if X is a (rough) vector field on M , then we can
form a function
ω(X) : M → R, p 7→ ωp (Xp ).
∂
If we write ω = ωi λi and X = X i ∂xi
in terms of local coordinates, then ω(X) has the
local coordinate representation
ω(X) = ωi X i .
Just as in the case of vector fields (see Proposition 7.2 and Proposition 7.4), there are
several ways to check smoothness of a covector field (see also Proposition 6.14).
Proposition 8.5 (Smoothness criteria for covector fields). Let M be a smooth manifold
and let ω : M → T∗ M be a rough covector field on M . The following are equivalent:
(a) ω is smooth.
82 Chapter 8. Differential Forms

(b) In every smooth chart, the component functions of ω are smooth.

(c) Each point of M is contained in some coordinate chart in which ω has smooth
component functions.

(d) For every X ∈ X∗ (M ), the function ω(X) : M → R is smooth.

(e) For every open subset U ⊆ M and every smooth vector field X on U , the function
ω(X) : U → R is smooth.

Proof. See [Exercise Sheet 13, Exercise 1].

Of course, since any open subset of a smooth manifold is again a smooth manifold,
Proposition 8.5 applies equally well to covector fields defined only on some open subset
of M .


Example 8.6. For any smooth chart U, (xi ) , the coordinate covector fields (λi ) defined
above constitute a local coframe over U , called a coordinate coframe. By Proposition 8.5,
every coordinate coframe is smooth, because its component functions in the given chart
are constants.

More generally, if (Ei ) is a (rough) local frame for TM over an open subset U ⊆
M ,
then there is a uniquely determined (rough) local coframe (εi ) over U such that εi |p is
the dual basis to (Ei |p ) for each p ∈ U , or equivalently εi (Ej ) = δji . This coframe is called
the coframe dual to (Ei ). Conversely, if (εi ) is a (rough) local coframe over an open subset
U ⊆ M , then there is a uniquely determined (rough) local frame (Ei ) for TM over U ,
called the frame dual to (εi ) and determined by εi (Ej ) = δji . For example, in a smooth
chart, the coordinate frame ∂x∂ i and the coordinate coframe (λi ) are dual to each other.

Lemma 8.7. Let M be a smooth manifold. If (Ei ) is a rough local frame over an open
subset U ⊆ M and if (εi ) is its dual coframe, then (Ei ) is smooth if and only if (εi ) is
smooth.

Proof. It suffices to show that for each p ∈ U , the frame

(Ei ) is smooth in a neighborhood
of p if and only if (εi ) is. Given p ∈ U , let V, (xi ) be a smooth coordinate chart such
that p ∈ V ⊆ U and write

∂
Ei = aki and εj = bjℓ λℓ
∂xk

for some matrices of real-valued functions aki and bjℓ defined on V . By virtue of Propo-
sitions 7.2 and 8.5, the vector fields Ei are smooth on V if and only if the functions aki
are smooth, and the covector fields εj are smooth on V if and only if the functions bℓj are
smooth. The fact that εj (Ei ) = δij implies that the matrices (aki ) and (bℓj ) are inverses to
each other. Since matrix inversion is a smooth map GL(n, R) → GL(n, R), we conclude
that either one of these matrix-valued functions is smooth if and only if the other one is
smooth.
 Section 8.1. Differential 1-Forms 83

8.1.4 The Differential of a Smooth Function

The most important application of covector fields is that they enable us to interpret in
a coordinate-independent way the partial derivatives of a smooth function as the compo-
nents of a covector field.
Let f ∈ C ∞ (M ). We define a covector field df , called the differential of f at p ∈ M ,
by
dfp (v) = vf, v ∈ Tp M.
Proposition 8.8. The differential of a smooth function is a smooth covector field.
Proof. It is straightforward to check that dfp ∈ T∗p M for all p ∈ M . To verify that df is
smooth we apply Proposition 8.5(d): for any X ∈ X(M ), the function df (X) is smooth,
because it is equal to Xf (see Proposition 7.4).
For a smooth real-valued function f : M → R on a smooth manifold M , we now have
two different definitions for the differential of f at p ∈ M . In Chapter 3 we defined dfp
as a linear map Tp M → Tf (p) R, while here we defined dfp as a covector at p ∈ M , i.e., a
linear map Tp M → R. These are really the same object, once we take into account the
canonical identification between Tf (p) R and R; one easy way to see this is to note that
both are represented in coordinates by the row matrix whose components are the partial
derivatives of f . (Let us verify this below for df defined as above.)
Let us compute the coordinate representation of df . Let (xi ) be smooth coordinates
on an open subset U ⊆ M and let (λi ) be the corresponding coordinate coframe on U .
Write df in coordinates as dfp = Ai (p) λi |p for some functions Ai : U → R. Then the
definition of df implies
 
∂ ∂ ∂f
Ai (p) = dfp i
= i
f= (p) ,
∂x p ∂x p ∂xi

which yields the following formula for the coordinate representation of df :

∂f
dfp = (p) λi p . (8.3)
∂xi
Thus, the component functions of df in any smooth coordinate chart are the partial
derivatives of (the coordinate representation of) f with respect to those coordinates. Due
to this, we can think of df as an analogue of the classical gradient (the vector field in
Rn whose components are the partial derivatives of the function), reinterpreted in a way
that makes coordinate-independent sense on a manifold.
If we apply (8.3) to the special case in which f is one of the coordinate functions
j
x : U → R, we obtain

∂xj
dxj p
= (p) λi p
= δij λi p
= λj p ;
∂xi
in other words, the coordinate vector field λj is none other than the differential dxj .
Therefore, (8.3) can be rewritten as
∂f
dfp = (p) dxi p , p ∈ U,
∂xi
84 Chapter 8. Differential Forms

or as an equation between covector fields instead of covectors

∂f
df = i
dxi . (8.4)
∂x
In particular, in the 1-dimensional case, this reduces to

df
df = dx .
dx
Thus, we have recovered the familiar classical expression for the differential of a function
f in coordinates. Henceforth, we abandon the notation λi for the coordinate coframe, and
use dxi instead.

Example 8.9. If
f : R2 → R, (x, y) 7→ x2 y cos x,
then
∂(x2 y cos x) ∂(x2 y cos x)
df = dx + dy
∂x ∂y
= (2xy cos x − x2 y sin x) dx + (x2 cos x) dy .

Proposition 8.10 (Properties of the differential). Let M be a smooth manifold and let
f, g ∈ C ∞ (M ). The following statements hold:

(a) If a, b ∈ R, then d(af + bg) = a df + b dg.

(b) d(f g) = f dg + g df .

(c) d(f /g) = (g df − f dg)/g 2 on the set where g ̸= 0.

(d) If J ⊆ R is an interval containing the image of f and if h : J → R is a smooth

function, then d(h ◦ f ) = (h′ ◦ f ) df .

(e) If f is constant, then df = 0. Conversely, if df = 0, then f is constant on each

connected component of M .

Proof. See [Exercise Sheet 13, Exercise 2].

8.1.5 Pullback of Covector Fields

Definition 8.11. Let F : M → N be a smooth map and let p ∈ M . The differential (or
tangent map) dFp : Tp M → TF (p) N yields a dual linear map dFp∗ : T∗F (p) N → T∗p M , called
the (pointwise) pullback by F at p (or the cotangent map of F at p) and characterized by

dFp∗ (ω)(v) = ω dFp (v) , ω ∈ T∗F (p) N, v ∈ Tp M.

Unlike vector fields, whose pushforwards are defined only in certain special cases (see,
e.g., Subsection 7.1.2), covector fields always pullback to covector fields.
 Section 8.1. Differential 1-Forms 85

Definition 8.12. Let F : M → N be a smooth map and let ω : N → T∗ N be a rough

covector field on N . We define a rough covector field F ∗ ω : M → T∗ M on M , called the
pullback of ω by F , by
(F ∗ ω)p = dFp∗ (ωF (p) ). (8.5)
It acts on a vector v ∈ Tp M by
(F ∗ ω)p (v) = ωF (p) dFp (v) .

Proposition 8.13. Let F : M → N be a smooth map and let ω be a (continuous) covector

field on N . If u : N → R is a continuous function, then
F ∗ (uω) = (u ◦ F )F ∗ ω.
If additionally u : N → R is smooth, then
F ∗ (du) = d(u ◦ F ).
Proof. Regarding the first statement, for any p ∈ M we have
(8.5)
F ∗ (uω)p == dFp∗ (uω)F (p) = dFp∗ u(F (p)) ωF (p)

lin.
(8.5)
== u(F (p)) dFp∗ ωF (p) == (u ◦ F )(p) (F ∗ ω)p

= (u ◦ F )(F ∗ ω) p ,

which implies the assertion.

Regarding the second statement, if p ∈ M and v ∈ Tp M , then

(8.5)
F ∗ (du) p (v) == dFp∗ (duF (p) ) (v)

dfn

== duF (p) dFp (v)
and the identific.
using ES4E1(b)
immediately

dfn of du

===== dFp (v) u
dfn of dFp
===== v(u ◦ F )
dfn of d(u◦F )
======= d(u ◦ F )p (v) ,
which yields the assertion.
Proposition 8.14. Let F : M → N be a smooth map and let ω be a (continuous) covector
field on N . Then F ∗ ω is a (continuous) covector field on M , and if ω is smooth, then so
is F ∗ ω.
Proof. Fix p ∈ M and choose smooth coordinates (y j ) for N in a neighborhood V of
F (p). Set U = F −1 (V ) and observe that U is a neighborhood of p in M . Writing ω in
coordinates as ω = ωj dy j for (continuous) functions on V and using Proposition 8.13
twice (for F |U ), we compute that
F ∗ ω = F ∗ (ωj dy j ) = (ωj ◦ F )F ∗ dy j = (ωj ◦ F ) d(y j ◦ F ) . (8.6)
In view of Proposition 8.8, this expression is continuous, and it is smooth when ω is
smooth, so we are done.
86 Chapter 8. Differential Forms

Formula (8.6) for the pullback of a covector field can also be written in the following
way:
F ∗ ω = (ωj ◦ F ) d(y j ◦ F ) = (ωj ◦ F ) dF j ,
where F j is the j-th component function of F in these coordinates. Using either of these
formulas, the computation of pullbacks in coordinates is quite simple.

Example 8.15. Consider the smooth map

F : R3 → R2 , (x, y, z) 7→ (x2 y, y sin z) = (u, v)

and the smooth covector field

ω = u dv + v du ∈ X∗ (R2 ) .

According to (8.6), we have

F ∗ ω = (u ◦ F ) d(v ◦ F ) + (v ◦ F ) d(u ◦ F )
= (x2 y) d(y sin z) + (y sin z) d(x2 y)
= (x2 y)(sin z dy + y cos z dz) + y sin z (2xy dx + x2 dy)
= (2xy 2 sin z) dx + (2x2 y sin z) dy + (x2 y 2 cos z) dz.

In other words, to compute F ∗ ω, all we need to do is substitute the component func-

tions of F for the coordinate functions of N everywhere they appear in ω.

Remark 8.16. Let F : M → N and G : N → P be smooth maps between smooth

manifolds (with or without boundary) and let η ∈ X∗ (P ). Then

∗
G ◦ F η = F ∗ (G∗ η).

Indeed, given p ∈ M and v ∈ Tp M , using [Exercise Sheet 4, Exercise 1(b)] we obtain



F ∗ (G∗ η) p (v) = (G∗ η)F (p) dFp (v) = ηG(F (p)) dGF (p) dFp (v)




 
∗ 
= η(G◦F )(p) d G ◦ F p (v) = G ◦ F η (v),
p

which yields the assertion.

8.1.6 Covector Fields and Submanifolds

In Subsection 7.1.3 we considered the conditions under which a (smooth) vector field
restricts to a submanifold. The restriction of a (smooth) covector field to a submanifold
is much simpler and will be briefly discussed below.
Let M be a smooth manifold, let S ⊆ M be an immersed submanifold and let ι : S ,→
M be the inclusion map. If ω ∈ X∗ (M ), then ι∗ ω ∈ X∗ (S). More precisely, given p ∈ S
and v ∈ Tp S, we have
(ι∗ ω)p v = ωp dιp (v) = ωp (v),


 Section 8.2. Differential k-Forms 87

since dιp : Tp S → Tp M is just the inclusion map under our usual identification of Tp S
with the subspace dιp (Tp S) of Tp M . Thus, ι∗ ω is just the restriction of ω to vectors
tangent to S. For this reason, ι∗ ω is often called the restriction of ω to S. Note, however,
that ι∗ ω might equal zero at a given point of S, even though considered as a covector field
on M , ω might not vanish there. For example:
Example 8.17. Consider ω = dy ∈ X∗ (R2 ) and let S : (y = 0) be the x-axis, considered
as an embedded submanifold of R2 . As a covector field on R2 , ω is clearly nonzero
everywhere, because one of its components is always equal to 1. However, the restriction
ι∗ ω of ω to S is identically zero, because y vanishes identically on S:

ι∗ ω = ι∗ dy = d(y ◦ ι) = 0.

To distinguish the two ways in which we might interpret the statement “ω vanishes
on S”, one usually says that ω vanishes along S (or vanishes at points of S) if ωp = 0 for
every p ∈ S. The weaker condition that ι∗ ω = 0 is expressed by saying that the restriction
of ω to S vanishes (or the pullback of ω to S vanishes).

8.2 Differential k-Forms

Definition 8.18. Let M be a smooth n-manifold and fix k ∈ N.
(a) We define the bundle of covariant k-tensors on M by
G
Tk (T∗ M ) := Tk (T∗p M )
p∈M

with the obvious projection map. It can be shown (exercise!) that it is a smooth
vector bundle of rank nk over M . Its (smooth) sections are called (smooth) covariant
k-tensor fields on M .
(b) The subset of Tk (T∗ M ) consisting of alternating k-tensors is defined as:
Vk ∗ G Vk
(T M ) := (T∗p M ).
p∈M

It can be shown (exercise!) that k (T∗ M ) is a
smooth subbundle of Tk (T∗ M ),

V
and thus it is a smooth vector bundle of rank nk over M . Its sections are called
(differential) k-forms on M ; they are (continuous) tensor fields whose value at each
point is an alternating k-tensor. The integer k is called the degree of the form. We
denote the vector space of smooth (differential) k-forms by

Ωk (M ) := Γ k (T∗ M ) .
V

⇝ A 0-form is a continuous real-valued function on M , because

(T∗p M ) ∼
V0 ∗ G V0 G
(T M ) = = R = M × R,
p∈M p∈M

see [Exercise Sheet 10, Exercise 3(c)].

88 Chapter 8. Differential Forms

⇝ A 1-form is a continuous covector field on M , because

(T∗p M ) ∼
V1 ∗ G V1 G
(T M ) = = T∗p M = T∗ M.
p∈M p∈M

The wedge product of two differential forms is defined pointwise:

(ω ∧ η)p = ωp ∧ ηp .

Thus, the wedge product of a k-form with an ℓ-form is a (k + ℓ)-form. In particular, if f

is a 0-form and if η is a k-form, then we interpret the wedge product f ∧ η to mean the
ordinary product f η; see (6.1).

Comment: If we define n
M
∗
Ω (M ) = Ωk (M ),
k=0

then the wedge product turns Ω∗ (M ) into an associative, anti-commutative, graded R-

algebra.


In any smooth chart U, (xi ) , a k-form ω can be written as
X′ X′
ω= ωI dxi1 ∧ . . . ∧ dxik = ωI dxI ,
I I

where the coefficients ωI are continuous functions defined on the coordinate domain U ,
we use dxI as an abbreviation for dxi1 ∧ . . . ∧ dxik (where I = (i1 , . . . , in )) and the
primed summation sign denotes a sum over only increasing multi-indices. According to
Proposition 6.14, ω is smooth if and only if the component functions ωI are smooth. Since
 
i1 ik ∂ ∂
dx ∧ . . . ∧ dx j
,..., j = δJI
∂x 1 ∂x k

by Lemma C.20 and Proposition C.25(c), the component functions ωI of ω are determined
by  
∂ ∂
ωI = ω ,..., i .
∂xi1 ∂x k

8.2.1 Pullback of k-Forms

If F : M → N is a smooth map and if ω is a differential k-form on N , then F ∗ ω is a
differential k-form on M , defined as follows:

(F ∗ ω)p (v1 , . . . , vk ) := ωF (p) dFp (v1 ), . . . , dFp (vk ) .

Lemma 8.19. The following statements hold:

(a) F ∗ : Ωk (N ) → Ωk (M ) is linear over R.

(b) F ∗ (ω ∧ η) = F ∗ ω ∧ F ∗ η.
 Section 8.2. Differential k-Forms 89

(c) In any smooth chart V, (y i ) for N , we have
X′  X′
F∗ ωI dy i1 ∧ . . . ∧ dy ik = (ωI ◦ F ) d(y i1 ◦ F ) ∧ . . . ∧ d(y ik ◦ F ).
I I

Proof. See [Exercise Sheet 14, Exercise 1].

