Differentiable Manifolds: Lecture Notes For Geometry 2

Differentiable manifolds
Lecture Notes for Geometry 2
Henrik Schlichtkrull
Department of Mathematics
University of Copenhagen
i
Preface
The purpose of these notes is to introduce and study differentiable mani-
folds. Differentiable manifolds are the central objects in differential geometry,
and they generalize to higher dimensions the curves and surfaces known from
Geometry 1. Together with the manifolds, important associated objects are
introduced, such as tangent spaces and smooth maps. Finally the theory
of differentiation and integration is developed on manifolds, leading up to
Stokes’ theorem, which is the generalization to manifolds of the fundamental
theorem of calculus.
These notes continue the notes for Geometry 1, about curves and surfaces.
As in those notes, the figures are made with Anders Thorup’s spline macros.
The notes are adapted to the structure of a course which stretches over 9
weeks. There are 9 chapters, each of a size that it should be possible to cover
in one week. For a shorter course one can read the first 6 chapters. The notes
were used for the first time in 2006. The present version has been revised,
and I would like to thank Andreas Aaserud, Hans Plesner Jakobsen and Kang
Li for many corrections and suggestions of improvements. Further revision
is undoubtedly needed, and comments and corrections will be appreciated.
Henrik Schlichtkrull
October, 2011
ii
Contents
1. Manifolds in Euclidean space. . . . . . . . . . . . . . . . . 1
1.1 Parametrized manifolds . . . . . . . . . . . . . . . . . 1
1.2 Embedded parametrizations . . . . . . . . . . . . . . . 3
1.3 Curves . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Chart and atlas . . . . . . . . . . . . . . . . . . . . 8
1.6 Manifolds . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 The coordinate map of a chart . . . . . . . . . . . . . . 11
1.8 Transition maps . . . . . . . . . . . . . . . . . . . . 13
1.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . 15
2. Abstract manifolds . . . . . . . . . . . . . . . . . . . . . 17
2.1 Topological spaces . . . . . . . . . . . . . . . . . . . 17
2.2 Abstract manifolds . . . . . . . . . . . . . . . . . . . 19
2.3 Examples . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Projective space . . . . . . . . . . . . . . . . . . . . 21
2.5 Product manifolds . . . . . . . . . . . . . . . . . . . 22
2.6 Smooth functions on a manifold . . . . . . . . . . . . . 23
2.7 Smooth maps between manifolds . . . . . . . . . . . . . 25
2.8 Lie groups . . . . . . . . . . . . . . . . . . . . . . . 28
2.9 Countable atlas . . . . . . . . . . . . . . . . . . . . 29
2.10 Whitney’s theorem . . . . . . . . . . . . . . . . . . . 30
2.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . 31
3. The tangent space . . . . . . . . . . . . . . . . . . . . . 33
3.1 The tangent space of a parametrized manifold . . . . . . . 33
3.2 The tangent space of a manifold in Rn . . . . . . . . . . 34
3.3 The abstract tangent space . . . . . . . . . . . . . . . 35
3.4 Equivalence of curves . . . . . . . . . . . . . . . . . . 37
3.5 The vector space structure . . . . . . . . . . . . . . . 38
3.6 Directional derivatives . . . . . . . . . . . . . . . . . 39
3.7 Action on functions . . . . . . . . . . . . . . . . . . . 40
3.8 The differential of a smooth map . . . . . . . . . . . . . 41
3.9 The standard basis . . . . . . . . . . . . . . . . . . . 45
3.10 Orientation . . . . . . . . . . . . . . . . . . . . . . 45
3.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . 46
4. Submanifolds . . . . . . . . . . . . . . . . . . . . . . . 49
4.1 Submanifolds in Rk . . . . . . . . . . . . . . . . . . . 49
4.2 Abstract submanifolds . . . . . . . . . . . . . . . . . 49
4.3 The local structure of submanifolds . . . . . . . . . . . . 51
4.4 Level sets . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 The orthogonal group . . . . . . . . . . . . . . . . . . 57
4.6 Domains with smooth boundary . . . . . . . . . . . . . 58
iii
4.7 Orientation of the boundary . . . . . . . . . . . . . . . 60

4.8 Immersed submanifolds . . . . . . . . . . . . . . . . . 62
4.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . 62
5. Topological properties of manifolds . . . . . . . . . . . . . . 65
5.1 Compactness . . . . . . . . . . . . . . . . . . . . . 65
5.2 Countable exhaustion by compact sets . . . . . . . . . . 66
5.3 Locally finite atlas . . . . . . . . . . . . . . . . . . . 67
5.4 Bump functions . . . . . . . . . . . . . . . . . . . . 68
5.5 Partition of unity . . . . . . . . . . . . . . . . . . . 69
5.6 Embedding in Euclidean space . . . . . . . . . . . . . . 70
5.7 Connectedness . . . . . . . . . . . . . . . . . . . . . 72
5.8 Connected manifolds . . . . . . . . . . . . . . . . . . 75
5.9 Components . . . . . . . . . . . . . . . . . . . . . . 76
5.10 The identity component of a Lie group . . . . . . . . . . 77
5.11 The Jordan-Brouwer theorem . . . . . . . . . . . . . . 79
5.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . 79
6. Vector fields and Lie algebras . . . . . . . . . . . . . . . . 81
6.1 Smooth vector fields . . . . . . . . . . . . . . . . . . 81
6.2 An equivalent formulation of smoothness . . . . . . . . . 83
6.3 Derivations . . . . . . . . . . . . . . . . . . . . . . 85
6.4 The tangent bundle . . . . . . . . . . . . . . . . . . . 85
6.5 The Lie bracket . . . . . . . . . . . . . . . . . . . . 87
6.6 Properties of the Lie bracket . . . . . . . . . . . . . . . 89
6.7 The Lie algebra of a Lie group . . . . . . . . . . . . . . 90
6.8 The tangent space at the identity . . . . . . . . . . . . 92
6.9 Homomorphisms of Lie groups . . . . . . . . . . . . . . 93
6.10 The Lie algebra of GL(n, R) . . . . . . . . . . . . . . . 94
6.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . 95
7. Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.1 The dual space . . . . . . . . . . . . . . . . . . . . . 97
7.2 The dual of a linear map . . . . . . . . . . . . . . . . 98
7.3 Tensors . . . . . . . . . . . . . . . . . . . . . . . . 99
7.4 Alternating tensors . . . . . . . . . . . . . . . . . . 101
7.5 The wedge product . . . . . . . . . . . . . . . . . . 104
7.6 The exterior algebra . . . . . . . . . . . . . . . . . 107
7.7 Exercises . . . . . . . . . . . . . . . . . . . . . . 107
8. Differential forms . . . . . . . . . . . . . . . . . . . . . 109
8.1 The cotangent space . . . . . . . . . . . . . . . . . 109
8.2 Covector fields . . . . . . . . . . . . . . . . . . . . 110
8.3 Differential forms . . . . . . . . . . . . . . . . . . . 113
8.4 Pull back . . . . . . . . . . . . . . . . . . . . . . 114
8.5 Exterior differentiation . . . . . . . . . . . . . . . . 116
iv
8.6 Exterior differentiation and pull back . . . . . . . . . . 122

8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . 122
9. Integration . . . . . . . . . . . . . . . . . . . . . . . 125
9.1 Null sets . . . . . . . . . . . . . . . . . . . . . . 125
9.2 Integration on Rn . . . . . . . . . . . . . . . . . . 126
9.3 Integration on a chart . . . . . . . . . . . . . . . . . 128
9.4 Integration on a manifold . . . . . . . . . . . . . . . 131
9.5 A useful formula . . . . . . . . . . . . . . . . . . . 133
9.6 Stokes’ theorem . . . . . . . . . . . . . . . . . . . 134
9.7 Examples from vector calculus . . . . . . . . . . . . . 137
9.8 Exercises . . . . . . . . . . . . . . . . . . . . . . 139
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Chapter 1
Manifolds in Euclidean space
In Geometry 1 we have dealt with parametrized curves and surfaces in

R2 or R3 . The definitions we have seen for the two notions are analogous to
each other, and we shall begin by generalizing them to arbitrary dimensions.
As a result we obtain the notion of a parametrized m-dimensional manifold
in Rn .
The study of curves and surfaces in Geometry 1 was mainly through
parametrizations. However, as it was explained, important examples of
curves and surfaces arise more naturally as level sets, for example the circle
{(x, y) | x2 + y 2 = 1} and the sphere {(x, y, z) | x2 + y 2 + z 2 = 1}. In order
to deal with such sets, we shall define a notion of manifolds, which applies to
subsets in Rn without the specification of a particular parametrization. The
new notion will take into account the possibility that the given subset of Rn
is not covered by a single parametrization. It is easy to give examples of sub-
sets of R3 that we conceive as surfaces, but whose natural parametrizations
do not cover the entire set (at least if we require the parametrizations to be
regular).
For example, we have seen that for the standard spherical coordinates on
the sphere there are two singular points, the poles. In order to have a regular
parametrization we must exclude these points. A variation of the standard
spherical coordinates with interchanged roles of y and z will have singular
poles in two other points. The entire sphere can thus be covered by spherical
coordinates if we allow two parametrizations covering different, overlapping
subsets of the sphere. Note that in contrast, the standard parametrization
of the circle by trigonometric coordinates is everywhere regular.
1.1 Parametrized manifolds

In the following m and n are arbitrary non-negative integers with m ≤ n.
Definition 1.1.1. A parametrized manifold in Rn is a smooth map σ: U →
Rn , where U ⊂ Rm is a non-empty open set. It is called regular at x ∈
U if the n × m Jacobi matrix Dσ(x) has rank m (that is, it has linearly
independent columns), and it is called regular if this is the case at all x ∈ U .
An m-dimensional parametrized manifold in Rn is a regular parametrized
manifold σ: U → Rn as above with U ⊂ Rm (that is, regularity is implied at
all points when we speak of the dimension).
2 Chapter 1
Clearly, a parametrized manifold with m = 2 and n = 3 is the same

as a parametrized surface, and the notion of regularity is identical to the
one introduced in Geometry 1. For m = 1 there is a slight difference with
the notion of parametrized curves, because in Geometry 1 we have required
a curve γ: I → Rn to be defined on an interval, whereas here we are just
assuming U to be an open set in R. Of course there are open sets in R
which are not intervals, for example the union of two disjoint open intervals.
Notice however, that if γ: U → Rn is a parametrized manifold with U ⊂ R,
then for each t0 ∈ U there exists an open interval I around t0 in U , and the
restriction of γ to that interval is a parametrized curve in the old sense. In
future, when we speak of a parametrized curve, we will just assume that it
is defined on an open set in R.
Perhaps the case m = 0 needs some explanation. By definition R0 is the
trivial vector space {0}, and a map σ: R0 → Rn has just one value p = σ(0).
By definition the map 0 → p is smooth and regular, and thus a 0-dimensional
parametrized manifold in Rn is a point p ∈ Rn .
Example 1.1.1 Let σ(u, v) = (cos u, sin u, cos v, sin v) ∈ R4 . Then
⎛ ⎞
− sin u 0
⎜ cos u 0 ⎟
Dσ(u, v) = ⎝ ⎠
0 − sin v
0 cos v
has rank 2, so that σ is a 2-dimensional parametrized manifold in R4 .
Example 1.1.2 The graph of a smooth function h: U → Rn−m , where
U ⊂ Rm is open, is an m-dimensional parametrized manifold in Rn . Let
σ(x) = (x, h(x)) ∈ Rn , then Dσ(x) is an n × m matrix, of which the first m
rows comprise a unit matrix. It follows that Dσ(x) has rank m for all x, so
that σ is regular.
Many basic results about surfaces allow generalization, often with proofs
analogous to the 2-dimensional case. Below is an example. By definition, a
reparametrization of a parametrized manifold σ: U → Rn is a parametrized
manifold of the form τ = σ ◦ φ where φ: W → U is a diffeomorphism of open
sets.
Theorem 1.1. Let σ: U → Rn be a parametrized manifold with U ⊂ Rm ,
and assume it is regular at p ∈ U . Then there exists a neighborhood of p in U ,
such that the restriction of σ to that neighborhood allows a reparametrization
which is the graph of a smooth function, where n − m among the variables
x1 , . . . , xn are considered as functions of the remaining m variables.
Proof. The proof, which is an application of the inverse function theorem for
functions of m variables, is entirely similar to the proof of the corresponding
result for surfaces (Theorem 2.11 of Geometry 1).
Manifolds in Euclidean space 3
1.2 Embedded parametrizations

We introduce a property of parametrizations, which essentially means that
there are no self intersections. Basically this means that the parametrization
is injective, but we shall see that injectivity alone is not sufficient to en-
sure the behavior we want, and we shall supplement injectivity with another
condition.
Definition 1.2.1. Let A ⊂ Rm and B ⊂ Rn . A map f : A → B which is
continuous, bijective and has a continuous inverse is called a homeomorphism.
The sets A and B are metric spaces, with the same distance functions
as the surrounding Euclidean spaces, and the continuity of f and f −1 is
assumed to be with respect to these metrics.
Definition 1.2.2. An embedded parametrized manifold is a regular parame-
trized manifold σ: U → Rn (with U ⊂ Rm ), such that σ: U → σ(U ) is a
homeomorphism.
In particular this definition applies to curves and surfaces, and thus we
can speak of embedded parametrized curves and embedded parametrized
surfaces.
In addition to being smooth and regular, the condition on σ is thus that
it is injective and that the inverse map σ(x) → x is continuous σ(U ) → U .
Since the latter condition of continuity is important in the following, we shall
elaborate a bit on it.
Definition 1.2.3. Let A ⊂ Rn . A subset B ⊂ A is said to be relatively
open if it has the form B = A ∩ W for some open set W ⊂ Rn .
For example, the interval B = [0; 1[ is relatively open in A = [0, ∞[, since
it has the form A∩W with W =]−1, 1[. As another example, let A = {(x, 0)}
be the x-axis in R2 . A subset B ⊂ A is relatively open if and only if it has
the form U × {0} where U ⊂ R is open (of course, no subsets of the axis
are open in R2 , except the empty set). If A is already open in Rn , then the
relatively open subsets are just the open subsets.
It is easily seen that B ⊂ A is relatively open if and only if it is open in
the metric space of A equipped with the distance function of Rn .
Let us return to the setting in Definition 1.2.2. The continuity of σ(x) → x
from σ(U ) to U means by definition that every open subset V ⊂ U has an
open preimage in σ(U ). The preimage of V by this map is σ(V ), hence the
condition is that V ⊂ U open implies σ(V ) ⊂ σ(U ) open. By the preceding
remark and definition this is equivalent to require that for each open V ⊂ U
there exists an open set W ⊂ Rn such that
σ(V ) = σ(U ) ∩ W. (1.1)

4 Chapter 1
The importance of this condition is illustrated in Example 1.2.2 below.

Notice that a reparametrization τ = σ ◦ φ of an embedded parametrized
manifold is again embedded. Here φ: W → U is a diffeomorphism of open
sets, and it is clear that τ is a homeomorphism onto its image if and only if
σ is a homeomorphism onto its image.
Example 1.2.1 The graph of a smooth function h: U → Rn−m , where
U ⊂ Rm is open, is an embedded parametrized manifold in Rn . It is regular
by Example 1.1.2, and it is clearly injective. The inverse map σ(x) → x is
the restriction to σ(U ) of the projection
Rn
x → (x1 , . . . , xm ) ∈ Rm
on the first m coordinates. Hence this inverse map is continuous. The open
set W in (1.1) can be chosen as W = V × Rn−m .
Example 1.2.2 Consider the parametrized curve γ(t) = (cos t, cos t sin t)
in R2 . It is easily seen to be regular, and it has a self-intersection in (0, 0),
which equals γ(k π2 ) for all odd integers k (see the figure below).
The interval I = ] − π2 , 3 π2 [ contains only one of the values k π2 , and the
restriction of γ to I is an injective regular curve. The image γ(I) is the full
set C in the figure below.
y
C = γ(I)
W
γ(V )
x
The restriction γ|I is not a homeomorphism from I to C. The problem occurs

in the point (0, 0) = γ( π2 ). Consider an open interval V =] π2 − , π2 + [ where
0 < < π. The image γ(V ) is shown in the figure, and it does not have
the form C ∩ W for any open set W ⊂ R2 , because W necessarily contains
points from the other branch through (0, 0). Hence γ|I is not an embedded
parametrized curve.
It is exactly the purpose of the homeomorphism requirement to exclude
the possibility of a ‘hidden’ self-intersection, as in Example 1.2.2. Based on
the example one can easily construct similar examples in higher dimension.
1.3 Curves
As mentioned in the introduction, we shall define a concept of manifolds
which applies to subsets of Rn rather than to parametrizations. In order
to understand the definition properly, we begin by the case of curves in R2 .
The idea is that a subset of R2 is a curve, if in a neighborhood of each of its
points it is the image of an embedded parametrized curve.
C p C ∩ W = γ(I)
W
x
Definition 1.3. A curve in R2 is a non-empty set C ⊂ R2 satisfying the

following for each p ∈ C. There exists an open neighborhood W ⊂ R2 of p,
an open set I ⊂ R, and an embedded parametrized curve γ: I → R2 with
image
γ(I) = C ∩ W. (1.2)
Example 1.3.1. The image C = γ(I) of an embedded parametrized curve

is a curve. In the condition above we can take W = R2 .
Example 1.3.2. A point {p} ⊂ R2 is not a curve, since there exists no
homeomorphism from an open set in R onto a point.
Example 1.3.3. The circle C = S 1 = {(x, y) | x2 + y 2 = 1} is a curve.
In order to verify the condition in Definition 1.3, let p ∈ C be given. For
simplicity we assume that p = (x0 , y0 ) with x0 > 0.
Let W ⊂ R2 be the right half plane {(x, y) | x > 0}, then W is an
open neighborhood of p, and the parametrized curve γ(t) = (cos t, sin t) with
t ∈ I =]− π2 , π2 [ is regular and satisfies (1.2). It is an embedded curve since the
inverse map γ(t) → t is given by (x, y) → tan−1 (y/x), which is continuous.
C p W
γ(I) = C ∩ W
Example 1.3.4. An 8-shaped set like the one in Example 1.2.2 is not
a curve in R2 . In that example we showed that the parametrization by
(cos t, cos t sin t) was not embedded, but of course this does not rule out that
6 Chapter 1
some other parametrization could satisfy the requirement in Definition 1.3.

