CHAPTER 4: FORMS ON MANIFOLDS

JUSTIN M. CURRY

Contents
1. Manifolds 2
1.1. The Notion of a Topological Space 2
1.2. The Hausdorff Axiom 3
1.3. Paracompactness 3
1.4. Manifolds Defined 4
1.5. Differentiable Manifolds 4
1.6. Smooth Mappings between Manifolds 7
1.7. Smooth Mappings between Arbitrary Subsets 8
1.8. Submanifolds 9
1.9. A More Intuitive Picture 9
1.10. Exercises 10
2. Tangent Spaces 12
2.1. Curves and Tangents 13
2.2. Mappings between Tangent Spaces 14
2.3. The Inverse Function Theorem 16
2.4. Submersions 17
2.5. Submanifolds and Regular Values 19
2.6. Immersions 20
2.7. Immersions and the Subset Picture 21
2.8. Exercises 23
3. Vector Fields on Manifolds 23
3.1. Push-Forwards and Pull-Backs 24
3.2. Smooth Vector Fields 24
3.3. Integral Curves 25
3.4. ODE Theory on Manifolds 25
3.5. Complete Vector Fields 26
3.6. One-Parameter Groups of Diffeomorphisms 28
3.7. Exercises 29
4. Differential Forms on Manifolds 31
4.1. Push-Forwards and Pull-Backs 31
4.2. Smooth k-forms 32
4.3. The Exterior Differentiation Operation 33
1
2 JUSTIN M. CURRY

4.4. The Interior Product and Lie Derivative Operation 34

4.5. Exercises in Symplectic Geometry: The Cotangent Bundle 35
5. Some Topological Results 36
5.1. The Support of a Function 36
5.2. Partitions of Unity 37
5.3. Proper Mappings 37
5.4. The Whitney imbedding theorem 38

It is our agenda in this chapter to extend to manifolds the results of

Chapters 2 and 3 and to reformulate and prove manifold versions of two
of the fundamental theorems of integral calculus: Stokes’ theorem and
the divergence theorem. In this first section we aim to introduce the
necessary background to understand the term “manifold”. In doing
so, we will be able to say much more sophisticated things than “a
manifold is something which locally looks like Rn ”. Other results in
differential topology beyond integral calculus will also become much
easier to handle once this first section is understood.

1. Manifolds
Before we can state the abstract definition of a manifold, we must
first restate some background material from topology, most notably,
the definition of a topological space.

1.1. The Notion of a Topological Space.

Definition 1.1. Let X be a set and J a collection of subsets of X.
If J satisfies the following properties, then we say that J defines a
topology on X or (X, J ) is a topological space. The properties are:
(a) X and ∅ are in J .
(b) ForSevery (finite or infinite) sub-collection U = {Uα , α ∈ I} of
J , α∈I Uα is in J .
(c) T
For every finite sub-collection U = {Ui , i = 1, . . . , n} of J ,
n
i=1 Ui is in J .

Definition 1.2. We say that a set U ∈ J is an open subset of X.

Many of you are most likely familiar with the concept of an open
set from an introductory analysis course, albeit defined in a different
manner than above. The open sets defined in analysis provide a nice
example of the metric space topology.
 CHAPTER 4: FORMS ON MANIFOLDS 3

Example 1.3. Let (X, d) be a non-empty metric space. Recall the

open ball of radius ǫ around a point x0 ∈ X is defined as the set
Bǫ (x0 ) = {x ∈ X, d(x, x0 ) < ǫ}. We then say that U ⊆ X is open, i.e.
belongs to the metric space topology Jd if and only if for every x0 ∈ U
there exists an ǫ > 0 such that Bǫ (x0 ) ⊂ U.
For those of you who haven’t had a course in topology and only a
course in analysis, one can assume that “topological space”=“metric
space”, that is, the metric space (X, d) induces a topology J on the set
X to make a topological space (X, J ) according to the above example.
However, topological spaces are more general than metric spaces as
there are topologies that don’t arise from any metric. It is important
to remember that a topological space is a set with extra data specified.
That extra data specifies what the open sets are. For more informa-
tion on topological spaces, please review Chapter 2 of James Munkres’
Topology.

1.2. The Hausdorff Axiom. One of the nice properties that metric
space topologies satisfy is known as the Hausdorff axiom. (It is an
exercise to show that all metrizable topologies are Hausdorff.)
Definition 1.4. (X, J ) is a Hausdorff topological space if for every
p1 , p2 ∈ X p1 6= p2 there exist open sets Ui i = 1, 2 with pi ∈ Ui with
U1 ∩ U2 = ∅.
1.3. Paracompactness. Another nice property that metrizable topolo-
gies have is known as paracompactness. First we offer a preliminary
definition.
Definition 1.5. A collection
S of compact sets Ci ⊆ X is an exhaustion
of X if Ci ⊆ Int(Ci+1 ) and Ci = X.
Definition 1.6. A space X is paracompact if there exists an exhaustion
of X by compact sets.
Example 1.7. The space X = Rn is paracompact. Simply build an
exhaustion via n-balls of increasing integer radii, i.e. Ci = Bin (0).
Lemma 1.8. Closed subsets of paracompact sets are paracompact.
Proof. If a space X is paracompact then there exists an exhaustion
{Ci} of X be compact sets. Suppose S ⊆ X is a closed subset. Since
closed subsets of compact sets are compact, we have that S is exhausted
by {Ci ∩ S}, which are compact. 
4 JUSTIN M. CURRY

1.4. Manifolds Defined. Recall that a homeomorphism is a map be-

tween topological spaces that is continuous and has a continuous in-
verse. Since continuity is defined in terms of open sets, we can think of
two homeomorphic topological spaces as topologically equivalent. As a
convention, we will just say X is a topological space instead of (X, J ).
We are now in a position to define what a manifold is.
Definition 1.9. Let X be a paracompact, Hausdorff space. X is an
n-dimensional manifold (or n-manifold for short) if for every p ∈ X
there exists an open neighborhood U of p in X, an open set V in Rn
and a homeomorphism ϕ : U → V .
Definition 1.10. The triple (U, V, ϕ) is called a chart.
The homeomorphism between U ⊆ X and V ⊆ Rn is a rigorous way
of saying that a manifold “locally looks like” Rn .
Definition 1.11. A chart (U, V, ϕ) is centered at p if ϕ(p) = 0 ∈ Rn .
Note that since we usually consider only one chart at a time, we can
always center it via translation.
1.5. Differentiable Manifolds. The above definition is the defini-
tion of what is called a topological manifold. For the purposes of this
book, we are interested almost exclusively with differentiable mani-
folds. Detailing this extra differentiability condition requires a little
extra work. Just as we began with a set and provided extra data to
define a topological space, we will have to start with a manifold and
specify extra structure to define a differentiable manifold. This extra
structure details how various charts interact with each other.
Let X be a topological manifold and (Ui , Vi , ϕi ) i = 1, 2 be two
charts. We use the following overlap diagram
U1 ∩ UI2
ϕ1
u II ϕ
uu II 2
uuu II
u II
uz u ϕ1,2 $
V1,2 / V2,1

to detail how these charts interact. Here V1,2 = ϕ1 (U1 ∩ U2 ) and V2,1 =
ϕ2 (U1 ∩ U2 ). These sets are open subsets of V1 and V2 and ϕ1,2 :=
ϕ2 ◦ ϕ−1
1 |V1,2 .

Remark 1.12. The map ϕ1,2 is a homeomorphism of V1,2 and V2,1 . It’s
inverse is ϕ2,1 = ϕ1 ◦ ϕ−1
2 |V2,1 .

We now state one of the crucial conditions involved in defining a

differentiable manifold.
 CHAPTER 4: FORMS ON MANIFOLDS 5

Definition 1.13. The charts (Ui , Vi , ϕ) i = 1, 2 are compatible if ϕ1,2

is a diffeomorphism. In which case ϕ2,1 = (ϕ1,2 )−1 is also a diffeomor-
phism.
Definition 1.14. An atlas is a collection of charts
A = {(Uα , Vα , ϕα ), α ∈ I}
S
such that any two charts in the collection are compatible and α∈I Uα =
X.
Definition 1.15. Let A be an atlas and (U, V, ϕ) a chart. (U, V, ϕ) is
said to be A-compatible if it is compatible with all charts in A.
Lemma 1.16. If (Ui , Vi, ϕi ) i = 1, 2 are A-compatible, then they are
compatible with each other.
Proof. The proof is left as an exercise. Hint: Let (Ui , Vi , ϕi) i = 1, 2, 3
be charts in A. Look at overlap diagrams of the form:
U1 ∩ U2 ∩MU3
q MMM
ϕ1 qqqq MMϕM2
qqq MMM
x qq
q ϕ1,2
M&
V1,2,3 / V2,1,3

Show that in this diagram ϕ1,2 = ϕ2,3 ◦ ϕ1,3 . 

Definition 1.17. An atlas A is complete or maximal if every A-
compatible chart is already in A.
Theorem 1.18. Given any atlas A there exists a complete atlas Â
containing A.
Proof. Take  to be the atlas consisting of all charts that are compat-
ible with the charts in A. By the above lemma, any two charts in Â
are compatible with each other, which, by definition, makes  an atlas
as well. By construction, Â is a complete atlas as well. 
Thus if we’re given an atlas, we can always avail ourselves of the
completion process to turn it into a complete atlas. We are now in a
position to state the additional “differentiable structure” that we must
put on a topological manifold to make it a differentiable manifold.
Definition 1.19. An n-dimensional differentiable manifold is a pair
(X, A) where X is an n-dimensional topological manifold with a com-
plete atlas A.
One of the simplest examples of a manifold of this type is the unit
circle S1 .
6 JUSTIN M. CURRY

Example 1.20 (The Unit Circle). Let X = S1 = {(x1 , x2 ) ∈ R2 , x21 +

x22 = 1} and U1 = S1 − (0, 1) and U2 = S1 − (0, −1). We would like
to show that the circle is a one-dimensional manifold. First we must
provide homeomorphisms to open subsets of R. This is accomplished
via stereographic projection. We first project from (0, 1) – also known
as the “north pole” – via ϕ1 by taking a line passing through (0, 1) and
any point p ∈ U1 and defining the image ϕ(p) to be the intersection
point of the line with the x-axis. Similarly, we can define ϕ2 to be
stereographic projection from the south pole (0, −1).
Exercise: Write down explicit formulas for ϕi i = 1, 2 and show that
they define homeomorphisms onto ϕi (Ui ) =: Vi = R. Then show that
V1,2 := ϕ1 (U1 ∩ U2 ) and V2,1 := ϕ2 (U1 ∩ U2 ) both equal R − 0 and
ϕ1,2 := ϕ2 ◦ ϕ−1 1
1 = x . Conclude that the pair (Ui , Vi , ϕi ) i = 1, 2 define
an atlas on S1 . This is known as The Mercator Atlas.
From this atlas we get a complete atlas by our completion process.
This makes S1 into a one-dimensional differentiable manifold.
Example 1.21 (The n-Sphere). As an exercise, see if you can extend
this construction to Sn , n ≥ 2. Hint: The construction is virtually the
x
same but the overlap map ϕ1,2 : Rn−1 r 0 → Rn−1 r 0 is ||x|| 2.