This lemma gives a computational rule for pullbacks of differential forms similar to
the one we developed earlier for covector fields, see (8.6).
Example 8.20. Consider the smooth function
F : R2 → R3 , (u, v) 7→ (u, v, u2 − v 2 ) = (x, y, z)
and the smooth 2-form
ω = y dx ∧ dz + x dy ∧ dz ∈ Ω2 (R3 ) .
Then
F ∗ ω = F ∗ y dx ∧ dz + x dy ∧ dz

= v du ∧ d(u2 − v 2 ) + u dv ∧ d(u2 − v 2 )
du∧du=0
= v du ∧ (2u du − 2v dv) + u dv ∧ (2u du − 2v dv) =====
dv∧dv=0

du∧dv=
= −2v 2 du ∧ dv + 2u2 dv ∧ du =====
−dv∧du

= −2(u2 + v 2 ) du ∧ dv.
Proposition 8.21 (Pullback formula for top forms). Let F : M → N be a smooth map
between smooth n-manifolds. If (xi ) and (y j ) are smooth coordinates on open subsets
U ⊆ M and V ⊆ N , respectively, and if u is a real-valued function on V , then the
following holds on U ∩ F −1 (V ):
F ∗ u dy 1 ∧ . . . ∧ dy n = (u ◦ F ) det DF dx1 ∧ . . . ∧ dxn ,



(8.7)
where DF represents the Jacobian matrix of F in these coordinates.
Proof. Since the fiber of n (T∗ M ) is spanned by dx1 ∧. . .∧dxn at each
V
∂ ∂

point, it suffices to
show that both sides of (8.7) agree when evaluated on ∂x1 , . . . , ∂xn . By Lemma 8.19(c)
we have
F ∗ u dy 1 ∧ . . . ∧ dy n = (u ◦ F ) d(y 1 ◦ F ) ∧ . . . ∧ d(y n ◦ F ),


| {z } | {z }
F1 Fn
so by Proposition C.25(c)(d) we obtain

 ∂ ∂ 
 ∂ ∂ 
F ∗ u dy 1 ∧ . . . ∧ dy n , . . . , = (u ◦ F ) dF 1
∧ . . . ∧ dF n
, . . . ,
∂x1 ∂xn ∂x1 ∂xn
  ∂  
= (u ◦ F ) det dF j
∂xi

1 n

 ∂ ∂ 
= (u ◦ F ) det DF dx ∧ . . . ∧ dx 1
,..., n ,
| {z ∂x ∂x }
=1
as desired.
90 Chapter 8. Differential Forms

Corollary 8.22. If U, (xi ) and U xj ) are overlapping smooth coordinate charts on
e , (e
a smooth manifold M , then the following identity holds on U ∩ U e:

xj 
 ∂e
x1 ∧ . . . ∧ de
de xn = det dx1 ∧ . . . ∧ dxn .
∂xi
Proof. Apply Proposition 8.21 for F = IdU ∩Ue , but using coordinates (xi ) in the domain
xj ) in the codomain.
and (e

8.2.2 The Exterior Derivative

We now define a natural differential operator on smooth forms, called the exterior deriva-
tive, which is a generalization of the differential of a function. More precisely, for each
smooth manifold M , we will show that there is a differential operator d : Ωk (M ) →
Ωk+1 (M ) satisfying d(dω) = 0 for all ω. P′ J
The definition of d on Euclidean space is straightforward: if ω = J ωJ dx is a
smooth k-form on an open subset U ⊆ Rn , its exterior derivative dω is defined to be the
following (k + 1)-form X  X
′ ′
J
d ωJ dx = dωJ ∧ dxJ , (8.8)
J J

where dωJ is the differential of the smooth function ωJ , see Subsection 8.1.4. In somewhat
more detail, this is
X  XX
′
j1 jk ∂ωJ i
d ωJ dx ∧ . . . ∧ dx = i
dx ∧ dxj1 ∧ . . . ∧ dxjk .
J J i
∂x

For instance, for a smooth 0-form f we have

∂f
df = dxi ,
∂xi
which is just the differential of f , see (8.4), while for a smooth 1-form ω = ωj dxj we
compute that
X  ∂ωj ∂ωi  i
dω = i
− j dx ∧ dxj .
i<j
∂x ∂x

In order to transfer this definition to manifolds, we first need to check that it satisfies
the following properties.

Proposition 8.23 (Properties of the exterior derivative on Rn ).

(a) d is R-linear.

(b) If ω is a smooth k-form and η is a smooth ℓ-form on an open subset U ⊆ Rn , then

d(ω ∧ η) = dω ∧ η + (−1)k ω ∧ dη.

(c) d ◦ d ≡ 0.
 Section 8.2. Differential k-Forms 91

(d) d commutes with pullbacks: If F : U ⊆ Rn → V ⊆ Rm is a smooth map between

open subsets of Euclidean spaces, and if ω ∈ Ωk (V ), then

F ∗ (dω) = d(F ∗ ω).

Proof.

(a) Follows immediately from the definition.

(b) Due to (a), it suffices to consider terms of the form ω = u dxI ∈ Ωk (U ) and η =
v dxJ ∈ Ωℓ (U ), where u, v ∈ C ∞ (U ).

 Claim: For any multi-index I we have

d(u dxI ) = du ∧ dxI .

 Proof: If I has repeated indices, then clearly d(u dxI ) = 0 = du ∧ dxI . Otherwise, let
σ be a permutation sending I to an increasing multi-index J. Then

d(u dxI ) = sgn(σ) d(u dxJ ) = sgn(σ) du ∧ dxJ = du ∧ dxI .

Using the claim, we compute

d(ω ∧ η) = d (u dxI ) ∧ (v dxJ )

= d(uv dxI ∧ dxJ )

dfn dv∧dxI =
== (v du + u dv) ∧ dxI ∧ dxJ =========
(−1)k dxI ∧dv

= (du ∧ dxI ) ∧ (v dxJ ) + (−1)k (u dxI ) ∧ (dv ∧ dxJ )

Claim
=== d(u dx}I ) ∧ (v| dx
| {z
J k
{z }) + (−1) (u dx}I ) ∧ d(v| dx
| {z
J
{z }) .
=ω =η =ω =η

(c) We first deal with the case of a smooth 0-form u:

 ∂u
i
 ∂ 2u dxi ∧dxi =0
d(du) = d i
dx = i j
dxi ∧ dxj ======
∂x ∂x ∂x
X  ∂ 2u ∂ 2u  i
= i ∂xj
− j ∂xi
dx ∧ dxj
i<j
∂x ∂x

=0
92 Chapter 8. Differential Forms

Let us now deal with the general case (u = ′J ωJ dxJ ∈ Ωk (U )):

P

X 
′
j1 jk
d(du) = d dωJ ∧ dx ∧ . . . ∧ dx
J

(a) X′ * 0 by jcase k=0

d(dω ) ∧ dx 1 ∧ . . . ∧ dxjk

==  +
(b)  J
J

0 by (b) and
by case k=0
X′  :
j1  jk


+ (−1) · dωJ ∧ d dx ∧
 . . . ∧ dx
J


= 0.

(d) Due to (a), it suffices to consider ω = u dxi1 ∧ . . . ∧ dxik . We have

Lemma 8.19(b)(c) &
F ∗ d(u dxi1 ∧ . . . ∧ dxik ) = F ∗ (du ∧ dxi1 ∧ . . . ∧ dxik ) ===========


P roposition 8.13

1
(∗)
= d(u ◦ F ) ∧ d(xi1 ◦ F ) ∧ . . . ∧ d(xik ◦ F ) ==

Lemma 8.19(c)
= d (u ◦ F ) d(xi1 ◦ F ) ∧ . . . ∧ d(xik ◦ F ) =========

= d F ∗ (u dxi1 ∧ . . . ∧ dxik ) .

Example 8.24. Let us compute the exterior derivatives of arbitrary 1-forms and 2-forms
on R3 .

• Any smooth 1-form ω on R3 can be written as

ω = P dx + Q dy + R dz

for some smooth functions P, Q, R on R3 . Using (8.8) and the fact that the wedge product
of any 1-form with itself is zero, we compute

dω = dP ∧ dx + dQ ∧ dy + dR ∧ dz
 
∂P ∂P ∂P
= dx + dy + dz ∧ dx +
∂x ∂y ∂z
   
∂Q ∂Q ∂Q ∂R ∂R ∂R
+ dx + dy + dz ∧ dy + dx + dy + dz ∧ dz
∂x ∂y ∂z ∂x ∂y ∂z
     
∂Q ∂P ∂R ∂P ∂R ∂Q
= − dx ∧ dy + − dx ∧ dz + − dy ∧ dz .
∂x ∂y ∂x ∂z ∂y ∂z
1
(∗): We have an expression of the form df ∧ η, where η = dg1 ∧ . . . ∧ dgk (with f = u ◦ F and
gℓ = xiℓ ◦ F ), so
p. 87 (b)
d(f η) === d(f ∧ η) == df ∧ η + (−1)0 f ∧ dη = df ∧ η,
since dη = 0 by (b) and (c).
 Section 8.2. Differential k-Forms 93

• Any smooth 2-form η on R3 can be written as

η = u dx ∧ dy + v dx ∧ dz + w dy ∧ dz

for some smooth functions u, v, w on R3 . Similarly, we compute

 
∂u ∂v ∂w
dη = − + dx ∧ dy ∧ dz .
∂z ∂y ∂x

Theorem 8.25 (Existence and uniqueness of exterior differentiation). Let M be a smooth

manifold. For each k ∈ N there are unique operators

d : Ωk (M ) → Ωk+1 (M ) ,

called exterior differentiation, satisfying the following properties:

(a) d is R-linear.

(b) If ω ∈ Ωk (M ) and η ∈ Ωℓ (M ), then

d(ω ∧ η) = dω ∧ η + (−1)k ω ∧ dη.

(c) d ◦ d ≡ 0.

(d) For f ∈ Ω0 (M ) = C ∞ (M ), df is the differential of f , given by df (X) = Xf .

In any smooth chart, d is given by (8.8).
Proof.
 Existence: Given ω ∈ Ωk (M ), for each smooth chart (U, φ) for M , we set
∗

dω := φ∗ d (φ−1 ) ω . (8.9)

This is well-defined, since for any other smooth chart (V, ψ) for M , the map φ ◦ ψ −1 is a
diffeomorphism between open subsets of Rn , so
∗
∗

ψ ∗ d (ψ −1 ) ω = (φ−1 ◦ φ)∗ ψ ∗ d (ψ −1 ) ω
| {z }
Id

∗ (φ−1 )∗ ψ ∗ =(ψ◦φ−1 )∗
= φ∗ (φ−1 )∗ ψ ∗ d (ψ −1 ) ω


============
& P roposition 8.23(d)

∗ ∗
= φ∗ d (ψ ◦ φ−1 ) (ψ −1 ) ω


| {z }
(ψ −1 ◦ ψ ◦ φ−1 )∗ =(φ−1 )∗

∗

= φ∗ d (φ−1 ) ω .

Moreover, d satisfies (a) - (d) by virtue of Proposition 8.23.

 Uniqueness: Suppose that d is any operator satisfying (a) – (d). We first show that d
is determined locally: if ω1 and ω2 are k-forms that agree on an open subset U ⊆ M , then
dω1 = dω2 on U . Indeed, let p ∈ U , set η := ω1 − ω2 and let ψ ∈ C ∞ (M ) be a smooth
94 Chapter 8. Differential Forms

bump function that is identically 1 on some neighborhood of p and supported in U . Then

ψη is identically zero, so (a) - (d) imply that 0 = d(ψη) = dψ ∧ η + ψ dη. Evaluating this
at p and using that ψ(p) = 1 and dψp = 0, we conclude that 0 = dηp = dω1 |p − dω2 |p ,
which proves the assertion.
Now, let ω ∈ Ωk (M ) and let (U, φ) be a smooth chart for M . Write ω in coordinates
as ′I ωI dxI . For any p ∈ U by means of a smooth bump function we construct global
P
smooth functions ω eI and xei on M that agree with ωI and dxi in a neighborhood of p.
By virtue of (a) – (d), together with the observation in the previous paragraph, it follows
that (8.8) holds at p. Since p was arbitrary, this d must be equal to the one we defined
above.

Comment: The preceding theorem can be summarized by saying that the differential on
functions extends uniquely to an anti-derivation of Ω∗ (M ) of degree +1 whose square is
zero.

Proposition 8.26 (Naturality of exterior derivative). If F : M → N is a smooth map,

then for each k the pullback map F ∗ : Ωk (N ) → Ωk (M ) commutes with d, i.e.,

F ∗ (dω) = d(F ∗ ω), ∀ ω ∈ Ωk (N ).

Proof. Applying Proposition 8.23(d) to the coordinate representation ψ ◦ F ◦ φ−1 of F

and using (8.9), on U ∩ F −1 (V ) we obtain
 
∗

F ∗ (dω) = F ∗ ψ ∗ d (ψ −1 ) ω
∗ ∗

= φ∗ (ψ ◦ F ◦ φ−1 ) d (ψ −1 ) ω
∗ ∗

= φ∗ d (ψ ◦ F ◦ φ−1 ) (ψ −1 ) ω
∗
= φ∗ d (φ−1 ) (F ∗ ω)

= d(F ∗ ω).

Definition 8.27. Let M be a smooth manifold and let ω ∈ Ωk (M ). We say that ω is

closed if dω = 0, and exact if there exists η ∈ Ωk−1 (M ) such that ω = dη.

Remark 8.28. Every exact form is closed, since d ◦ d ≡ 0, but the converse does not hold
in general, see Example 11.27. However, it can be shown that closed forms are locally
exact (but not necessarily globally), so the question of whether a given closed form is
exact depends on global properties of the manifold.
CHAPTER 9
MANIFOLDS WITH BOUNDARY

We briefly discuss manifolds with boundary. They play a central role in the theory of
integration on manifolds, which will be developed in Chapter 11.

9.1 Topological Manifolds with Boundary

Definition 9.1. The closed n-dimensional upper half-space Hn ⊆ Rn is defined as

Hn = (x1 , . . . , xn ) ∈ Rn | xn ≥ 0 .


The interior and the boundary of Hn as a subset of Rn are denoted by Int Hn and ∂Hn ,
respectively.
If n > 0, then

Int Hn = (x1 , . . . , xn ) ∈ Rn xn > 0 ,



∂Hn = (x1 , . . . , xn ) ∈ Rn xn = 0 ,


whereas if n = 0, then
H0 = R0 = {0} and ∂H0 = ∅ .
Definition 9.2. An n-dimensional topological manifold with boundary is a second-coun-
table, Hausdorff topological space M in which every point has a neighborhood homeo-
morphic either to an open subset of Rn or to a (relatively) open subset of Hn .
An open subset U ⊆ M together with a map φ : U → Rn that is a homeomorphism
onto an open subset of Rn or Hn is called a chart for M . When it is necessary to make the
distinction, we call (U, φ) an interior chart for M if φ(U ) is an open subset of Rn (which
includes the case of an open subset of Hn that does not intersect ∂Hn ), and a boundary
chart for M if φ(U ) is a open subset of Hn such that φ(U ) ∩ ∂Hn ̸= ∅.
A point p ∈ M is called an interior point of M if it is in the domain of some interior
chart, and a boundary point of M if it is in the domain of a boundary chart that sends p to
∂Hn . The set of all boundary points of M is denoted by ∂M and is called the boundary of
M , while the set of all interior points of M is denoted by Int M and is called the interior
of M .

95
96 Chapter 9. Manifolds with Boundary

Theorem 9.3 (Topological invariance of the boundary). If M is a topological manifold

with boundary, then each point of M is either a boundary point or an interior point, but
not both. Thus, ∂M and Int M are disjoint sets whose union is M .

Example 9.4.

(1) Every interval in R is a connected topological 1-manifold with boundary, whose man-
ifold boundary consists of its endpoints (if any).
n
(2) The closed unit ball B ⊆ Rn is a connected topological n-manifold with boundary,
whose (manifold) boundary is Sn−1 and whose interior is Bn ; see [Lee13, Problem 1.11].

Proposition 9.5. Let M be a topological manifold with boundary.

(a) Int M is an open subset of M and a topological n-manifold without boundary.

(b) ∂M is a closed subset of M and a topological (n − 1)-manifold without boundary.

(c) M is a topological manifold (in the sense of Definition 1.1) if and only if ∂M = ∅.

Proof. Exercise!

9.2 Smooth Manifolds with Boundary

If U is an open subset of Hn , then a map F : U → Rk is said to be smooth if for each
x ∈ U there exists an open subset U e ⊆ Rn containing x and a smooth map Fe : U e → Rk
that agrees with F on Ue ∩ U . If F is such a map, then the restriction of F to U ∩ Int Hn is
smooth in the usual sense. By continuity, all partial derivatives of F at points of U ∩Int Hn
are determined by their values in Int Hn , and thus in particular are independent of the
choice of extension.

Definition 9.6. Let M be a topological manifold with boundary. A smooth structure for
M is defined to be a maximal smooth atlas (a collection of charts whose domains cover M
and whose transition maps (and their inverses) are smooth in the sense just described).
With such a structure, M is called a smooth manifold with boundary.

In the following lengthy remark we collect some basic definitions and facts about
smooth manifolds with boundary, referring to [Lee13] for further information.

Remark 9.7.