That this is not the case can be seen from Lemma 1.3 below.
It is of importance to exclude sets like this, because there is not a well
defined tangent line in the point p of self-intersection. If a parametrization
is given, we can distinguish the passages through p, and thus determine a
tangent line for each branch. However, without a chosen parametrization
both branches have to be taken into account, and then there is not a unique
tangent line in p.
The definition of a curve allows the following useful reformulation.

Lemma 1.3. Let C ⊂ R2 be non-empty. Then C is a curve if and only if it
satisfies the following condition for each p ∈ C:
There exists an open neighborhood W ⊂ R2 of p, such that C ∩ W is the
graph of a smooth function h, where one of the variables x1 , x2 is considered
a function of the other variable.
Proof. Assume that C is a curve and let p ∈ C. Let γ: I → R2 be an embedded
parametrized curve satisfying (1.2) and with γ(t0 ) = p. By Theorem 1.1, in
the special case m = 1, we find that there exists a neighborhood V of t0
in I such that γ|V allows a reparametrization as a graph. It follows from
(1.1) and (1.2) that there exists an open set W ⊂ R2 such that γ(V ) =
γ(I) ∩ W = C ∩ W ∩ W . The set W ∩ W has all the properties desired of
W in the lemma.
Conversely, assume that the condition in the lemma holds, for a given
point p say with
C ∩ W = {(t, h(t)) | t ∈ I},
where I ⊂ R is open and h: I → R is smooth. The curve t → (t, h(t)) has
image C ∩W , and according to Example 1.2.1 it is an embedded parametrized
curve. Hence the condition in Definition 1.3 holds, and C is a curve.
The most common examples of plane curves are constructed by means of
the following general theorem, which frees us from finding explicit embedded
parametrizations that satisfy (1.2). For example, the proof in Example 1.3.3,
that the circle is a curve, could have been simplified by means of this theorem.
Recall that a point p ∈ Ω, where Ω ⊂ Rn is open, is called critical for a
differentiable function f : Ω → R if
fx 1 (p) = · · · = fx n (p) = 0.
Theorem 1.3. Let f : Ω → R be a smooth function, where Ω ⊂ R2 is open,

and let c ∈ R. If it is not empty, the set
C = {p ∈ Ω | f (p) = c, p is not critical }

is a curve in R2 .
Proof. By continuity of the partial derivatives, the set of non-critical points in
Ω is an open subset. If we replace Ω by this set, the set C can be expressed as
a level curve {p ∈ Ω | f (p) = c}, to which we can apply the implicit function
theorem (see Geometry 1, Corollary 1.5). It then follows from Lemma 1.3
that C is a curve.
Example 1.3.5. The set C = {(x, y) | x2 + y 2 = c} is a curve in R2 for each

c > 0, since it contains no critical points for f (x, y) = x2 + y 2 .
Example 1.3.6. Let C = {(x, y) | x4 − x2 + y 2 = 0}. It is easily seen that
this is exactly the image γ(I) in Example 1.2.2. The point (0, 0) is the only
critical point in C for the function f (x, y) = x4 − x2 + y 2 , and hence it follows
from Theorem 1.3 that C \ {(0, 0)} is a curve in R2 . Since (0, 0) is critical,
the theorem does not apply to C itself, but as mentioned in Example 1.3.4,
it can be seen from Lemma 1.3 that it is not a curve.
1.4 Surfaces
We proceed in the same fashion as for curves.
Definition 1.4. A surface in R3 is a non-empty set S ⊂ R3 satisfying the
following property for each point p ∈ S. There exists an open neighborhood
W ⊂ R3 of p and an embedded parametrized surface σ: U → R3 with image
σ(U ) = S ∩ W. (1.3)
z
W
S
σ(U ) = S ∩ W
Example 1.4.1. The image S = σ(U ) of an embedded parametrized surface

is a surface in R3 . In the condition above we can take W = R3 .
Example 1.4.2. The 2-sphere S 2 = {(x, y, z) | x2 + y 2 + z 2 = 1} and the
cylinder {(x, y, z) | x2 + y 2 = 1} are surfaces. Rather than providing for
each point p an explicit embedded parametrization satisfying (1.3), we use
the following theorem, which is analogous to Theorem 1.3.
8 Chapter 1
Theorem 1.4. Let f : Ω → R be a smooth function, where Ω ⊂ R3 is open,

and let c ∈ R. If it is not empty, the set
S = {p ∈ Ω | f (p) = c, p is not critical } (1.4)
is a surface in R3 .
Proof. The proof, which combines Geometry 1, Corollary 1.6, with Lemma
1.4 below, is entirely similar to that of Theorem 1.3.
Example 1.4.3. Let us verify for the sphere that it contains no critical
points for the function f (x, y, z) = x2 + y 2 + z 2 . The partial derivatives
are fx = 2x, fy = 2y, fz = 2z, and they vanish simultaneously only at
(x, y, z) = (0, 0, 0). This point does not belong to the sphere, hence it is a
surface. The verification for the cylinder is similar.
Lemma 1.4. Let S ⊂ R3 be non-empty. Then S is a surface if and only if
it satisfies the following condition for each p ∈ S:
There exist an open neighborhood W ⊂ R3 of p, such that S ∩ W is
the graph of a smooth function h, where one of the variables x1 , x2 , x3 is
considered a function of the other two variables.
Proof. The proof is entirely similar to that of Lemma 1.3.
1.5 Chart and atlas

As mentioned in the introduction there exist surfaces, for example the
sphere, which require several, in general overlapping, parametrizations. This
makes the following concepts relevant.
Definition 1.5. Let S be a surface in R3 . A chart on S is an injective
regular parametrized surface σ: U → R3 with image σ(U ) ⊂ S. A collection
of charts σi : Ui → R3 on S is said to cover S if S = ∪i σi (Ui ). In that case
the collection is called an atlas of S.
Example 1.5.1. The image S = σ(U ) of an embedded parametrized surface
as in Example 1.4.1 has an atlas consisting just of the chart σ itself.
Example 1.5.2. The map
σ(u, v) = (cos v, sin v, u), u, v ∈ R
is regular and covers the cylinder S = {(x, y, z) | x2 + y 2 = 1}, but it is not

injective. Let
U1 = {(u, v) ∈ R2 | −π < v < π}, U2 = {(u, v) ∈ R2 | 0 < v < 2π},
and let σi denote the restriction of σ to Ui for i = 1, 2. Then σ1 and σ2 are

both injective, σ1 covers S with the exception of a vertical line on the back
where x = −1, and σ2 covers with the exception of a vertical line on the front
where x = 1. Together they cover the entire set and thus they constitute an
atlas.
z
y
x
Example 1.5.3. The spherical coordinate map
σ(u, v) = (cos u cos v, cos u sin v, sin u), − π2 < u < π

2
, −π < v < π,
and its variation
σ̃(u, v) = (cos u cos v, sin u, cos u sin v), − π2 < u < π

2, 0 < v < 2π,
are charts on the unit sphere. The restrictions on u and v ensure that they
are regular and injective. The chart σ covers the sphere except a half circle
(a meridian) in the xz-plane, on the back where x ≤ 0, and the chart σ̃
similarly covers with the exception of a half circle in the xy-plane, on the
front where x ≥ 0 (half of the ‘equator’). As seen in the following figure the
excepted half-circles are disjoint. Hence the two charts together cover the
full sphere and they constitute an atlas.
z
y
x
Theorem 1.5. Let S be a surface. There exists an atlas of it.

Proof. For each p ∈ S we choose an embedded parametrized surface σ as in
Definition 1.4. Since a homeomorphism is injective, this parametrization is
a chart on S. The collection of all these charts is an atlas.
10 Chapter 1
1.6 Manifolds
We now return to the general situation where m and n are arbitrary
integers with 0 ≤ m ≤ n.
Definition 1.6.1. An m-dimensional manifold in Rn is a non-empty set
S ⊂ Rn satisfying the following property for each point p ∈ S. There exists
an open neighborhood W ⊂ Rn of p and an m-dimensional embedded (see
Definition 1.2.2) parametrized manifold σ: U → Rn with image σ(U ) = S∩W.
The surrounding space Rn is said to be the ambient space of the manifold.
Clearly this generalizes Definitions 1.3 and 1.4, a curve is a 1-dimensional
manifold in R2 and a surface is a 2-dimensional manifold in R3 .
Example 1.6.1 The case m = 0. It was explained in Section 1.1 that a
0-dimensional parametrized manifold is a map R0 = {0} → Rn , whose image
consists of a single point p. An element p in a set S ⊂ Rn is called isolated
if it is the only point from S in some neighborhood of p, and the set S is
called discrete if all its points are isolated. By going over Definition 1.6.1 for
the case m = 0 it is seen that a 0-dimensional manifold in Rn is the same as
a discrete subset.
Example 1.6.2 If we identify Rm with {(x1 , . . . , xm , 0 . . . , 0)} ⊂ Rn , it is
an m-dimensional manifold in Rn .
Example 1.6.3 An open set Ω ⊂ Rn is an n-dimensional manifold in Rn .
Indeed, we can take W = Ω and σ = the identity map in Definition 1.6.1.
Example 1.6.4 Let S ⊂ S be a relatively open subset of an m-dimensional
manifold in Rn . Then S is an m-dimensional manifold in Rn .
The following lemma generalizes Lemmas 1.3 and 1.4.
Lemma 1.6. Let S ⊂ Rn be non-empty. Then S is an m-dimensional
manifold if and only if it satisfies the following condition for each p ∈ S:
There exist an open neighborhood W ⊂ Rn of p, such that S ∩ W is the
graph of a smooth function h, where n − m of the variables x1 , . . . , xn are
considered as functions of the remaining m variables.
Proof. The proof is entirely similar to that of Lemma 1.3.
Theorem 1.6. Let f : Ω → Rk be a smooth function, where k ≤ n and where
Ω ⊂ Rn is open, and let c ∈ Rk . If it is not empty, the set
S = {p ∈ Ω | f (p) = c, rank Df (p) = k}
is an n−k-dimensional manifold in Rn .
Proof. Similar to that of Theorem 1.3 for curves, by means of the implicit
function theorem (Geometry 1, Corollary 1.6) and Lemma 1.6.
A manifold S in Rn which is constructed as in Theorem 1.6 as the set of

solutions to an equation f (x) = c is often called a variety. In particular, if the
equation is algebraic, which means that the coordinates of f are polynomials
in x1 , . . . , xn , then S is called an algebraic variety.
Example 1.6.5 In analogy with Example 1.4.3 we can verify that the m-
sphere
S m = {x ∈ Rm+1 | x21 + · · · + x2m+1 = 1}
is an m-dimensional manifold in Rm+1 .
Example 1.6.6 The set
S = S 1 × S 1 = {x ∈ R4 | x21 + x22 = x23 + x24 = 1}
is a 2-dimensional manifold in R4 . Let

x21 + x22
f (x1 , x2 , x3 , x4 ) = ,
x23 + x24
then
2x1 2x2 0 0
Df (x) =
0 0 2x3 2x4
and it is easily seen that this matrix has rank 2 for all x ∈ S.
Definition 1.6.2. Let S be an m-dimensional manifold in Rn . A chart on

S is an m-dimensional injective regular parametrized manifold σ: U → Rn
with image σ(U ) ⊂ S, and an atlas is a collection of charts σi : Ui → Rn
which cover S, that is, S = ∪i σi (Ui ).
As in Theorem 1.5 it is seen that every manifold in Rn possesses an atlas.
1.7 The coordinate map of a chart
In Definition 1.6.2 we require that σ is injective, but we do not require

that the inverse map is continuous, as in Definition 1.6.1. Surprisingly, it
turns out that the inverse map has an even stronger property, it is smooth
in a certain sense.
Definition 1.7. Let S be a manifold in Rn , and let σ: U → S be a chart.

If p ∈ S we call (x1 , . . . , xm ) ∈ U the coordinates of p with respect to σ when
p = σ(x). The map σ −1 : σ(U ) → U is called the coordinate map of σ.
12 Chapter 1
Theorem 1.7. Let σ: U → Rn be a chart on a manifold S ⊂ Rn , and let

p0 ∈ σ(U ) be given. The coordinate map σ −1 allows a smooth extension,
defined on an open neighborhood of p0 in Rn .
More precisely, let q0 ∈ U with p0 = σ(q0 ). Then there exist open neigh-
borhoods W ⊂ Rn of p0 and V ⊂ U of q0 such that
σ(V ) = S ∩ W, (1.5)
and a smooth map ϕ: W → V such that
ϕ(σ(q)) = q, σ(ϕ(x)) = x (1.6)
for all q ∈ V and all x ∈ S ∩ W .
Proof. Let W ⊂ Rn be an open neighborhood of p0 such that S ∩ W is

parametrized as a graph, as in Lemma 1.6. Say the graph is of the form
σ̃(x1 , . . . , xm ) = (x, h(x)) ∈ Rn ,
where Ũ ⊂ Rm is open and h: Ũ → Rn−m smooth, then S ∩ W = σ̃(Ũ ). Let

π(x1 , . . . , xn ) = (x1 , . . . , xm ), then
σ̃(π(p)) = p
for each point p = (x, h(x)) in S ∩ W .