The process of giving a manifold a differentiable or smooth structure,

may seem a little haphazard, as there are several different atlases that
a given manifold could have. The following example illustrates an
alternative atlas for Sn .

Example 1.22 (The Coordinate Atlas). In the previous exercise you

were asked to extend the Mercator construction to the n-sphere. That
atlas is in some ways the most efficient in that it requires only two
charts. In this example we construct an atlas that requires 2n + 2
charts. Our initial data is as follows:

X = Sn = {x ∈ Rn+1 , ||x|| = 1}
Ui+ = {(x1 , . . . , xn+1 ) ∈ Sn , xi > 0}
Ui− = {(x1 , . . . , xn+1 ) ∈ Sn , xi < 0}
V = Bn = {y ∈ Rn , ||y|| < 1}

ϕ± ± n n
i : Ui ⊂ S → V ⊂ R to be the projection map defined by

ϕ±
i (x1 , . . . , xn+1 ) = (x1 , . . . , x̂i , . . . , xn+1 ).

In the case of S2 ⊂ R3 , the maps would be ϕ± 1 (x, y, z) = (y, z),

± ±
ϕ2 (x, y, z) = (x, z) and ϕ3 (x, y, z) = (x, y). The inverse in the general
 CHAPTER 4: FORMS ON MANIFOLDS 7

case is
s X
(ϕ± −1
i ) (x1 , . . . , x̂i , . . . , xn+1 ) = (x1 , . . . , ± 1 − x2j , . . . , xn+1 ).
j6=i

Exercise: Compute ϕ± ± −1
i ◦(ϕj ) and show that they are smooth. Deduce
n
that S is a differentiable manifold.
The previous several exercises may seem like a great deal of effort
to show that the familiar n-sphere is a manifold – it is by definition a
subset of Euclidean space! The next example exhibits a manifold that
is not easily visualized as a subset of a larger ambient space.
Example 1.23 (Real Projective Space). RPn , n-dimensional real pro-
jective space, is defined as the space of one-dimensional subspaces
(lines) in Rn+1 . Define the map
π
Sn / RPn
that sends a vector v to its span < v >.
Exercise: Check this is a two-to-one map. Check that π : Ui+ ֒→ RPn
is injective and onto a set Ui . We then have the charts defined with
homeomorphisms
π −1 ϕ+
i
ϕi : Ui / Ui+ / V.
n
These form an atlas, making RP into a manifold. More details about
this example can be found in the exercises at the end of the section.
Example 1.24. If (X, A) is a manifold and U is an open subset of X,
then U is a manifold in its own right.
Proof. Let (Uα , Vα , ϕα ), α ∈ I be the elements of an atlas A. Let
Uα′ = Uα ∩ U,
Vα′ = ϕα (Uα ∩ U),
ϕ′α = ϕ|Uα∩U .
Observe that (Uα′ , Vα′ , ϕ′α ), α ∈ I define an atlas A′ on U. 
1.6. Smooth Mappings between Manifolds.
Notation 1.25 (Conventions for the Chapter). From now on we will
adopt the convention that “manifolds” will actually mean “differen-
tiable manifold”. Furthermore, we will simply refer to “a manifold X”,
assuming that (X, A) is implicitly understood. Also in this spirit when
we refer to a chart (U, V, ϕ) we’ll assume without saying that (U, V, ϕ)
is in A.
8 JUSTIN M. CURRY

Definition 1.26. Let X be an n-dimensional manifold and f : X →

Rm a continuous map. We say that f is C ∞ if for every chart (U, V, ϕ)
the map f ◦ ϕ−1 : V → Rm is C ∞ .
Remark 1.27. The assumption that f is continuous is not necessary to
the definition, because the definition of smoothness implies continuity;
however, it is convenient to have that the pre-image of open sets is
open before doing further work.
The map f ◦ ϕ−1 is sometimes called the coordinate representation
of f as it is the function of the coordinates on the open set, V ⊆ Rn .
Thus the above definition says that a function on a manifold is smooth
if and only if it is smooth in every coordinate representation of that
function. A similar result is true for maps between manifolds.
Definition 1.28. Let X and Y be n and m-dimensional manifolds
respectively and f : X → Y a continuous map. We say f is C ∞ if for
every p ∈ X and for every chart (U, V, ϕ) on Y ϕY ◦ f : f −1 (U) → Rm
is C ∞ .
Note that since f is continuous f −1 (U) is open in X and is thus a
manifold in its own right. Putting together the above two definitions
yields the following, more general, definition.
Definition 1.29. Let X and Y be n and m-dimensional manifolds
respectively and f : X → Y a continuous map. We say f is C ∞ if for
every p ∈ X there exist charts (UX , VX , ϕX ) and (UY , VY , ϕY ) about p
and f (p) respectively so that ϕY ◦ f ◦ ϕ−1 ∞
X is C .

Remark 1.30 (For Every vs. There Exists). Traditionally the difference
between “for every” and “there exists” in mathematical definitions is
night and day, but in the case of smooth manifold theory this is not
true. In particular, in the above several definitions, we could have
simply required that there exist one chart where the required function
is C ∞ . The fact that all the charts in the atlas are assumed to be
compatible then gives us for free that the function is C ∞ in every chart!
Lemma 1.31 (Composition of smooth maps is smooth). Assume Xi
for i = 1, 2, 3 are manifolds. If fi : Xi → Xi+1 for i = 1, 2 are C ∞
maps, then f2 ◦ f1 is a C ∞ map.
Proof. Left as an exercise for the reader. 
1.7. Smooth Mappings between Arbitrary Subsets. Assume Mi
i = 1, 2 are manifolds and Xi a subset of Mi with f : X1 → X2 a
continuous map.
 CHAPTER 4: FORMS ON MANIFOLDS 9

Definition 1.32. f is a C ∞ map if for every p ∈ X1 there exists a

neighborhood O of p in M1 and a C ∞ map f˜ : O → M2 such that
f = f˜ on X1 ∩ O
Theorem 1.33. Assume Mi i = 1, 2, 3 are manifolds with subsets Xi .
If fi : Xi → Xi+1 i = 1, 2 are C ∞ maps, then f2 ◦ f1 : X1 → X3 is C ∞
as well.
Exercise 1.34. Prove the preceding Theorem.
Definition 1.35. For Xi ⊆ Mi i = 1, 2 f : X1 → X2 is a diffeomor-
phism if f is 1 − 1, onto and f and f −1 are C ∞ maps.
1.8. Submanifolds.
Definition 1.36. Let M = M n be an n-dimensional manifold and X
a subset of M. X is a k-dimensional submanifold of M if for every
p ∈ X there exists an open neighborhood U of p in X, an open set V
in Rk and a diffeomorphism ϕ : U → V .
Claim 1.37. The collection of all triples (U, V, ϕ) where U is an open
set in X, V an open set in Rk and ϕ : U → V a diffeomorphism,
defines an atlas A on X.
Proof. Given the charts (Ui , Vi , ϕi ) i = 1, 2
U1 ∩ UI2
ϕ1
u II ϕ
uu II 2
uuu II
u II
uz u ϕ1,2 $
V1,2 / V2,1

we see that if ϕ1 and ϕ2 are diffeomorphisms and ϕ1,2 := ϕ2 ◦ ϕ−1

1 |V1,2 ,
then ϕ1,2 is automatically a diffeomorphism. 
1.9. A More Intuitive Picture. Although manifolds such as RPn
are not easily visualized as subsets of Euclidean space – thus lending
support to the “intrinsic” viewpoint of manifolds – many manifolds
that we will encounter later on are naturally submanifolds of RN for
some N.
In fact we will describe below (see §2.7) a result which says that
every manifold is a submanifold of RN for some N. In other words no
generality is lost by regarding manifolds as being submanifolds of RN .
This viewpoint we will call the “subset” or “extrinsic” picture of man-
ifolds to contrast it with the intrinsic picture of manifolds that we’ve
been exposing in this chapter. A clear exposition of this “extrinsic”
picture can be found in Spivak’s Calculus on Manifolds or Munkres’
Analysis on Manifolds. One of its advantages is that their definition of
10 JUSTIN M. CURRY

“manifold” completely avoids the complications of charts and atlases:

If M1 and M2 are Euclidean spaces, all one needs to make sense of
the definitions (1.37) and (1.41) in the previous section is elementary
freshman calculus. On the other hand this viewpoint also has its dis-
advantages. For instance it is explained in §5.4 how to realize RP 2 as
a submanifold of R6 , but the “extrinsic” picture of RP 2 as a subman-
ifold of R6 is not nearly illuminating as the intrinsic picture of RP 2
described in §1. (This remark applies to a lot of other examples as
well, for instance the example discussed in §1.10, exercise 7.)
1.10. Exercises.
(1) Products. Assume that M1 , . . . , Mk are topological manifolds
of dimensions n1 , . . . , nk respectively. Prove that the product
manifold M1 × · · · × Mk is a topological manifold of dimension
(n1 + · · · + nk ). Assume that the Mi are smooth manifolds and
prove that the resulting product is a smooth manifold as well.
(2) Graphs. Let f : U ⊆ Rn → Rk be a continuous function.
Define the graph of f to be the set
Gf = {(x, y) ∈ Rn × Rk , x ∈ U, y = f (x)}
and prove that it is a manifold of dimension n. Hint – How
does this set come equipped with its own chart? Now assume
that f is C ∞ . Prove that the graph is a smooth manifold.
(3) Spheres and Atlases. Review the construction of the Merca-
tor atlas and the coordinate atlas for the n-sphere. Show that
the two atlases are compatible.
(4) Atlas Revisited. Let M be a set and let U = {Ui , i ∈ I} be
a collection of subsets of X. Suppose that for each i ∈ I one is
given an open set Vi in Rn and a bijective map ϕi : Ui → Vi .
We will
S call U an atlas if
• Ui = M
• ϕi (Ui ∩ Uj ) =: Vi,j is an open subset of Vi for all i, j ∈ I
• In the overlap diagram below, the bottom arrow is a dif-
feomorphism.
Ui ∩ Uj
w GG
ϕi ww GG ϕj
ww GG
ww GG
w{ w G#
ϕi,j
Vi,j / Vj,i