(1) Cf. Chapter 2: Smoothness of a map F : M → N between manifolds with boundary

is defined in the same way (see Definition 2.4), with the usual understanding that a map
whose domain is a subset of Hn is smooth if it admits an extension to a smooth map in
a neighborhood of each point, and a map whose codomain is a subset of Hn is smooth if
it is smooth as a map into Rn .
Smooth partitions of unity exist on smooth manifolds with boundary.
 Section 9.2. Smooth Manifolds with Boundary 97

(2) Cf. Chapter 3: If M is a smooth n-manifold with boundary, then the tangent space
Tp M to M at p ∈ M is defined in the same way (see Definition 3.4), and it is an n-
dimensional R-vector space. For any smooth chart (U, (xi )) containing p, the coordinate
vectors
∂ ∂
1
,..., n
∂x p ∂x p

(where ∂x∂n p
should be interpreted as a one-sided derivative when p ∈ ∂M ) form a basis
for Tp M .
Let M be a smooth manifold with boundary and let p ∈ ∂M . It is intuitively evident
that the vectors in Tp M can be separated in three classes: those tangent to the boundary,
those pointing inward, and those pointing outward. Formally, we make the following
definition:

Definition: If p ∈ ∂M , then the vector v ∈ Tp M \ Tp ∂M it said to be inward-pointing

if for some ε > 0 there exists a smooth curve γ : [0, ε) → M such that γ(0) = p and
γ ′ (0) = v, and it is called outward-pointing if there exists such a curve with domain
(−ε, 0].

Proposition: Let M is a smooth manifold with boundary, p ∈ ∂M , and (xi ) be any

smooth boundary coordinates defined on a neighborhood of p. The inward-pointing vec-
tors in Tp M are precisely those with positive xn -component, the outward-pointing ones
are those with negative xn -component, and the ones tangent to ∂M are those with zero
xn -component. Thus, Tp M is the disjoint union of Tp ∂M , the set of inward-pointing
vectors, and the set of outward-pointing vectors. Finally, v ∈ Tp M is inward-pointing if
and only if −v is outward-pointing.

The differential of a smooth map F : M → N between manifolds with boundary is

defined in the same way (see Definition 3.6) and has the same representation in coordinates
bases.

(3) Cf. Chapter 4: Submersions, immersions, embeddings and local diffeomorphisms are
defined in the same way (see Definitions 4.2 and 4.7(b)), and there is a version of the rank
theorem in this setting (see [Lee13, Theorem 4.15 and Problem 4.3]).

(4) Cf. Chapter 5: Immersed and embedded submanifolds of smooth manifolds with
boundary are defined in the same way (see 5.1 and 5.12) and are themselves smooth
manifolds with (possibly empty) boundary.

⇝ For properties of (immersed) submanifolds with boundary, see [Lee13, Chapter 5,

Submanifolds with Boundary].
⇝ For a version of the regular level set theorem in this setting (cf. Theorem 5.9), see
[Lee13, Problem 5.23].

Theorem: If M is a smooth n-manifold with boundary, then with the subspace topology,
∂M is a topological (n − 1)-manifold (without boundary), and has a unique smooth
structure such that it is a properly embedded submanifold of M .
98 Chapter 9. Manifolds with Boundary

(5) Cf. Chapter 7: The tangent bundle of a smooth n-fold with boundary is defined in
the same way (see Definition 3.11) and it is a smooth vector bundle of rank n over the
given manifold (see Proposition 6.4). Vector fields are also defined in the same way (see
Definition 7.1), but flows in this setting need to be treated with extra care (see [Lee13,
Chapter 9, Flows and Flowouts on Manifolds with Boundary]).

Proposition: If M is a smooth manifold with boundary, then there exists a smooth

vector field on M whose restriction to ∂M is everywhere inward-pointing, and one whose
restriction to ∂M is everywhere outward-pointing.

(6)
Vk Cf. Chapter 8: The cotangent bundle T∗ M (respectively the k-th exterior power
(T∗ M ) of the cotangent bundle) of a smooth n-manifold M with boundary is defined in
the same way (see Definition 8.2, respectively
Definition 8.18(b)), and it is a smooth vec-
tor bundle of rank n (respectively of rank nk ) over M (see Proposition 8.3, respectively
Definition 8.18(b)). Differential k-forms (0 ≤ k ≤ n) are also defined in the same way
(see Definition 8.18(b)), and so does their exterior derivative as well (see Theorem 8.25).
CHAPTER 10
ORIENTATIONS

The purpose of this chapter is to introduce a subtle but important property of smooth
manifolds, called orientation. An orientation of a line or a curve is simply a choice of direc-
tion along it. For 2-dimensional manifolds, an orientation is essentially a choice of which
rotational direction should be considered “clockwise” and which “counterclockwise”. For
3-dimensional ones, it is a choice between “left-handedness” and “right-handedness”. The
general definition of an orientation is an adaptation of these everyday concepts to arbitrary
dimensions.

10.1 Orientations of Vector Spaces

In this section we discuss orientations of vector spaces. We are all familiar with certain
informal rules for singling out preferred ordered bases of R1 , R2 , and R3 . We usually
choose a basis for R1 that points to the right (i.e., in the positive direction). A natural
family of preferred ordered bases for R2 consists of those for which the rotation from the
first vector to the second is in the counterclockwise direction. And every student of vector
calculus encounters “right-handed” bases in R3 : these are the ordered bases (E1 , E2 , E3 )
with the property that when the fingers of your right hand curl from E1 to E2 , your thumb
points in the direction of E3 .
Although “to the right”, “counterclockwise”, and “right-handed” are not mathematical
terms, it is easy to translate the rules for selecting preferred bases of R1 , R2 , and R3 into
rigorous mathematical language: in all three cases, the preferred bases are the ones whose
transition matrices from the standard basis have positive determinants.
In an abstract vector space for which there is no canonical basis, we no longer have any
way to determine which bases are “correctly oriented”. For example, if V is the vector
space of polynomials in one real variable of degree at most 2, who is to say which of the
ordered bases (1, x, x2 ) and (x2 , x, 1) is “right-handed”? All we can say in general is what
it means for two bases to have the “same orientation”. Thus, we are led to introduce the
following definition.

Definition 10.1. Let V be a real vector space of dimension n ≥ 1. We say that two or-
dered bases (E1 , . . . , En ) and (E
e1 , . . . , E
en ) for V are consistently oriented if the transition

99
100 Chapter 10. Orientations

matrix (Bij )1≤i,j≤n , defined by X

Ei = Bij E
ej ,
j

has positive determinant.

Exercise 10.2: Show that being consistently oriented is an equivalence relation on the
set of all ordered bases of V , and show that there are exactly two equivalence classes.
Definition 10.3. Let V be a real vector space.
• If dimR V = n ≥ 1, we define an orientation for V as an equivalence class of ordered
bases. If (E1 , . . . , En ) is any ordered basis for V , then we denote the orientation
that it determines by [E1 , . . . , En ], and the opposite orientation by −[E1 , . . . , En ].

• If dimR V = 0, we define an orientation for V to be simply a choice of one of the

numbers ±1.
Definition 10.4. A vector space together with a choice of orientation is called an oriented
vector space. If V is oriented, then any ordered basis (E1 , . . . , En ) that is in the given
orientation is said to be positively oriented (or simply oriented ). Any ordered basis that
is not in the given orientation is said to be negatively oriented .
Example 10.5. Consider the Euclidean space V = Rn . The orientation [e1 , . . . , en ] of
Rn determined by the standard basis {e1 , . . . , en } is called the standard orientation. You
should convince yourself that, in our usual way of representing the axes graphically, an
oriented basis for R1 is one that points to the right; an oriented basis for R2 is one for which
the rotation from the first basis vector to the second is counterclockwise; and an oriented
basis for R3 is a right-handed one. (These can be taken as mathematical definitions for
the words “right”, “counterclockwise”, and “right-handed”.) The standard orientation
for R0 is defined to be +1.
There is an important connection between orientations and alternating tensors, which
is expressed in the following proposition.
Proposition 10.6. Let V be a real vector space of dimension n. Each nonzero element
ω ∈ Λn (V ∗ ) determines an orientation Oω of V as follows: if n ≥ 1, then Oω is the set
of ordered bases (E1 , . . . , En ) for V such that ω(E1 , . . . , En ) > 0, while if n = 0, then Oω
is +1 if ω > 0, and −1 if ω < 0. Moreover, two nonzero n-covectors on V determine the
same orientation if and only if each is a positive multiple of the other.
Proof. The 0-dimensional case is immediate, since a nonzero element of Λ0 (V ∗ ) is just a
nonzero real number (as it is a function R0 → R). Thus, we may assume that n ≥ 1. Let
ω be a nonzero element of Λn (V ∗ ), and denote by Oω the set of ordered bases on which ω
gives positive values. We need to show that Oω is exactly one equivalence class.
Suppose (Ei ) and (E ej ) are any two ordered bases for V , and let B : V → V be the
linear map sending Ej to E ej for all j. This means that the matrix representation of B
with respect to (Ei ) on the source and (E ej ) on the target is the transition matrix between
the two bases. By Proposition C.22 we obtain

ω(E
e1 , . . . , E
en ) = ω(BE1 , . . . , BEn ) = (det B) ω(E1 , . . . , En ).
 Section 10.2. Orientations of Smooth Manifolds 101

It follows that the basis (E ej ) is consistently oriented with (Ei ) if and only if ω(E
e1 , . . . , E
en )
and ω(E1 , . . . , En ) have the same sign, which is the same as saying that Oω is one equiv-
alence class. The last statement then follows easily (and is thus left as an exercise).

Definition 10.7. If V is an oriented n-dimensional real vector space and if ω is an n-

covector that determines the orientation of V as described in Proposition 10.6, then we
say that ω is a (positively) oriented n-covector .

For example, the n-covector ε1...n = ε1 ∧ · · · ∧ εn is positively oriented for the standard
orientation on Rn ; see Lemma C.20(c).
Recall that if V is an n-dimensional real vector space, then the vector space Λn (V ∗ ) is
1-dimensional by Proposition C.21. Proposition 10.6 shows that choosing an orientation
for V is equivalent to choosing one of the two components of Λn (V ∗ ) \ {0}. This formu-
lation also works for 0-dimensional vector spaces, and explains why we have defined an
orientation of a 0-dimensional space in the way we did.

10.2 Orientations of Smooth Manifolds

In this section we briefly discuss the theory of orientations of smooth manifolds. They
have numerous applications, most notably in the theory of integration on manifolds, see
Chapter 11.

Definition 10.8. Let M be a smooth manifold with or without boundary. A pointwise

orientation on M is defined to be a choice of orientation of each tangent space.

By itself, this is not a very useful concept, because the orientations at nearby points
may have no relation to each other. For example, a pointwise orientation on Rn might
switch randomly from point to point between the standard orientation and its opposite.
In order for pointwise orientations to have some relationship with the smooth structure,
we need an extra condition to ensure that the orientations of nearby tangent spaces are
consistent with each other.

Definition 10.9. Let M be a smooth manifold with or without boundary, endowed with
a pointwise orientation. If (Ei ) is a local frame for T M over an open subset U ⊆ M ,
then we say that (Ei ) is positively oriented (or simply oriented ) if (E1 |p , . . . , En |p ) is a
positively oriented ordered basis for Tp M at each point p ∈ U ; see Definition 10.4. A
negatively oriented frame for T M over U ⊆ M is defined analogously.

Definition 10.10. Let M be a smooth manifold with or without boundary (of dimension
n ≥ 1).

(a) A pointwise orientation on M is said to be continuous if every point of M is in the

domain of an oriented local frame for T M .

(b) An orientation of M is a continuous pointwise orientation.

(c) We say that M is orientable if there exists an orientation for it; otherwise we say
that M is nonorientable.
102 Chapter 10. Orientations

Exercise 10.11: Let M be an oriented smooth manifold with or without boundary of

dimension n ≥ 1. Show that every local frame with connected domain is either positively
oriented or negatively oriented. Moreover, show that the connectedness assumption is
necessary.

Example 10.12. We give here some examples of orientable and nonorientable manifolds.

(1) Every parallelizable1 manifold is orientable. Indeed, if (E1 , . . . , En ) is a smooth

global frame for M , then we define a pointwise orientation on M by declaring the basis
(E1 |p , . . . , En |p ) for Tp M to be positively oriented at each p ∈ M , and it is clear that
this pointwise orientation is continuous, because every point of M is in the domain of the
oriented smooth global frame (Ei ). Therefore, for each n ∈ N, the Euclidean space Rn is
orientable.

(2) For each n ∈ N, the unit n-sphere Sn ⊆ Rn+1 is orientable. Indeed, this follows
from Proposition 10.21, because Sn is a hypersurface in Rn+1 , to which the vector field
N = xi ∂/∂xi is nowhere tangent. We define the standard orientation of Sn to be the one
determined by N . (The standard orientation of S0 is the one that assigns the orientation
+1 to the point +1 ∈ S0 and −1 to the point −1 ∈ S0 .) Alternatively, this follows from
Proposition 10.23, because Sn is the boundary of the closed unit ball. (It can be checked
that the orientation thus induced on Sn is the standard one.)

(3) The so-called Möbius band is nonorientable; see [Lee13, Examples 10.3 and 15.38].

Definition 10.13. An oriented manifold (with or without boundary) is an orderer pair

(M, O), where M is an orientable smooth manifold (with or without boundary) and O is
a choice of orientation for M . For each p ∈ M , the orientation of Tp M determined by O
is denoted by Op .

If M is zero-dimensional, then this definition just means that an orientation of M

is a choice of ±1 attached to each of its points. The continuity condition is vacuous in
this case, and the notion of oriented frames is not useful. Clearly, every 0-manifold is
orientable.

10.2.1 Two Ways of Specifying Orientations

The following two propositions, namely Proposition 10.14 and Proposition 10.18, give
ways of specifying orientations on manifolds that are more practical to use than the
definition.

Proposition 10.14 (The orientation determined by an n-form). Let M be a smooth

n-manifold with or without boundary. Any nonvanishing n-form ω on M determines a
unique orientation of M for which ω is positively oriented at each point. Conversely, if
M is given an orientation, then there is a smooth nonvanishing n-form on M that is
positively oriented at each point.
1
A smooth manifold M with or without boundary which admits a smooth global frame or, equivalently,
whose tangent bundle T M is the trivial smooth vector bundle of rank dim M (see [Exercise Sheet 10,
Exercise 5]) is called parallelizable. Note that the Euclidean space Rn is parallelizable, and it can also be
shown that S1 , S3 and S7 are the only spheres that are parallelizable.
 Section 10.2. Orientations of Smooth Manifolds 103

Proof.
“⇒”: Let ω be a nonvanishing n-form on M . By Proposition 10.6, ω defines a pointwise
orientation on M , so it remains to show that it is continuous. Since this is trivially true
for n = 0, we may assume that n ≥ 1. Given p ∈ M , let (Ei ) be any local frame for T M
over a connected open neighborhood U of p in M , and let (εi ) be the dual coframe. The
expression for ω in this frame over U is
ω = f ε 1 ∧ . . . ∧ εn
for some continuous function f on U . The fact that ω is nonvanishing means that f is
nonvanishing, and thus by Lemma C.20(c) we obtain
ωp (E1 |p , . . . , En |p ) = f (p) ̸= 0 for all p ∈ U.
Since U is connected, it follows that this expression is either always positive or always
negative on U , and therefore the given frame is either positively oriented or negatively
oriented. If the latter case holds, then we can replace E1 by −E1 to obtain a new frame
that is positively oriented. Hence, the pointwise orientation determined by ω is continuous.
“⇒”: We refer to [Lee13, Proposition 15.5] for the details.
Due to Proposition 10.14, we may now give the following definition.
Definition 10.15. Let M be a smooth n-manifold with or without boundary. Any
nonvanishing n-form on M is called an orientation form. If M is oriented and if ω is an
orientation form determining the given orientation, then we also say that ω is positively
oriented (or simply oriented ).
If M is zero-dimensional, then a nonvanishing 0-form (i.e., a nonvanishing smooth
real-valued function) on M assigns the orientation +1 to points where it is positive and
−1 to points where it is negative.
Remark 10.16. It is straightforward to check (see Proposition 10.6) that if ω and ω
e are
two positively oriented smooth n-forms on M , then ωe = f ω for some strictly positive
smooth real-valued function f on M .
Definition 10.17.


(a) A smooth coordinate chart U, (xi ) on an oriented smooth manifold with or without
boundary is said to be positively oriented (or simply oriented ) if the coordinate
frame (∂/∂xi ) is positively oriented, and negatively oriented if the coordinate frame
(∂/∂xi ) is negatively oriented; see Definition 10.9.
(b) A smooth atlas {(Uα , φα )} for a smooth manifold M with or without boundary is
said to be consistently oriented if for each α, β, the transition map φβ ◦ φ−1
α has
positive Jacobian determinant everywhere on φα (Uα ∩ Uβ ).
Proposition 10.18 (The orientation determined by a coordinate atlas). Let M be a
smooth manifold with or without boundary of dimension n ≥ 1. Given any consistently
oriented smooth atlas for M , there exists a unique orientation for M with the property
that each chart in the given atlas is positively oriented. Conversely, if M is oriented and
either ∂M = ∅ or n > 1, then the collection of all oriented smooth charts is a consistently
oriented atlas for M .
104 Chapter 10. Orientations

Proof. Assume first that M has a consistently oriented smooth atlas. Each chart in the
atlas determines a pointwise orientation at each point of its domain. Wherever two of
the charts overlap, the transition matrix between their respective coordinate frames is
the Jacobian matrix of the transition map (see the bottom of p. 27 and (3.6)), which has
positive determinant by assumption, so they determine the same pointwise orientation at
each point. The pointwise orientation on M thus determined is continuous, because each
point of M is in the domain of an oriented coordinate frame.
Conversely, assume that M is oriented and either ∂M = ∅ or n > 1. Each point is
in the domain of a smooth chart with connected domain, and if the chart is negatively
oriented (see Exercise 10.11), then we can replace x1 with −x1 to obtain a new chart that
is positively oriented. The fact that all these charts are positively oriented guarantees that
their transition maps have positive Jacobian determinants, so they form a consistently
oriented atlas.2

Exercise 10.19: Let M be a connected, orientable, smooth manifold with or without

boundary. Show that M has exactly two orientations. Moreover, if two orientations of M
agree at one point, then they are equal.