Since σ is continuous, the subset U1 = σ −1 (W ) of U is open. The map
π ◦ σ: U1 → Ũ is smooth, being composed by smooth maps, and it satisfies
σ̃ ◦ (π ◦ σ) = σ. (1.7)
By the chain rule for smooth maps we have the matrix product equality
D(σ̃)D(π ◦ σ) = Dσ,
and since the n × m matrix Dσ on the right has independent columns, the
determinant of the m × m matrix D(π ◦ σ) must be non-zero in each x ∈ U1
(according to the rule from matrix algebra that rank(AB) ≤ rank(B)).
By the inverse function theorem, there exists open sets V ⊂ U1 and Ṽ ⊂ Ũ
around q0 and π(σ(q0 )), respectively, such that π ◦ σ restricts to a diffeomor-
phism of V onto Ṽ . Note that σ(V ) = σ̃(Ṽ ) by (1.7).
σ(U ) S
σ̃(Ũ )
σ σ̃ π
π◦σ
V Ṽ
U Ũ
Let W̃ = W ∩ π −1 (Ṽ ) ⊂ Rn . This is an open set, and it satisfies
S ∩ W̃ = S ∩ W ∩ π −1 (Ṽ ) = σ̃(Ũ ) ∩ π −1 (Ṽ ) = σ̃(Ṽ ) = σ(V ).
The map ϕ = (π ◦ σ)−1 ◦ π: W̃ → V is smooth and satisfies (1.6).

Corollary 1.7. Let σ: U → S be a chart on a manifold S in Rn . Then σ
is an embedded parametrized manifold, and the image σ(U ) is relatively open
in S.
Proof. For each q0 ∈ U we choose open sets V ⊂ U and W ⊂ Rn , and a
map ϕ: W → V as in Theorem 1.7. The inverse of σ is the restriction of the
smooth map ϕ, hence in particular it is continuous. Furthermore, the union
of all these sets W is open and intersects S exactly in σ(U ). Hence σ(U ) is
relatively open, according to Definition 1.2.3.
It follows from the corollary that every chart on a manifold satisfies the
condition in Definition 1.6.1 of being imbedded with open image. This does
not render that condition superfluous, however. The point is that once it is
known that S is a manifold, then the condition is automatically fulfilled for
all charts on S.
1.8 Transition maps

Since the charts in an atlas of a manifold S in Rn may overlap with each
other, it is important to study the change from one chart to another. The
map σ2−1 ◦ σ1 : x → x̃, which maps a set of coordinates x in a chart σ1 to the
coordinates of the image σ1 (x) with respect to another chart σ2 , is called
14 Chapter 1
the transition map between the charts. We will show that such a change of
coordinates is just a reparametrization.
Let Ω ⊂ Rk be open and let f : Ω → Rn be a smooth map with f (Ω) ⊂ S.
Let σ: U → Rn be a chart on S, then the map σ −1 ◦ f , which is defined on
f −1 (σ(U )) = {x ∈ Ω | f (x) ∈ σ(U )} ⊂ Rk ,
is called the coordinate expression for f with respect to σ.

Lemma 1.8. The set f −1 (σ(U )) is open and the coordinate expression is
smooth from this set into U .
Proof. Since f is continuous into S, and σ(U ) is open by Corollary 1.7, it
follows that the inverse image f −1 (σ(U )) is open. Furthermore, if an element
x0 in this set is given, we let p0 = f (x0 ) and choose V , W and ϕ: W → V
as in Theorem 1.7. It follows that σ −1 ◦ f = ϕ ◦ f in a neighborhood of x0 .
The latter map is smooth, being composed by smooth maps.
Theorem 1.8. Let S be a manifold in Rn , and let σ1 : U1 → Rn and σ2 : U2 →
Rn be charts on S. Let
Vi = σi−1 (σ1 (U1 ) ∩ σ2 (U2 )) ⊂ Ui
for i = 1, 2. These are open sets and the transition map
σ2−1 ◦ σ1 : V1 → V2 ⊂ Rm
is a diffeomorphism.
σ2 (U )
σ1 (U )
S ⊂ Rn
σ1 σ2
U1 ⊂ R m U2 ⊂ R m
V1 σ2−1 ◦ σ1 V2
Proof. Immediate from Lemma 1.8.

1.9 Exercises
1 Consider the parametrized curve
t t2
γ(t) = (− , − )
1 + t3 1 + t3
with t ∈] − ∞, −1[ ∪ ] − 1, ∞[. Verify that γ is a bijection onto the set
C = {(x, y) ∈ R2 | x3 + xy + y 3 = 0}.
2 Let γ be as above, with t ∈ I where −1 ∈ / I ⊂ R. In each of the following

cases I = Ii , prove or disprove that γ is an embedded parametrized curve.
1) I1 =] − ∞, −1[,
2) I2 =] − 1, 0[,
3) I3 =]0, ∞[,
4) I4 =] − 1, ∞[.
3 Let C ⊂ R2 be a curve. Prove that C has no interior points.
4 Verify that the set C in Exercise 1 above is not a curve in R2 .
5 Let c ∈ R \ {0} and put
C = {(x, y) ∈ R2 | x3 + xy + y 3 = c}.
1
Verify that C is a curve in R2 if c = 27 .
1
6 Let c = 27
and let C be as above. Prove or disprove that it is a curve in
R2 .
7 Prove Lemma 1.4.
8 Find an atlas for the cylinder (Example 1.5.2), consisting of a single chart.
9 Let S = {(x, y, z) | z = 0} be the xy-plane, let U = {(u, v) ∈ R2 | u > v},
and let σ : U → R3 be given by σ(u, v) = (u + v, uv, 0). Verify that σ
is a chart on S. After that, replace the condition u > v by u = v in the
definition of U , and prove that σ is no longer a chart.
10 Give an example of a level set S = {(x, y, z) | f (x, y, z) = 0} which is a
surface in R3 , and yet there exists a point in S for which ( ∂f , ∂f , ∂f ) =
∂x ∂y ∂z
(0, 0, 0).
11 Let S = {(x, y, z) | xyz = c} where c ∈ R. Show that S is a surface in
R3 if c = 0. Determine S explicitly when c = 0 and prove that it is not a
surface.
16 Chapter 1
12 Let C ⊂ R2 be a curve in R2 , and put S = C × R = {(x, y, z) | (x, y) ∈ C}.

Prove that S is a surface in R3 (it is called a generalized cylinder).
13 Let S ⊂ R3 . Show that S is a surface if and only if the following holds.
For each p ∈ S there exists an open neighborhood W ⊂ R3 of p and a
smooth function f : W → R such that
(i) S ∩ W = {(x, y, z) ∈ W | f (x, y, z) = 0},
(ii) (fx (p), fy (p), fz (p)) = (0, 0, 0).
Propose (and prove) a similar statement for curves in R2 .
14 Verify that {x | x21 + x22 − x23 − x24 = 1} is a 3-dimensional manifold in R4 .
15 Let M ⊂ Rn be an m-dimensional manifold in Rn , and let N ⊂ M be a
subset. Prove that N is also an m-dimensional manifold in Rn if and only
if it is relatively open in M (see Definition 1.2.3)
16 Consider the set M of points (x, y, z) in R3 which satisfy both equations
x2 + y 2 + z 2 = 1 and x2 = yz 2 .
Show that P = (0, −1, 0) is isolated in M (that is, there is a neighborhood

in R3 , in which P is the only point from M ). Find another point Q in M ,
such that M \ {P, Q} is a manifold in R3 .
17 Let C be a curve in R2 .
a. Let f : Ω → R2 be a smooth map, where Ω ⊂ R3 is open. Let S =
f −1 (C), and assume that Df (p) has rank 2 for all p ∈ S. Prove that S is
a surface in R3 (hint: Apply Exercise 13 above in order to describe C as a
level set (locally). Now consider the composed map).
b. Assume that x > 0 for all (x, y) ∈ C, and let

S = {(x, y, z) | ( x2 + y 2 , z) ∈ C}.
Prove that this is a surface, and describe it in words.

Chapter 2
Abstract manifolds
The notion of a manifold S defined in the preceding chapter assumes S to

be a subset of a Euclidean space Rn . However, a more axiomatic and abstract
approach to differential geometry is possible, and in many ways preferable.
Of course, a manifold in Rn must satisfy the axioms that we set up for an
abstract manifold. Our axioms will be based on properties of charts.
From the point of view of differential geometry the most important prop-
erty of a manifold is that it allows the concept of a smooth function. We
will define this notion and the more general notion of a smooth map between
abstract manifolds.
2.1 Topological spaces

Since the aim of differential geometry is to bring the methods of differential
calculus into geometry, the most important property that we wish an abstract
manifold to have is the possibility of differentiating functions on it. However,
before we can speak of differentiable functions, we must be able to speak of
continuous functions. In this preliminary section we will briefly introduce the
abstract framework for that, the structure of a topological space. Topological
spaces is a topic of general topology, here we will just introduce the most
essential notions. Although the framework is more general, the concepts we
introduce will be familiar to a reader who is acquainted with the theory of
metric spaces.
Definition 2.1.1. A topological space is a non-empty set X equipped with
a distinguished family of subsets, called the open sets, with the following
properties:
1) the empty set and the set X are both open,
2) the intersection of any finite collection of open sets is again open,
3) the union of any collection (finite or infinite) of open sets is again open.
Example 2.1.1 In the Euclidean spaces X = Rk there is a standard notion
of open sets, and the properties in the above axioms are known to hold. Thus
Rk is a topological space.
Example 2.1.2 Let X be a metric space. Again there is a standard notion
of open sets in X, and it is a fundamental result from the theory of metric
spaces that the family of all open sets in X has the properties above. In this
fashion every metric space is a topological space.
18 Chapter 2
Example 2.1.3 Let X be an arbitrary set. If we equip X with the collection

consisting just of the empty set and X, it becomes a topological space. We
say in this case that X has the trivial topology. In general this topology does
not result from a metric, as in Example 2.1.2. The topology on X obtained
from the collection of all subsets, is called the discrete topology. It results
from the discrete metric, by which all non-trivial distances are 1.
The following definitions are generalizations of well-known definitions in
the theory of Euclidean spaces, and more generally, metric spaces.
Definition 2.1.2. A neighborhood of a point x ∈ X is a subset U ⊂ X
with the property that it contains an open set containing x. The interior
of a set A ⊂ X, denoted A◦ , is the set of all points x ∈ A for which A is a
neighborhood of x.
Being the union of all the open subsets of A, the interior A◦ is itself an
open set, according to Definition 2.1.1.
Definition 2.1.3. Let A ⊂ X be a subset. It is said to be closed if its
complement Ac in X is open. In general, the closure of A, denoted Ā, is the
set of all points x ∈ X for which every neighborhood contains a point from
A, and the boundary of A is the set difference ∂A = Ā \ A◦ , which consists
of all points with the property that each neighborhood meets with both A
and Ac .
It is easily seen that the closure Ā is the complement of the interior (Ac )◦
of Ac . Hence it is a closed set. Likewise, the boundary ∂A is closed.
Definition 2.1.4. Let X and Y be topological spaces, and let f : X → Y
be a map. Then f is said to be continuous at a point x ∈ X if for every
neighborhood V of f (x) in Y there exists a neighborhood U of x in X such
that f (U ) ⊂ V , and f is said to be continuous if it is continuous at every
x ∈ X.
Lemma 2.1.1. The map f : X → Y is continuous if and only if the inverse
image of every open set in Y is open in X.
Proof. The proof is straightforward.
Every (non-empty) subset A of a metric space X is in a natural way a
metric space of its own, when equipped with the restriction of the distance
function of X. The open sets in this metric space A are the relatively open
sets, that is, the sets A ∩ W where W is an open subset of X (see Definition
1.2.3). This observation has the following generalization.
Lemma 2.1.2. Let X be a topological space, and let A ⊂ X be non-empty.
Then A is a topological space of its own, when equipped with the collection of
all subsets of A of the form A ∩ O, where O is open in X.
Abstract manifolds 19
Proof. The conditions in Definition 2.1.1 are easily verified.

A subset A of a topological space is always assumed to carry the topology
of Lemma 2.1.2, unless otherwise is mentioned. It is called the induced (or
relative) topology, and the open sets are said to be relatively open.
If A ⊂ X is a subset and f is a map A → Y , then f is said to be continuous
at x ∈ A, if it is continuous with respect to the induced topology. It is easily
seen that if f : X → Y is continuous at x ∈ A, then the restriction f |A : A → Y
is also continuous at x.
Definition 2.1.5. Let X and Y be topological spaces, and let A ⊂ X and
B ⊂ Y . A map f : A → B which is continuous, bijective and has a continuous
inverse is called a homeomorphism (compare Definition 1.2.1).
Finally, we mention the following important property of a topological
space, which is often assumed in order to exclude some rather peculiar topo-
logical spaces.
Definition 2.1.6. A topological space X is said to be Hausdorff if for every
pair of distinct points x, y ∈ X there exist disjoint neighborhoods of x and y.
Every metric space is Hausdorff, because if x and y are distinct points,
then their mutual distance is positive, and the open balls centered at x and
y with radius half of this distance will be disjoint by the triangle inequality.
On the other hand, equipped with the trivial topology (see example 2.1.3),
a set of at least two elements is not a Hausdorff topological space.
2.2 Abstract manifolds

Let M be a Hausdorff topological space, and let m ≥ 0 be a fixed natural
number.
Definition 2.2.1. An m-dimensional smooth atlas of M is a collection
(Oi )i∈I of open sets Oi in M such that M = ∪i∈I Oi , together with a collec-
tion (Ui )i∈I of open sets in Rm and a collection of homeomorphisms, called
charts, σi : Ui → Oi = σi (Ui ), with the following property of smooth transi-
tion on overlaps:
For each pair i, j ∈ I the map σj−1 ◦ σi is smooth from the open set
σi−1 (Oi ∩ Oj ) ⊂ Rm to Rm .
Example 2.2 Let S ⊂ Rn be an m-dimensional manifold in Rn (see Defini-

tion 1.6.1), which we equip with an atlas as in Definition 1.6.2 (as mentioned
below the definition, such an atlas exists). It follows from Corollary 1.7 that
for each chart σ the image O = σ(U ) is open in S and σ: U → O is a home-
omorphism. Furthermore, it follows from Theorem 1.8 that the transition
maps are smooth. Hence this atlas on S is a smooth atlas according to
Definition 2.2.1.
20 Chapter 2
In the preceding example S was equipped with an atlas as in Definition

1.6.2, but one must keep in mind that there is not a unique atlas associated
with a given manifold in Rn . For example, the use of spherical coordinates is
just one of many ways to parametrize the sphere. If we use a different atlas
on S, it is still the same manifold in Rn . In order to treat this phenomenon
abstractly, we introduce an equivalence relation for different atlases on the
same space M .
Definition 2.2.2. Two m-dimensional smooth atlases on M are said to be
compatible, if every chart from one atlas has smooth transition on its overlap
with every chart from the other atlas (or equivalently, if their union is again
an atlas).
It can be seen that compatibility is an equivalence relation. An equiv-
alence class of smooth atlases is called a smooth structure. It follows from
Theorem 1.8 that all atlases (Definition 1.6.2) on a given manifold S in Rn
are compatible. The smooth structure so obtained on S is called the standard
smooth structure.
Definition 2.2.3. A differentiable manifold (or just a manifold) of dimen-
sion m, is a Hausdorff topological space M , equipped with an m-dimensional
smooth atlas. Compatible atlases are regarded as belonging to the same ma-
nifold (the precise definition is thus that a manifold is a Hausdorff topological
space equipped with a smooth structure). A chart on M is a chart from any
atlas compatible with the structure.
It is clear from Example 2.2 that manifolds in Rn are differentiable mani-
folds. To emphasize the contrast we also use the term abstract manifold for
the manifolds in Definition 2.2.3.
It is often required of a differentiable manifold that it should have a count-
able atlas (see Section 2.9). We do not require this here.
2.3 Examples
Example 2.3.1 Let M be an m-dimensional real vector space. Fix a basis
v1 , . . . , vm for M , then the map
σ: (x1 , . . . , xm ) → x1 v1 + · · · + xm vm
is a linear bijection Rm → M . We equip M with the distance function

so that this map is an isometry, then M is a metric space. Furthermore,
the collection consisting just of the map σ, is an atlas. Hence M is an m-
dimensional differentiable manifold.
If another basis w1 , . . . , wm is chosen, the atlas consisting of the map
τ : (x1 , . . . , xm ) → x1 w1 + · · · + xm wm
is compatible with the previous atlas. The transition maps σ −1 ◦ τ and