(a) Prove that M can be equipped with a topology by decree-

ing that U ⊆ M is open if and only if ϕi (U ∩ Ui ) is open
for all i.
 CHAPTER 4: FORMS ON MANIFOLDS 11

(b) Show that if M is equipped with this topology then the

atlas above really is an atlas in the sense defined in the
previous section. Deduce that M is a manifold.
(5) Real Projective Space. RPn – Let π : Rn+1 r 0 → RPn be
the map that sends a vector to its span, i.e.
π(x1 , . . . , xn+1 ) = [x1 , . . . , xn+1 ]
where the brackets denote homogeneous coordinates. Thus, for
a given x ∈ Rn+1 r 0, π(x) = [x] denotes the equivalence class
of vectors contained in the same one-dimensional subspace gen-
erated by x.
(a) Define Ui to be the set of projected points [x1 , . . . , xn+1 ]
where xi 6= 0. Show that the map
∼
=
ϕi : Ui / Rn
defined by
x1 xi−1 xi+1 xn+1
ϕi ([x1 , . . . , xn+1 ]) = ( ,..., , ,..., )
xi xi xi xi
is indeed a homeomorphism.
(b) Check that U is an atlas in the sense of Exercise 4.
(6) Real Projective Space Re-visited. Another approach to
defining the manifold structure on RPn . The 2 − 1 map Sn →
RPn .
(7) Grassmannians Let Gr(k, n) denote the space of all k-dimen-
sional subspaces of an n-dimensional vector space, i.e Rn . Note
that this is a generalization of projective space, since Gr(1, n +
1) ∼
= RPn .
Let ei for i = 1, . . . , n be the standard basis vectors of Rn . For
every k-element subset, I, of {1, . . . , n} let VI = span{ej , i ∈/ I}.
Let UI be the set of all k-dimensional subspaces of V of Rn , i.e.
V ∈ Gr(k, n), with V ∩ VI = {0}.
We will sketch below a proof of the following result:
Theorem 1.38. The UI ’s define an atlas in the sense of Exer-
cise 4.
(a) Suppose V is a k-dimensional subspace of Rn . Let wr =
(a1,r , . . . , an,r ) for r = 1, . . . , k be a basis of V and let A
be the k × n matrix whose row vectors are the wr ’s. Let
AI be the k × k minor of A whose columns are the vectors
vi = (ai,1 , . . . , ai,k ) i ∈ I. Show that detAI 6= 0 if and only
if V ∩ VI = {0}.
12 JUSTIN M. CURRY

Hint: Show that if detAI = 0 there exists a vector (c1 , . . . ,

k
cPk) = c ∈ PR r {0} with AI c = 0 and conclude that
cr wr = i∈I / λj ej .
(b) Let V ∩ VI = {0}. Show that there is a unique basis,
wr = (a1,r , . . . , an,r ) for r = 1, . . . , k, of V for which the
k × k matrix AI is the identity matrix.
(c) For each V ∈ UI choose a basis wr = (a1,r , . . . , an,r ) r =
1, . . . , k such that AI is the identity matrix. Let AV be the
k × (n − k) dimensional matrix obtained from the k × n
matrix A = [ai,j ] by deleting the minor AI . Show that the
map V ∈ UI → AV is a bijective map
∼
=
φI : UI / Mk,n−k

(d) Show that, equipped with these maps, the collection {UI , I ⊆
{1, . . . , n}} is an atlas in the sense of Exercise 4.
Hints:
(i) Suppose V ∈ UI ∩ UJ and let wr = (a1,r , . . . , an,r ) r =
1, . . . , k be a basis of V . Show that φI (V ) = Mk,n−k is the
matrix A−1 I A with its I-minor deleted and φJ (V ) is the
−1
matrix AJ A with its J-minor deleted.
(ii) For each A ∈ Mk,n let ρI (A) ∈ Mk,n−k be the matrix A with
its I-th minor deleted and for each B ∈ Mk,n−k let γI (B) be
the matrix obtained from B by inserting the k × k identity
matrix as the I-th minor of γI (B). Show that B ∈ φI (UI ∩
UJ ) if and only if the J-th minor of γI (B) is non-singular
and show that if γI (B) = A, then φJ ◦φ−1 −1
I (B) = ρJ (AJ A).
−1 ∞
Conclude that φJ ◦ φI is a C map.

2. Tangent Spaces
The entire point of introducing differentiable manifolds is to have
manifolds that one can “do Calculus on”. In order for this to make sense
we need a concept of a derivative. Differentiation, as understood in
elementary calculus courses, involves tangent-making. Essentially the
same is true for the manifold case with a few subtleties. In particular,
if we adopt the intrinsic viewpoint of manifolds, how does one make
sense of a space “tangent” to a manifold when there isn’t a larger
ambient space for the manifold and tangent space to live in? Just as
we have presented two pictures of manifolds, we must now develop their
corresponding versions of tangent spaces.
 CHAPTER 4: FORMS ON MANIFOLDS 13

Notation 2.1. We will now use M and N to refer to differentiable

manifolds instead of the previously used X and Y . Occasionally M n
will be used to denote an n-dimensional differentiable manifold M.
2.1. Curves and Tangents.
Definition 2.2. A curve in M is a pair (γ, I) where I ⊆ R is an open
interval and γ : I → M is a smooth map. We will denote the set of
curves passing through a point p ∈ M by
Cp (M) := {(γ, I), 0 ∈ I, γ(0) = p}.
Remark 2.3. Notice that in the above definition we have given the point
p ∈ M the “special” position of γ(t) for t = 0. Note that as long as
a curve γ passes through the point p for some tp ∈ I, we can then
re-parametrize and obtain γ̃ = γ(t − tp ).
As a convention, we consider (γ1 , I1 ) ≡ (γ2 , I2 ) if and only if γ1 = γ2
on a subinterval I of I1 ∩ I2 with 0 ∈ I.
Remark 2.4. For the initiated reader Cp (M) is called the set of “germs
of C ∞ maps γ : (R, 0) → (M, p).”
Since the tangent space only depends locally on the manifold, we
will focus on the set of curves passing through neighborhood U of
p ∈ M, and, via the equivalence relation “≡” identify Cp (U) with
Cp (M). Furthermore, we want to capture the notion of a tangent
space in terms of vectors tangent to curves. In order to do so we will
temporarily assume that U ⊆ Rn — translating back to the intrinsic
manifold viewpoint later.
Definition 2.5. Assume (γi , Ii ) ∈ Cp (U), i = 1, 2 for U ⊆ Rn . We say
that (γ1 , I1 ) and (γ2 , I2 ) are tangent at p if dγdt1 (0) = dγdt2 (0).
Note that in this setup γi : Ii → U ⊆ Rn are vector-valued functions
and so their derivatives can be thought of as tangent vectors.
Definition 2.6. Two curves in Cp (U) are equivalent if they are tangent
at p, i.e.
dγ1 dγ2
(γ1 , I1 ) ∼p (γ2 , I2 ) ⇔ (0) = (0).
dt dt
Exercise 2.7. Check that this does indeed define an equivalence rela-
tion. Hint – Check that ∼p is reflexive, symmetric and transitive.
Definition 2.8. The tangent space to a point p ∈ U consists of the set
of equivalence classes of curves, i.e.
Tp U := Cp (U) ∼p
14 JUSTIN M. CURRY

Remark 2.9 (Functoriality). The map

dγ
(γ, I) → (0)
dt
induces a bijective map
Tp (U) → Rn .
Now let V be an open subset of Rm and f : (U, p) → (V, q = f (p)) a
C ∞ map. This induces a map
f˜ : Cp (U) → Cq (V )
(γ, I) 7→ (f ◦ γ, I)
which, in turn, induces a map on tangent spaces, i.e. gives rise to a
diagram

f˜
Cp (U) / Cq (V )

 dfp 
Tp U / Tq V

 Jf (p) 
Rn / Rm
We now would like to translate these definitions to the intrinsic man-
ifold setting. In particular, two curves on a manifold are tangent if and
only if their coordinate representations are tangent.
Definition 2.10. Assume M is a manifold, with chart (U, V, ϕ) cen-
tered at p ∈ M, then (γi, Ii ) ∈ Cp (M), i = 1, 2 are tangent at p if ϕ ◦ γ1
and ϕ ◦ γ2 are tangent at ϕ(p) = 0 ∈ Rn .
Exercise 2.11. Show that this definition is independent of the choice of
chart. Hint – Use overlap diagrams and the diagram from the previous
remark.
As before, tangency is an equivalence relation ∼p , thus allowing us
to define the tangent space to a manifold in an identical manner.
Definition 2.12. Tp (M) := Cp (M) ∼p
2.2. Mappings between Tangent Spaces.
Claim 2.13. If M m , N n are manifolds and f : (M, p) → (N, q) a
smooth map, then the map f˜ : Cp (M) → Cp (N) preserves tangency.
 CHAPTER 4: FORMS ON MANIFOLDS 15