10.2.2 Orientations of Hypersurfaces

If M is an oriented smooth manifold and if S is an immersed submanifold of M (with
or without boundary), then S might not inherit an orientation from M , even if S is
embedded. Clearly, it is not sufficient to restrict an orientation form from M to S, since
the restriction of an n-form to a manifold of lower dimension must necessarily be zero.
For example, the Möbius band (see Example 10.12(3)) is nonorientable, even though it
can be embedded in R3 , which is orientable.
However, when S is an immersed or embedded hypersurface in M (i.e., a codimension
1-submanifold of M ), it is sometimes possible to use an orientation on M to induce an
orientation on S; see Proposition 10.21 below for the details. We first need to introduce
the following definitions.

Definition 10.20. Let M be a smooth manifold with or without boundary and let S ⊆ M
be an immersed submanifold with or without boundary. A vector field along S is a section
of the ambient tangent bundle T M |S , i.e., a continuous map N : S → T M with the
property that Np ∈ Tp M for every p ∈ S. Such a vector field is said to be nowhere
tangent to S if Np ∈ Tp M \ Tp S for all p ∈ S; cf. Subsection 7.1.3.

Note that any vector field on M restricts to a vector field along S (not necessarily
tangent to S), but in general not every vector field along S is of this form, see Lemma 6.11.

Proposition 10.21. Let M be an oriented smooth n-manifold with or without boundary,

let S be an immersed hypersurface with or without boundary in M , and let N be a vector
field along S which is nowhere tangent to S. Then S has a unique orientation such that for
each p ∈ S, (E1 , . . . , En−1 ) is an oriented basis for Tp S if and only if (Np , E1 , . . . , En−1 )
is an oriented basis for Tp M .
2
This does not work for boundary charts when dim M = n = 1, because of our convention that the
last coordinate is nonnegative in a boundary chart.
 Section 10.2. Orientations of Smooth Manifolds 105

Proof. See [Lee13, Proposition 15.21].

Figure 10.1: The orientation induced by a nowhere tangent vector field

Note that not every hypersurface admits a nowhere tangent vector field, see for instance
[Lee13, Problem 15.6]. However, the following result gives a sufficient condition that holds
in many cases.

Corollary 10.22. If M is an oriented smooth manifold and if S ⊆ M is a regular level

set of a smooth function f : M → R, then S is orientable.

Proof. See [Lee13, Proposition 15.23].

10.2.3 Boundary Orientations

If M is a smooth manifold with boundary ∂M ̸= ∅, then ∂M is an embedded hypersurface
without boundary in M (see the Theorem in Remark 9.7(4)) and there always exists a
smooth outward-pointing vector field along ∂M (see the Proposition in Remark 9.7(5)).
Since such a vector field is nowhere tangent to ∂M (see the Proposition in Remark 9.7(2)),
it determines an orientation on ∂M by Proposition 10.21, provided that M is oriented.
The following proposition shows that this orientation is independent of the choice of an
outward-pointing vector field along ∂M , and it is called the induced orientation or the
Stokes orientation on ∂M .

Proposition 10.23 (The induced orientation on a boundary). Let M be an oriented

smooth n-manifold with boundary, where n ≥ 1. Then ∂M is orientable, and all outward-
pointing vector fields along ∂M determine the same orientation on ∂M .

Proof. See [Lee13, Proposition 15.24].

Example 10.24. We determine the induced orientation on ∂Hn when Hn itself has the
standard orientation inherited from Rn . We can identify ∂Hn with Rn−1 under the corre-
spondence
(x1 , . . . , xn−1 , 0) ↔ (x1 , . . . , xn−1 ).
Since the vector field −∂/∂xn is outward-pointing along Hn , the standard coordinate
frame for Rn−1 is positively oriented for ∂Hn if and only if [−∂/∂xn , ∂/∂x1 , . . . , ∂/∂xn−1 ]
is the standard orientation for Rn ; see Proposition 10.21. This orientation satisfies

−∂/∂xn , ∂/∂x1 , . . . , ∂/∂xn−1 = − ∂/∂xn , ∂/∂x1 , . . . , ∂/∂xn−1

   

= (−1)n ∂/∂x1 , . . . , ∂/∂xn−1 , ∂/∂xn .

 

Thus, the induced orientation on ∂Hn is equal to the standard orientation on Rn−1 when
n is even, but it is opposite to the standard orientation when n is odd. In particular, the
standard coordinates on ∂Hn ≈ Rn−1 are positively oriented if and only if n is even.
106 Chapter 10. Orientations

10.2.4 Orientations and Smooth Maps

Definition 10.25. Let M and N be oriented smooth manifolds with or without boundary
and let F : M → N be a local diffeomorphism.

• If both M and N are positive-dimensional, then we say that F is orientation-

preserving if for each p ∈ M , the isomorphism dFp : Tp M → TF (p) N takes positively
oriented bases of Tp M to positively oriented bases of TF (p) N , and orientation-
reversing if it takes positively oriented bases of Tp M to negatively oriented bases
of TF (p) N .

• If both M and N are zero-dimensional, then we say that F is orientation-preserving

if for every p ∈ M , the points p and F (p) have the same orientation, and it is
orientation-reversing if they have opposite orientation; see the paragraph after Def-
inition 10.13.

Remark 10.26. A composition of orientation-preserving maps is also orientation-preser-

ving.

Lemma 10.27. Let M and N be oriented positive-dimensional smooth manifolds with or

without boundary and let F : M → N be a local diffeomorphism. Show that the following
are equivalent:

(a) F is orientation-preserving.

(b) With respect to any positively oriented smooth charts for M and N , the Jacobian
matrix of F has positive determinant.

(c) If ω is any positively oriented orientation form for N , then F ∗ ω is a positively

oriented orientation form for M .

Proof. Exercise!
Here is another important method for constructing orientations.

Proposition 10.28 (The pullback orientation). Let M and N be smooth manifolds with
or without boundary. If F : M → N is a local diffeomorphism and if N is oriented, then
M has a unique orientation, called the pullback orientation induced by F , such that F is
orientation-preserving.

Proof. For each p ∈ M there is a unique orientation on Tp M that makes the isomorphism
dFp : Tp M → TF (p) N orientation-preserving. This defines a pointwise orientation on M ;
provided that it is continuous, it is the unique orientation on M with respect to which F
is orientation-preserving. To see that it is continuous, just choose a smooth orientation
form ω of N using Proposition 10.14 (so that ω is positively oriented) and note that F ∗ ω
is a smooth orientation form for M , determining by construction and by Proposition 10.14
the above pointwise orientation on M , which is thus continuous, as desired.
CHAPTER 11
INTEGRATION ON MANIFOLDS

11.1 Line Integrals

Another important application of covector fields (cf. Subsection 8.1.4) is to make coordi-
nate-independent sense of the notion of line integrals, which generalize ordinary integrals
to the setting of curves in manifolds.

Definition 11.1. Let M be a smooth manifold with or without boundary. A curve

segment in M is a continuous curve γ : [a, b] → M whose domain is a compact interval.
It is a smooth curve segment in M if it is smooth when [a, b] is considered as a manifold
with boundary (or, equivalently, if γ has an extension to a smooth curve defined in a
neighborhood of each endpoint). It is a piecewise smooth curve segment in M if there
exists a finite partition a0 = a < a1 < · · · < ak−1 < ak = b of [a, b] such that γ|[ai−1 ,ai ] is
smooth1 for every 1 ≤ i ≤ k.

Definition 11.2. Let M be a smooth manifold with or without boundary. Let ω be a

smooth covector field on M . If γ : [a, b] → M is a piecewise smooth curve segment, then
the line integral of ω over γ is defined to be the real number
Z k Z
X
ω := γ ∗ ω,
γ i=1 [ai−1 ,ai ]

where [ai−1 , ai ], 1 ≤ i ≤ k, are subintervals of [a, b] on which γ is smooth. If t denotes
∗

the standard coordinate on R, then the smooth covector field ωi := γ ∗ ω = γ|[ai−1 ,ai ] ω
on [ai−1 , ai ] can be written as ωi = fi (t) dt for some smooth function fi : [ai−1 , ai ] → R,
so the integral of ωi over [ai−1 , ai ] is given by
Z Z ai
ωi = fi (t) dt.
[ai−1 ,ai ] ai−1

1
Continuity of γ means that γ(t) approaches the same value as t approaches any of the points ai (other
than a0 or ak ) from the left or the right. Smoothness of γ in each subinterval means that γ has one-sided
velocity vectors at each such ai when approaching from the left or the right, but these one-sided velocities
need not be equal.

107
108 Chapter 11. Integration on Manifolds

Therefore,
Z k Z
X ai
ω= fi (t) dt.
γ i=1 ai−1

Proposition 11.3 (Properties of line integrals). Let M be a smooth manifold with or

without boundary. Let γ : [a, b] → M be a piecewise smooth curve segment in M , and let
ω, ω1 , ω2 ∈ X∗ (M ). The following statements hold:

(a) For any c1 , c2 ∈ R we have

Z Z Z


c1 ω1 + c2 ω2 = c1 ω1 + c2 ω2 .
γ γ γ

(b) If γ is a constant map, then

Z
ω = 0.
γ

(c) If γ1 := γ|[a,c] and γ2 := γ|[c,b] , where a, b, c ∈ R with a < c < b, then

Z Z Z
ω= ω+ ω.
γ γ1 γ2

(d) If F : M → N is any smooth map and if η ∈ X∗ (N ), then

Z Z
∗
F η= η.
γ F ◦γ

Proof.
(a) Follows immediately from the corresponding property of usual integrals.
(b) Since γ is constant, for any p ∈ [a, b] we have dγp = 0, and thus

(γ ∗ ω)p (v) = ωγ(p) dγp (v) = 0 for any v ∈ Tp [a, b],

which implies that γ ∗ ω = 0. Therefore,

Z Z
ω= γ ∗ ω = 0.
γ [a,b]

(c) Follows immediately from the corresponding property of usual integrals.

(d) By Remark 8.16 we deduce that
Z Z Z Z
∗ ∗ ∗ ∗


F η= γ F η = (F ◦ γ) η = η.
γ [a,b] [a,b] F ◦γ
 Section 11.1. Line Integrals 109

Example 11.4. Consider the smooth covector field ω on M = R2 \{0} given by

x dy − y dx
ω=
x2 + y 2
and the smooth curve segment
γ : [0, 2π] → M, t 7→ (cos t, sin t).
The line integral of ω over γ equals
Z 2π Z 2π
cos t(cos t dt) − sin t(− sin t dt)
Z Z
∗
ω= γ ω= = dt = 2π.
γ [0,2π] 0 sin2 t + cos2 t 0

Definition 11.5. Let M be a smooth manifold with or without boundary. If γ : [a, b] →

M and γ e : [c, d] → M are piecewise smooth curve segments in M , then we say that γ e is
a reparametrization of γ if γ e = γ ◦ φ for some diffeomorphism φ : [c, d] → [a, b]. If φ is
an increasing function (i.e., t1 < t2 =⇒ φ(t1 ) < φ(t2 )), then we say that γ e is a forward
reparametrization of γ, while if φ is a decreasing function (i.e., t1 < t2 =⇒ φ(t1 ) > φ(t2 )),
then we say that γ e is a backward reparametrization of γ. (More generally, with obvious
modifications one can allow φ to be piecewise smooth.)
Lemma 11.6 (Diffeomorphism invariance of the integral). Let ω be a smooth covector
field on the compact interval [a, b] ⊆ R and let φ : [c, d] → [a, b] be a diffeomorphism. We
have Z
Z


 ω if φ is increasing ,
 [a,b]
φ∗ ω = Z
[c,d] 
− ω if φ is decreasing .


[a,b]

Proof. Denote by s, resp. t, the standard coordinate on [c, d], resp. [a, b]. Then ω can
be written as ωt = f (t) dt for some smooth function f : [a, b] → R,
and now (8.4) and
(8.6) show that φ∗ ω has the coordinate expression (φ∗ ω)s = f φ(s) φ′ (s) ds. Inserting
this into the definition of the line integral and using the change of variables formula for
ordinary integrals, we obtain
Z b

Z Z d


 f (t) dt if φ is increasing,
 a
φ∗ ω = f φ(s) φ′ (s) ds =


[c,d] c  Z b

− f (t) dt if φ is decreasing,


a

which yields the statement.

Proposition 11.7 (Parameter independence of line integrals). Let M be a smooth man-
ifold with or without boundary, let ω ∈ X∗ (M ), and let γ be a piecewise smooth curve
segment in M . For any reparametrization γ e of γ we have
 Z
Z


 ω if γe is a forward reparametrization ,
 γ
ω= Z
γ 
− ω if γ e is a backward reparametrization .
e 

γ
110 Chapter 11. Integration on Manifolds

Proof. Exercise! (First deal with the case when γ is smooth using Lemma 11.6 and
Remark 8.16, and then treat the general case using Proposition 11.3(c).)
Proposition 11.8. Let M be a smooth manifold with or without boundary and let ω ∈
X∗ (M ). If γ : [a, b] → M is a piecewise smooth curve segment in M , then the line integral
of ω over γ can also be expressed as the ordinary integral
Z Z b
ωγ(t) γ ′ (t) dt.


ω=
γ a

Proof. See [Lee13, Proposition 11.38].

Theorem 11.9 (Fundamental theorem for line integrals). Let M be a smooth manifold
with or without boundary. Let f ∈ C ∞ (M ) and let γ be a piecewise smooth curve segment
in M . Then Z



df = f γ(b) − f γ(a) .
γ

Proof. Suppose first that γ is smooth. By combining Proposition 11.8, [Exercise Sheet
13, Exercise 3(a)] and the fundamental theorem of calculus we obtain
Z Z b Z b
′


′


df = dfγ(t) γ (t) dt = f ◦ γ (t) = f ◦ γ (b) − f ◦ γ (a).
γ a a

Suppose now that γ is merely piecewise smooth and consider a finite partition a0 =
a < a1 < · · · < ak−1 < ak = b of [a, b] such that γ|[ai−1 ,ai ] is smooth for every 1 ≤ i ≤ k.
In view of Proposition 11.3(c), applying the above argument on each subinterval and
summing, we find that
Z k 
X




df = f γ(ai ) − f γ(ai−1 ) = f γ(b) − f γ(a) ,
γ i=1

because the contributions from all the interior points cancel.

Example 11.10. Consider the smooth covector field
ω = 2xy 3 dx + 3x2 y 2 dy ∈ X∗ (R2 ) = Ω1 (R2 ).
Note that ω is exact, since ω = df for the function f : R2 → R, (x, y) 7→ x2 y 3 . We now
compute the line integral of ω along the arc of the parabola y = x2 from (0, 0) to (1, 1).
Since the latter can be parametrized by the smooth curve segment γ : [0, 1] → R2 , t 7→
(t, t2 ), by Theorem 11.9 we obtain
Z Z

: 1
: 0
ω = df =  f γ(1)
 − 
f γ(0)
 = 1.
γ γ

This can also be verified by a direct computation as follows: We have

γ ∗ ω = 2t(t2 )3 dt + 3t2 (t2 )2 d(t2 ) = 2t7 dt + 6t7 dt = 8t7 dt,
and hence Z Z Z 1  1
dfn ∗
ω == γ ω= 8t7 dt = t8 0 = 1.
γ [0,1] 0
 Section 11.2. Integration of Differential Forms 111

11.2 Integration of Differential Forms

We first define the integral of a differential form over a domain in Euclidean space, and
then we show how to use diffeomorphism invariance and smooth partitions of unity to
extend this definition to n-forms on oriented n-manifolds. The key feature of this definition
is that it is invariant under orientation-preserving diffeomorphisms. After developing the
general theory of integration of differential forms on oriented manifolds, we state (without
complete proof) one of the most important theorems in differential geometry: Stokes’
theorem. It is a far-reaching generalization of the fundamental theorem of calculus and
of the fundamental theorem for line integrals (Theorem 11.9), as well as of the classical
theorems of vector calculus, such as Green’s theorem (Theorem 11.23).

11.2.1 Integration in Rn
Definition 11.11. Let D ⊆ Rn be a domain of integration (i.e., a bounded subset of Rn
whose boundary has n-dimensional measure zero, such as a rectangle according to [Lee13,
Proposition C.18]), and let ω be a continuous n-form on D. Since ω can be written as
ω = f dx1 ∧ . . . ∧ dxn for some continuous function f : D → R, we define the integral of ω
over D to be the usual integral
Z Z Z Z
1 n 1 n
ω= f dx ∧ . . . ∧ dx := f dx . . . dx = f dV .
D D D D

(In simple terms, to compute the integral of a form such as f dx1 ∧ . . . ∧ dxn , just
“erase the wedges”.)

Definition 11.12. Let U be an open subset of Rn or Hn and let ω be a compactly

supported n-form on U . We define
Z Z
ω := ω,
U D

where D ⊆ Rn or Hn is any domain of integration containing supp ω, and ω is extended to

be zero on the complement of its support. Note that Definition 11.12 does not depend on
the choice of domain of integration, and the right-hand side reduces to Definition 11.11.