τ −1 ◦ σ are linear, hence smooth. In other words, the smooth structure of M
is independent of the choice of the basis.
Example 2.3.2 This example generalizes Example 1.6.4. Let M be a
differentiable manifold, and let M be an open subset. For each chart
σi : Ui → Oi ⊂ M , let Oi = M ∩ Oi , this is an open subset of M , and
the collection of all these sets cover M . Furthermore, Ui = σi−1 (Oi ) is open
in Rm , and the restriction σi of σi to this set is a homeomorphism onto its
image. Clearly the transition maps (σj )−1 ◦ σi are smooth, and thus M is a
differentiable manifold with the atlas consisting of all these restricted charts.
Example 2.3.3 Let M = R equipped with the standard metric. Let σ(t) =
t3 for t ∈ U = R, then σ is a homeomorphism U → M . The collection of
this map alone is an atlas on M . The corresponding differential structure on
R is different from the standard differential structure, for the transition map
σ −1 ◦ i between σ and the identity is not smooth at t = 0 (although there
exists a diffeomorphism, see Definition 2.6.1, between M and standard R).
Example 2.3.4 Let X be an arbitrary set, equipped with the discrete topol-
ogy. For each point x ∈ X, we define a map σ: R0 → X by σ(0) = x. The
collection of all these maps is a 0-dimensional smooth atlas on X.
2.4 Projective space

In this section we give an example of an abstract manifold constructed
without a surrounding space Rn .
Let M = RPm be the set of 1-dimensional linear subspaces of Rm+1 . It
is called real projective space, and can be given the structure of an abstract
m-dimensional manifold as follows.
Assume for simplicity that m = 2. Let π: x → [x] = Rx denote the
natural map of R3 \ {0} onto M , and let S ⊂ R3 denote the unit sphere. The
restriction of π to S is two-to-one, for each p ∈ M there are precisely two
elements ±x ∈ S with π(x) = p. We thus have a model for M as the set of
all pairs of antipodal points in S.
We shall equip M as a Hausdorff topological space as follows. A set A ⊂ M
is declared to be open if and only if its preimage π −1 (A) is open in R3 \ {0}
(or equivalently, if π −1 (A) ∩ S is open in S). We say that M has the quotient
topology relative to R3 \{0}. The conditions for a Hausdorff topological space
are easily verified. It follows immediately from Lemma 2.1.1 that the map
π: R3 \ {0} → M is continuous, and that a map f : O → Y , defined on a
subset O ⊂ M , is continuous if and only if f ◦ π is continuous π −1 (O) → Y .
For i = 1, 2, 3 let Oi denote the subset {[x] | xi = 0} in M . It is open
since π −1 (Oi ) = {x | xi = 0} is open in R3 . Let Ui = R2 and let σi : Ui → M
be the map defined by
σ1 (u) = [(1, u1 , u2 )], σ2 (u) = [(u1 , 1, u2 )], σ3 (u) = [(u1 , u2 , 1)]

22 Chapter 2
for u ∈ R2 . It is continuous since it is composed by π and a continuous map

R2 → R3 . Moreover, σi is a bijection of Ui onto Oi , and M = O1 ∪ O2 ∪ O3 .
Theorem 2.4.1. The collection of the three maps σi : Ui → Oi forms a
smooth atlas on M .
Proof. It remains to check the following.
1) σi−1 is continuous Oi → R2 . For example
x2 x3
σ1−1 (p) = ( , )
x1 x1
when p = π(x). The ratios xx21 and xx31 are continuous functions on R3 \ {x1 =
0}, hence σ1−1 ◦ π is continuous.
2) The overlap between σi and σj satisfies smooth transition. For example
1 u2
σ1−1 ◦ σ2 (u) = ( , ),
u1 u1
which is smooth R2 \ {u | u1 = 0} → R2 .
2.5 Product manifolds

If M and N are metric spaces, the Cartesian product M × N is again a
metric space with the distance function
d((m1 , n1 ), (m2 , n2 )) = max(dM (m1 , m2 ), dN (n1 , n2 )).
Likewise, if M and N are Hausdorff topological spaces, then the product

M ×N is a Hausdorff topological space in a natural fashion with the so-called
product topology, in which a subset R ⊂ M × N is open if and only if for each
point (p, q) ∈ R there exist open sets P and Q of M and N respectively, such
that (p, q) ∈ P × Q ⊂ R (the verification that this is a topological space is
quite straightforward).
Example 2.5.1 It is sometimes useful to identify Rm+n with Rm × Rn . In
this identification, the product topology of the standard topologies on Rm
and Rn is the standard topology on Rm+n .
Let M and N be differentiable manifolds of dimensions m and n, respec-
tively. For each chart σ: U → M and each chart τ : V → N we define
σ × τ: U × V → M × N by σ × τ (x, y) = (σ(x), τ (y)).

Theorem 2.5. The collection of the maps σ × τ is an m + n-dimensional

smooth atlas on M × N .
Proof. The proof is straightforward.
We call M × N equipped with the smooth structure given by this atlas for
the product manifold of M and N . The smooth structure on M ×N depends
only on the smooth structures on M and N , not on the chosen atlases.
Notice that if M is a manifold in Rk and N is a manifold in Rl , then
we can regard M × N as a subset of Rk+l in a natural fashion. It is easily
seen that this subset of Rk+l is an m + n-dimensional manifold (according
to Definition 1.6.2), and that its differential structure is the same as that
provided by Theorem 2.5, where the product is regarded as an ‘abstract’ set.
Example 2.5.2 The product S 1 × R is an ‘abstract’ version of the cylinder.
As just remarked, it can be regarded as a subset of R2+1 = R3 , and then it
becomes the usual cylinder.
The product S 1 ×S 1 , which is an ‘abstract’ version of the torus, is naturally
regarded as a manifold in R4 . The usual torus, which is a surface in R3 , is
not identical with this set, but there is a natural bijective correspondence.
2.6 Smooth maps in Euclidean spaces

We shall now define the important notion of a smooth map between man-
ifolds. We first study the case of manifolds in Rn .
Notice that the standard definition of differentiability in a point p of a
map f : Rn → Rl requires f to be defined in an open neighborhood of p in
Rn . Notice also that in order to define the higher derivatives at p, we need f
to be differentiable in a neighborhood of p. Recall that f is said to be smooth
at p if there exists an open neighborhood of p on which f is differentiable to
all orders.
The previous definition does not make sense for a map defined on an m-
dimensional manifold in Rn , because in general a manifold is not an open
subset of Rn .
Definition 2.6.1. Let X ⊂ Rn and Y ⊂ Rl be arbitrary subsets. A map
f : X → Y is said to be smooth at p ∈ X, if there exists an open set W ⊂ Rn
around p and a smooth map F : W → Rl which coincides with f on W ∩ X.
The map f is called smooth if it is smooth at every p ∈ X.
If f is a bijection of X onto Y , and if both f and f −1 are smooth, then f
is called a diffeomorphism.
A smooth map F as above is called a local smooth extension of f . In
order to show that a map defined on a subset of Rn is smooth, one thus
has to find such a local smooth extension near every point in the domain
of definition. If f is smooth at p then f agrees with a continuous function
F in a neighborhood of p, hence f is continuous at p. We observe that the
24 Chapter 2
new notion of smoothness agrees with the previous one when f is defined
on an open set X in Rn . We also observe that the notion of smoothness
is independent of the target set Y , that is, f is smooth into Y if and only
if it is smooth, considered just as a map into Rl . Finally, we note that if
p ∈ X ⊂ X ⊂ Rn and f : X → Rl is smooth at p, then the restriction f |X is
also smooth at p.
Definition 2.6.2. Let S ⊂ Rn and S̃ ⊂ Rl be manifolds. A map f : S → S̃
is called smooth if it is smooth according to Definition 2.6.1 with X = S and
Y = S̃.
In particular, the above definition can be applied with S̃ = R. A smooth
map f : S → R is said to be a smooth function, and the set of these is denoted
C ∞ (S). It is easily seen that C ∞ (S) is a vector space when equipped with
the standard addition and scalar multiplication of functions. We remark that
if F ∈ C ∞ (W ) where W ⊂ Rn is open, then f = F |S ∈ C ∞ (S) if S ⊂ W .
Example 2.6.1 The functions x → xi where i = 1, . . . , n are smooth func-
tions on Rn . Hence they restrict to smooth functions on every manifold
S ⊂ Rn .
Example 2.6.2 Let S ⊂ R2 be the circle {x | x21 + x22 = 1}, and let
Ω = S \ {(−1, 0)}. The function f : Ω → R defined by f (x1 , x2 ) = 1+x x2
1
is a
smooth function, since it is the restriction of the smooth function F : W → R
defined by the same expression for x ∈ W = {x ∈ R2 | x1 = −1}.
Example 2.6.3 Let S ⊂ Rn be an m-dimensional manifold, and let σ: U →
S be a chart. It follows from Theorem 1.7 that σ −1 is smooth σ(U ) → Rm .
Lemma 2.6. Let p ∈ X ⊂ Rn and Y ⊂ Rm . If f : X → Y is smooth at p,
and g: Y → Rl is smooth at f (p), then g ◦ f : X → Rl is smooth at p.
Proof. Let F : W → Rm be a local smooth extension of f around p. Likewise
let G: V → Rl be a local smooth extension of g around f (p) ∈ Y . The set
W = F −1 (V ) is an open neighborhood of p, and G◦(F |W ) is a local smooth
extension of g ◦ f at p.
The definition of smoothness that we have given for manifolds in Rn uses
the ambient space Rn . In order to prepare for the generalization to abstract
manifolds, we shall now give an alternative description.
Theorem 2.6. Let f : S → Rl be a map and let p ∈ S. If f is smooth at p,
then f ◦ σ: U → Rl is smooth at σ −1 (p) for each chart (σ, U ) on S around p.
Conversely, if f ◦ σ is smooth at σ −1 (p) for some chart σ around p, then f
is smooth at p.
Here the smoothness of f ◦ σ refers to the ordinary notion as explained
above Definition 2.6.1.
Proof. The first statement is immediate from Lemma 2.6. For the converse,
assume f ◦ σ is smooth at σ −1 (p) and apply Lemma 2.6 and Example 2.6.3
to f |σ(U) = (f ◦ σ) ◦ σ −1 . It follows that f |σ(U) is smooth at p. Since σ(U )
is relatively open in S (by Corollary 1.7), a local smooth extension of f |σ(U)
is also a local smooth extension of f , hence f is smooth at p.
2.7 Smooth maps between abstract manifolds

Inspired by Theorem 2.6, we can now generalize to abstract manifolds. It
follows from that theorem that the following notion agrees with the previous
one for manifolds in Rn .
Definition 2.7.1. Let M be a differentiable manifold of dimension m. A
map f : M → Rl is called smooth at p ∈ M if for some chart (σ, U ) around
p, the map f ◦ σ: U → Rl is smooth at q = σ −1 (p). The map is called smooth
if it is smooth at every p ∈ M .
Using the smoothness of transitions it is seen that once f ◦ σ is smooth
at σ −1 (p) for some chart around p, then this will be the case for every chart
around p. Notice that if f : M → Rl is smooth at p then it is continuous at p,
since in a neighborhood it can be written as the composition (f ◦σ)◦σ −1, and
σ −1 is continuous by definition. The set of all smooth functions f : M → R
is denoted C ∞ (M ). This is a vector space when equipped with the standard
addition and scalar multiplication of functions.
Notice also that if Ω ⊂ M is open, then Ω is a differentiable manifold of
its own (see Example 2.3.2), and hence the preceding definitions make sense
also for functions f : Ω → Rl .
Example 2.7.1 Let σ: U → M be a chart on a differentiable manifold M .
Since σ −1 ◦ σ is obviously smooth, it follows that σ −1 is smooth σ(U ) → Rm .
We have defined what it means for a map from a manifold to be smooth,
and we shall now define what smoothness means for a map into a manifold.
As before we begin by considering manifolds in Euclidean space. Let S
and S̃ be manifolds in Rn and Rl , respectively, and let f : S → S̃. It was
defined in Definition 2.6.2 what it means for f to be smooth. We will give
an alternative description.
Let σ: U → S and σ̃: Ũ → S̃ be charts on S and S̃, respectively, where
U ⊂ Rm and Ũ ⊂ Rk are open sets. For a map f : S → S̃, we call the map
σ̃ −1 ◦ f ◦ σ: x → σ̃ −1 (f (σ(x))), (2.1)
the coordinate expression for f with respect to the charts (see the following
figure). It is defined for all x ∈ U for which f (σ(x)) ∈ σ̃(Ũ ), that is, it is
defined on the set
σ −1 (f −1 (σ̃(Ũ ))) ⊂ U, (2.2)
and it maps into Ũ .
26 Chapter 2
Theorem 2.7. Let f : S → S̃ be a map and let p ∈ S. Let σ and σ̃ be

charts on S and S̃, respectively, around p and f (p). Then f is smooth at
p (according to Definition 2.6.2) if and only if p is an interior point of the
subset f −1 (σ̃(Ũ )) of S, and σ̃ −1 ◦ f ◦ σ is smooth at σ −1 (p).
p f f (p)
S ⊂ Rn S̃ ⊂ Rl
σ σ̃
U ⊂ Rm Ũ ⊂ Rk
σ̃ −1 ◦ f ◦ σ
Note that the statement that p is interior is of importance, since the set
(2.2), on which σ̃ −1 ◦f ◦σ is defined, then contains an open set around σ −1 (p),
so that the usual notion of smoothness can be applied to this map at this
point.
Proof. Assume that f is smooth at p. It was remarked below Definition 2.6.1
that then it is continuous at p. Since σ̃(Ũ ) is open in S̃ by Corollary 1.7, it
follows that f maps a neighborhood of p in S into σ̃(Ũ ), hence p is interior
as stated.
It follows from Theorem 2.6 that f ◦ σ is smooth at q = σ −1 (p), and from
Lemma 2.6 and Example 2.6.3 that the composed map σ̃ −1 ◦ f ◦ σ is smooth
at q.
For the converse we just have to note that the identity
f ◦ σ = σ̃ ◦ (σ̃ −1 ◦ f ◦ σ),
shows that f ◦ σ is smooth at q. This implies that f is smooth at p, according

to Theorem 2.6.
By using the formulation of smoothness in Theorem 2.7, we can now gener-
alize the notion. Let M and M̃ be differentiable manifolds, and let f : M → M̃
be a map. Assume σ: U → σ(U ) ⊂ M and σ̃: Ũ → σ̃(Ũ ) ⊂ M̃ are charts on
the two manifolds, then as before we call the map
σ̃ −1 ◦ f ◦ σ,
which is defined on (2.2), the coordinate expression for f with respect to the
given charts.
Definition 2.7.2. Let f : M → M̃ be a map between abstract manifolds,

and let p ∈ M . Then f is called smooth at p if there exists a chart σ: U → M
around p, and a chart σ̃: Ũ → M̃ around f (p), such that p is an interior point
of f −1 (σ̃(Ũ )) ⊂ M and such that the coordinate expression σ̃ −1 ◦ f ◦ σ is
smooth at σ −1 (p). The map is called smooth if it is smooth at every p ∈ M .
A bijective map f : M → M̃ , is called a diffeomorphism if f and f −1 are
both smooth.
Again we notice that if f : M → M̃ is smooth at p, then it is continuous

at p. This follows immediately from the definition above, by writing f =
σ̃ ◦ (σ̃ −1 ◦ f ◦ σ) ◦ σ −1 in a neighborhood of p.
It follows from Theorem 2.7 that the notion of smoothness is the same as
before if M = S ⊂ Rn and M̃ = S̃ ⊂ Rl . Likewise, there is no conflict with
Definition 2.7.1 in case M is abstract and M̃ = Rl , since in that case we can
use the identity map for σ̃: Rl → Rl .
It is easily seen that the composition of two smooth maps between differ-
entiable manifolds is again smooth.
Example 2.7.2 Let M, N be finite dimensional vector spaces of dimension

m and n, respectively. These are differentiable manifolds, according to Ex-
ample 2.3.1. Let f : M → N be a linear map. If we choose a basis for each
space and define the corresponding charts as in Example 2.3.1, then the co-
ordinate expression for f is a linear map from Rm to Rn (given by the matrix
that represents f ), hence smooth. It follows from Definition 2.7.2 that f is
smooth. If f is bijective, its inverse is also linear, and hence in that case f
is a diffeomorphism.
Example 2.7.3 Let σ: U → M be a chart on an m-dimensional differentiable