Proof. Suppose γi ∈ Cp (M) i = 1, 2 are tangent. By definition, in

some chart (UM , VM , ϕM ) centered at p, d(ϕM ◦ γ1 )/dt(0) = d(ϕM ◦
γ2 )/dt(0). If (UN , VN , ϕN ) is a chart on N such that f (UM ) ⊆ UN , we
see that
d(ϕN ◦ f ◦ γi )/dt(0) = d(ϕN ◦ f ◦ ϕ−1
M )|0 d(ϕM ◦ γi )/dt(0)

are equal for i = 1, 2 so tangency is preserved. Note that γ̂i := ϕM ◦

γi : R → Rm so these derivatives are well-defined. The coordinate
representative of f is fˆ = ϕN ◦ f ◦ ϕ−1
M so that derivative is also well-
defined. 
Remark 2.14 (Functoriality for Mappings of Manifolds). Let (U, V, ϕ)
be a chart centered at p and (U ′ , V ′ , ϕ′ ) at chart centered at f (p) = q,
with f (U) ⊆ U ′ .
f
U / U′
ϕ ∼
= ∼
= ϕ′
 g 
V / V′
Here g = ϕ′ ◦ f ◦ ϕ−1 , which, as before, induces a map on the set of
curves:
f˜
Cp (U) / Cq (U ′ )
ϕ̃ ∼
= ∼
= ϕ̃′
 g̃

C0 (V ) / C0 (V ′ )
Applying the equivalence relation from before then induces a map on
the tangent spaces:
dfp
Tp (U) / Tq (U ′ )
dϕp ∼
= ∼ ′
= dϕp
 dg0

T0 (V ) / T0 (V ′ )

Claim 2.15. The map, dfp ,is independent of the choice of chart.
Proof. For any n-dimensional manifold M we get the overlap diagram

U1 ∩ UI2
ϕ1
u II ϕ
uu II 2
uuu II
u II
uz u ϕ1,2 $
V1,2 / V2,1
16 JUSTIN M. CURRY

and
Tp M
GG
(dϕ1 )p www GG(dϕ2 )p
ww GG
w ∼
w ∼ GG
{ww = (dϕ1,2 )0 = G#
T0 Rn / T0 Rn

∼
= ∼
=
 Jϕ1,2 (0) 
Rn / Rn
Since the maps ϕi i = 1, 2 are diffeomorphisms, their derivatives are
bijections and hence so is (dϕ1 , 2). 
Consider now the case where M n is a manifold and (U, V, ϕ) is a
chart centered at p. The map dϕp allows us to put coordinates on
the tangent space, Tp M, just as ϕ does for the underlying manifold.
Moreover by claim 2.15 this vector space structure is independent of
charts. In other words
Definition 2.16. Tp M becomes a vector space by requiring dϕp to be
linear.
We will leave for you to check the following two claims.
Claim 2.17. If f : (M, p) → (N, q) is a smooth map, then dfp :
Tp M → Tq N is a linear map of vector spaces.
Claim 2.18. If f : (M, p) → (N, q) is a diffeomorphism, then dfp :
Tp M → Tq N is an isomorphism of vector spaces.
This latter claim actually has a stronger cousin known as the inverse
function theorem. Studying maps of tangent spaces reveals important
properties of the underlying maps and vice-versa.
2.3. The Inverse Function Theorem. Assume U1 , U2 are open sets
in Rn and f : (U1 , p1 ) → (U2 , p2 ) is a C ∞ map. The Jacobi mapping
∂fi
Jf (p) : Rn → Rn Jf (p1 ) = [ ∂x j
(p1 )] is the derivative of f at p1 , which
is the best linear approximation to the map. We have the following
result:
Theorem 2.19 (Inverse Function Theorem for Rn ). The map Jf (p1 )
is bijective if and only if f maps a neighborhood of p1 in U1 diffeomor-
phically onto a neighborhood of p2 in U2 .
From the diagram
dfp1
Tp1 U1 / Tp2 U2
∼
= ∼
=
 Jf (p1 ) 
Rn / Rn
 CHAPTER 4: FORMS ON MANIFOLDS 17

we know that Jf (p1 ) bijective ⇔ dfp1 bijective.

Theorem 2.20 (Inverse Function Theorem for Manifolds). Assume
that M1n , M2n are manifolds and f : (M1 , p1 ) → (M2 , p2 ) is a smooth
map. dfp1 : Tp1 M1 → Tp2 M2 is bijective if and only if f maps a neigh-
borhood of p1 diffeomorphically onto a neighborhood of p2 .
Proof. Assume (Ui , Vi, ϕi ) i = 1, 2 are charts centered at pi with f (U1 ) ⊂
U2 . Setting g := ϕ2 ◦ f ◦ ϕ−11 we get the commutative diagram

f
U1 / U2
∼
= ∼
=
 g 
V1 / V2
which induces the corresponding diagram on the tangent spaces
dfp1
Tp1 M1 / Tp2 M2
∼
= ∼
=
 dg0 
T0 V1 / T0 V2
We know that the vertical arrows are bijections by the chain rule.
Accordingly, dfp1 is bijective if and only if dg0 bijective. If dg0 is bi-
jective than g is a local diffeomorphism at 0 by the inverse function
theorem for Rn and thus f is a local diffeomorphism at pi . 

2.4. Submersions. The Inverse Function Theorem provides useful in-

formation about mappings by looking at their closest linear approxi-
mation – the derivative. The assumption that df is bijective is a very
strong assumption and we should consider ways of weakening this as-
sumption. In this subsection we consider the case when df is merely
onto (submersions). In the next subsection we consider when df is 1−1
(immersions).
Since the derivative is a linear mapping, it is natural to assume that
some linear algebra is involved. Let’s review some basic facts.
Definition 2.21 (Canonical Submersion). For k < n we define the
canonical submersion
π : Rn → Rk
to be the map π(x1 , . . . , xn ) = (x1 , . . . , xk ).
We then have the following nice linear algebra result regarding sub-
mersions:
18 JUSTIN M. CURRY

Proposition 2.22. If A : Rn → Rk is onto, there exists a bijective

linear map B : Rn → Rn such that AB = π.
Proof. The proof is an exercise for the reader. Hint: Show that one
can choose a basis, v1 , . . . , vn of Rn such that
Avi = ei , i = 1, . . . , k
is the standard basis of Rk and
Avi = 0 , i > k.
Let e1 , . . . , en be the standard basis of Rn and set Bei = vi . 
Now let us return to the manifold setting.
Definition 2.23. Assume M1m , M2n are manifolds with m ≥ n and
f : (M1 , p1 ) → (M2 , p2 ) is a smooth map. f is a submersion at p1 if
and only if dfp1 is onto.
Theorem 2.24. If f is a submersion at p1 then there exist charts
(Ui , Vi , ϕi) centered at pi such that f (U1 ) ⊆ U2 and the following dia-
gram commutes
f
U1 / U2
∼
= ∼
=
 
π
V1 / V2

Proof. Choose any charts (Ui , Vi, ϕi ) centered at pi such that f (U1 ) ⊂
U2 . Setting g := ϕ2 ◦ f ◦ ϕ−1
1 as before

f
U1 / U2
∼
= ∼
=
 g 
V1 / V2
Check that if dfp1 is onto then dg0 is onto. Hint – Chain Rule. Now
we can make a linear change of coordinates in Rm such that Jg (0) = π.
Let G : (V1 , 0) → (Rm , 0) be the map sending x = (x1 , . . . , xm ) to
G(x) = (g1 (x), . . . , gn (x), xn+1 , . . . , xm ). Recognize that
 
Jg (0) 0
JG (0) =
0 I
and since Jg (0) = π we have that JG (0) = I and π ◦ G = g. By the
inverse function theorem we know that G is a diffeomorphism of V1
 CHAPTER 4: FORMS ON MANIFOLDS 19

onto a neighborhood W1 of 0 in Rm . This results in the diagram

f
U1 / U2
ϕ1 ϕ2
 g 
V1 / V2
=
||
∼ |
G = |||
π
 ||
W1
Replacing ϕ1 with G ◦ ϕ1 provides the desired diagram:
f
U1 / U2
∼
= ∼
=
 π

V1 / V2


2.5. Submanifolds and Regular Values. One important class of

submanifolds are “pre images of regular values of mappings”. More
explicitly:
Definition 2.25. Assume M1n , M2m are manifolds with n > m and
f : M1 → M2 is a smooth map. We say q ∈ M2 is a regular value of f
if for every p ∈ M1 , p ∈ f −1 (q), f is a submersion at p.
Theorem 2.26. Let k = n − m and let X = f −1 (q), then X is a
k-dimensional submanifold of M1 .
Proof. Putting p1 = p and p2 = q, the canonical submersion theorem
says that there exist charts (Ui , Vi , ϕi ) centered at pi such that f (U1 ) =
U2 and
f
U1 / U2
∼
= ∼
=
 π

V1 / V2
is a commutative diagram and π(x1 , . . . , xn ) = (xk+1 , . . . , xn ). Thus
π −1 (0) = {x ∈ V1 , xk+1 = · · · = xn = 0}
= V 1 ∩ Rk
and ϕ maps U1 ∩ f −1 (p2 ) bijectively onto V1 ∩ Rk . Since X = f −1 (p2 )
X ∩ U1 = U is diffeomorphic to an open subset V = V1 ∩ Rk of Rk . 
20 JUSTIN M. CURRY

Example 2.27 (The n-Sphere). If M = Rn+1 and f : Rn+1 → R

f (x) = x21 +· · ·+x2n+1 , then f −1 (1) = Sn . Check that 1 ∈ R is a regular
value of f thus verifying that Sn is an n-dimensional submanifold of
Rn+1 .
Example 2.28 (Orthogonal Matrices). We begin with the following
two manifolds
2
M1 = Mn = Rn
M2 = Sn = Rn(n+1)
where Mn denotes the space of all n × n real matrices and Sn denotes
the space of real symmetric matrices. We then have the map f : M1 →
M2 defined by f (A) = At A. Check that the n×n identity matrix I ∈ Sn
is a regular value of f . Conclude that the space of n × n orthogonal
matrices On = f −1 (I) is an n(n−1)/2 dimensional submanifold of Mn .
2.6. Immersions. Again, we recall what is meant by immersions in
the linear algebra sense.
Definition 2.29 (Canonical Immersion). For k < n we define the
canonical immersion
ι : Rk → Rn
to be the map ι(x1 , . . . , xk ) = (x1 , . . . xk , 0, . . . 0).
Exercise 2.30. Verify that the inclusion map defined above is the
transpose of the projection map, i.e. π t = ι.
Checking the previous exercise leads to a quick proof that every
injective map can be put into its canonical form.
Proposition 2.31. If A : Rk → Rn is one–one, there exists a bijective
linear map C : Rn → Rn such that CA = ι.
Proof. The rank of [ai,j ] is equal to the rank of [aj,i ], so if if A is one–one,
there exists a bijective linear map B : Rn → Rn such that At B = π.
Letting C = B t and taking transposes we get ι = π t = CB 
Definition 2.32. Let f : (M1 , p1 ) → (M2 , p2 ) be a smooth map as
before. f is an immersion at p1 if and only if dfp1 is injective, i.e. 1 − 1.
Theorem 2.33. If f is a immersion at p1 then there exist charts
(Ui , Vi , ϕi) centered at pi such that f (U1 ) ⊆ U2 and the following dia-
gram commutes
f
U1 / U2
∼
= ∼
=
 