Proposition 11.13. Let D and E be open domains of integration in Rn or Hn , and let

G : D → E be a smooth map that restricts to an orientation-preserving or orientation-
reversing diffeomorphism D → E. If ω is an n-form on E, then
Z
Z


 ω, if G is orientation-preserving,
 E
G∗ ω = Z
D 
− ω , if G is orientation-reversing.


E

Proof. Follows from the (usual) change of variables formula ([Lee13, Theorem C.26])
and the pullback formula for n-forms (Proposition 8.21), taking also Lemma 10.27 into
account.
112 Chapter 11. Integration on Manifolds

As we cannot guarantee that arbitrary open or compact subsets are domains of inte-
gration, we need the following lemma in order to extend Proposition 11.13 to compactly
supported n-forms defined on open subsets.
Lemma 11.14. If U is an open subset of Rn or Hn and if K is a compact subset of U ,
then there is an open domain of integration D such that

K ⊆ D ⊆ D ⊆ U.

Proof. See [Lee13, Lemma 16.2].

Proposition 11.15. Let U and V be open subsets of Rn or Hn , and let G : U → V be
an orientation-preserving or orientation-reversing diffeomorphism. If ω is a compactly
supported n-form on V , then
Z
Z


 ω, if G is orientation-preserving,
 V
∗
Gω= Z
U 
−

 ω , if G is orientation-reversing.
V

Proof. By Lemma 11.14 there is an open domain of integration E such that

supp ω ⊆ E ⊆ E ⊆ V.

(See Figure 11.1.) Since diffeomorphisms take interiors to interiors, boundaries to bound-
aries, and sets of measure zero to sets of measure zero, we infer that D := G−1 (E) ⊆
U is an open domain of integration containing supp(G∗ ω). We conclude by Proposi-
tion 11.13.

Figure 11.1: Diffeomorphism invariance of the integral of a form on an open subset

Using the above proposition we can now make sense of the integral of a differential
n-form over an oriented n-manifold.

11.2.2 Integration on Manifolds

Definition 11.16. Let M be an oriented smooth n-manifold with or without boundary
and let ω be an n-form on M , where n ≥ 1. Suppose that ω is compactly supported in
the domain of a single smooth chart (U, φ) for M that is either positively or negatively
oriented. We define the integral of ω over M to be
Z Z
∗
ω=± (φ−1 ) ω (11.1)
M φ(U )

with the positive sign for a positively oriented chart, and the negative sign otherwise.
∗
(See Figure 11.2.) Since (φ−1 ) ω is a compactly supported n-form on the open subset
φ(U ) ⊆ Rn or Hn , its integral is defined as in Definition 11.12.
 Section 11.2. Integration of Differential Forms 113

Figure 11.2: The integral of a form over a manifold

R
Proposition 11.17. If M and ω are as above, then M
ω does not depend on the choice
of smooth chart whose domain contains supp ω.
Proof. Let (U, φ) and (U e be two smooth charts such that supp ω ⊆ U ∩ U
e , φ) e . If both
e ◦ φ−1 : φ(U ∩ U
charts are similarly oriented, then φ e ) → φ(U
e ∩U e ) is an orientation-
preserving diffeomorphism (see the proof of Proposition 10.18 and Lemma 10.27), so
Z Z Z
−1 ∗ −1 ∗ P roposition 11.15 ∗ ∗
(φe )ω= (φ
e ) ω ========== (φe ◦ φ−1 ) (φe−1 ) ω
φ(
eU e) φ(
eU e ∩U ) φ(U ∩U
e)
Z Z
∗ ∗ ∗
= (φ−1 ) φ
e∗ (φ
e−1 ) ω = (φ−1 ) ω .
φ(U ∩U
e) | {z } φ(U )
= Id∗

If the charts are oppositely oriented, then the two definitions given by (11.1) have opposite
signs, but is compensated by the fact that φ e ◦ φ−1 is orientation-reversing, so Proposi-
tion 11.15 introduces an
R extra negative sign into the above computation. In either case,
the two definitions of M ω agree.
To integrate over an entire manifold, we combine this definition with a partition of
unity.
Definition 11.18. Let M be an oriented smooth n-manifold with or without boundary
and let ω be a compactly supported n-form on M , where n ≥ 1. Let {Ui } be a finite open
cover of supp ω by domains of positively or negatively oriented smooth charts2 , and let
{ψi } be a smooth partition of unity subordinate to this cover. We define the integral of
ω over M to be Z XZ
ω= ψi ω . (11.2)
M i M

Since for each i the n-form ψi ω is compactly supported in Ui , each of the terms in this
(finite) sum is well defined according to our previous discussion.
The following proposition shows that the integral is well defined.
Proposition 11.19. The definition (11.2) does not depend on the choice of open cover
or partition of unity.
Proof. Let {U
ej } be another open cover of supp ω by domains of positively or negatively
oriented smooth charts, and let {ψej } be a subordinate smooth partition of unity. Since
Z Z X  XZ
ψi ω = ψj ψi ω =
e ψej ψi ω for every i,
M M j j M

we obtain XZ XZ
ψi ω = ψej ψi ω.
i M i,j M

2
The reason we allow for negatively oriented charts is that it may not be possible to find positively
oriented boundary charts on a 1-manifold with boundary, as noted in the proof of Proposition 10.18.
114 Chapter 11. Integration on Manifolds

Each term in this last sum is the integral of a form that is compactly supported in the
domain of a single smooth chart (e.g. in Ui ), so by Proposition 11.17 each term is well
defined, regardless
R of which coordinate map we use to compute it. The same argument,
starting with M ψej ω instead, shows that
XZ XZ
ψej ω = ψej ψi ω.
j M i,j M
R
Thus, both definitions yield the same value for M
ω.
We have a special definition in the zero-dimensional case. The integral of a compactly
supported 0-form (i.e., a real-valued function) f over an oriented 0-manifold M is defined
to be the sum Z X
f := ±f (p), (11.3)
M p∈M

where we take the positive sign at points where the orientation is positive and the negative
sign otherwise. The assumption that f is compactly supported implies that there are only
finitely many non-zero terms in this sum.
If S ⊆ M is an oriented immersed k-dimensional manifold (with or without boundary)
and
R if ω Ris a∗k-form on M whose restriction to S is compactly supported, then we interpret
S
ω as S ι ω, where ι : S ,→ M is the inclusion map. In particular, if M is a compact,
R n-manifold with boundary and if ω is an (n − 1)-form on M , then we
oriented, smooth
can interpret ∂M ω unambiguously as the integral of ι∗ ω over ∂M , where ∂M is always
understood to have the induced (Stokes) orientation; see Proposition 10.23.
Proposition 11.20 (Properties of integrals). Let M and N be nonempty oriented smooth
n-manifolds with or without boundary, and let ω and η be compactly supported n-forms
on M .
(a) Linearity: If a, b ∈ R, then
Z Z Z
aω + bη = a ω+b η.
M M M

(b) Orientation reversal: If −M denotes M with opposite orientation, then

Z Z
ω=− ω.
−M M

(c) Positivity: If ω is a positively oriented orientation form, then

Z
ω > 0.
M

(d) Diffeomorphism invariance: If F : N → M is an orientation-preserving or an

orientation-reversing diffeomorphism, then
Z
Z


 ω, if F is orientation-preserving,
 M
F ∗ω = Z
N 
−

 ω, if F is orientation-reversing.
M
 Section 11.2. Integration of Differential Forms 115

Proof.
(a) Exercise.
(b) Exercise (follows from the usual change of variables formula).
(c) Since ω is a positively oriented orientation form on M , if (U, φ) is a positively oriented
∗
smooth chart, then (φ−1 ) ω is a positive function times dx1 ∧ . . . ∧ dxn (while if (U, φ) is
negatively oriented, then it is a negative function times the Rsame form); see the proof of
Proposition 10.14. Therefore, each term in (11.2) defining M ω is nonnegative, with at
least one strictly positive term, proving thus (c).
(d) It suffices to treat the case when ω is compactly supported in a single positively
or negatively oriented smooth chart. If (U, φ) is a positively oriented such
chart and
if F is orientation-preserving, then it is easy to check that F −1 (U ), φ ◦ F is an ori-
ented smooth chart on N whose domain contains supp(F ∗ ω), so the result follows from
Proposition 11.15. The remaining cases follow from this result and (b).

11.2.3 Stokes’ theorem

We now state the cental result in theory of integration on manifolds, Stokes’ theorem.
However, we do not provide its complete proof; instead, we refer to [Lee13, Theorem
16.11] for the details.
Theorem 11.21 (Stokes’ theorem). Let M be an oriented smooth n-manifold with bound-
ary and let ω be a compactly supported smooth (n − 1)-form on M . Then
Z Z
dω = ω.
M ∂M

Here, ∂M is understood to have the induced (Stokes) orientation, and the ω on the
right-hand side is to be interpreted as ι∗∂M ω. If ∂M = ∅, then the right-hand side is to be
interpreted as 0. When M is 1-dimensional, the right-hand integral is just a finite sum,
see (11.3).
Proof of the case M = R2 . We have to show that
Z
dω = 0, where ω = f dx + g dy ∈ Ω1c (R2 ).
R2

Since f and g have compact support, we may pick r > 0 such that both supp(f ) and
supp(g) are contained in the interior of the square [−r, r] × [−r, r]. Then
Z Z  
∂f ∂g Fubini
dω = dy ∧ dx + dx ∧ dy ====
R2 R2 ∂y ∂x
Z rZ r Z rZ r
∂f ∂g
=− (x, y) dx dy + (x, y) dx dy
−r −r ∂y −r −r ∂x
Z r Z r
 y=r  x=r
=− f (x, y) y=−r dx + g(x, y) x=−r dy
−r | {z } −r | {z }
=0 =0

=0
116 Chapter 11. Integration on Manifolds

11.2.4 Applications of Stokes’ theorem

Example 11.22. Let M be
a smooth manifold. Let γ : [a, b] ,→ M be a smooth embed-
ding, so
 that S := γ [a, b] is an embedded 1-submanifold with 0-dimensional boundary
∂S = γ(a), γ(b) in M (and γ is a diffeomorphism onto its image S). If we give S the
orientation (via the differential dγp , p ∈ [a, b]) such that γ is orientation-preserving, then
for any f ∈ C ∞ (M ) we obtain
Z Z Z
Def inition 11.2 ∗

Def inition 11.18
df ========= γ df ==========−1
df
γ [a,b] γ : chart S
Z
T heorem 11.21 (11.3)


========= f === f γ(b) − f γ(a) ,
∂S

because the boundary orientation at γ(a) is −1, while at γ(b) is +1. Thus, Stokes’
theorem reduces to the fundamental theorem for line integrals (Theorem 11.9) in this
case. In particular, when γ : [a, b] ,→ R is the inclusion map, then Stokes’ theorem is just
the fundamental theorem of calculus.
Theorem 11.23 (Green’s theorem). Let D ⊆ R2 be a compact regular domain (i.e.,
properly embedded codimension-0 submanifold with boundary), and let P , Q be smooth
real-valued functions on D. Then
Z   Z
∂Q ∂P
− dx dy = P dx + Q dy.
D ∂x ∂y ∂D

Proof. Apply Stokes’ theorem to the 1-form P dx + Q dy.

In particular, with P (x, y) = −y and Q(x, y) = x, we compute the area of D:
Z
1
A(D) = (x dy − y dx) .
2 ∂D
Corollary 11.24 (Integrals of exact forms). If M is a compact, oriented, smooth n-
manifold without boundary, then the integral of any exact n-form over M is zero:
Z
dω = 0 if ∂M = ∅.
M

Corollary 11.25 (Integrals of closed forms over boundaries). If M is a compact, oriented,

smooth n-manifold with boundary, then the integral over ∂M of any closed (n − 1)-form
on M is zero: Z
ω = 0 if dω = 0 on M.
∂M

Corollary 11.26. Let M be a smooth n-manifold with or without boundary, let S ⊆ M

be an oriented, compact, smooth
R k-dimensional submanifold (without boundary), and let
ω be a closed k-form on M . If S ω ̸= 0, then both of the following are true:
(a) ω is not exact on M .

(b) S is not the boundary of an oriented, compact, smooth submanifold with boundary
in M .
 Section 11.2. Integration of Differential Forms 117

Proof.
(a) If ω were exact on M , then ω = dη for some (k − 1)-form η on M , so we have
Z Z Z Z
dfn ∗ ∗ P roposition 8.26 Corollary 11.24
0 ̸= ω == ιS ω = ιS (dη) ========== d(ι∗S η) ========= 0,
S S S S

which is a contradiction.
(b) Argue again by contradiction and invoke Corollary 11.25.

Example 11.27. Consider the smooth covector field

x dy − y dx
ω= 2 2
on M = R2 \{0}.
x +y
Now, we will show that ω is closed, but not exact. Indeed:

• Setting
x y
f (x, y) = and g(x, y) = − ,
x2 + y2 x2 + y2
so that ω = f dy + g dx, we compute that
∂f ∂g
dω = df ∧ dy + dg ∧ dx = dx ∧ dy + dy ∧ dx
∂x ∂y
y 2 − x2 y 2 − x2
= dx ∧ dy + dy ∧ dx
(x2 + y 2 )2 (x2 + y 2 )2
= 0,

which shows that ω is closed.

• If ω were exact, then there would exist a smooth function f : M → R such that ω = df ,
so by Stokes’ theorem and the fact that ∂γ = ∅ we would then obtain
Z Z Z
ω = df = f = 0,
γ γ ∂γ

which contradicts Example 11.4. Therefore, ω is not exact.


Finally, if (r, θ) are polar coordinates on the right half-plane H = (x, y) | x > 0 ⊆ M ,
then we may compute the polar coordinate expression for ω ∈ X∗ (M ) as follows: Since
x = r cos θ and y = r sin θ, we have
r cos θ r sin θ
ω= 2
d(r sin θ) − d(r cos θ)
r r2
cos θ sin θ
= (sin θ dr + r cos θ dθ) − (cos θ dr − sin θ dθ)
r r
= dθ.
APPENDIX A
THE REAL PROJECTIVE SPACE

Most of the smooth manifolds that we encountered in this course were intrinsically sub-
spaces of some Euclidean space Rn . However, the set-up of the general theory (that is,
endowing topological manifolds with a smooth structure) is designed precisely so as to
allow our objects of study to come along as abstract spaces, rather than requiring them
to be subsets of some Rn . Hence, it would be nice to see an example of a smooth manifold
which takes advantage of this abstract set-up. An elementary, yet important, example is
the real projective space RPn , which will be described in this appendix.

The underlying set of RPn :

Let n ∈ N∗ . There is a natural group action of R× := R \ {0} on Rn+1 \ {0} given by

R× × (Rn+1 \ {0}) → Rn+1 \ {0}

(λ, x) 7→ λx.

As with any group action, we can form the quotient set, whose points are the orbits of the
action. Concretely, we define the real projective space of dimension n, denoted by RPn ,
to be the quotient of the above action, i.e.,

RPn := Rn+1 \ {0} / R× .

Note that RPn comes equipped with a natural surjection

π : Rn+1 \ {0} → RPn

x 7→ [x] := R× · x.

In particular, notice that points of RPn are in one-to-one correspondence with one-
dimensional subspaces of Rn+1 : if [x] ∈ RPn , then [x] ∪ {0} = R · x is the one-dimensional
subspace of Rn+1 generated by x, while if L is any one-dimensional subspace of Rn+1 ,
then L \ {0} = [x] for any x ∈ L \ {0}. (This is the geometric picture you should have in
mind when thinking about RPn .) If

x = (x0 , . . . , xn )

119
120 Appendix A. The Real Projective Space

is a point of Rn+1 \ {0}, then we denote by

π(x) = [x] = [x0 : . . . : xn ]

the corresponding point of RPn . Note that [x0 : . . . : xn ] = [y0 : . . . : yn ] if and only if
there exists λ ̸= 0 such that λxi = yi for all i.

The topology of RPn :

By definition, RPn is a quotient of Rn+1 \ {0}, and the latter can be equipped with its
natural Euclidean topology. Recall that, in general, there is a procedure with which the
quotient of some topological space can be equipped with a natural topology. Concretely,
one can easily show that the collection

TRPn := U ⊆ RPn | π −1 (U ) ⊆ Rn+1 \ {0} is open



is a topology on RPn . Moreover, if we endow RPn with this topology, then the quotient
map π : Rn+1 \ {0} → RPn is continuous, and a map f : RPn → X from RPn to some
topological space X is continuous if and only if so is the composite map f ◦ π. The same
is true for any subset A ⊆ RPn endowed with the subspace topology.
At this point, there are several things that need to be checked about the topological
space RPn .

Exercise A.1: Show that RPn is Hausdorff by going through the following steps:

(i) Show that the quotient map π : Rn+1 \ {0} → RPn is open.

(ii) Show that the set

e := (x, y) ∈ Rn+1 \ {0} × Rn+1 \ {0} | [x] = [y]




∆



is closed in Rn+1 \ {0} × Rn+1 \ {0} .

(iii) Show that the set

[x], [x] ∈ RPn × RPn | [x] ∈ RPn



∆ :=

is closed in RPn × RPn .

(iv) Conclude that RPn is Hausdorff.