manifold M . It follows from the assumption of smooth transition on overlaps
that σ is smooth U → M , if we regard U as an m-dimensional manifold (with
the identity chart). By combining this observation with Example 2.7.1, we
see that σ is a diffeomorphism of U onto its image σ(U ) (which is open in
M by Definition 2.2.1).
Conversely, every diffeomorphism g of a non-empty open subset V ⊂ Rm
onto an open subset in M is a chart on M . Indeed, by the definition of
a chart given at the end of Section 2.2, this means that g should overlap
smoothly with all charts σ in an atlas of M , that is g −1 ◦ σ and σ −1 ◦ g
should both be smooth (on the sets where they are defined). This follows
from the preceding observation about compositions of smooth maps.
In particular, it follows that if f : M → M̃ is a diffeomorphism and σ is a
chart on M , then f ◦ σ is a chart on M̃ .
28 Chapter 2
2.8 Lie groups
Definition 2.8.1. A Lie group is a group G, which is also a differentiable

manifold, such that the group operations
(x, y) → xy, x → x−1 ,
are smooth maps from G × G, respectively G, into G.
Example 2.8.1 Every finite dimensional real vector space V is a group,

with the addition of vectors as the operation and with neutral element 0.
The map (x, y) → x + y is linear V × V → V , hence it is smooth (see
Example 2.7.2). Likewise x → −x is smooth and hence V is a Lie group.
Example 2.8.2 The set R× of non-zero real numbers is a 1-dimensional
Lie group, with multiplication as the operation and with neutral element
1. Likewise the set C× of non-zero complex numbers is a 2-dimensional Lie
group, with complex multiplication as the operation. The smooth structure
is determined by the chart (x1 , x2 ) → x1 + ix2 . The product
xy = (x1 + ix2 )(y1 + iy2 ) = x1 y1 − x2 y2 + i(x1 y2 + x2 y1 )
is a smooth function of the entries, and so is the inverse

x1 − ix2
x−1 = (x1 + ix2 )−1 = .
x21 + x22
Example 2.8.3 Let G = SO(2), the group of all 2 × 2 real matrices which
are orthogonal, that is, they satisfy the relation AAt = I, and which have
determinant 1. The set G is in one-to-one correspondence with the unit circle
in R2 by the map
x1 x2
(x1 , x2 ) → .
−x2 x1
If we give G the smooth structure so that this map is a diffeomorphism,
then it becomes a 1-dimensional Lie group, called the circle group. The
multiplication of matrices is given by a smooth expression in x1 and x2 , and
so is the inversion x → x−1 , which only amounts to a change of sign on x2 .
Example 2.8.4 Let G = GL(n, R), the set of all invertible n × n matrices.
It is a group, with matrix multiplication as the operation. It is a manifold
in the following fashion. The set M(n, R) of all real n × n matrices is in
2
bijective correspondence with Rn and is therefore a manifold of dimension
n2 . The subset G = {A ∈ M(n, R) | det A = 0} is an open subset, because
the determinant function is continuous. Hence G is a manifold.
Furthermore, the matrix multiplication M(n, R) × M(n, R) → M(n, R) is
given by smooth expressions in the entries (involving products and sums),
hence it is a smooth map. It follows that the restriction to G × G is also

smooth.
Finally, the map x → x−1 is smooth G → G, because according to
Cramer’s formula the entries of the inverse x−1 are given by rational func-
tions in the entries of x (with the determinant in the denominator). It follows
that G = GL(n, R) is a Lie group.
Example 2.8.5 Let G be an arbitrary group, equipped with the discrete
topology (see Example 2.3.4). It is a 0-dimensional Lie group.
Theorem 2.8. Let G ⊂ GL(n, R) be a subgroup which is also a manifold in
2
M(n, R) Rn . Then G is a Lie group.
Proof. It has to be shown that the multiplication is smooth G × G → G.
According to Definition 2.6.1 we need to find local smooth extensions of the
multiplication map and the inversion map. This is provided by Example
2
2.8.4, since GL(n, R) is open in Rn .
Example 2.8.6 The group SL(2, R) of 2 × 2-matrices of determinant 1 is a
3-dimensional manifold in R4 , since
Theorem
1.6 can be applied with f equal
to the determinant function det c d = ad − bc. Hence it is a Lie group.
a b
2.9 Countable atlas

The definition of an atlas of a differentiable manifold leaves no limitation
on the size of the family of charts. Sometimes it is useful that the atlas is
not too large. In this section we introduce an assumption of this nature. In
particular, we shall see that all manifolds in Rn satisfy the assumption.
Recall that a set A is said to be countable if it is finite or in one-to-one
correspondence with N.
Definition 2.9. An atlas of a differentiable manifold M is said to be
countable if the set of charts in the atlas is countable.
In a topological space X, a base for the topology is a collection of open
sets V with the property that for every point x ∈ X and every neighborhood
U of x there exists a member V in the collection, such that x ∈ V ⊂ U . For
example, in a metric space the collection of all open balls is a base for the
topology.
Example 2.9 As just mentioned, the collection of all open balls B(y, r) is a
base for the topology of Rn . In fact, the subcollection of all balls, for which
the radius as well as the coordinates of the center are rational, is already a
base for the topology. For if x ∈ Rn and a neighborhood U is given, there
exists a rational number r > 0 such that B(x, r) ⊂ U . By density of the
rationals, there exists a point y ∈ B(x, r/2) with rational coordinates, and
by the triangle inequality x ∈ B(y, r/2) ⊂ B(x, r) ⊂ U . We conclude that in
30 Chapter 2
Rn there is a countable base for the topology. The same is then true for any
subset X ⊂ Rn , since the collection of intersections with X of elements from
a base for Rn , is a base for the topology in X.
Lemma 2.9. Let M be a differentiable manifold. Then M has a countable
atlas if and only if there exists a countable base for the topology.
A topological space, for which there exists a countable base, is said to be
second countable.
Proof. Assume that M has a countable atlas. For each chart σ: U → M in
the atlas there is a countable base for the topology of U , according to the
example above, and since σ is a homeomorphism it carries this to a countable
base for σ(U ). The collection of all these sets for all the charts in the atlas,
is then a countable base for the topology of M , since a countable union of
countable sets is again countable.
Assume conversely that there is a countable base (Vk )k∈I for the topology.
For each k ∈ I, we select a chart σ: U → M for which Vk ⊂ σ(U ), if such
a chart exists. The collection of selected charts is clearly countable. It
covers M , for if x ∈ M is arbitrary, there exists a chart σ (not necessarily
among the selected) around x, and there exists a member Vk in the base with
x ∈ Vk ⊂ σ(U ). This member Vk is contained in a chart, hence also in a
selected chart, and hence so is x. Hence the collection of selected charts is a
countable atlas.
Corollary 2.9. Let S be a manifold in Rn . There exists a countable atlas
for S.
Proof. According to Example 2.9 there is a countable base for the topol-
ogy.
In Example 2.3.4 we introduced a 0-dimensional smooth structure on an
arbitrary set X, with the discrete topology. Any basis for the topology must
contain all singleton sets in X. Hence if the set X is not countable, there
does not exist any countable atlas for this manifold.
2.10 Whitney’s theorem

The following is a famous theorem, due to Whitney. The proof is too
difficult to be given here. In Section 5.6 we shall prove a weaker version of
the theorem.
Theorem 2.10. Let M be an abstract manifold of dimension m, and assume
there exists a countable atlas for M . Then there exists a diffeomorphism of
M onto a manifold in R2m .
For example, it can be shown that the projective space RP2 is diffeomor-
phic with a 2-dimensional manifold in R4 . Notice that the assumption of a
countable atlas cannot be avoided, due to to Corollary 2.9.
The theorem could give one the impression that if we limit our interest
to manifolds with a countable atlas, then the notion of abstract manifolds
is superfluous. This is not so, because in many circumstances it would be
very inconvenient to be forced to perceive a particular smooth manifold as a
subset of some high-dimensional Rk . The abstract notion frees us from this
limitation.
1.11 Exercises
1 Prove Lemma 2.1.1.

2 Let X be a finite set. Prove that it is a Hausdorff topological space in
one and only one way. If X has 2 elements, in how many ways is it a
topological space?
3 Equip R with the family of all intervals ]a, ∞[ where a ∈ R, including also
∅ and R. Verify that it is a topological space, but not Hausdorff.
4 Let X, Y be topological spaces, and let f, g: X → Y be continuous maps.
Prove that if Y is Hausdorff, then {x ∈ X | f (x) = g(x)} is closed.
Give an example which shows the conclusion can fail if Y is not Hausdorff
(for example, Y could be a two element set with the trivial topology).
5 Verify that compatibility of atlasses is an equivalence relation (see Defini-
tion 2.2.3). Prove that the union of all the atlasses in a given equivalence
class is an atlas (called the maximal atlas for the structure).
6 Prove that if A ⊂ S 2 is open, then π(A) is open in RP 2 .
7 The natural injection of S 1 into the equator of S 2 induces a map f : RP 1 →
RP 2 . Prove that the image is closed in RP 2 , and that f is a homeomor-
phism onto it.
8 a. Let X be a topological space with an equivalence relation ∼, and let
M = X/ ∼ denote the set of equivalence classes, with π: X → M the
natural map. A set A ⊂ M is said to be open if and only if the preimage
π −1 (A) is open in X. Verify that M is a topological space (called the
quotient space). – Unfortunately, it need not be Hausdorff, even if X is.
b. Define a relation on X = R2 by (s, t) ∼ (s , t ) if and only if s − s is

an integer and t = (−1)s −s t. Verify it is an equivalence relation. Let M
be the quotient space. Prove that it is Hausdorff.
c. For each a ∈ R let Ua =]a, a + 1[×R ⊂ R2 and let σa : Ua → M be the
restriction of the map π from above to this set. Prove that the collection
of all these maps is an atlas on M .
9 Let CP 1 denote the set of 1-dimensional (complex) subspaces of the com-
plex vector space C2 . Turn it into an abstract 2-dimensional manifold.
32 Chapter 2
10 Verify that if M is a manifold in Rk and N is a manifold in Rl , then M ×N

is a manifold in Rk+l , and its smooth structure is identical to that of the
product manifold.
11 Let U ⊂ Rm be an open set, and let h1 : U → Rk1 and h2 : U → Rk2 be
smooth maps. Let S1 ⊂ Rm+k1 and S1 ⊂ Rm+k2 denote the graphs of the
two maps. Prove that the map (x, h1 (x)) → (x, h2 (x)) is a diffeomorphism
of S1 onto S2 .
12 Let S 2 denote the unit sphere in R3 , and define a map f : S 2 → S 2 by
f (x, y, z) = (x cos(z) + y sin(z), x sin(z) − y cos(z), z)
Prove that f is a diffeomorphism.

13 Let F : R3 → R3 be given by
F (x, y, z) = (xy, yz, zx),
and define f : RP 2 → R3 by f (±(x, y, z)) = F (x, y, z) for (x, y, z) ∈ S 2 .

Prove that f is smooth, but not injective.
14 Verify that the set of 2 × 2 upper triangular matrices with non-zero deter-
minant is a Lie group (with matrix multiplication as the product).
15 Let f1 : M1 → M̃1 and f2 : M2 → M̃2 be smooth maps. Verify that f1 × f2
is smooth M1 × M2 → M̃1 × M̃2 . Let G1 , G2 be Lie groups. Verify that
the product G1 × G2 is a Lie group.
Chapter 3
The tangent space

In this chapter the tangent space for a differentiable manifold is defined.
Let us recall from Geometry 1 that for a parametrized surface σ: U → R3
we defined the tangent space at point x ∈ U as the linear subspace in R3
spanned by the derived vectors σu and σv at x. In this chapter we shall
generalize the concept to abstract manifolds. We first consider manifolds in
Rn . Here we can still define the tangent space with reference to the ambient
space Rn . For abstract manifolds we give an axiomatic definition, which
reflects the essential properties from the previous case. The main problem is
then to prove existence of the tangent space.
3.1 The tangent space of a parametrized manifold

Let σ: U → Rn be a parametrized manifold, where U ⊂ Rm is open.
Definition 3.1. The tangent space Tx0 σ of σ at x0 ∈ U is the linear subspace
of Rn spanned by the columns of the n × m matrix Dσ(x0 ).
Assume that σ is regular at x0 . Then the the columns of Dσ(x0 ) are
linearly independent, and they form the so-called standard basis for the space
Tx0 σ. If v ∈ Rm then Dσ(x0 )v is the linear combination of these basis vectors
with the coordinates of v as coefficients. Hence the map defined by
dσx0 : v → Dσ(x0 )v (3.1)
is a linear isomorphism of R onto the m-dimensional tangent space Tx0 σ.
m
Observe that the tangent space is a ‘local’ object, in the sense that if
two parametrized manifolds σ: U → Rn and σ : U → Rn are equal on some
neighborhood of a point x0 ∈ U ∩ U , then Tx0 σ = Tx0 σ .
From Geometry 1 we recall the following result, which is easily generalized
to the present setting.
Theorem 3.1. The tangent space is invariant under reparametrization. In
other words, if φ: W → U is a diffeomorphism of open sets in Rm and τ =
σ ◦ φ, then Ty0 τ = Tφ(y0 ) σ for all y0 ∈ W .
Proof. The proof is essentially the same as in Geometry 1, Theorem 2.7.
By the chain rule, the matrix identity Dτ (y0 ) = Dσ(φ(y0 ))Dφ(y0 ) holds.
Equivalently,
dτy0 (v) = dσφ(y0 ) (Dφ(y0 )v) (3.2)
for all v ∈ Rm . Hence Ty0 τ ⊂ Tφ(y0 ) σ. The opposite inclusion follows by the
same argument with φ replaced by its inverse.
34 Chapter 3
3.2 The tangent space of a manifold in Rn

Let S ⊂ Rn be an m-dimensional manifold (see Definition 1.6.1).
Definition 3.2. The tangent space of S at a point p ∈ S is the m-dimensional
linear space
Tp S = Tx0 σ ⊂ Rn , (3.3)
where σ: U → S is an arbitrary chart on S with p = σ(x0 ) for some x0 ∈ U .
Notice that it follows from Theorem 3.1 that the tangent space Tp S =
Tx0 σ is actually independent of the chart σ, since any other chart around p
will be a reparametrization, according to Theorem 1.8 (more precisely, the
reparametrization takes place on the overlap of the two charts). Observe also
the change of notation, in that the foot point p of Tp S belongs to S ⊂ Rn in
contrast to the previous foot point x0 ∈ Rm of Tx0 σ.
With the generalization to abstract manifolds in mind we shall give a
different characterization of the tangent space Tp S, which is also closely
related to a property from Geometry 1 (Theorem 2.4).
We call a parametrized smooth curve γ: I → Rn a parametrized curve on
S if the image γ(I) is contained in S.
Theorem 3.2. Let p ∈ S. The tangent space Tp S is the set of all vectors in
Rn , which are tangent vectors γ (t0 ) to a parametrized curve on S, γ: I → S,
with γ(t0 ) = p for some t0 ∈ I.
Proof. Let σ: U → S be a chart with p = σ(x0 ) for some x0 ∈ U . Then
Tp S = Tx0 σ, and dσx0 (see (3.1)) is a linear isomorphism of Rm onto Tp S.
Given a parametrized curve γ: I → Rn on S with γ(t0 ) = p, we define a
parametrized curve μ in Rm by μ = σ −1 ◦ γ for t in a neighborhood of t0 .
This is the coordinate expression for γ with respect to σ. It follows from
Lemma 1.8 that μ is smooth, and since γ = σ ◦ μ, it follows from the chain
rule that
γ (t0 ) = Dσ(x0 )μ (t0 ) = dσx0 (μ (t0 )) ∈ Tp S. (3.4)
To obtain the opposite inclusion we define for each v ∈ Rm a parametrized
curve on S by
γv (t) = σ(x0 + tv)
for t close to 0 (so that x0 + tv ∈ U ). As before it follows from the chain rule
that
γv (0) = dσx0 (v). (3.5)
Since dσx0 maps onto Tp S, we conclude that every vector in this space is
obtained as the tangent vector γv (0) for some v ∈ Rm .
We have described the tangent space at p as the space of tangent vectors
to curves through p on S. However, it should be emphasized that the corre-
spondence between curves and tangent vectors is not one-to-one. In general
many different curves through p give rise to the same tangent vector.
The tangent space 35
Example 3.2.1 Let Ω ⊂ Rn be an open subset, then Ω is an n-dimensional

manifold in Rn , with the identity map as a chart (see Example 1.6.3). The
tangent space is Rn , and the map (3.1) is the identity map.
Example 3.2.2 Suppose S ⊂ Rn is the n − k-dimensional manifold given
by an equation f (p) = c as in Theorem 1.6, where in addition it is required
that Df (p) has rank k for all p ∈ S. Then we claim that the tangent space
Tp S is the null space for the matrix Df (p), that is,
Tp S = {v ∈ Rn | Df (p)v = 0}.
Here we recall that by a fundamental theorem of linear algebra, the dimension

of the null space is n − k when the rank of Df (p) is k.
The claim can be established as follows by means of Theorem 3.2. If v
is a tangent vector, then v = γ (t0 ) for some parametrized curve on S with
p = γ(t0 ). Since γ maps into S, the function f ◦ γ is constant (with the value
c), hence (f ◦ γ) (t0 ) = 0. It follows from the chain rule that Df (p)v = 0.
Thus the tangent space is contained in the null space. Since both are linear
spaces of dimension n − k, they must be equal.
Example 3.2.3 Let C denote the ∞-shaped set in Examples 1.2.2 and 1.3.6.
The parametrized curve γ(t) = (cos t, cos t sin t) is a parametrized curve on
C, and it passes through (0, 0) for each t = k π2 with k an odd integer. If C was
a curve in R2 , then γ (k π2 ) would belong to T(0,0) C, according to Theorem
3.2. But the vectors γ ( π2 ) = (−1, −1) and γ (3 π2 ) = (1, −1) are linearly
independent, and hence T(0,0) C cannot be a 1-dimensional linear space. Thus
we reach a contradiction, and we conclude (again) that C is not a curve.
3.3 The abstract tangent space