ι
V1 / V2
 CHAPTER 4: FORMS ON MANIFOLDS 21

Proof. Choose any charts (Ui , Vi, ϕi ) centered at pi such that f (U1 ) ⊂
U2 . Setting g := ϕ2 ◦ f ◦ ϕ−1
1 as before
f
U1 / U2
∼
= ∼
=
 g 
V1 / V2
Check that if dfp1 is injective then dg0 is injective. Now we can make
a linear change of coordinates in Rn such that Jg (0) = ι. Let Ṽ ⊆ Rn
be the open set {(x1 , . . . , xn ) ∈ Rn , (x1 , . . . , xm ) ∈ V } and let G : Ṽ →
Rn be the map sending x = (x1 , . . . , xn ) to G(x) = g(x1 , . . . , xm ) +
(0, . . . , 0, xm+1 , . . . , xn ). Check that JG (0) = I and G ◦ ι = g.
By the inverse function theorem G maps a neighborhood, W , of 0
in Ṽ diffeomorphically onto a neighborhood of 0 in V2 . In the dia-
gram above lets replace V2 by (V2 )new = G(W ) and V1 by (V1 )new =
g −1(G(W )). Shrinking U1 and U2 if necessary we get the diagram:
f
U1 / U2
ϕ1 ϕ2
 g 
V1 B / V2
O
BB
BBι ∼
BB= G
B
W
Replacing ϕ2 by G−1 ◦ ϕ2 and letting (V2 )new = W provides the desired
diagram:
f
U1 / U2
∼
= ∼
=
 
ι
V1 / V2

2.7. Immersions and the Subset Picture.
Theorem 2.34. Assume M is an n-dimensional manifold. If X a k-
dimensional submanifold then the inclusion map ι : X ֒→ M is a 1 − 1
proper immersion.
Proof. Left as an exercise for the reader. Here are some hints: Let p ∈
X and U an open neighborhood of p in X, V an open neighborhood of
0 in Rk and ϕ : U → V a diffeomorphism. Shrinking U if necessary, we
can extend ϕ to a C ∞ map ϕ̃ : Ũ → V where Ũ is an open neighborhood
22 JUSTIN M. CURRY

of p in M and ϕ̃ is a C ∞ map. Show that ϕ = ϕ̃ ◦ ι and deduce the

injectivity of dιp from the chain rule and bijectivity of dϕp . 
A corollary of this result and the canonical immersion theorem is
the existence of adapted charts for submanifolds. These charts give
us the nice coordinate property that a k-dimensional submanifold only
requires k coordinates of the original n-dimensional chart.
Corollary 2.35. Let M n be a manifold and X ⊂ M a closed subset of
M. X is an k-dimensional submanifold of M if for every p ∈ X there
exists a chart (U, V, ϕ) centered at p and an open neighborhood W of 0
in Rk such that
(2.36) ϕ(U ∩ X) = ι(W ).
By projecting onto the first k-coordinates, we get a homeomorphism
ψ : U ∩ X → W with the defining property
ϕ
U ∩ XH /V
HH
HH }}>
HH }}
ψ HH# }}} ι
}
W
Definition 2.37. A chart (U, V, ϕ) with the property 2.36 is called an
adapted chart for the submanifold X.
We alluded in §1.9 to the “intrinsic” definition of manifolds via charts
and atlases and contrasted this with the “extrinsic” definition in which
manifolds are, by assumption, “submanifolds of RN ”. A celebrated
theorem of Hassler Whitney, the Whitney imbedding theorem asserts
that every n-dimensional manifold can be imbedded diffeomorphically
into R2n+1 as an n-dimensional submanifold, thus showing that these
two definitions are compatible. We won’t attempt to prove Whitney’s
theorem; however in §5 we will sketch a (weak form of) a special case
of this theorem, and leave for you, as an exercise, to fill in the details.
We’ll ask you to show that every compact n-dimensional manifold is,
up to diffeomorphism, a submanifold of some RN . (However, we won’t
require you to show that N = 2n + 1.)
Remark 2.38 (Tangent Space is a Vector Subspace). Assume that M is
an n-dimensional manifold and X is a k-dimensional submanifold with
ι : X ֒→ M the inclusion map. For every p ∈ X the map
dιp : Tp X → Tp M
is injective so by identifying Tp X with its image under this map we can
think of Tp X as a vector subspace of Tp M.
 CHAPTER 4: FORMS ON MANIFOLDS 23

The above remark is visually helpful when M = Rn and so Tp M ∼ =

n k
R . In this case we can think of the tangent space to X at p, Tp X, as a
k-dimensional plane sitting inside a copy of Rn centered at p ∈ X ⊂ Rn .
For manifolds X realized as inverse images of regular values under
a mapping f , the vector subspace structure of Tp X can be defined in
terms of the linear mapping dfp .
Claim 2.39. Assume M1 , M2 are manifolds, f : M1 → M2 a map with
regular value q ∈ M2 defining a submanifold X = f −1 (q). If p ∈ X
then
Tp X = ker dfp .
Proof. Composing f with the inclusion map defines a map f ◦ ι : X →
M2 . Since for all points p ∈ X, f (p) = q it defines a constant map on
X, thus d(f ◦ι)p = dfp ◦dιp = 0. Consequently, Im dιp ⊆ ker dfp proving
that Tp X ⊆ ker dfp . To prove equality, we use a dimension argument.
Since
dim X = dim M1 − dim M2
= dim Tp M1 − dim Tq M2
= dim ker dfp
and dim X = dim Tp X, we have that dim Tp X = dim ker dfp and thus
Tp X = ker dfp . 
Example 2.40 (Skew-Symmetric Matrices). Recall that we showed
that On = f −1 (I) where f : Mn → Sn is the map f (A) = At A. Show
that at the identity matrix I ∈ Mn dfI (A) = A + At . Conclude that
TI On = {A ∈ Mn , At = −A}.
2.8. Exercises.
(1) Let f : X → Y be a C ∞ map and define Z := graph f . Show
that the tangent space to Z at a point (x, y) is
Tx,y Z = {v, dfx(v), v ∈ Tx }.
(2) Let X and Y be manifolds and X × Y their product. Show
that:
Tx,y X × Y ∼
= Tx X × Ty Y

3. Vector Fields on Manifolds

Definition 3.1. Assume Mn is an n-dimensional manifold. A vector
field on M is a function v that associates with every point p a vector
v(p) ∈ Tp M.
24 JUSTIN M. CURRY

3.1. Push-Forwards and Pull-Backs.

Definition 3.2. Let X and Y be manifolds and f : X → Y a C ∞
mapping. Given a vector field, v, on X and a vector field, w, on Y ,
we’ll say that v and w are f -related if, for all p ∈ X and q = f (p)
(3.3) (df )p v(p) = w(q) .
Definition 3.4 (Push-Forward for Vector Fields). Assume X and Y
are manifolds and f : X → Y is a diffeomorphism. The push-forward
of a vector field v on X to a vector field w =: f∗ v on Y is defined
point-wise by equation 3.3.
Definition 3.5 (Pull-Back for Vector Fields). Similarly, given a vector
field, w, on Y we can define a vector field, v, on X by applying the
same construction to the inverse diffeomorphism, f −1 : Y → X. We
will call the vector field (f −1 )∗ w the pull-back of w by f and denote it
by f ∗ w.
Remark 3.6 (Functoriality). Let X, Y and Z be manifolds and f :
X → Y and g : Y → Z diffeomorphisms. If v is a vector field on X,
then
(g ◦ f )∗ v = g∗ (f∗ v)
is a vector field on Z. Similarly, if w is a vector field on Z, then
(g ◦ f )∗ w = f ∗ (g ∗ w)
is a vector field on X.
Exercise 3.7. Describe why the assumption that f : X → Y is a
diffeomorphism is necessary for the definition of push-forwards and
pull-backs for vector fields. Later we will see that pull-backs for k-
forms only require C ∞ maps.
3.2. Smooth Vector Fields. With the push-forward/pull-back ter-
minology in place, we now have a convenient language for discussing
smooth vector fields. If a vector field v is defined on an open subset U
of X then we know, by definition, that for every p ∈ U there exists an
open neighborhood O of p and a diffeomorphism ϕ : O → V ⊆ Rn .
Since Rn is a manifold, and V ⊆ Rn a submanifold, we can just set
Y = V and f = ϕ in the definition of a push-forward to obtain the
vector field ϕ∗ v on V . Following our previous strategy, we will define
a smooth vector field as one whose push-forward on Rn is smooth.
Definition 3.8. A vector field v on U ⊆ X is C ∞ if ϕ∗ v is C ∞ .
In similar fashion we can define a smooth vector field globally if it is
smooth locally for all points on the manifold.
 CHAPTER 4: FORMS ON MANIFOLDS 25

Definition 3.9. A vector field v on a manifold X n is C ∞ if, for every

point p ∈ X, v is C ∞ on a neighborhood of p.
Exercise 3.10. Show that these definitions are independent of the
choice of chart.
Proposition 3.11. Assume v is a C ∞ vector field on X and f : X → Y
is a diffeomorphism, then f∗ v is C ∞ . Similarly, if w is a C ∞ vector field
on Y then f ∗ w is a C ∞ vector field on X.
3.3. Integral Curves. Assume M = M n is an n-dimensional mani-
fold. Recall that a smooth curve passing through p ∈ M is a C ∞ map
γ : I ⊂ R → M satisfying γ(0) = p. The derivative at 0, or tangent
map at 0, is dγ0 : T0 R → Tp M. If we apply dγ0 to the unit vector
u = 1, we obtain
dγ
(0) = dγ0 (1).
dt
Recall that a vector field v just assigns to points on the manifold to vec-
tors in the corresponding tangent space, i.e. v : M → T M. Similarly,
by fixing the unit vector as we have done above and letting the point
of evaluation t to vary we obtain a map dγt : t ∈ I ⊂ R → Tγ(t) M. If a
vector field and a curve agree on this map, we call the curve an integral
curve.
Definition 3.12. γ : I → M is an integral curve of v if for all a ∈ I
dγ
(a) = v(γ(a)).
dt
3.4. ODE Theory on Manifolds.
Proposition 3.13. Let X and Y be manifolds and f : X → Y a C ∞
map. If v and w are vector fields on X and Y which are f -related, then
integral curves of v get mapped by f onto integral curves of w.
Proof. If the curve, γ : I → X is an integral curve of v we have to show
that f ◦ γ : I → Y is an integral curve of w. If γ(t) = p and q = f (p)
then by the chain rule
w(q) = dfp (v(p)) = dfp (dγt (~u))
= d(f ◦ γ)t (~u) .