[Hint: Use (iii) and that the collection

U × V | U, V ∈ TRPn


is a basis for the topology of RPn × RPn by definition of the product topology.]
 121

Solution:
(i) Note that we have
[
π −1 π(U ) =


λ · U.
λ∈R×

As multiplication by a scalar λ ∈ R× is a homeomorphism, the sets λ · U are open,

and thus π −1 (π(U )) is open as well. By definition of the quotient topology, we
conclude that π(U ) is open.
(ii) Notice that [x] = [y] if and only if the 2 × (n + 1) matrix with lines x and y has
submaximal rank. By the solution to part (c) of [Exercise Sheet 2, Exercise 4], the
set ∆
e of such matrices is closed.

(iii) As π is open, the map

π × π : (Rn+1 \ {0}) × (Rn+1 \ {0}) → RPn × RPn
is open as well (it suffices to check that it maps basis elements U × V to open
subsets). It is straightforward to see that
e c ) = ∆c
π × π(∆
where •c denotes the complement. Hence, ∆c is open, and thus ∆ is closed.
(iv) Let [x] ̸= [y] be two distinct points of RPn . Then ([x], [y]) ∈ ∆c , and as ∆c is open,
there exist open subset U, V of RPn such that
([x], [y]) ∈ U × V ⊆ ∆c .
Now notice that U ∩ V = ∅, because otherwise U × V would contain a point of the
diagonal. Hence [x] and [y] can be separated by open subsets, i.e. RPn is Hausdorff.
Exercise A.2: Show that RPn is second-countable.
[Hint: Use Exercise 1(i).]

Solution: Let B be a countable basis for Rn+1 \ {0}. Set

B′ := π(B) | B ∈ B


and notice that as π is open, this is a collection of open subsets of RPn . Let us show that
B′ is a basis for the topology of RPn . To this end, let U ⊆ RPn be open and [x] ∈ U a
point. Then x ∈ π −1 (U ), and thus there exists B ∈ B such that x ∈ B ⊆ π −1 (U ). But
then [x] ∈ π(B) ⊆ U . Hence, B′ is a countable basis for RPn .

Exercise A.3: Show that RPn is locally Euclidean of dimension n as follows.

(i) For each 0 ≤ i ≤ n, set
Ui := [x0 : . . . : xn ] | xi ̸= 0 ⊆ RPn .


Show that Ui is open, and that

n
[
n
RP = Ui .
i=0
122 Appendix A. The Real Projective Space

(ii) For each 0 ≤ i ≤ n, consider the map

φi : Ui → Rn
 
x0 xi−1 xi+1 xn
[x0 : . . . : xn ] 7→ ,..., , ,..., .
xi xi xi xi

Show first that φi is well-defined, and subsequently that it is a homeomorphism.

Conclude that RPn is locally Euclidean of dimension n.

Solution:

(i) Notice that

π −1 (Ui ) = {(x0 , . . . , xn ) | xi ̸= 0} ⊆ Rn+1
is open, and thus Ui is open by definition of the quotient topology.
Now, note that for any [x] ∈ RPn we have x ∈ Rn+1 \ {0}, and thus there exists an
index i ∈ {0, . . . , n} such that xi ̸= 0. Hence [x] ∈ Ui , and as [x] was arbitrary, we
infer that
[n
n
RP = Ui .
i=0

(ii) The ratio xj /xi is invariant under scaling x, and thus φi is well-defined. To check
that it is continuous, it suffices to check that φi ◦ π is a continuous map from
π −1 (Ui ) = Rn+1 n
xi ̸=0 to R , but this is straightforward by the defining formula. Finally,
to show that φi is a homeomorphism, we construct a continuous inverse. Consider
the map

Ψi : Rn → π −1 (Ui )
(y1 , . . . , yn ) 7→ (y1 , . . . , yi , 1, yi+1 , . . . , yn ),

which is clearly continuous, and set ψi = π ◦ Ψi . This is continuous, since it is a

composition of continuous maps, and it is straightforward to see that φi and ψi are
mutually inverse. Hence, φi is a homeomorphism with inverse φ−1 i = ψi .

Exercise A.4:

(i) Show that RPn is connected.

(ii) Show that the restriction of π to S n ⊆ Rn+1 \ {0} is still surjective. Conclude that
RPn is compact.

By the above exercises we infer that RPn is an n-dimensional topological manifold ,

which is additionally compact and connected.
Before continuing the study of RPn , a few words S about the open subsets Ui defined
in Exercise 3(i) are in order. The open cover RPn = ni=0 Ui is called the standard open
cover of RPn . The equality, for example, φn ([x]) = y, means that the line corresponding
to [x] meets the plane Rn ×{1} at the point (y, 1). The complement of Un consists of those
 123

lines which do not intersect the plane Rn × {1}, which (as you may convince yourself) are
precisely the lines contained in Rn × {0}. Hence, we may somewhat suggestively write

RPn = Un ⊔ P Rn × {0} ∼ = Rn ⊔ RPn−1 .

We may thus regard RPn as a compactification of Rn by adding the points of RPn−1 ,

which from this point of view are often called points at infinity. In particular, the real
projective line RP1 (n = 1) may be regarded a one-point compactification of the real
line R1 , obtained by adding to it a “point at infinity”, and the real projective plane RP2
(n = 2) may be viewed as a compactification of the real plane R2 by adding to it a “line
at infinity”.

The smooth structure of RPn :

The standard open cover
n
[
n
RP = Ui ,
i=0

together with the homeomorphisms

 
n x0 xi−1 xi+1 xn
φi : Ui → R , [x0 : . . . : xn ] 7→ ,..., , ,..., , 0 ≤ i ≤ n,
xi xi xi xi

determine an atlas of RPn . According to Proposition 1.8(a), to obtain a smooth structure

n

on RP , it only remains to check that the charts (Ui , φi ) 0≤i≤n are smoothly compatible.

Exercise A.5: Let 0 ≤ i < j ≤ n. Show that the transition map from (Ui , φi ) to (Uj , φj )
is a diffeomorphism by computing that

φj ◦ φ−1 n n
i : Rxj ̸=0 → Rxi+1 ̸=0
1
(x1 , . . . , xn ) 7→ (x1 , . . . , xi , 1, xi+1 , . . . , xj−1 , xj+1 , . . . , xn ) ,
xj

and

φi ◦ φ−1 n
j : Rxi+1 ̸=0 → Rxj ̸=0
n

1
(x1 , . . . , xn ) 7→ (x1 , . . . , xi , xi+2 , . . . , xj , 1, xj+1 , . . . , xn ) .
xi+1

Solution: It is a straightforward albeit a tedious calculation to verify the formulas. Once

verified, it is immediate that the transition functions are smooth.

It follows from Exercise 5 that

n
ARPn := (Ui , φi )

i=0

is a smooth atlas for RPn , and the induced by Proposition 1.8(a) smooth structure on
RPn is referred to as the standard one. Thus, we now have a smooth manifold, namely
RPn , which is not intrinsically defined as a subset of Rn !
124 Appendix A. The Real Projective Space

Comment: A posteriori, Whitney’s embedding theorem (see Appendix B) asserts that

there is a smooth embedding RPn ,→ R2n (and the exponent 2n is in fact minimal if n
is a power of 2), so we can realize the smooth manifold RPn as a submanifold of R2n .
However, it would be very awkward if we were only able to speak about RPn as a smooth
manifold once we find such an embedding, so the flexibility of defining it abstractly is
certainly very helpful.

Further exercises about RPn :

Exercise A.6: Prove the following assertions:

(i) The quotient map π : Rn+1 \ {0} → RPn is smooth.

(ii) A map F : RPn → M to a smooth manifold M is smooth if and only if the composite
map F ◦ π : Rn+1 \ {0} → M is smooth.

Solution: See [Exercise Sheet 3, Exercise 5].

Exercise A.7: Show that the map

F : Rn → RPn , (x1 , . . . , xn ) 7→ [x1 : · · · : xn : 1]

is a diffeomorphism onto a dense open subset of RPn .

Solution: See [Exercise Sheet 3, Exercise 6].

Exercise A.8: Let P : Rn+1 \ {0} → Rk+1 \ {0} be a smooth map, and suppose that for
some d ∈ Z we have P (λx) = λd P (x) for all λ ∈ R× and x ∈ Rn+1 \ {0}. Show that the
map Pe : RPn → RPk given by Pe [x] = [P (x)] is well-defined and smooth.

Exercise A.9: Show that RP1 ∼

= S1 as smooth manifolds.
[Hint: To define an appropriate map, it might be helpful to use the identifications R2 ∼
=C
1 ∼
and S = {z ∈ C | |z| = 1}.]

Exercise A.10: Show that the quotient map π : Rn+1 \ {0} → RPn is a smooth submer-
sion, and that the kernel of the differential dπp : Tp Rn+1 \ {0} → T[p] RPn is the subspace

generated by p.

Solution: See [Exercise Sheet 7, Exercise 1(b)].

Exercise A.11: Consider the smooth map

F : R2 → RP2 , (x, y) 7→ [x : y : 1]

and the smooth vector field X on R2 defined by

∂ ∂
X = −y +x .
∂x ∂y

Show that there is a smooth vector field Y on RP2 that is F -related to X, and compute
its coordinate representation in terms of each of the charts defined in Exercise 3(ii).
APPENDIX B
SARD’S THEOREM AND WHITNEY’S THEOREMS

Theorem B.1 (Sard’s theorem). If F : M → N is a smooth map between smooth mani-

folds, then the set of the critical values of F has measure zero in N .

⇝ “almost all” c ∈ N are regular values of F ⇒

⇒ “almost all” level sets F −1 (c) of F are properly embedded submanifolds of M of

dimension dim M − dim N .

Theorem B.2 (Whitney’s embedding theorem). Every smooth n-manifold admits a

proper smooth embedding into R2n+1 .

⇝ Every smooth n-manifold is diffeomorphic to a properly embedded submanifold of

R2n+1 .
(Use Whitney’s embedding theorem, Proposition 5.3, Claim 3 from the proof of
Proposition 4.6 and [Exercise Sheet 8, Exercise 1(b)].)

Theorem B.3 (Whitney’s immersion theorem). Every smooth n-manifold admits a smooth
immersion into R2n .

The above two theorems are sometimes referred to as the easy or weak Whitney
embedding and immersion theorems, because Whitney obtained later the following im-
provements.

Theorem B.4 (Strong Whitney’s embedding theorem). Given n ≥ 1, every smooth n-

manifold admits a proper smooth embedding into R2n .

Theorem B.5 (Strong Whitney’s immersion theorem). Given n ≥ 2, every smooth n-

manifold admits a smooth immersion into R2n−1 .

For the proofs of all the above results, as well as a discussion of sets of measure zero
(in Rn or in smooth manifolds) we refer to [Lee13, Chapter 6 and Appendix C].

125
APPENDIX C
MULTILINEAR ALGEBRA

1 The Dual of a Vector Space

Definition C.1. Let V be a finite-dimensional real vector space.
(a) A covector on V is a real-valued linear functional on V , i.e., a linear map ω : V → R.
(b) The set of all covectors on V is a real vector space under the obvious operations
of pointwise addition and scalar multiplication. It is denoted by V ∗ and called the
dual space of V .
The next proposition expresses the most important fact about V ∗ (in the finite-
dimensional case).
Proposition C.2. Let V be a real vector space of dimension n. Given any basis (E1 , . . . , En )
for V , consider the covectors ε1 , . . . , εn ∈ V ∗ defined by
εi (Ej ) = δji .
Then (ε1 , . . . , εn ) is a basis for V ∗ , called the dual basis to (Ej ). In particular,
dimR V = dimR V ∗ .
Proof. Exercise!
In general, if (Ej ) is a basis for V and if (εi ) is its dual basis, then for any vector
v = v j Ej ∈ V we have
εi (v) = v j εi (Ej ) = v j δji = v i .
Thus, the i-th basis covector εi picks out the i-th component of a vector with respect to
the basis (Ej ).
More generally, we can express an arbitrary covector ω ∈ V ∗ in terms of the dual basis
as
ω = ωi εi ,
where the i-th component is determined by ωi = ω(Ei ). Thus, the action of the given
covector ω ∈ V ∗ on a vector v = v j Ej ∈ V is
ω(v) = ωi v j εi (Ej ) = ωi v i .

127
128 Appendix C. Multilinear Algebra

Definition C.3. Let V and W be real vector spaces and let A : V → W be a linear map.
The dual map of A is the linear map A∗ : W ∗ → V ∗ defined by
(A∗ ω)(v) := ω(Av), ω ∈ W ∗ , v ∈ V.
It is straightforward to check that the dual map satisfies the following properties:
(a) (A ◦ B)∗ = B ∗ ◦ A∗ .
(b) (IdV )∗ = IdV ∗ .
Proposition C.4. The assignment that sends a vector space to its dual space and a linear
map to its dual linear map is a contravariant functor from the category of real vector spaces
to itself.
Another important fact about the dual of a finite-dimensional vector space is the
following.
Proposition C.5. Let V be a finite-dimensional real vector space. For any given v ∈ V ,
define a linear functional ξ(v) by
ξ(v) : V ∗ → R
ω 7→ ξ(v)(ω) := ω(v).
Then ξ(v) ∈ (V ∗ )∗ ; that is, ξ(v) is a linear functional on V ∗ . Moreover, the map
ξ : V → (V ∗ )∗
v 7→ ξ(v)
is an R-linear isomorphism, which is canonical (it is defined without reference to any
basis).
Proof. The proof that both ξ(v) and ξ are R-linear maps are left as exercises. Since by
Proposition C.2 we have
dim V = dim V ∗ = dim(V ∗ )∗ ,
it suffices to prove that ξ is injective. To this end, let v ∈ V be non-zero, complete it to
a basis {v = E1 , E2 , . . . , En } of V , and let (εi ) be its dual basis. Then
ξ(v)(ε1 ) = ε1 (v) = ε1 (E1 ) = 1,
so ξ(v) ̸= 0. Therefore, ker ξ = 0; in other words, ξ is injective, as desired.
Due to Proposition C.5, the real number ω(v) obtained by applying a covector ω to a
vector v is sometimes denoted by either of the more symmetric-looking notations ⟨ω, v⟩
or ⟨v, ω⟩; both expressions can be thought of either as the action of the covector ω ∈ V ∗
on the vector v ∈ V , or as the action of the linear functional ξ(v) ∈ V ∗∗ on the element
ω ∈ V ∗ . There should be no cause for confusion with the use of the same angle bracket
notation for inner products: whenever one of the arguments is a vector and the other a
covector, the notation ⟨ω, v⟩ is always to be interpreted as the natural pairing between
vectors and covectors, not as an inner product.
There is also a symmetry between bases and dual bases for a finite-dimensional vector
space V : any basis for V determines a dual basis for V ∗ , and conversely, any basis for V ∗
determines a dual basis for V ∗∗ ∼= V . If (εi ) is the basis for V ∗ dual to a basis (Ej ) for V ,
then (Ej ) is the basis dual to (εi ), because both statements are equivalent to the relation
⟨εi , Ej ⟩ = δji .
 Section 2. Multilinear Maps and Tensors 129

2 Multilinear Maps and Tensors

In the preceding section, we defined and briefly examined the dual of a vector space
(in the finite-dimensional case), which is the space of real-valued linear functions on the
given vector space. A natural, and from the point of view of (differential) geometry very
important, generalization is to consider functions with several arguments, which are linear
in each individual argument. These are called multilinear functions.

Definition C.6. Let V1 , . . . , Vk and W be real vector spaces. A map F : V1 ×· · ·×Vk → W

is called multilinear if it is linear as a function of each variable separately when the others
are held fixed; that is, if 1 ≤ i ≤ k is arbitrary, and if we are given elements vi , vi′ ∈ Vi
and real numbers a, a′ ∈ R, then

F (v1 , . . . , avi + a′ vi′ , . . . , vk ) = aF (v1 , . . . , vi , . . . , vk ) + a′ F (v1 , . . . , vi′ , . . . , vk ).

Denote by L(V1 , . . . , Vk ; W ) the set of multilinear maps from V1 × · · · × Vk to W , and

note that L(V1 , . . . , Vk ; W ) has the structure of a real vector space. In the special case
when V1 = . . . = Vk = V and W = R, we often call an element of the space L(V, . . . , V ; R)
a k-multilinear function on V ; see Definition C.11.
Now, if the target space is W = R, then there is a simple operation with which one
can succesively build multilinear maps.

Definition C.7. Let V1 , . . . , Vk and W1 , . . . , Wl be real vector spaces, and consider F ∈

L(V1 , . . . , Vk ; R) and G ∈ L(W1 , . . . , Wl ; R). The function

F ⊗ G : V 1 × · · · × V k × W1 × · · · × Wl → R
(v1 , . . . , vk , w1 , . . . , wl ) 7→ F (v1 , . . . , vk ) G(w1 , . . . , wl )

is called the tensor product of F and G.

Exercise C.8:

(a) Show that, given F and G as above, the function F ⊗ G is multilinear, that is,

F ⊗ G ∈ L(V1 , . . . , Vk , W1 , . . . , Wl ; R).

(b) Show that the tensor product operation

− ⊗ − : L(V1 , . . . , Vk ; R) × L(W1 , . . . , Wl ; R) → L(V1 , . . . , Vk , W1 , . . . , Wl ; R)

(F, G) 7→ F ⊗ G

is bilinear, i.e., multilinear with two variables, and associative, i.e., for any multilin-
ear real-valued functions F, G, H, we have F ⊗ (G ⊗ H) = (F ⊗ G) ⊗ H.

Given a finite-dimensional real vector space V , we described in Section 1 how to obtain

a basis for the dual space V ∗ = L(V ; R) from a basis for V . With the above operation at
hand, we may now generalize this to the space L(V1 , . . . , Vk ; R).
130 Appendix C. Multilinear Algebra

Proposition C.9. Let V1 , . . . , Vk be R-vector spaces of dimensions n1 , . . . nk , respectively.