Consider an abstract m-dimensional manifold M , and let σ: U → M be
a chart. It does not make sense to repeat Definitions 3.1 and 3.2, because
the n × m-matrix Dσ(x0 ) is not defined. Instead, we shall give an abstract
definition, which also takes the map (3.1) and the relation (3.2) into account.
Definition 3.3. Let p ∈ M . A tangent space of M at p is an m-dimensional
real vector space T together with a rule, which assigns in a compatible way
to each chart (σ, U ) on M around p a linear isomorphism
dσx0 : Rm → T (3.6)
called the differential of σ at x0 = σ −1 (p) ∈ U . The requested compatibility

is that for all pairs of charts around p, we have
dτy0 (v) = dσx0 (D(σ −1 ◦ τ )(y0 )v) (3.7)
for all v ∈ Rm . Here D(σ −1 ◦ τ )(y0 ) is the Jacobian matrix of σ −1 ◦ τ at

y0 = τ −1 (p).
36 Chapter 3
We see that for a manifold S ⊂ Rn , the space Tp S is a tangent space,

according to this abstract definition. The linear isomorphisms dσx0 are given
by (3.1), and the compatibility (3.7) follows from (3.2). We can thus regard
the differential in (3.6) as an abstract version of the map (3.1).
However, we need to address some important issues before this abstract
definition can be of any use. These are the issues of existence and uniqueness.
The uniqueness will be dealt with immediately. The existence is shown in
the following sections (see Theorem 3.5).
Lemma 3.3. Let T1 and T2 be tangent spaces of M at p, with associated
isomorphisms
di σx0 : Rm → Ti , (i = 1, 2).
There exists a unique linear isomorphism ϕ: T1 → T2 such that
ϕ ◦ d1 σx0 = d2 σx0 (3.8)
for all charts σ around p.

Proof. Select an arbitrary chart σ around p and define ϕ: T1 → T2 by (3.8),
that is, by ϕ = d2 σx0 ◦ (d1 σx0 )−1 . This is a linear isomorphism, and it is
clearly the only candidate for ϕ when (3.8) is required. If we prove that ϕ is
independent of σ then (3.8) follows for all other charts.
Let τ be another chart around p. Since the Jacobian matrix in (3.7) only
depends on τ and σ (and not on T1 and T2 ), these compatibility relations
imply
(d1 σx0 )−1 ◦ d1 τy0 = (d2 σx0 )−1 ◦ d2 τy0 .
It follows that d2 τy0 ◦ (d1 τy0 )−1 = d2 σx0 ◦ (d1 σx0 )−1 .
It follows that a tangent space is unique up to isomorphism. We illustrate
the use of the tangent space with the following analogue of Theorem 3.2.
We define a parametrized curve on M to be a smooth map γ: I → M ,
where I ⊂ R is open. By definition (see Definition 2.7.2), this means that
σ −1 ◦ γ is smooth for all charts on M . As before, if σ is a chart on M , we
call μ = σ −1 ◦ γ the coordinate expression of γ with respect to σ. If p ∈ M
is a point, a parametrized curve on M through p, is a parametrized curve on
M together with a point t0 ∈ I for which p = γ(t0 ).
Theorem 3.3. Let γ be a parametrized curve on M through p, and let (σ, U )
be a chart around p, say with p = σ(x0 ) = γ(t0 ). Let μ denote the coordinate
expression of γ.
If T is a tangent space of M at p, then the vector γ (t0 ) ∈ T , defined by
γ (t0 ) = dσx0 (μ (t0 )) (3.9)
and called the tangent vector of γ at t0 , is independent of the chart σ.

Every vector in T occurs in this fashion as the tangent vector for some
parametrized curve γ on M through p.
For the case M = S, compare the definition in (3.9) with (3.4).
Proof. If τ is another chart around p and ν = τ −1 ◦ γ the corresponding
coordinate expression, then
ν = τ −1 ◦ γ = (τ −1 ◦ σ) ◦ (σ −1 ◦ γ) = (τ −1 ◦ σ) ◦ μ
on the overlap. The chain rule implies
ν (t0 ) = D(τ −1 ◦ σ)(x0 )μ (t0 ). (3.10)
It now follows from (3.7) that dτy0 (ν (t0 )) = dσx0 (μ (t0 )).
For the last statement we essentially just repeat from the proof of Theorem
3.2. For each v ∈ Rm let γv (t) = σ(x0 + tv) for t close to 0. This is a
parametrized curve on M through p. In fact, in this case μ(t) = x0 + tv,
hence μ (0) = v and γv (0) = dσx0 (v). It follows that v → γv (0) is exactly
our isomorphism dσx0 of Rm onto T .
3.4 Equivalence of curves

We shall use Theorem 3.3 as motivation for the existence proof of the
tangent space.
We define the following relation between curves through a fixed point
p ∈ M . Let γ1 : I1 → M and γ2 : I2 → M be parametrized curves on M
through p, say with p = γ1 (t1 ) = γ2 (t2 ), and let σ be a chart around p, say
p = σ(x0 ) and with coordinate expressions μi = σ −1 ◦ γi for the two curves.
We call γ1 and γ2 tangential at p, if the coordinate expressions satisfy
μ1 (t1 ) = μ2 (t2 ). (3.11)
Lemma 3.4. Being tangential at p is an equivalence relation on curves on

M through p. It is independent of the chosen chart σ.
Proof. The first statement is easily seen. If τ is another chart with coordinate
expressions νi = τ −1 ◦ γi for the two curves, then it follows from (3.10) that
νi (ti ) = D(τ −1 ◦σ)(x0 ) μi (ti ) for each curve. Hence the relation (3.11) implies
the similar relation for ν1 and ν2 .
We write the equivalence relation as γ1 ∼p γ2 , and we interprete it as an
abstract equality between the ‘tangent vectors’ (which have not been defined)
of the two curves at p.
Definition 3.4. The set of ∼p -classes of parametrized curves γ through p on
M is denoted by Tp M . The equivalence class of γ is denoted by [γ]p ∈ Tp M .
38 Chapter 3
In the next section we prove that Tp M can be equipped with a structure

as a vector space, and with maps dσx0 : Rm → Tp M so that it becomes a
tangent space to M at p in the sence of Definition 3.3. For this reason Tp M
will be called the tangent space of M at p, and the equivalence class [γ]p will
be denoted γ (t0 ) and called the tangent vector to γ at p.
The following observation is of importance, because it shows that Tp M is
a ‘local’ object, that is, it only depends on the structure of M in a vicinity
of p. If M ⊂ M is an open subset, then M is a differentiable manifold
in itself, according to Example 2.3.2. Obviously a parametrized curve on
M can also be regarded as a parametrized curve on M , and the notion of
two curves being tangential at a point p ∈ M is independent of how we
regard the curves. It follows that there is a natural inclusion of Tp M in
Tp M . Conversely, a parametrized curve γ: I → M through p is tangential
at p to its own restriction γ|I where I = γ −1 (M ), and the latter is also a
parametrized curve on M . It follows that Tp M = Tp M .
3.5 The vector space structure

The elements of Tp M are classes of curves, and they are not easily seen as
vectors for any reasonable vector space structure. We will remedy this and
at the same time prove that Tp M is a tangent space.
Theorem 3.5. Let M be an abstract manifold and let p ∈ M .
(i) Let σ: U → M be a chart on M with p = σ(x0 ). For each element
v ∈ Rm let γv (t) = σ(x0 + tv) for t close to 0. The map dσx0 : Rm → Tp M
defined by v → [γv ]p is a bijection.
(ii) There exists a unique structure on Tp M as a vector space, such that
dσx0 is a linear isomorphism for every chart σ around p.
(iii) Equipped with this structure and these isomorphisms, the space Tp M
becomes a tangent space to M at p (see Definition 3.3).
Proof. (i) Notice that σ −1 ◦ γv (t) = x0 + tv, and (σ −1 ◦ γv ) (0) = v. It follows
that if γv1 ∼p γv2 then v1 = v2 . Hence dσx0 is injective. To see that it is
surjective, let an equivalence class [γ]p ∈ Tp M be given, say with γ(t0 ) = p.
Let μ = σ −1 ◦ γ be the coordinate expression with respect to σ, and put
v = μ (t0 ). Then γv ∼p γ. Hence dσx0 (v) = [γ]p , and the map is surjective.
(ii) By means of the bijection dσx0 : v → [γv ]p we can equip Tp M as an
m-dimensional vector space. We simply select some chart σ and impose the
unique structure on Tp M so that dσx0 is an isomorphism for this chart. In
other words, we define addition by the requirement that dσx0 (v) +dσx0 (w) =
dσx0 (v + w) for all v, w ∈ Rm , and likewise for scalar multiplication.
Since the linear structure has been defined with reference to a selected
chart σ, the linarity of dτy0 for all other charts around p must be shown. We
will prove that the relation (3.7) holds, that is,
dτy0 (w) = dσx0 (D(σ −1 ◦ τ )(y0 )w) (3.12)
for all w ∈ Rm . Since multiplication by D(σ −1 ◦ τ )(y0 ) is a linear map, it

follows from (3.12) that dτy0 is linear.
We now prove (3.12). For w ∈ Rm we consider the parametrized curve
δw (t) = τ (y0 + tw) on M . By definition, dτy0 (w) is the equivalence class
[δw ]p of δw . Let v = D(σ −1 ◦ τ )(y0 )w, then the claim in (3.12) amounts
to [δw ]p = [γv ]p . We rewrite δw in σ-coordinates as δw (t) = σ(μ(t)), where
μ(t) = σ −1 ◦ τ (y0 + tw). Then by the chain rule
μ (0) = D(σ −1 ◦ τ )(y0 )w = v.
It follows that the curves γ1 = δw and γ2 (t) = γv obey the relation (3.11).
Hence γ1 ∼p γ2 and (3.12) follows.
(iii) This was established in the course of (ii).
We have introduced a vector structure on Tp M , but if M = S where S ⊂
R then Tp S already has such a structure, inherited from the surrounding Rn .
n
However, it follows from Lemma 3.3 that the two structures are isomorphic.
Example 3.5 Let V be an m-dimensional real vector space regarded as an
abstract manifold as in Example 2.3.1. We will show that for each element
p ∈ V there exists a natural isomorphism L: V → Tp V . For each vector
r ∈ V we define L(r) to be the ∼p -class of the line p + tr with direction r,
regarded as a curve through p. In order to prove that L is an isomorphism of
vector spaces, we choose a basis v1 , . . . , vm for V , and let σ: Rm → V denote
the corresponding isomorphism (x1 , . . . , xm ) → x1 v1 + · · · + xm vm , which
constitutes a chart on V (see Example 2.3.1). For x ∈ Rm and r = σ(x) we
have, with the notation of Theorem 3.5,
p + tr = p + tσ(x) = σ(x0 + tx) = γx (t)
and hence dσx0 (x) = [γx ]p = L(r). Since dσx0 is an isomorphism of vector
spaces Rm → Tp V (by Theorem 3.5(ii)), and since x → r is an isomorphism
Rm → V , we conclude that r → L(r) is an isomorphism V → Tp V .
3.6 Directional derivatives

An important property of tangent vectors is that they can be brought to
act on smooth functions as a kind of differentiation operators.
We recall the following concept from multivariable calculus. For a function
f : Ω → R on an open set Ω ⊂ Rn , which is differentiable at a point p ∈ Ω,
we define the directional derivative in direction X ∈ Rn as the derivative at
0 of the function t → f (p + tX), that is, we differentiate f along the line
through p with direction X. We denote the directional derivative as follows:
d
Dp,X (f ) = f (p + tX) ∈ R. (3.13)
dt t=0
40 Chapter 3
The directional derivatives in the directions of the canonical basis vectors

e1 , . . . , en are the partial derivatives
∂f
Dp,ei = (p),
∂xi
the entries of the 1 × n-matrix Df (p). The general directional derivative is
related to the partial derivatives through the formula, obtained from applying
the chain rule to f (p + tX),
n
∂f
Dp,X (f ) = Df (p)X = ai (p), (3.14)
i=1
∂xi
where X = (a1 , . . . , an ) is regarded as a column. In particular, we see that
DX (f ) depends linearly on both X and f .
The following expression, easily derived from the chain rule, will be useful
later. Let g: Rm → Rn be differentiable with g(q) = p, then the directional
derivative of the composed function f ◦ g is
Dq,X (f ◦ g) = Dp,Dg(q)X (f ), X ∈ Rm . (3.15)
3.7 Action on functions

We shall now extend the concept of directional derivatives to abstract
manifolds. Let M be an abstract manifold, and let Tp M be the tangent
space at some point p ∈ M .
Lemma 3.7.1. Let f : M → R be smooth at p, and let Z ∈ Tp M . Let (σ, U )
be a chart around p, and let x0 = σ −1 (p) ∈ U and X = dσx−1
0
(Z) ∈ Rm . The
number
Dx0 ,X (f ◦ σ) ∈ R
is independent of the chart σ.
Proof. If τ is another chart, say with p = τ (y0 ) and Z = dτy0 (Y ), then by
writing f ◦ τ = f ◦ σ ◦ (σ −1 ◦ τ ) in a neighborhood of y0 , we see from (3.15)
that
Dy0 ,Y (f ◦ τ ) = Dx0 ,D(σ−1 ◦τ )(y0 )Y (f ◦ σ) = Dx0 ,X (f ◦ σ),
where (3.7) was used in the last step.
Definition 3.7. Let f , Z, σ, x0 and X be as in Lemma 3.7.1. The

directional derivative of f at p in direction Z is
Dp,Z (f ) = Dx0 ,X (f ◦ σ).
According to lemma 3.7.1, Dp,Z is independent of σ. In particular, it
follows that for an open set M ⊂ Rn the definitions of Dp,Z f , by (3.13)
and by Definition 3.7, agree. Notice also that Dp,Z f is linear, with respect
to both Z (use Theorem 3.5(ii)) and f . If we use the model for Tp M from
Theorem 3.5, that it consists of equivalence classes of curves through p, we
can describe the directional derivative alternatively as follows.
Theorem 3.7. Let Z ∈ Tp M . The directional derivative of f ∈ C ∞ (M )

satisfies
Dp,Z (f ) = (f ◦ γ) (t0 ),
for all parametrized curves γ on M with p = γ(t0 ) and Z = γ (t0 ).
Proof. Fix a chart σ around p, and let μ be the coordinate expression for γ
with respect to σ. Then since γ (t0 ) = Z, we have by definition (see Theorem
3.3) that Z = dσx0 (X) where p = σ(x0 ) and X = μ (t0 ). Since f ◦γ = f ◦σ◦μ
we obtain from the chain rule
(f ◦ γ) (t0 ) = D(f ◦ σ)(x0 )μ (t0 ) = Dx0 ,X (f ◦ σ) = Dp,Z (f ),
where (3.14) was used in the second step.