From this result it follows that the local existence, uniqueness and
“smooth dependence on initial data” results about vector fields that we
described in §2.1 of Chapter 2 are true for vector fields on manifolds.
More explicitly, let U ⊆ X be the domain of some chart and (U, V, ϕ).
26 JUSTIN M. CURRY

Since V ⊆ Rn these results are true for the vector field w = ϕ∗ v and
hence, since w and v are ϕ-related, they are true for v. In particular,
we have recaptured the following important theorems for differential
equations on manifolds:
Proposition 3.14 (Local Existence). For every p ∈ U there exists an
integral curve, γ(t) , −ǫ < t < ǫ, of v with γ(0) = p.
Proposition 3.15 (Local Uniqueness). Let γi : Ii → U i = 1, 2 be
integral curves of v and let I = I1 ∩ I2 . If γ2 (a) = γ1 (a) for some a ∈ I
then γ1 ≡ γ2 on I. Furthermore, there exists a unique integral curve,
γ : I1 ∪ I2 → U with γ = γ1 on I1 and γ = γ2 on I2 .
Proposition 3.16 (Smooth Dependence on Initial Data). For every
p ∈ U there exists a neighborhood, O of p in U, an interval (−ǫ, ǫ) and
a C ∞ map, h : O × (−ǫ, ǫ) → U such that for every p ∈ O the curve
γp (t) = h(p, t) , −ǫ < t < ǫ ,
is an integral curve of v with γp (0) = p.
3.5. Complete Vector Fields.
Definition 3.17. A C ∞ curve γ : [0, b) → M is a maximal integral
curve of v if it can’t be extended to an interval [0, b′ ] where b′ > b, i.e.
there is no integral curve γ ′ : [0, b′ ] → M such that γ = γ ′ |[0,b) .
Definition 3.18. A sequence of points pi ∈ M i = 1, 2, . . . tends to
infinity in M if for every compact set C ⊆ M there exists an i0 such
that pi ∈/ C for i > i0 .
Theorem 3.19. If γ : [0, b) → M is a maximal integral curve of v
then either of the following is true:
(a) b = +∞
(b) There exists a sequence {ti } ∈ [0, b) with ti → b and the image
sequence {γ(ti )} tends to infinity in M.
Proof. (See Birkhoff and Rota page 172.) Suppose not. Then there
exists a compact set C such that for all t ∈ [0, b) γ(t) ∈ C and hence
there exists a sequence ti ∈ [0, b) and a point p ∈ C such that ti → t
and γ(ti ) → p. Let U be a open neighborhood of p and
h : U × (−ǫ, ǫ) → M
a C ∞ map such that for all q ∈ U γq (t) = h(q, t) −ǫ < t < ǫ is an
integral curve of p with γq (0) = q. Since γ(ti ) → p and ti → b there
 CHAPTER 4: FORMS ON MANIFOLDS 27

exists a ti such that ti > b − 2ǫ and γ(ti ) ∈ U. Set c = ti and q = γ(ti ).

Then the curve γ̃(t) : [0, b + 2ǫ ) → M defined by

γ(t) 0≤t≤c
γ̃(t) =
γq (t + c) c ≤ t ≤ b + 2ǫ
is an integral curve extending γ. 
If M is compact then we can, in fact, say more. Because an integral
curve cannot, by definition, leave the manifold, then if the manifold is
compact, it can never tend to infinity. By the previous theorem, this
curve exists for all future time. Vector fields that have integral curves
for all time are called complete.
Definition 3.20. A vector field v is complete if for every p ∈ M there
exists an integral curve γp (t) −∞ < t < ∞ satisfying γp (0) = p.
Theorem 3.21. If M is compact, then, for any vector field v on M,
v is complete.
Proof. By the theorem above there exists for every p ∈ M an integral
curve γp : [0, ∞) → M with γp (0) = p. Noting that integral curves
of −v are just integral curves of v reparametrized by setting t → −t.
Accordingly, by the theorem above, for every p ∈ M an integral curve
γp : (−∞, 0] → M of v with γp (0) = p and putting these two results
together an integral curve of v γp : (−∞, ∞) → M with γp (0) = p. 
Complete vector fields naturally give rise to a map that allows us to
“flow” open sets on a manifold based on how the individual integral
curves evolve in time for each point. Such a map f : M × R → M is
obtained by setting
f (p, t) := γp (t)
and then defining the associated map ft : M → M to be a function of
points rather than time, i.e.
ft (p) := f (p, t).
One can visualize this map as placing a tracer particle in some deter-
ministic fluid flow and watching the particle, originating at some point
p, get carried after some time s to a point fs (p). Allowing the map to
begin with fs (p) and flow for some time t should obtain the same result
as starting at p and flowing for time t+s, since the flow is deterministic.
Claim 3.22. If v is complete and ft : M → M is defined as above,
then ft ◦ fs = ft+s .
28 JUSTIN M. CURRY

Proof. Let q = fs (p). Then the curves γ1 (t) = fs+t (p) −∞ < t < ∞
and γ2 (t) = ft (q) −∞ < t < ∞ are both integral curves of v with initial
point γ1 (0) = γ2 (0) = q and hence they coincide for all time. 
Theorem 3.23. If M is compact then f : M × R → M is C ∞ .
Proof. By smooth dependence on initial data there exists for every
p ∈ M a neighborhood Up of p and an ǫp > 0 such that
f : Up × (−ǫp , ǫp ) → M
is C ∞ . The sets Up for p ∈ M are a covering of M, which is compact.
There exists a finite subcover Upi i = 1, . . . , k of M. Now let ǫ = min ǫpi .
Since the sets Upi × (−ǫ, ǫ) cover M × (−ǫ, ǫ) and f restricted to each
of these sets is C ∞ ,
f : M × (−ǫ, ǫ) → M
∞
is a C map. Now note that for any T > 0 there exists a large positive
integer N such that T /N < ǫ. Using the fact that
fT = fT /N ◦ · · · ◦ fT /N
and since each fT /N is C ∞ , the composition f : M × (−T, T ) → M is
C ∞ . Since T is arbitrary, we conclude that f : M × R → M is C ∞ . 
Remark 3.24. With a little effort, the assumption that M is compact
in Theorem 3.23 can be weakened. See Exercises 12 and 13 at the end
of the section for more details.
3.6. One-Parameter Groups of Diffeomorphisms. You may have
observed from Claim 3.22 that the functions ft are parameterized by
the additive group (R, +). In particular, f0 = IdM is the identity
mapping and for any t ∈ R there exists a −t ∈ R so that ft ◦ f−t =
ft−t = f0 = IdM , so the set ft has an identity element and inverses –
making it a group. The result stating that f : M × R → M is C ∞ gives
us that ft : M → M is C ∞ for any t as well. Consequently, both ft and
its inverse (ft )−1 = f−t are C ∞ – making it a diffeomorphism.
Definition 3.25. The set of functions
{ft , −∞ < t < ∞}
is called a one-parameter group of diffeomorphisms generated by v.
If the vector field v is not complete then it does not generate a
one-parameter group of diffeomorphisms. We have shown that if the
underlying manifold M is compact, then any vector field is complete,
but if M is not compact, then it admits a great deal of incomplete
vector fields.
 CHAPTER 4: FORMS ON MANIFOLDS 29

∂
Example 3.26. Let M = R and v = x2 ∂x . Exercise – What are the
integral curves? Where are they undefined?
Proposition 3.27. If a vector field v is compactly supported then it
is complete – and thus generates a one-parameter group of diffeomor-
phisms.
Proof. The proof is identical to the case when M is compact. See
above. 
3.7. Exercises.
(1) Let X ⊆ R3 be the paraboloid, x3 = x21 + x22 and let w be the
vector field
∂ ∂ ∂
w = x1 + x2 + 2x3 .
∂x1 ∂x2 ∂x3
(a) Show that w is tangent to X and hence defines by
restriction a vector field, v, on X.
(b) What are the integral curves of v?
(2) Let S 2 be the unit 2-sphere, x21 + x22 + x23 = 1, in R3 and let w
be the vector field
∂ ∂
w = x1 − x2 .
∂x2 ∂x1
(a) Show that w is tangent to S 2 , and hence by restriction
defines a vector field, v, on S 2 .
(b) What are the integral curves of v?
(3) As in problem 2 let S 2 be the unit 2-sphere in R3 and let w be
the vector field
 