(j) (j)
n

For each 1 ≤ j ≤ k, let E1 , . . . , Enj be a basis of Vj , and denote by ε1(j) , . . . , ε(j)j the
corresponding dual basis of Vj∗ . Then the set
n o
B := ε(1) ⊗ · · · ⊗ ε(k) | 1 ≤ i1 ≤ n1 , . . . , 1 ≤ ik ≤ nk
i1 ik

is a basis for L(V1 , . . . , Vk ; R), which therefore has dimension n1 . . . nk .

Proof. First, given F ∈ L(V1 , . . . , Vk ; R), define for each multi-index I = (i1 , . . . , ik ) with
1 ≤ ij ≤ nj for all 1 ≤ j ≤ k, a number FI ∈ R by
 
(1) (k)
FI := F Ei1 , . . . , Eik .

Also, use the short-hand notation

ε⊗I := εi(1)
1
⊗ · · · ⊗ εi(k)
k
.

We will show that X

F = FI ε⊗I ,
I

where the sum is taken over all multi-indices as above, and thereby show that B spans
L(V1 , . . . , Vk ; R). To this end, take (v1 , . . . , vk ) ∈ V1 × · · · × Vk . For integers ij between 1
i (j) (j)

and nj , let vjj ∈ R be the coefficient of vj with respect to the basis E1 , . . . , Enj , i.e.,

i i
vjj = ε(j)
j
(vj ).

Then by the multilinearity of F we have

X   X
(1) (k)
F (v1 , . . . , vk ) = v1i1 · · · vkik F Ei1 , . . . , Eik = v1i1 · · · vkik FI .
I I

On the other hand, we have

" #
X X X
FI ε⊗I (v1 , . . . , vk ) = FI ε⊗I (v1 , . . . , vk ) = v1i1 · · · vkik FI .
I I I

⊗I
agree at any k-tuple and thus are equal, so B indeed spans
P
Hence F and I FI ε
L(V1 , . . . , Vk ; R).
Finally, in order to see that B is linearly independent, suppose that we have
X
λI ε⊗I = 0
I

for
 some real numbers  λI ∈ R indexed by multi-indices I. Evaluating both sides at
(1) (k)
Ei1 , . . . , Eik for some fixed multi-index I = (i1 , . . . , ik ), we obtain by the same com-
putation as above that λI = 0. Hence, B is linearly independent.
 Section 2. Multilinear Maps and Tensors 131

The proof of Proposition C.9 shows also that the components Fi1 ...ik of a multilinear
function F in terms of the basis elements in B are given by
 
(1) (k)
Fi1 ...ik = F Ei1 , . . . , Eik .

Thus, F is completely determined by its action on all possible sequences of basis vectors.

Remark C.10. You might have already encountered the abstract construction of the ten-
sor product of vector spaces. If so, then regarding the above discussion (which shows that
the real vector space L(V1 , . . . , Vk ; R) can be viewed as the set of all linear combinations
of objects of the form ω 1 ⊗ · · · ⊗ ω k , where ω i ∈ Vi∗ are covectors), one should remark
the following: given finite-dimensional real vector spaces V1 , . . . , Vk , there is a canonical
isomorphism
V1∗ ⊗ · · · ⊗ Vk∗ ∼= L(V1 , . . . , Vk ; R),
which is induced by the multilinear map

Φ : V1∗ × . . . × Vk∗ → L(V1 , . . . , Vk ; R)

Φ ω 1 , . . . , ω k (v1 , . . . , vk ) := ω 1 ⊗ · · · ⊗ ω k (v1 , . . . , vk )

= ω 1 (v1 ) · · · ω k (vk ).

Under this canonical isomorphism, abstract tensors correspond to the concrete tensor
product of multilinear functions defined above. As it is a natural isomorphism, we may
use the expression V1∗ ⊗· · ·⊗Vk∗ as a notation for L(V1 , . . . , Vk ; R) (this is a typical example
of slight abuse of notation, where one identifies naturally isomorphic objects). Finally,
using Proposition C.5, we also obtain a canonical identification

V1 ⊗ · · · ⊗ Vk ∼
= L(V1∗ , . . . , Vk∗ ; R).

Therefore, we may view the above construction as a concrete construction of the abstract
tensor product.

Let us now turn our attention to various spaces of multilinear functions on a finite-
dimensional real vector space that naturally appear in (differential) geometry.

Definition C.11. Let V be a finite-dimensional real vector space. For any integer k ≥ 1,
we denote by T k (V ∗ ) the space of k-multilinear functions on V , i.e.,

T k (V ∗ ) := L(V, . . . , V ; R) ∼
= |V ∗ ⊗ .{z
.. ⊗ V ∗.
| {z } }
k times k copies

By convention, we also define T 0 (V ∗ ) := R. The elements of T k (V ∗ ) are often referred to

as covariant k-tensors on V .

Observe that every linear functional ω : V → R is (trivially) multilinear, so a covariant

1-tensor is just a covector on V . Thus,

T 1 (V ∗ ) = V ∗ .
132 Appendix C. Multilinear Algebra

According to Proposition C.9, we obtain a basis for T k (V ∗ ) as follows. Assume that

V has dimension n, let (E1 , . . . , En ) be a basis for V and denote by (ε1 , . . . , εn ) the dual
basis for V ∗ . For a multi-index I = (i1 , . . . , ik ), where 1 ≤ ij ≤ n for all j, define the
elementary covariant k-tensor ε⊗I by the formula

ε⊗I := εi1 ⊗ · · · ⊗ εik

(see the proof of Proposition C.9) and for an integer m ∈ Z≥1 , denote by [m] the set
{1, . . . , m}. Then the set
 ⊗I
ε | I ∈ [n][k]

is a basis for T k (V ∗ ); in particular, we have

dimR T k (V ∗ ) = nk .

Therefore, every covariant k-tensor α ∈ T k (V ∗ ) can be written uniquely in the form

α = αI ε⊗I = αi1 ...ik εi1 ⊗ · · · ⊗ εik ,

where the nk coefficients αI = αi1 ...ik are determined by

αi1 ...ik = α(Ei1 , . . . , Eik ).

For example, T 2 (V ∗ ) is the space of bilinear forms on V – note that a covariant 2-tensor
on V is simply a real-valued bilinear function of two vectors – and every bilinear form on
V can be written as β = βij εi ⊗ εj for some uniquely determined n × n matrix (βij ).

Definition C.12. For a covariant k-tensor α ∈ T k (V ∗ ) and a permutation σ ∈ Sk , denote

by σ α the covariant k-tensor given by

σ
α: V × · · · × V → R


(v1 , . . . , vk ) 7→ α vσ(1) , . . . , vσ(k) .

In the following two sections we will discuss two important subspaces of T k (V ∗ ),

namely the subspaces of symmetric resp. alternating covariant k-tensors. Both are de-
scribed by the way that a permutation of the arguments of the given covariant k-tensor
changes its value. A significant application of symmetric tensors in the theory of smooth
manifolds is in the form of Riemannian metrics. Loosely speaking, a Riemannian metric is
a choice of an inner product on each tangent space of the given manifold, varying smoothly
from point to point, and allows one to define geometric concepts such as lenghts, angles
and distances on the manifold. Riemannian metrics will not be discussed in this course,
and this is the main reason why the discussion about symmetric tensors in Section 3 will
be kept to a minimum. On the other hand, differential forms will be discussed thoroughly
in Lecture 13 and Lecture 14 of this course. They constitute a significant application of
alternating tensors in smooth manifold theory, and they will be presented in Section 4.
 Section 3. Symmetric Tensors 133

3 Symmetric Tensors
In all probability, you have already encountered the concept of inner product on a finite-
dimensional real vector space V . It is a bilinear map ⟨·, ·⟩ : V ×V → R which is symmetric
and positive definite; in particular, ⟨·, ·⟩ is a covariant 2-tensor on V , having the additional
property that its value is unchanged when the two input arguments are exchanged; namely,
we have ⟨v1 , v2 ⟩ = ⟨v2 , v1 ⟩ for any v1 , v2 ∈ V . We now generalize this notion to any
covariant k-tensor on V .
Definition C.13. Let V be a finite-dimensional real vector space.
(a) A covariant k-tensor α ∈ T k (V ∗ ) on V is said to be symmetric if its value is un-
changed by interchanging any pair of its arguments; namely, for all v1 , . . . , vk ∈ V
and all 1 ≤ i < j ≤ k, we have
α(v1 , . . . , vi , . . . , vj , . . . , vk ) = α(v1 , . . . , vj , . . . , vi , . . . , vk ).

(b) The set of symmetric covariant k-tensors on V is denoted by Σk (V ∗ ). It is clearly a

linear subspace of T k (V ∗ ). By convention, we define Σ0 (V ∗ ) := R, and we also note
that Σ1 (V ∗ ) = T 1 (V ∗ ) = V ∗ .
Exercise C.14: We define a projection Sym : T k (V ∗ ) → Σk (V ∗ ), called symmetrization,
by the formula
1 Xσ
Sym(α) := α,
k! σ∈S
k
σ
where α was defined in Definition C.12. Show that Sym is well-defined and linear, and
that the following are equivalent:
(a) α is symmetric,
(b) α = σ α for all σ ∈ Sk ,
(c) α = Sym(α).

4 Alternating Tensors
Recall that the determinant may be regarded as a function det : Rn ×· · ·×Rn → R, taking
as input n column vectors with n entries each, and having as output the determinant of
the n × n matrix formed by these n column vectors. This map is multilinear, so det is
a covariant n-tensor on Rn . Moreover, it has the property that its value changes sign
whenever two of its input entries are interchanged; in other words, det is an alternating
n-tensor. We now generalize this notion to arbitrary covariant k-tensors.
Definition C.15. Let V be a finite-dimensional real vector space.
(a) A covariant k-tensor α ∈ T k (V ∗ ) on V is said to be alternating (or anti-symmetric
or skew-symmetric) if its value changes sign whenever any two of its arguments are
interchanged; namely, for all v1 , . . . , vk ∈ V and 1 ≤ i < j ≤ k, we have
α(v1 , . . . , vi , . . . , vj , . . . , vk ) = −α(v1 , . . . , vj , . . . , vi , . . . , vk ).
134 Appendix C. Multilinear Algebra

(b) The set of alternating covariant k-tensors on V is denoted by Λk (V ∗ ). It is clearly a

linear subspace of T k (V ∗ ) and its elements of Λk (V ∗ ) are also called exterior forms,
multicovectors or k-covectors. By convention, we define Λ0 (V ∗ ) := R, and we also
note that Λ1 (V ∗ ) = T 1 (V ∗ ) = V ∗ .

Note that every covariant 2-tensor β can be expressed as a sum of an alternating and
a symmetric tensor, because
1
1

β(v, w) = β(v, w) − β(w, v) + β(v, w) + β(w, v)
2 2
= α(v, w) + σ(v, w),

where
1
β(v, w) − β(w, v) ∈ Λ2 (V ∗ )


α(v, w) :=
2
is an alternating 2-tensor on V and
1
β(v, w) + β(w, v) ∈ Σ2 (V ∗ )


σ(v, w) :=
2
is a symmetric 2-tensor on V . However, this is not true for tensors of higher rank, as the
following exercise demonstrates.

Exercise C.16: Let (e1 , e2 , e3 ) be the standard dual basis for (R3 )∗ . Show that e1 ⊗e2 ⊗e3
is not equal to a sum of an alternating tensor and a symmetric tensor.
Recall that there is a group homomorphism sgn : Sk → {±1}, which maps a permuta-
tion σ ∈ Sk to 1 if it is a product of an even number of transpositions (even permutation),
and to −1 otherwise (odd permutation). We may use it to describe alternating tensors as
follows.

Exercise C.17: We define a projection Alt : T k (V ∗ ) → Λk (V ∗ ), called alternation, by

the formula
1 X
Alt(α) := (sgn σ) σ α,
k! σ∈S
k

σ
where α was defined in Definition C.12. Show that Alt is well-defined and linear, and
that the following are equivalent:

(a) α is alternating,

(b) α = (sgn σ) σ α for all σ ∈ Sk ,

(c) α = Alt(α),

(d) α(v1 , . . . , vk ) = 0 whenever v1 , . . . , vk ∈ V are linearly dependent,

(e) α(v1 , . . . , vk ) = 0 whenever there are i ̸= j such that vi = vj .

Example C.18. Let us explicitly compute Alt for 1-, 2- and 3-tensors.

• If α is a 1-tensor, then Alt(α) = α.

 Section 4. Alternating Tensors 135

• If β is a 2-tensor, then
1

Alt(β)(u, v) = β(u, v) − β(v, u) .
2

• If γ is a 3-tensor, then
1
Alt(γ)(u, v, w) = γ(u, v, w) + γ(v, w, u) + γ(w, u, v)
6

−γ(v, u, w) − γ(u, w, v) − γ(w, v, u) .

4.1 Elementary Alternating Tensors

Recall that for any basis of V , we described an induced basis of T k (V ∗ ) in terms of tensor
products of elements of the dual basis; cf. Proposition C.9. We obtain here a similar
description for a basis of Λk (V ∗ ).
Let V be a real vector space of dimension n, let (E1 , . . . , En ) be a basis for V ,
and denote by (ε1 , . . . , εn ) the corresponding dual basis for V ∗ . For a multi-index I =
(i1 , . . . , ik ) ∈ [n][k] , define the elementary alternating k-tensor (or elementary k-covector )
εI by the formula
εI := k! Alt ε⊗I ,

where
ε⊗I = εi1 ⊗ · · · ⊗ εik ∈ T k (V ∗ )
is the elementary k-tensor. Therefore, if v1 , . . . , vk ∈ V , then the value of εI at the k-tuple
(v1 , . . . , vk ) is given by the formula
X
εI (v1 , . . . , vk ) = (sgn σ) ε⊗I vσ(1) , . . . , vσ(k)


σ∈Sk
X Y
εij vσ(j)


= (sgn σ)
σ∈Sk 1≤j≤k
 
εi1 (v1 ) · · · εi1 (vk )
= det  ... ... ..  .

. 
ik ik
ε (v1 ) · · · ε (vk )

In other words, to compute εI (v1 , . . . , vk ), we write the coefficients of (v1 , . . . , vk ) with

respect to the basis (E1 , . . . , En ) of V in the form of a n × k-matrix, we consider the k × k
submatrix formed by the lines i1 , . . . , ik , and then we compute its determinant.

Example C.19. In terms of the standard dual basis (e1 , e2 , e3 ) for (R3 )∗ , we have
 1 
13 v w1
e (v, w) = det 3 = v 1 w3 − v 3 w1 ,
v w3

since v = v 1 e1 + v 2 e2 + v 3 e3 and w = w1 e1 + w2 e2 + w3 e3 , and

e123 (v, w, z) = det(v, w, z).

136 Appendix C. Multilinear Algebra

Since Alt : T k (V ∗ ) → Λk (V ∗ ) is surjective, we know that εI | I ∈ [n][k] is a gener-



ating set of Λk (V ∗ ). To extract from k ∗

 Iit a basis[k]of Λ (V ), we need the following lemma,
which describes the redundancy of ε | I ∈ [n] . In order to state it nicely, we need to
introduce the following notation: for a multi-index I ∈ [n][k] and a permutation σ ∈ Sk ,
denote by Iσ the multi-index

Iσ = iσ(1) , . . . , iσ(k) .
Also, denote by δJI the following generalization of the Kronecker-delta to multi-indices
I, J ∈ [n][k] :
(
sgn σ if neither I nor J have repeated entries and J = Iσ for some σ ∈ Sk ,
δJI :=
0 if I or J have repeated entries or J is not a permutation of I.

and observe that  

δji11 . . . δji1k
 . .
. . ...  .

δJI = det 

..

ik ik
δj1 . . . δjk
Lemma C.20. With the same notation as in the preceeding paragraph, the following
statements hold:
(a) If I has a repeated index, then εI = 0.
(b) If J = Iσ for some σ ∈ Sk , then εJ = (sgn σ) εI .
(c) For I, J ∈ [n][k] we have
εI Ej1 , . . . , Ejk = δJI .