If M = S, a manifold in Rn , we can describe the directional derivative in
yet another way, which invokes the surrounding space.
Lemma 3.7.2. Let f : S → R be smooth at p, and let F : W → R be a local
smooth extension. Let Z ∈ Tp S be a tangent vector, which we regard as an
element in Rn . Then
Dp,Z f = Dp,Z F, (3.16)
where the directional derivative on the right hand side is defined as in (3.13).
Proof. Let γ be a parametrized curve on S with p = γ(t0 ) and Z = γ (t0 ).
Then Dp,Z f = (f ◦ γ) (t0 ) by Theorem 3.7. Since γ maps into S we have
f ◦γ = F ◦γ in a neighborhood of t0 , and hence (f ◦γ) (t0 ) = (F ◦γ) (t0 ). On
the other hand, if we regard Rn as an n-dimensional differentiable manifold
with the identity chart, and γ as a parametrized curve on it, then it also
follows from Theorem 3.7 that Dp,Z F = (F ◦ γ) (t0 ).
Example 3.7 Let S be the unit sphere in R3 , let p = (0, 0, 1) ∈ S and

let X ∈ Tp S denote the tangent vector X = (1, 3, 0). Consider the function
f (x, y, z) = xz on S. The directional derivative is
∂ ∂
DX f = [ + 3 ](xz)(p) = 1.
∂x ∂y
This follows from Lemma 3.7.2, since F (x, y, z) = xz, (x, y, z) ∈ R3 , is a

smooth extension of f .
3.8 The differential of a smooth map

Let M , N be differentiable manifolds of dimension m and n, respectively,
and let p ∈ M . Let f : M → N be smooth. The differential, also called
tangent map, of f at p will now be defined. We shall define it as a linear map
42 Chapter 3
from Tp M to Tf (p) N . It serves as a generalization of the Jacobian n × m-

matrix Df (p) of a map f : Rm → Rn (which, in turn, generalizes the ordinary
derivative f of f : R → R).
If M and N are open subsets of Rm and Rn , then by definition the dif-
ferential dfp is the linear map Rm → Rn which has Df (p) as its matrix with
respect to the canonical bases. Observe that the notation is in accordance
with (3.1), where it is used for f = σ, a parametrized surface. The differential
satisfies
f (p + h) − f (p) = dfp (h) + o(h),
which means that
|f (p + h) − f (p) − dfp (h)|
→0
|h|
as h → 0. The interpretation is that the linear map dfp is an approximation
to the increment of f at p.
Notice that if m = 1 and M ⊂ R is open, then the linear map dfp : R → Rn
is completely determined by its value dfp (1) ∈ Rn . This value is exactly the
derivative f (p) ∈ Rn , which constitutes the n × 1-matrix Df (p). This is for
example the case when f = γ is a curve in N .
The important chain rule is valid for the differential of a composed map
f ◦ g, where g: L → M and f : M → N are smooth maps between open sets
L ⊂ Rl , M ⊂ Rm and N ⊂ Rn . It reads
d(f ◦ g)p = dfg(p) ◦ dgp , (3.17)
and it is an immediate consequence of the corresponding rule for the matrices.

Assume now that M and N are arbitrary differentiable manifolds, and let
p∈M
Lemma 3.8.1. Let f : M → N be a smooth map. There exists a unique
linear map dfp : Tp M → Tf (p) N such that for all charts σ and η around p and
f (p),
(dηu0 )−1 ◦ dfp ◦ dσx0 = d(η −1 ◦ f ◦ σ)x0 : Rm → Rn (3.18)
where σ(x0 ) = p and η(u0 ) = f (p).
Proof. We select two charts σ and η, and use (3.18) to define dfp . We need
to verify that it is independent of the choices of σ and η.
Suppose then that two other charts τ and ζ on M and N are given, with
τ (y0 ) = p and ζ(v0 ) = f (p). Then from (3.14) we read
dτy0 = dσx0 ◦ d(σ −1 ◦ τ )y0 and dηu0 = dζv0 ◦ d(ζ −1 ◦ η)u0 .
The desired identity of dfp with the expression (3.18) for τ and ζ now follows
by applying the chain rule (3.17) to
ζ −1 ◦ f ◦ τ = (ζ −1 ◦ η) ◦ (η −1 ◦ f ◦ σ) ◦ (σ −1 ◦ τ ).
Definition 3.8. The map dfp : Tp M → Tf (p) N in Lemma 3.8.1 is called the
differential of f .
The chain rule for manifolds reads as follows. If g: L → M and f : M → N
are smooth maps between differentiable manifolds, then
d(f ◦ g)p = dfg(p) ◦ dgp . (3.19)
It is an easy consequence of (3.17), applied to the coordinate expressions of

f and g from (3.18).
If we use the model for the tangent space from Theorem 3.5, we can de-
scribe the differential alternatively as follows. Note that if γ is a parametrized
curve on M through p, then f ◦ γ is a parametrized curve on N through f (p).
Theorem 3.8. The differential dfp of f at p satisfies
dfp (γ (t0 )) = (f ◦ γ) (t0 ) (3.20)
for all parametrized curves γ on M through p = γ(t0 ).

Proof. Choose charts σ and η as above, and let μ = σ −1 ◦ γ denote the
coordinate expression of γ with respect to σ. Then λ = (η −1 ◦ f ◦ σ) ◦ μ is the
coordinate expression of f ◦ γ with respect to η. Now γ (t0 ) = dσx0 (μ (t0 ))
and (f ◦ γ) (t0 ) = dηu0 (λ (t0 )). The chain rule implies λ (t0 ) = d(η −1 ◦ f ◦
σ)x0 (μ (t0 )). By (3.18)
dfp (γ (t0 )) = dηu0 (d(η −1 ◦ f ◦ σ)x0 (μ (t0 ))) = dηu0 (λ (t0 )),
so that (3.20) follows.

Example 3.8.1 We have seen in Example 2.7.3 that a chart σ is a smooth
map U → M . We will determine its differential dσx0 , which is a smooth map
from Rm (the tangent space of U at x0 ) to Tp M . Let v ∈ Rm . Then v is the
tangent vector at 0 to the curve x0 + tv in U , and hence by Lemma lemij
dσx0 (v) is the tangent vector at 0 to the curve σ(x0 +tv). This curve is exactly
the curve γv , and hence dσx0 (v) = [γv ]p . We conclude that the differential
dσx0 is exactly the isomorphism that was denoted such in Theorem 3.5.
Example 3.8.2 Let f : M → R be a smooth function on M . The tangent
space Ty R of R at any point y ∈ R is identified with R (see Example 3.2.1).
The differential dfp is therefore a linear map Tp M → R. By comparing the
expression (3.20) with Theorem 3.7 we see that dfp (X) = Dp,X f .
Assume that S and T are manifolds in Rk and Rl , respectively, and let
f : S → T be smooth. Then we can describe the differential dfp in yet another
way, which invokes the surrounding spaces. Recall (see Definition 2.6.1) that
f has a local smooth extension F : W → Rl in a neighborhood of each point
p ∈ S.
44 Chapter 3
Lemma 3.8.2. Let F : W → Rl be a local smooth extension of F at p, and

let dFp : Rk → Rl be its differential at p. Then dFp (Tp S) ⊂ Tf (p) T and dfp
equals the restriction of dFp to Tp S.
Proof. Let γ be a parametrized curve on M through p, then dfp (γ (t0 )) =

(f ◦ γ) (t0 ) by Theorem 3.8. Similarly, by regarding γ as a curve in the
surrounding space, dFp (γ (t0 )) = (F ◦ γ) (t0 ). Since f ◦ γ = F ◦ γ, the lemma
follows.
The differential of the identity map M → M is the identity map id: Tp M →

Tp M . Hence it follows from the chain rule that if f : M → N is a diffeomor-
phism, then
d(f −1 )f (p) ◦ dfp = d(f −1 ◦ f )p = id .
From this relation and the similar one with opposite order of f and f −1 , we
see that when f is a diffeomorphism, then dfp is bijective with (dfp )−1 =
d(f −1 )f (p) . We thus obtain that the differential of a diffeomorphism is a
linear isomorphism between tangent spaces. In particular, two manifolds
between which there exists a diffeomorphism, must have the same dimension.
Example 3.8.3 Let f : V → W be a linear map between finite dimensional

vector spaces, as in Example 2.7.2. In Example 3.5 we have identified the
tangent space Tp V of a vector space V with V itself. By applying this
identification for both V and W , we shall see that the differential at p ∈ V
of a linear map f : V → W is the map f itself. An element v ∈ V = Tp V is
the tangent vector to the curve t → p + tv at t = 0. This curve is mapped
to t → f (p + tv) = f (p) + tf (v) by f , and the tangent vector at t = 0 of the
latter curve on W is exactly f (v). Hence dfp (v) = f (v) as claimed.
In Section 3.7 an action of tangent vectors by directional derivations was

defined. We will now determine the differential in this picture.
Lemma 3.8.3. Let X ∈ Tp M and let f : M → N be smooth. Then
Df (p),dfp (X) (ϕ) = Dp,X (ϕ ◦ f )
for all ϕ ∈ C ∞ (N ).
Proof. Let γ be a representative of X. Then f ◦ γ is a representative of

dfp (X) and hence by Theorems 3.7 and 3.8
Ddfp (X) (ϕ) = (ϕ ◦ f ◦ γ) (t0 ) = DX (ϕ ◦ f ).

3.9 The standard basis

Let M be a differentiable manifold. Given a chart σ with p = σ(x0 ), we
obtain a basis for Tp M from Theorem 3.5, by taking the isomorphic image
dσx0 (ei ) of each of the standard basis vectors e1 , . . . , em for Rm (in that
order), that is, the basis vectors will be the equivalence classes of the curves
t → σ(x0 + tei ). This basis for Tp M is called the standard basis with respect
to σ:
dσx0 (e1 ), . . . , dσx0 (em ).
These basis vectors, which depend on the chosen chart σ, can be seen as
abstract analogs of the basis vectors
σx 1 , σx 2
for the tangent space of a parametrized surface in R3 .

The standard basis vectors determine the following operators D1 , . . . , Dm
of directional differentiation. Let Di = Dp,dσx0 (ei ) . Then by Definition 3.7
∂
Di f = (f ◦ σ)(x0 ). (3.21)
∂xi
∂
Because of (3.21) it is customary to denote Di by ∂xi
.
3.10 Orientation
Let V be a finite dimensional vector space. Two ordered bases (v1 , . . . , vn )
and (ṽ1 , . . . , ṽn ) are said to be equally oriented if the transition matrix S,
whose columns are the coordinates of the vectors v1 , . . . , vn with respect
to the basis (ṽ1 , . . . , ṽn ), has positive determinant. Being equally oriented
is an equivalence relation among bases, for which there are precisely two
equivalence classes. The space V is said to be oriented if a specific class has
been chosen, this class is then called the orientation of V , and its member
bases are called positive. The Euclidean spaces Rn are usually oriented by the
class containing the standard basis (e1 , . . . , en ). For the null space V = {0}
we introduce the convention that an orientation is a choice between the signs
+ and −.
Example 3.10.1 For a two-dimensional subspace V of R3 it is customary
to assign an orientation by choosing a normal vector N . The positive bases
(v1 , v2 ) for V are those for which (v1 , v2 , N ) is a positive basis for R3 (in
other words, it is a right-handed triple).
Let σ be a chart on a differentiable manifold M , then the tangent space
is equipped with the standard basis (see Section 3.8) with respect to σ. For
each p ∈ σ(U ) we say that the orientation of Tp M , for which the standard
basis is positive, is the orientation induced by σ.
46 Chapter 3
Definition 3.10. An orientation of a differentiable manifold M is an

orientation of each tangent space Tp M , p ∈ M , such that there exists an
atlas of M in which all charts induce the given orientation on each tangent
space.
The manifold is called orientable if there exists an orientation. If an ori-
entation has been chosen we say that M is an oriented manifold and we call
a chart positive if it induces the proper orientation on each tangent space.
A diffeomorphism map f : M → N between oriented manifolds of equal di-
mension is said to be orientation preserving if for each p ∈ M , the differential
dfp maps positive bases for Tp M to positive bases for Tf (p) N .
Example 3.10.2 Every manifold M , of which there exists an atlas with
only one chart, is orientable. The orientation induced by that chart is of
course an orientation of M .
Example 3.10.3 Suppose S ⊂ Rn is the n − k-dimensional manifold given
by an equation f (p) = c as in Theorem 1.6. As seen in Example 3.2.2 the
tangent space at p ∈ S is the null space
Tp S = {X ∈ Rn | Df (p)X = 0}.
for the k × n matrix Df (p). Let w1 , . . . , wk ∈ Rn denote the rows of Df (p),
in the same order as they appear in the matrix. We can then define an
orientation of Tp S by declaring an ordered basis (v1 , . . . , vn−k ) positive if the
combined basis (v1 , . . . , vn−k , w1 , . . . , wk ) for Rn is positive with respect to
the standard order. It can be shown that this is an orientation of S, which
is thus orientable.
For a surface in R3 given by f (x) = c, where f is a scalar valued function,
this means that we define the orientation by means of the normal given by
the gradient vector w = ( ∂f , ∂f , ∂f ) of f .
∂x ∂y ∂z
Not all manifolds are orientable however, the most famous example being
the two-dimensional Möbius band. A model for it in R3 can be made by
gluing together the ends of a strip which has been given a half twist. It is
impossible to make a consistent choice of orientations, because the band is
‘one-sided’. Choosing an orientation in one point forces it by continuity to
be given in neighboring points, etcetera, and eventually we are forced to the
opposite choice in the initial point.
3.11 Exercises
1 Let
M = {(x, y, z) ∈ R3 | x2 + y 2 + z 2 = 1 and x2 = yz 2 }
and put C = M \ {(0, ±1, 0)} (see also Exercise 1.16).
a. Determine Tp C where p = (0, 0, 1).
b. Prove that C ∪ {(0, 1, 0)} is not a curve in R3 .
2 Let S = {(x, y, z) | x2 + y 2 = 1}, p = (1, 0, 0) ∈ S and X = e2 = (0, 1, 1).

a. Verify that X ∈ Tp S.
b. Let f (x, y, z) = y − z. Determine the directional derivative LX f at p.
c. Same as before, but now with f (cos θ, sin θ, z) = z − θ.
d. Determine a smooth function f on S for which LX f = 1.
3 Let M and N be differentiable manifolds, and let p ∈ M , q ∈ N be given.
Establish a natural isomorphism between the vector spaces Tp M × Tq N
and T(p,q) (M × N ). Verify that the isomorphism amounts to the identity
map, when M ⊂ Rk , N ⊂ Rl and Rk × Rl is identified with Rk+l .
4 Let π: R3 \{0} → RP 2 be the projection map. Show that there is a natural
isomorphism of vector spaces between Tp S 2 and Tπ(p) RP 2 for each p ∈ S 2 .
5 Consider the following curves on RP 2 :
α(t) = π(cos t, sin t, 0),