∂ ∂ ∂ ∂
w= − x3 x1 + x2 + x3
∂x3 ∂x1 ∂x2 ∂x3
(a) Show that w is tangent to S 2 and hence by restriction
defines a vector field, v, on S 2 .
(b) What do its integral curves look like?
(4) Let S 1 be the unit circle, x21 +x22 = 1, in R2 and let X = S 1 ×S 1
in R4 with defining equations
f1 = x21 + x22 − 1 = 0

f2 = x23 + x24 − 1 = 0 .
(a) Show that the vector field
 
∂ ∂ ∂ ∂
w = x1 − x2 + λ x4 − x3 ,
∂x2 ∂x1 ∂x3 ∂x4
30 JUSTIN M. CURRY

λ ∈ R, is tangent to X and hence defines by restriction a

vector field, v, on X.
(b) What are the integral curves of v?
(c) Show that Lw fi = 0.
(5) For the vector field, v, in problem 4, describe the one-parameter
group of diffeomorphisms it generates.
(6) Let X and v be as in problem 1 and let f : R2 → X be the
map, f (x1 , x2 ) = (x1 , x2 , x21 + x22 ). Show that if u is the vector
field,
∂ ∂
u = x1 + x2 ,
∂x1 ∂x2
then f∗ u = v.
(7) Let X be a submanifold of X in RN and let v and w be the
vector fields on X and U. Denoting by ι the inclusion map of
X into U, show that v and w are ι-related if and only if w is
tangent to X and its restriction to X is v.
(8) Let X be a submanifold of RN and U an open subset of RN
containing X, and let v and w be the vector fields on X and
U. Denoting by ι the inclusion map of X into U, show that
v and w are ι-related if and only if w is tangent to X and its
restriction to X is v.
(9)An elementary result in number theory asserts
Theorem 3.28. A number, λ ∈ R, is irrational if and only if
the set
{m + λn , m and n integers}
is a dense subset of R.
Let v be the vector field in problem 4. Using the theorem
above prove that if λ/2π is irrational then for every integral
curve, γ(t), −∞ < t < ∞, of v the set of points on this curve
is a dense subset of X.
(10) Let v be a vector field on X and ϕ : X → R, a C ∞ function.
Show that if the function
(3.29) Lv ϕ = ι(v) dϕ
is zero ϕ is constant along integral curves of v.
(11) Suppose that ϕ : X → R is proper. Show that if Lv ϕ = 0, v is
complete.
Hint: For p ∈ X let a = ϕ(p). By assumption, ϕ−1 (a) is
compact. Let ρ ∈ C0∞ (X) be a “bump” function which is one
on ϕ−1 (a) and let w be the vector field, ρv. By Theorem 3.27,
w is complete and since
 CHAPTER 4: FORMS ON MANIFOLDS 31

Lw ϕ = ι(ρv) dϕ = ρι(v) dϕ = 0
ϕ is constant along integral curves of w. Let γ(t), −∞ < t <
∞, be the integral curve of w with initial point, γ(0) = p. Show
that γ is an integral curve of v.
(12) Show that in Theorem 3.23 the hypothesis that M is compact
can be replaced with v compactly supported.
(13) Let Ci i = 1, 2, . . . be an exhaustion of M by compact sets and
ρi ∈ Co∞ (M) a bump function which is 1 on Ci . By 12, Theorem
3.23 is true for each of the vector fields vi := ρi v. Conclude that
Theorem 3.23 is true for v itself. Here are a few hints:
(a) Let γ : [0, T ] → M be an integral curve of v. Show that
for some i γ([0, T ]) is contained in IntCi .
(b) Conclude that γ is an integral curve of vi .
(c) Let (fi )t , −∞ < t < ∞ be the one-parameter group of
diffeomorphisms associated with vi . Show that there exists
a neighborhood, U, of γ(0) with the property
(fi )t (U) ⊆ IntCi
for 0 ≤ t ≤ T .
(d) Conclude that, on the set U × [0, T ), fi = f .
(e) Conclude from the foregoing that f : U × [0, T ] → M is a
C ∞ map.
(f) Finally, noting that p = γ(0) can be an arbitrary point on
M and T an arbitrary positive real number, conclude that
f : M × (−∞, ∞) → M
is C ∞ .

4. Differential Forms on Manifolds

Definition 4.1. Let X be a manifold. A differential k-form is a func-
tion ω that assigns to each p ∈ X an element ω(p) of Λk (Tp∗ X).
4.1. Push-Forwards and Pull-Backs. Recall that in the previous
section we defined the terms push-forward and pull-back for vector
fields via a diffeomorphism f . For differential forms the situation is
even nicer. Just as in §2.5 we can define the pull-back operation on
forms for any C ∞ map f : X → Y . Specifically: Let ω be a k-form on
Y . For every p ∈ X, and q = f (p) the linear map
dfp : Tp X → Tq Y
32 JUSTIN M. CURRY

induces by (1.8.2) a pull-back map

(dfp )∗ : Λk (Tq∗ ) → Λk (Tp∗ )
and, as in §2.5, we’ll define the pull-back, f ∗ ω, of ω to X by defining
it at p by the identity
(4.2) (f ∗ ω)(p) = (dfp )∗ ω(q) .
Notice that, in contrast to vector fields where we defined push-
forwards first, pull-backs of k-forms are defined first for any C ∞ map,
independent of push-forwards. However, in both scenarios, we can get
push-forwards as pull-backs of the inverse map. To summarize:
Definition 4.3 (Pull-backs and Push-forwards). Suppose X and Y
are manifolds, ω a k-form in Y , with f : X → Y a C ∞ map where
f (p) = q. The pull-back of ω is defined point-wise as
(f ∗ ω)(p) = (dfp )∗ ω(q).
Now suppose that η is a k-form on X and g : X → Y is a diffeomor-
phism, g(p) = q then the push-forward of η is
(g∗ η)(q) = (g −1∗ η)(q) = (dgq−1 )∗ η(p).
Remark 4.4 (Functoriality). Let X, Y and Z be manifolds and f :
X → Y and g : Y → Z C ∞ maps. Then if ω is a k-form on Z
f ∗ (g ∗ ω) = (g ◦ f )∗ ω .
Now assume that ω is a k-form on X and f : X → Y , g : Y → Z are
diffeomorphisms. Observe that
(g ◦ f )∗ ω = [(g ◦ f )−1 ]∗ ω
= (f −1 ◦ g −1)∗ ω
= g −1∗ (f −1∗ ω)
= g∗ (f∗ ω).
4.2. Smooth k-forms. As with vector fields, smooth k-forms are k-
forms whose local coordinate representations are smooth.
Definition 4.5. Assume U ⊆ X is the domain of some chart (U, V, ϕ)
and ω is a k-form on U. ω is C ∞ if ϕ∗ ω is C ∞ on V .
Extending this definition globally provides us with the definition:
Definition 4.6. A k-form ω on X is C ∞ if, for every point p ∈ X, ω
is C ∞ on some neighborhood of p.
Exercise 4.7. As with vector fields, show that these definitions are
independent of choice of chart.
 CHAPTER 4: FORMS ON MANIFOLDS 33

Proposition 4.8. Assume Xi i = 1, 2 are manifolds and f : X1 → X2

is a C ∞ map. If ω is a C ∞ k-form on Y , then f ∗ ω is a C ∞ k-form on
X.
Proof. For pi ∈ Xi i = 1, 2 and p2 = f (p1 ) let Ui be neighborhoods of
pi , shrinking U1 if necessary so that f (U1 ) ⊆ U2 . Assuming (Ui , Vi , ϕi)
i = 1, 2 are charts in their respective atlases, then the map g := ϕ2 ◦
f ◦ ϕ−1 ∞
1 : V1 → V2 is C . Since ω is a C
∞
k-form on Y , ϕ2∗ ω is C ∞ , so
ϕ1∗ (f ∗ ω) = (f ◦ ϕ−1 )∗ ω = (ϕ−1 ∗ ∗
2 ◦ g) = g (ϕ2∗ ω)

is C ∞ . 
4.3. The Exterior Differentiation Operation.
Definition 4.9. Let ω be a C ∞ k-form on a manifold X. For p ∈ X
take a chart (U, V, ϕ) and define the exterior derivative of ω, dω, on X
by the formula
(4.10) dω := (ϕ−1 )∗ dϕ∗ ω .
We now wish to show that this definition is independent of chart.
Assume p ∈ X is contained in two charts (Ui , Vi , ϕi ) i = 1, 2 and define
ψ := ϕ2 ◦ ϕ−1
1 . Observe that

dϕ2∗ ω = d(ψ ◦ ϕ1 )∗ ω
= dψ∗ ϕ1∗ ω
= ψ∗ dϕ1∗ ω
= ϕ2∗ ϕ−1
1∗ dϕ1∗ ω
hence

ϕ−1 −1
2∗ dϕ2∗ ω = ϕ1∗ dϕ1∗ ω .

Note that we have used the fact that ψ∗ d = dψ∗ , i.e. the push-
forward commutes with the exterior derivative d. This follows from
the fact that the pull-back commutes with the exterior derivative and
defining the push-forward as the pull-back via the inverse map. Recall
that for ω = f dxI and a C ∞ map G, G∗ dω = G∗ (df ∧ dxI ) = d(f ◦ G) ∧
d(xI ◦ G) = dG∗ (f dxI ).
We can therefore, define the exterior derivative, dω, globally by
defining it to be equal to (4.10) on every chart.
It’s easy to see from the Definition 4.9 that this exterior differenti-
ation operation inherits from the exterior differentiation operation on
open subsets of Rn the properties from Section 2.3 of Chapter 2:
I. For ω1 , ω2 in Ωk (U) d(ω1 + ω2 ) = dω1 + dω2 .
34 JUSTIN M. CURRY

II. For ω1 ∈ Ωk (U) and ω2 ∈ Ωℓ (U) d(ω1 ∧ ω2 ) = dω1 ∧ ω2 +

(−1)k ω1 ∧ dω2 .
III. For ω ∈ Ωk (U) d(dω) = 0.
Note that for zero forms, i.e., C ∞ functions, f : X → R, df is the
“intrinsic” df defined in Section 2.1, i.e., for p ∈ X dfp is the derivative
of f
dfp : Tp X → R
viewed as an element of Λ1 (Tp∗ X).

4.4. The Interior Product and Lie Derivative Operation. Given

a k-form, ω ∈ Ωk (X) and a C ∞ vector field, w, we will define the interior
product

(4.11) ι(v)ω ∈ Ωk−1 (X) ,

as in §2.4, by setting
(ι(v)ω)p = ι(vp )ωp
and the Lie derivative

(4.12) Lv ω = Ωk (X)
by setting

(4.13) Lv ω = ι(v) dω + dι(v)ω .