Proof. Exercise!
Lemma C.20 tells us that from the generating set εI | I ∈ [n][k] of Λk (V ∗ ), we may


discard all those εI ’s for which I has a repeated index, and for any I having no re-
peated index, we need only take one element from the set {εIσ | σ ∈ Sk } and dis-
card the rest. A nice choice is thus the following: notice that for any multi-index I
having no repeated indices, there exists a unique permutation σ ∈ Sk such that Iσ is
strictly increasing, i.e., iσ(1) < · · · < iσ(k) . Therefore, according to Lemma C.20, the set
ε | I ∈ [n][k] is strictly increasing still generates Λk (V ∗ ), and there is no obvious re-
 I

dundancy in it. Essentially due to Lemma C.20(c), this set is linearily independent, and
thus we obtain the following result:
Proposition C.21. With the same notation as above, the set
 I
ε | I ∈ [n][k] is a strictly increasing multi-index

is a basis for Λk (V ∗ ). In particular, we have

 
k ∗ n n!
dimR Λ (V ) = = ,
k k!(n − k)!
and
Λk (V ∗ ) = {0} for k > n.
 Section 4. Alternating Tensors 137

Proof. Assume first that k > n. Since then every k-tuple of vectors is linearly dependent,
it follows from Exercise C.17(d) that Λk (V ∗ ) = {0}.
Assume now that k ≤ n. We need to show that

E := εI | I ∈ [n][k] is a strictly increasing multi-index



is linearly independent and spans Λk (V ∗ ). The fact that E generates Λk (V ∗ ) was already
discussed above. Suppose now that we have some linear relation
X
λ I εI = 0
I∈[n][k] strictly increasing

for some λI ∈ R. If we fix a strictly increasing multi-index J ∈ [n][k] , then evaluating the
above relation at (Ej1 , . . . , Ejk ) gives λJ = 0 according to Lemma C.20(c). Thus, E is
linearly independent. In conclusion, E is a basis of Λk (V ∗ ), as desired.
In particular, if V is a real vector space of dimension n, then the above proposition
implies that Λn (V ∗ ) is 1-dimensional, spanned by the elementary n-covector ε(1,...,n) . As
discussed in the beginning of this subsection, ε(1,...,n) sends an n-tuple (v1 , . . . , vn ) to the
determinant of the matrix (vji )1≤i,j,≤n , where vji = εi (vj ) is the i-th component of vj with
respect to the chosen basis of V . Note that when V = Rn with the standard basis, the
2
covector ε(1,...,n) (which by definition is a function from (Rn )n = Rn to R) is precisely the
usual determinant function.
One consequence of this observation is the following useful description of the behavior
of an n-covector on an n-dimensional vector space under linear maps. Recall that if
T : V → V is a linear map, then the determinant of T is defined to be the determinant of
the matrix representation of T with respect to any basis (recall that any two such matrix
representation are conjugations of each other and hence have the same determinant, so
this is well-defined).

Proposition C.22. Let V be an n-dimensional real vector space and let ω ∈ Λn (V ∗ ). If

T : V → V is any linear map and if v1 , . . . , vn ∈ V are arbitrary vectors, then

ω(T v1 , . . . , T vn ) = (det T ) ω(v1 , . . . , vn ). (•)

j
Proof. Let (Ei ) be any basis for V , and let (εi ) be the dual basis. P j Denote by (Ti )1≤i,j≤n the
matrix of T with respect to this basis, and set Ti = T Ei = j Ti Ej . By Proposition C.21,
we can write ω = cε(1,...,n) for some c ∈ R. Since both sides of (•) are multilinear
functions of (v1 , . . . , vn ), it suffices to verify the identity when the vi ’s are basis vectors.
Furthermore, since both sides are alternating, by Lemma C.20 we only need to check the
case (v1 , . . . , vn ) = (E1 , . . . , En ). In this case, the right-hand side of (•) is

(det T ) c ε(1,...,n) (E1 , . . . , En ) = c det T.

On the other hand, the left-hand side of (•) reduces to

ω(T E1 , . . . , T En ) = c ε(1,...,n) (T1 , . . . , Tn ) = c det (εj (Ti ))1≤i,j≤n = c det (Tij )1≤i,j≤n .

which is thus equal to the right-hand side.

138 Appendix C. Multilinear Algebra

4.2 The Wedge Product

Recall that for any covariant tensors α ∈ T k (V ∗ ) and β ∈ T l (V ∗ ) we defined the covariant
(k + l)-tensor α ⊗ β; see Definition C.7. This allowed us to build ’higher’ covariant tensors
out of lower ones, and also to describe a basis for T k (V ∗ ) in terms of tensor products of
elements of a dual basis. We now describe a similar construction for alternating tensors.
Definition C.23. Let V be a finite-dimensional real vector space, and let ω ∈ Λk (V ∗ )
and η ∈ Λl (V ∗ ) be alternating tensors on V . The wedge product (or exterior product) of
ω and η is denoted by ω ∧ η and is defined to be the (k + l)-covector given by the formula
(k + l)!
ω ∧ η := Alt(ω ⊗ η).
k!l!
As ⊗ is bilinear and Alt is linear, the map − ∧ − : Λk (V ∗ ) × Λl (V ∗ ) → Λk+l (V ∗ ) is
bilinear. It is therefore natural to examine what the wedge product looks like on basis
vectors. This also motivates the somewhat mysterious normalization factor (k + l)!/(k!l!),
because we have the following result.
Lemma C.24. Let V be a finite-dimensional real vector space, and let (ε1 , . . . , εn ) be
a basis for V ∗ . For any multi-indices I = (i1 , . . . , ik ) and J = (j1 , . . . , jl ) we have the
formula
εI ∧ εJ = εI⌢J ,
where I ⌢ J = (i1 , . . . , ik , j1 , . . . , jl ) is the (k + l)-multi-index obtained by concatenating
I and J.
Proof. By multilinearity, as in the proof of Proposition C.9, it suffices to show that

εI ∧ εJ (Ep1 , . . . , Epk+l ) = εI⌢J (Ep1 , . . . , Epk+l ) (⋆)

for any sequence of basis vectors (Ep1 , . . . , Epk+l ). We do this by considering several cases.
Case 1: The multi-index P = (p1 , . . . , pk+l ) has a repeated index. Then by part (e) of
Exercise C.17, both sides of (⋆) evaluate to 0.
Case 2: P contains an index that does not appear in either I or J. In this case, the
right-hand side of (⋆) is zero by part (c) of Lemma C.20. Similarly, each term in the
expansion of the left-hand side of (⋆) involves either I or J evaluated on a sequence of
basis vectors that is not a permutation of I or J, respectively, so the left-hand side is also
zero.
Case 3: P = I ⌢ J and P has no repeated indices. In this case, the right-hand side of
(⋆) is equal to 1, again by part (c) of Lemma C.20, so we need to show that the left-hand
side is also equal to 1. By definition,

εI ∧ εJ (Ep1 , . . . , Epk+l ) =
(k + l)!
= Alt(εI ⊗ εJ )
k!l!
1 X
= (sgn σ)εI (Epσ(1) , . . . , Epσ(k) )εJ (Epσ(k+1) , . . . , Epσ(k+l) ).
k!l! σ∈S
k+l
 Section 4. Alternating Tensors 139

By Lemma C.20 again, the only terms in the sum above that give nonzero values are
those in which σ permutes the first k indices and the last l indices of P separately. In
other words, σ must be of the form σ = τ η, where τ ∈ Sk acts by permuting {1, . . . , k}
and η ∈ Sl acts by permuting {k + 1, . . . , k + l}. Since then sgn σ = (sgn τ )(sgn η), we
have

εI ∧ εJ (Ep1 , . . . , Epk+l ) =
1 X
= (sgn τ )(sgn η) εI (Epτ (1) , . . . , Epτ (k) ) εJ (Epk+η(1) , . . . , Epk+η(l) )
k!l! τ ∈S
k
η∈Sl
! !
1 X 1 X
= (sgn τ ) εI (Epτ (1) , . . . , Epτ (k) ) (sgn η) εJ (Epk+η(1) , . . . , Epk+η(l) )
k! τ ∈S l! η∈S
k l
I J



= Alt(ε )(Ep1 , . . . , Epk ) Alt(ε )(Epk+1 , . . . , Epk+l )
= εI (Ep1 , . . . , Epk ) εJ (Epk+1 , . . . , Epk+l )
=1

where we used that Alt fixes alternating tensors by Exercise C.17, and again used part
(c) of Lemma C.20 (recall that we are in the case P = I ⌢ J).

Case 4: P is a permutation of I ⌢ J and has no repeated indices. In this case,

applying a permutation to P brings us back to Case 3. As both sides of (⋆) are alternating,
the effect of this permutation is to multiply both sides by the same sign. Hence the result
holds in this final case as well.
This completes the proof of the lemma.
Together with the bilinearity of − ∧ −, this gives the following properties of the wedge
product.
Proposition C.25. Let ω, η, ξ be multicovectors on a finite-dimensional real vector space
V . Then we have the following properties:
(a) Associativity:
ω ∧ (η ∧ ξ) = (ω ∧ η) ∧ ξ.

(b) Anticommutativity: if ω ∈ Λk (V ∗ ) and η ∈ Λl (V ∗ ), then

ω ∧ η = (−1)kl η ∧ ω.

(c) If (ε1 , . . . , εn ) is a basis of V ∗ and I = (i1 , . . . , ik ) a multi-index, then

εi 1 ∧ . . . ∧ εi k = εI .

(d) For any ω 1 , . . . , ω k ∈ V ∗ and v1 , . . . , vk ∈ V we have

 
ω 1 ∧ . . . ∧ ω k (v1 , . . . , vk ) = det ω j (vi ) 1≤i,j≤k .

Proof. Exercise!
140 Appendix C. Multilinear Algebra

Due to Proposition C.25(c), we generally use the notations εI and εi1 ∧ . . . ∧ εik
interchangably.
An element η ∈ Λk (V ∗ ) is said to be decomposable if it can be expressed in the form
η = ω 1 ∧ . . . ∧ ω k for some covectors ω 1 , . . . , ω k ∈ V ∗ . Note that not every k-covector
is decomposable when k > 1; however, it follows from Proposition C.21 and Proposi-
tion C.25(c) that every k-covector can be written as a linear combination of decomposable
ones.
 BIBLIOGRAPHY

[Lee09] J. M. Lee, Manifolds and differential geometry, Graduate Studies in Mathematics,

vol. 107, American Mathematical Society, Providence, RI, 2009.

[Lee13] , Introduction to smooth manifolds, 2nd ed., Graduate Texts in Mathe-

matics, vol. 218, Springer, New York, 2013.

141
 INDEX

atlas, 4 interior chart, 95

complete, 4 smooth, 6
maximal, 4 smooth, negatively oriented, 103
smooth, 4 smooth, positively oriented, 103
smooth, consistently oriented, 103 coordinate coframe, 82
coordinate covector field, 80
boundary coordinate cube, 2
chart, 95 smooth, 6
of a manifold, 95 coordinate domain, 2
point, 95 smooth, 6
topological invariance of, 96 coordinate frame, 64
bump function coordinate map, 2
continuous, 14 smooth, 6
smooth, 14 coordinate neighborhood, 2
bundle of covariant k-tensors, 87 smooth, 6
coordinate representation
closed map, 35 of a function, 9
closed map lemma, 36 of a map, 10
coframe coordinate vector, 25
dual, 82 coordinate vector field, 64
global, 81 coordinates
local, 81 local, 2
compactly supported cotangent bundle, 80
function, 13 cotangent map, 84
components cotangent space
of a (rough) section, 61 to a manifold, 79
of a tangent vector, 25 covariant k-tensor fields, 87
coordinate covector, 127
coframe, 82 decomposable, 140
coordinate ball, 2 positively oriented n-covector, 101
smooth, 6 covector field
coordinate basis, 25 component functions of, 81
coordinate chart, 2, 95 continuous, 81
boundary chart, 95 coordinate, 80

143
144 Index

pullback of, vanishes, 87 basis, 127

restriction of, 87 map, 128
restriction of, vanishes, 87 space, 127
rough, 81
smooth, 81 exterior derivative
vanishes along a submanifold, 87 naturality, 94
vanishes at points of a submanifold, 87 of a k-form on Rn , 90
critical point of a k-form on a smooth manifold, 93
of a map, 48 exterior differentiation, 93
critical value
flow, 73
of a map, 49
domain, 73
cube
fundamental theorem on, 74
closed, 39
generated by a vector field, 75
open, 39
global, 73
curve
infinitesimal generator of, 74
continuous, 30
maximal, 73
reparametrization, 109
of a vector field, 75
segment, 107
flow domain, 73
segment, piecewise smooth, 107
frame
segment, smooth, 107
coordinate, 64
smooth, 31
dual, 82
defining function global, 60
local, 50 local, 60
of an embedded submanifold, 50 negatively oriented, 101
defining map positively oriented, 101
local, 50 smooth, 60
of an embedded submanifold, 50 function supported in set, 13
derivation
geometric tangent space, 19
at a point, 20, 22
geometric tangent vector, 19
of C ∞ (M ), 66
global flow, 73
diffeomorphic manifolds, 12
global trivialization
diffeomorphism, 12
of a vector bundle, 56
local, 37
graph
differential
of a continuous function, 2
global, 30
of a smooth function, 6
of a function, 83
of a smooth map, 46
of a smooth map, 23, 97
graph coordinates, 3
differential 1-form, 81
Green’s theorem, 115
differential k-form, 87
closed, 94 homeomorphism, 1
degree of, 87 local, 37
exact, 94 hypersurface
wedge product, 88 embedded, 45
differential n-form
orientation form, 103 integral
positively oriented, 103 of a compactly supported n-form on a
directional derivative in Rn , 19 manifold, 113
domain of integration, 111 of an n-form over a domain of
dual integration in Rn , 111
 Index 145

of an n-form over an open subset in on the tangent bundle, 30

Rn , 111
integral curve open map, 35
maximal, 70 orientable smooth manifold, 101
of a vector field, 70 orientation
interior determined by a coordinate atlas, 103
chart, 95 determined by an n-form, 102
of a manifold, 95 induced by a local diffeomorphism
point, 95 (pullback), 106
inverse function theorem induced on ∂M , 105
for manifolds, 37 of a real vector space, 100
of a smooth manifold, 101
Jacobian matrix, 26 pointwise, continuous, 101
pointwise, of a smooth manifold, 101
level set, 48 Stokes, on ∂M , 105
of a constant-rank map, 49 orientation form, 103
regular, 49 oriented
Lie bracket, 68 consistently, ordered bases, 99
coordinate formula for, 68 consistently, smooth atlas, 103
naturality of, 69 frame, negatively, 101
of vector fields tangent to a frame, positively, 101
submanifold, 69 negatively, smooth chart, 103
line integral, 107 positvely, smooth chart, 103
fundamental theorem for, 110 vector space, 100
parameter independence of, 109 vector space, negatively, 100
local k-slice condition, 47 vector space, positively, 100
local defining function, 50
local defining map, 50 partition of unity, 13
local diffeomorphism, 37 smooth, 13
orientation-preserving, 106 partitions of unity
orientation-reversing, 106 smooth, 96
local embedding theorem, 41 projective space, 119
local homeomorphism, 37 as a smooth manifold, 123
local section theorem, 41 as a topological manifold, 122
local slice criterion, 47 proper map, 35, 36
local trivialization pullback
of a vector bundle, 55 of a covector field, 85
smooth, 56 pointwise, 84
locally finite, 13
quotient map, 42
manifold quotient topology, 42
ambient, 45
smooth, 5 rank
smooth, nonorientable, 101 constant, 33
smooth, orientable, 101 full, 33
smooth, with boundary, 96 of a smooth map, 33
topological, 1 rank theorem, 38
topological, with boundary, 95 global, 40
rectangle
natural coordinates closed, 39
on the cotangent bundle, 81 open, 39
146 Index

regular level set, 49 Stokes orientation, 105

regular level set theorem, 50 Stokes’ theorem, 115
regular point subbundle, 62
of a map, 48 smooth, 62
regular value submanifold
of a map, 49 codimension, 45, 50
reparametrization of a curve, 109 embedded, 45
backward, 109 immersed, 50
forward, 109 open, 7
properly embedded, 46
Sard’s theorem, 125 support
section of a function, 13
component functions, 61 of a vector field, 63
global, 59
linearly independent, 60 tangent bundle, 28
local, 41 ambient, 59
of a map, 41 tangent covector
of a vector bundle, 59 to a manifold, 79
rough, 59 tangent map, 23
smooth, 59 tangent space
spanning, 60 geometric, 19
zero, 59 to a smooth manifold, 22
slice, 47 to a smooth manifold with boundary,
slice chart, 47 97
slice coordinates, 47 to a submanifold, 52–54
smooth embedding, 33 to an open submanifold, 24
smooth function, 9 tangent vector
smooth immersion, 33 geometric, 19
smooth map, 10, 96 inward-pointing, 97
on arbitrary subset, 15 outward-pointing, 97
on open subset of Hn , 96 to a smooth manifold, 22
smooth structure topological embedding, 33
determined by an atlas, 5 transition function between smooth local
on topological manifold, 5 trivializations, 57
on topological manifold with boundary, transition map, 4
96 trivialization
smooth submersion, 33 global, 56
smoothly compatible charts, 4 local, 55
space smooth local, 56
Hausdorff, 1
locally Euclidean, 1 unit n-sphere, Sn , 3
second-countable, 1 unit ball, Bn , 3
standard open cover of RPn , 122 upper half-space
standard orientation boundary, 95
of Sn , 102 closed, 95
of Rn , 100 interior, 95
standard smooth structure open, 95
on Sn , 7
on Rn , 6 vector bundle, 55
on RPn , 123 base, 55
 Index 147

fiber, 55 component functions of, 64

global frame, 60 continuous, 63
global trivialization, 56 coordinate, 64
local frame, 60 Euler, 64
local trivialization, 55 flow, 75
product bundle, 56 integral curve of, 70
projection, 55 nowhere tangent to a submanifold, 104
restriction of, 58 pushforward, 67
section of, 59 rough, 63
smooth, 56 smooth, 63
smooth (local or global) frame, 60 tangent to a submanifold, 67
smooth subbundle of, 62 tangent to a submanifold at a point, 67
smoothly trivial, 56 velocity vector of curve, 31
subbundle of, 62
total space, 55 Whitney’s embedding theorem, 125
trivial, 56 strong, 125
Whitney sum, 58 weak, 125
zero section, 59 Whitney’s immersion theorem, 125
vector field strong, 125
along a submanifold, 104 weak, 125
compactly supported, 63
complete, 76 zero set, 48