It’s easily checked that these operations satisfy the identities (2.4.2)–
(2.4.8) and (2.4.12)–(2.4.13) (since, just as in §2.4, these identities are
deduced from the definitions (4.11) and (3.3) by purely formal manip-
ulations). Moreover, if v is complete and

ft : X → X , −∞ < t < ∞

is the one-parameter group of diffeomorphisms of X generated by v the

Lie derivative operation can be defined by the alternative recipe
 
d ∗
(4.14) Lv ω = f ω (t = 0)
dt t
as in (2.5.22). (Just as in §2.5 one proves this by showing that the
operation (4.14) has the properties (2.12) and (2.13) and hence that it
agrees with the operation (4.13) provided the two operations agree on
zero-forms.)
 CHAPTER 4: FORMS ON MANIFOLDS 35

4.5. Exercises in Symplectic Geometry: The Cotangent Bun-

dle. The following is an extremely important example of a manifold.
It arises naturally in classical mechanics, by considering an N-particle
system constrained to a certain configuration space X (X is typically
a submanifold of R3N ). A concrete example is the pendulum, whose
configuration space is simply X = S1 . The phase space – or position-
momentum space – is then described by the cotangent bundle, which
we will define below.
Definition 4.15. Let M = M 2n be an even dimensional manifold and
ω ∈ Ω2 (M). We say ω is a symplectic form if dω = 0 (ω is closed) and
(ω ∧ · · · ∧ ω)p = ωp∧n 6= 0 for all p ∈ M (ω is non-degenerate).
Definition 4.16. If M = M 2n and ω is a symplectic form, then we
call (M, ω) a symplectic manifold.
(1) Cotangent Bundle as a Manifold. Let X be an n-dimensional
manifold,
T ∗ X = {(p, ζ), p ∈ X, ζ ∈ Tp∗ X}
the cotangent bundle, and π : T ∗ X → X the projection map
(p, ζ) 7→ p. Show that if (U, V, ϕ) is a chart on X and T ∗ U =
π −1 (U) is the subset of T ∗ X “sitting over U” then one gets a
bijective map
ϕ̃ : Ũ := T ∗ U → V × Rn∗ =: Ṽ
mapping (p, ζ) to (q, η) where ϕ(p) = q η ∈ Tq∗ Rn ∼ = Rn∗ ζ =
(dϕp )∗ η. Show that the collection (Ũ , Ṽ , ϕ̃) is an atlas in the
sense of Exercise 4. Conclude that the cotangent bundle of a
manifold, T ∗ X, is a manifold.
(2) Canonical One Form. Let M = T ∗ X be the cotangent bun-
dle of a manifold. For each point q = (p, ζ) ∈ M, let αq be the
element of Tq∗ M defined by the identity αq = (dπ)∗q ζ. Note that
dπq is a map from Tq M → Tp X so its transpose goes in the
opposite direction. Show that the assignment q ∈ M 7→ αq ∈
Tq∗ M defines a one-form. (This form is called the canonical
one-form on M.)
(3) Functoriality. Show that if Xi are manifolds, f : X1 → X2
a diffeomorphism, and Mi = T ∗ Xi i = 1, 2 are the respective
cotangent bundles, then f gives rise to a diffeomorphism g :
M1 → M2 where

p2 = f (p1 )
g(p1 , ζ1 ) = (p2 , ζ2 ) ⇔
ζ1 = (dfp1 )∗ ζ2
36 JUSTIN M. CURRY

Show that if α1 and α2 are the canonical one-forms on M1 and

M2 then g ∗α2 = α1 . Hint: Look at
dg
Tq1 M / Tq2 M
dπq1 dπq2
 df

Tp1 X1 / Tp2 X2

which after applying transposes reverses the vertical and hori-

zontal arrows.
(4) Show that if X = Rn and M = T ∗ Rn = Rn × Rn∗ , with linear
coordinates, xP1 , . . . , xn , ξ1 , . . . , ξn , then the canonical one-form
α is equal to i ξi dxi . Conclude that α is smooth.
(5) Let M = T ∗ X and let ω = −dα be the exterior derivative of
the canonical one-form. Show that ω is symplectic. Hint: By
functoriality it suffices to check this for X = Rn .
(6) Hamiltonian Vector Fields. Let H be a C ∞ function on
M. Show that there is a unique vector field vH on M with the
defining property
ι(vH )ω = dH.
Hint: By linear algebra, check that ωpn 6= 0 if and only if the
map Tp M → Tp∗ M v 7→ ι(v)ωp is bijective. Alternative Hint:
By functoriality, check the special case X = Rn .
(7) Conservation of Energy. Suppose vH is complete. Let ft :
M → M be the one-parameter group of diffeomorphisms. From
ι(vH )ω = dH conclude that ft∗ H = H, i.e. the Hamiltonian –
“energy” – is conserved. Hint: Show LvH H = 0, ft∗ ω = ω.
(8) Conservation of Volume. Following the previous exercise,
from ι(vH )ω = 0 conclude that ft∗ ω n = ω n .

5. Some Topological Results

5.1. The Support of a Function.
Definition 5.1. Assume that U is an open subset of Rn and f : U → R
a C ∞ map. Define the support of a function f to be the closed set
supp(f ) := {p ∈ U, f (p) 6= 0}.
Definition 5.2. A function is said to be compactly supported if its
support is compact. For U ⊆ Rn , the space of compactly supported
C ∞ functions defined on U is denoted Co∞ (U).
 CHAPTER 4: FORMS ON MANIFOLDS 37

Remark 5.3. Assume X is a manifold and U ⊆ X. If f ∈ Co∞ (U) then

we can regard f as being in Co∞ (X), i.e. we can extend f to a function
on all of X by setting f ≡ 0 on X r U.
5.2. Partitions of Unity. We are now poised to introduce the use
of partitions of unity. Requisite to their definition is the existence of
so-called bump functions.
Lemma 5.4. Let X be an n-dimensional manifold. For every point
p ∈ X and open set U ⊆ X containing p there exists a function f ∈
Co∞ (U) such that f ≡ 1 on a neighborhood of p and 0 ≤ f ≤ 1.
Proof. Shrinking U if necessary, we can assume U is the domain of a
chart ϕ : U → V . Let q = ϕ(p). The result is true for V, q and hence
for U, p.

Theorem 5.5. If X is a manifold with an open cover {Uα , α ∈ I},
then there exist functions ρi ∈ Co∞ (X), i ∈ N such that
(1) 0 ≤ ρi ≤ 1
(2) for every compact set C there exists an i0 such that supp(ρi ) ∩
C
P= ∅ for i > i0
(3) i ρi = 1
(4) For every i there exists an αi ∈ I such that supp(ρi ) ⊆ Uαi .
Proof. Let C1 ⊆ C2 ⊆ · · · be a compact exhaustion of X. Pick a
point p ∈ Ci r Int(Ci−1 ). p ∈ Uα for some α ∈ I and there exists
a bump function fp ∈ Co∞ ((Int(Ci+1 ) − Ci−2 ) ∩ Uα ) with fp ≡ 1 on
a neighborhood Up of p. The sets Up , p ∈ Ci r Int(Ci−1 ) cover Ci r
Int(Ci−1 ). Since Ci rInt(Ci−1 ) is compact there exists a finite subcover
Upk , 1 ≤ k ≤ Ni . Let fk,i = fpk , i.e. fk = fk,1 1 ≤ k ≤ N1 , fk+N1 =
fk,2 1 ≤ k ≤ N2 , etc. The functions fi have all the properties desired
above except the third property, and to arrange that they have this
property simply renormalize, setting ρi = Pfifi . 

5.3. Proper Mappings.

Lemma 5.6. If Mi i = 1, 2 are manifolds and f : M1 → M2 is a proper
C ∞ map, then
(a) f (M1 ) is a closed subset of M2
(b) Given p1 ∈ M1 and U1 a neighborhood of p1 in M1 there exists
a neighborhood U2 of the image point p2 = f (p1 ) in M2 such
that f −1 (U2 ) ⊂ U1
38 JUSTIN M. CURRY

Theorem 5.7. If f : M1 → M2 is a 1 − 1 proper immersion and X =

f (M1 ), then X is a closed submanifold of M2 and f is a diffeomorphism
of M1 onto X.
Proof. Let p1 ∈ M1 and p2 = f (p1 ). By the canonical immersion
theorem, there exists a chart (Ui , Vi , ϕi ) centered at pi such that U1 ⊂
f −1 (U2 ) and
f
U1 / U2
∼
= ∼
=
 
ι
V1 / V2
where ι : Rk → Rn is the canonical immersion
ι(x1 , . . . , xk ) = (x1 , . . . , xk , 0, . . . , 0)
(k and n are the dimensions of M1 and M2 ). By the above lemma
we can shrink U2 such that f −1 (U2 ) = U1 and hence ι(V1 ) = ϕ2 (U2 ∩
X). Thus ι−1 ◦ ϕ2 : U2 ∩ X → V1 is a diffeomorphism of the open
neighborhood U = U2 ∩ X of p2 in V onto the open subset V1 of
Rk . 
5.4. The Whitney imbedding theorem. The following exercises
outline a proof of the Whitney Embedding Theorem for compact man-
ifolds.
(1) Partitions of Unity on Compact sets. Let X be a compact
set and Uα α = 1, . . . m an open cover. If ρi i = 1, . . . k is a
partition of unity subordinate to the open cover, show that this
partition of unity can be chosen so that m = k and the support
of ρi is contained in Ui .
(2) Whitney Embedding for Compact Manifolds. Let X be a
compact n-manifold. Suppose (Ui , Vi, ϕi ) are charts on X whose
domains are a cover of X. Show that there is a 1 − 1 immersion
f : X → RnN +N . Hint: Let ρi i = 1, . . . , N be a partition of
unity as in Exercise 1. Show that the map f : X → RnN +N
defined by
f (x) = (ρ1 ϕ1 (x), . . . , ρN ϕN (x), ρ1 , . . . , ρN )
is such a 1 − 1 immersion. Sub-Hint: Show that if p ∈ Ui and
dfp (v) = 0 then (dρi )p (v) = 0 and ρi (p)(dϕi)p (v) = 0.
(3) Whitney Embedding for RP2 . Consider the map γ : S2 →
R6 defined by
(x1 , x2 , x3 ) → (x21 , x22 , x23 , x1 x2 , x1 x3 , x2 , x3 ).
 CHAPTER 4: FORMS ON MANIFOLDS 39

(a) Show that this map is an immersion.

(b) Show that it is 2 − 1 i.e. that
γ(x1 , x2 , x3 ) = γ(−x1 , −x2 , −x3 ).
(c) Conclude that there is a 1 − 1 immersion γ̃ : RP2 → R6
with the property
γ̃([x1 , x2 , x3 ]) = γ(x1 , x2 , x3 )
for (x1 , x2 , x3 ) ∈ S2 .