β(t) = π(− cos t, 0, sin t),
γ(t) = π(1, t, t).
At t = 0 they all pass through the same point p. Determine a non-trivial

linear relation between their equivalence classes at this point (we know
that such a relation exists, since the tangent space is 2-dimensional).
6 Let M be a differentiable manifold, and M an open subset (see Example
2.3.2). Show:
a. There is a natural linear isomorphism of Tx M onto Tx M for each
x ∈ M , and it is the differential of the inclusion map M → M .
b. If f : M → N is a smooth map into a differentiable manifold, then the
restriction of f to M is also smooth, and it has the same differential as f
at every x ∈ M .
c. If f : N → M is a map from a differentiable manifold into M with
f (N ) ⊂ M , then f is smooth as a map into M if and only if it is smooth
into M , and the two maps have the same differential at each y ∈ N .
7 Let M, N be differentiable manifolds.
a. Let μ: M ×N → M and ν: M ×N → N be the projection maps (p, q) →
p and (p, q) → q. Verify that the isomorphism T(p,q) M × N → Tp M × Tq N
constructed in Exercise 3 is Z → (dμ(p,q) (Z), dν(p,q)(Z)).
b. For a given element q ∈ N let φq : M → M × N be defined by φq (x) =
(x, q) for x ∈ M . Define ψ p : N → M × N similarly by ψ p (y) = (p, y) for
p ∈ M . Verify that these maps are smooth, and that the inverse of the
above mentioned isomorphism is (X, Y ) → dφqp (X) + dψqp (Y ).
48 Chapter 3
8 Let M, N be differentiable manifolds, and let f : M → N be smooth. Prove

the following inverse function theorem for manifolds.
a. Let p ∈ M be given, and assume that the differential dfp : Tp M →
Tf (p) N is an isomorphism of vector spaces (hence, in particular dim M =
dim N ). Then there exist open sets U and V around p and f (p), respec-
tively, such that f restricts to a diffeomorphism U → V .
b. Assume that dfp is bijective for all p ∈ M . Then f (M ) is open in N ,
hence a manifold. If in addition f is injective, then it is a diffeomorphism
of M onto this manifold f (M ).
9 Let Ω ⊂ S 2 be an open hemisphere. Verify that the restriction to Ω of the
projection π: S 2 → RP 2 is a diffeomorphism onto its image.
10 Let G be a group with neutral element e, which at the same time is a
differentiable manifold, such that the multiplication map m: G × G → G
is smooth.
a. Show that left multiplication, lg : x → gx and right multiplication
rg : x → xg are diffeomorphisms of G to itself, for all g ∈ G.
b. Show that when T(x,y) G×G is identified with Tx G×Ty G as above, then
the differential dm(e,e) : T(e,e) G × G → Te G is the map (X, Y ) → X + Y
(Hint: consider the cases Y = 0 and X = 0 separately, and apply linearity)
c. Show that the map (x, y) → (xy, y) is smooth and bijective G × G →
G × G, and determine the inverse Φ: G × G → G × G.
d. Show that Φ is smooth in a neighborhood of (e, e), and conclude that
the inversion i: g → g −1 is smooth in a neighborhood of e.
e. Prove that the inversion i is everywhere smooth (hint: consider rg ◦i◦lg ).
Conclude: The assumption for x → x−1 is superfluous in the definition of
a Lie group.
Chapter 4
Submanifolds
The notion of a submanifold of a differentiable manifold will now be de-

fined. In fact, there exist two different notions of submanifolds, ‘embed-
ded submanifolds’ and ‘immersed submanifolds’. Here we shall focus on the
stronger notion of embedded submanifolds, and for simplicity we will just
use the term ‘submanifold’.
4.1 Submanifolds in Rk
Definition 4.1. Let S be a manifold in Rk . A submanifold of S is a subset

T ⊂ S, which is also a manifold in Rk .
Lemma 4.1. Let T be a submanifold of S ⊂ Rk . Then dim T ≤ dim S and
Tp T ⊂ Tp S for all p ∈ T . Moreover, the inclusion map i: T → S is smooth,
and its differential dip at p ∈ T is the inclusion map Tp T → Tp S.
Proof. The inclusion map i: T → S is the restriction of the identity map
id: Rk → Rk , hence it is smooth, and by Lemma 3.8.2 its differential at p is
the restriction of the differential of id. The differential of the identity map is
the identity map, so we conclude that dip : Tp T → Tp S is the identity map. In
particular, then Tp T ⊂ Tp S. The inequality of dimensions is an immediate
consequence.
Example 4.1 The circle {(x, y, 0) | x2 + y 2 = 1} is a one-dimensional

manifold in R3 (for example by Theorem 1.6 with f (x, y, z) = (x2 + y 2 , z)
and c = (1, 0)). It is a submanifold (the equator) of the 2-sphere {(x, y, z) |
x2 + y 2 + z 2 = 1}.
4.2 Abstract submanifolds

Let M be an abstract manifold. A naive generalization of the definition
above would be that a submanifold of M is subset of M , which is an abstract
manifold of its own. This however, would not be a feasible definition, because
it does not ensure any compatibility between the differential structures of M
and the subset. In Definition 4.1 the compatibility, which is reflected in
Lemma 4.1, results from the position of both manifolds inside an ambient
space Rk . For the general definition we use that lemma as inspiration.
Recall from Lemma 2.1.2 that every subset of a topological space is equip-
ped as a topological space with the induced topology.
50 Chapter 4
Definition 4.2. Let M be an abstract manifold. An (abstract) submanifold

of M is a subset N ⊂ M which is an abstract manifold on its own such that:
(i) the topology of N is induced from M ,
(ii) the inclusion map i: N → M is smooth, and
(iii) the differential dip : Tp N → Tp M is injective for each p ∈ N .
In this case, the manifold M is said to be ambient to N . In particular,
since dip is injective, the dimension of N must be smaller than or equal to
that of M . By Definition 3.8, the differential dip maps the equivalence class
of a curve γ through p on N to the equivalence class of the same curve,
now regarded as a curve on the ambient manifold (formally the new curve is
i ◦ γ). Based on the assumption in (iii) that this map is injective, we shall
often regard the tangent space Tp N as a subspace of Tp M .
Example 4.2.1 Let M be an m-dimensional real vector space, regarded as
an abstract manifold (see Example 2.3.1). Every linear subspace N ⊂ M is
then an abstract submanifold. The inclusion map N → M is linear, hence it
is smooth and has an injective differential in each point p ∈ N (see Examples
2.7.2 and 3.8.3).
Example 4.2.2 It follows directly from Lemma 4.1 that a submanifold T
of a manifold S in Rk is also an abstract submanifold of S, when S and T
are regarded as abstract manifolds.
Example 4.2.3 A non-empty open subset of an m-dimensional abstract
manifold M is a submanifold. Indeed, as mentioned in Example 2.3.2 the
subset is an abstract manifold of its own, also of dimension m. The conditions
(i)-(iii) are easily verified in this case. Conversely, it follows by application of
the inverse function theorem to the inclusion map i, that every m-dimensional
abstract submanifold of M is an open subset of M .
Example 4.2.4 Let M = R2 with the standard differential structure, and
let N ⊂ M be the x-axis, equipped with standard topology together with
the non-standard differential structure given by the chart τ (s) = (s3 , 0) (see
Example 2.3.3). The inclusion map i is smooth, since i ◦ τ : s → (s3 , 0)
is smooth into R2 . Its differential at s = 0 is 0, hence (iii) fails, and N
(equipped with τ ) is not a submanifold of M .
Notice that the property of being a submanifold is transitive, that is, if N
is a submanifold of M , and M is a submanifold of L, then N is a submanifold
of L. This follows from application of the chain rule to the composed inclusion
map.
The following lemma deals with the special case where the ambient mani-
fold is M = Rk .
Lemma 4.2. Let N ⊂ Rk be a subset, equipped with the induced topology.
Then N is a submanifold of Rk in the abstract sense if and only if it is
Submanifolds 51
a manifold in Rk . In this case the smooth structure of N as an abstract

submanifold agrees with that as a manifold in Rk .
Proof. The implication ‘if’ follows from Example 4.2.2 with S = Rk . For the
converse we assume that N carries the structure of an abstract submanifold
of Rk , and we have to verify the properties in Definition 1.6.1, and that the
smooth structure is the same.
Let σ: U → N be a chart (in the sense of an abstract manifold). We claim
that σ is an embedded parametrized manifold with an image which is open
in N . Since N is covered by such charts, all the mentioned properties will
follow from this claim.
It is part of the axioms for abstract manifolds that σ is a homeomorphism
onto an open subset of N . Since it is required in (i) that N carries the induced
topology, this means that σ(U ) = N ∩ W as required in Definition 1.6.1.
We can regard σ as a map into Rk , by composing it with the inclusion map
N → Rk . The latter map is smooth by assumption (ii), hence σ: U → Rk is
smooth.
It remains to be seen that σ is regular, which means that the columns
of the Jacobian matrix Dσ(p) are linearly independent. Since the differen-
tial dσp : Rn → Rk is the linear map represented by Dσ(p), an equivalent
requirement is that this map is injective.
By the chain rule the differential of σ: U → Rk is composed by the differ-
ential of σ: U → N and the differential of the inclusion map. The differential
of σ: U → N is an isomorphism according to Example 3.8.1, and the differ-
ential dip of N → Rk is injective by assumption (iii). Hence the differential
of σ: U → Rk is injective.
The following corollary ensures that we do not have a conflict of defini-
tions. Because of this we usually drop the word ‘abstract’ and just speak of
submanifolds.
Corollary 4.2. Let S be a manifold in Rk , and let N ⊂ S be a subset. Then
N is an abstract submanifold according to Definition 4.2 if and only if it is
a submanifold according to Definition 4.1.
Proof. The implication ‘if’ was remarked in Example 4.2.2. To see the con-
verse, assume that N is an abstract submanifold of S, according to Definition
4.2. By the remark about transitivity before Lemma 4.2, N is an abstract
submanifold of Rk , so by Lemma 4.2 it is a manifold in Rk . Since it is also
contained in S, it is a submanifold according to Definition 4.1.
4.3 The local structure of submanifolds

The following theorem, which shows that submanifolds are very well be-
haved, explains why they are said to be ‘embedded’. It shows that locally
52 Chapter 4
all n-dimensional submanifolds of an m-dimensional differentiable manifold

look similar to the simplest example of an n-dimensional manifold in Rm ,
the ‘copy’
{(x1 , . . . , xn , 0, . . . , 0)}
of Rn inside Rm (Example 1.6.2).
In the statement of the following theorem, we shall regard Rn as a subset
of Rm (for m ≥ n), with the canonical inclusion
j(x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0) ∈ Rm .
Theorem 4.3. Let M be an m-dimensional differentiable manifold, and

N ⊂ M a non-empty subset. Let n ≤ m.
If N is an n-dimensional submanifold, then for each p ∈ N there exist a
number > 0 and a chart σ: U =] − , [m → M with σ(0) = p, such that the
following holds.
(1) σ(U ) ∩ N = σ(U ∩ Rn )
(2) The restriction of σ to U ∩ Rn is a chart on N .
Conversely, if for each p ∈ N there exist a chart σ: U → M with σ(0) = p
and
σ(U ) ∩ N = σ(U ∩ Rn ),
then N is an n-dimensional submanifold of M , and the restricted maps
σ|U∩Rn constitute the charts of an atlas for it.
Rm
U σ
N σ(U )
M
R n p
Example 4.3 Let M = {(x, y, z) | x2 + y 2 + z 2 = 1} and let N ⊂ M be the

equator. We have seen in Example 4.1 that it is a submanifold. The sper-
ical coordinates σ(u, v) = (cos u cos v, sin u cos v, sin v) have the properties
required in the theorem for p = (1, 0, 0) if we choose for example = π2 .
Proof. The proof is based on a clever application of the inverse function
theorem. Assume N is a submanifold, let p ∈ N , and choose arbitrary charts
σ̃: U → M and τ : V → N about p. We may assume that 0 ∈ U with σ̃(0) = p
and 0 ∈ V with τ (0) = p.
In order to attain the properties (1) and (2), we would like to arrange that
V = U ∩Rn and τ = σ̃|V . We will accomplish this by means of a modification
Submanifolds 53
of σ̃, and by shrinking the sets U and V . The modification will be of the
form σ = σ̃ ◦ Φ, with Φ a suitably defined diffeomorphism of a neighborhood
of 0 in Rm .
Let
Ψ = σ̃ −1 ◦ τ : Rn → Rm ,
then Ψ is defined in a neighborhood of 0, and we have Ψ(0) = 0. Notice that
we can view Ψ as the coordinate expression σ̃ −1 ◦ i ◦ τ for the inclusion map
in these coordinates around p.
The Jacobian matrix DΨ(0) of Ψ at 0 is an m × n matrix, and since dip
is injective this matrix has rank n. By a reordering of the coordinates in Rm
we may assume that
A
DΨ(0) =
B
where A is an n × n matrix with non-zero determinant, and B is an arbitrary
(m − n) × n matrix.
Define Φ: Rm → Rm on a neighborhood of 0 by Φ(x, y) = Ψ(x) + (0, y) for
x ∈ Rn , y ∈ Rm−n . The Jacobian matrix of Φ at 0 has the form

A 0
DΦ(0) =
B I
where I denotes the (m − n) × (m − n) identity matrix. This Jacobian matrix

has non-zero determinant and hence the inverse function theorem implies
that Φ is a diffeomorphism in a neighborhood of 0. Notice that we have
defined Φ so that Ψ = Φ ◦ j. Hence with coordinates σ = σ̃ ◦ Φ about p we
obtain
σ ◦ j = σ̃ ◦ Φ ◦ j = σ̃ ◦ Ψ = τ,
in a neighborhood of 0. With our modified chart σ we have thus arranged
that σ ◦ j = τ .
Let > 0 be sufficiently small so that ] − , [m and ] − , [n are contained
in all the mentioned neighborhoods of 0 in Rm and Rn , respectively. Let
U =] − , [m and V =] − , [n , and take the restriction of σ and τ to these
sets. Then j(V ) = U ∩Rn , hence τ (V ) = σ(j(V )) ⊂ σ(U ) ∩N , and (2) holds.
In order to arrange equality in (1) we shall replace by a smaller value. It
is easily seen that such a change will not destroy the validity of the already
established inclusion τ (V ) ⊂ σ(U ) ∩ N , nor does it destroy property (2).
Since τ (V ) is open in N , which carries the induced topology, there exists
an open set W ⊂ M such that τ (V ) = W ∩ N . Choose > 0 such that
> and σ(U ) ⊂ W , where U =] − , [m . Let V =] − , [n . We claim
that then τ (V ) = σ(U ) ∩ N . The inclusion ⊂ has already been remarked,
so let u ∈ U be an element with σ(u) ∈ N . Since τ (V ) = W ∩ N there
exists v ∈ V such that σ(u) = τ (v), and hence σ(u) = σ(j(v)). Since σ is
54 Chapter 4
injective it then follows that u = j(v), from which we conclude that v ∈ V .

We have thus shown that σ(U ) ∩ N ⊂ τ (V ). The equality in (1) has been
established.
Conversely, assume there exists a chart σ of the mentioned type around
each point p ∈ N . Then the maps τ = σ ◦ j constitute an atlas on N . The
condition of smooth transition maps follows from the same condition for M .
Hence N can be given a smooth structure. The inclusion map, expressed in
the local coordinates, is σ −1 ◦ i ◦ τ = j. It follows that it is smooth and has
an injective differential, so that N is a submanifold.
Corollary 4.3.1. If N ⊂ M is a submanifold of a differentiable manifold,
then there exists for each p ∈ N an open neighborhood W of p in M and a
smooth map π: W → N such that π(x) = x for all x ∈ W ∩ N .
Proof. Choose a chart σ around p as in the theorem, and let W = σ(U ) and
π(σ(x1 , . . . , xm )) = τ (x1 , . . . , xn ) for x ∈ U .
Corollary 4.3.2. Let L, M be differentiable manifolds, let N ⊂ M be a
submanifold, and let f : L → N be a map. Then f is smooth L → M if and
only if it is smooth L → N .
Proof. The statement ‘if’ follows from the fact that f = i◦f , where i: N → M
is the inclusion map, which is smooth. The statement ‘only if’ follows from
the fact that, in the notation of the preceding corollary, f = π ◦ f , where
π: M → N is defined and smooth in a neighborhood of p.
A smooth structure on a subset N ⊂ M , with respect to which N is a
submanifold of M , will be called a submanifold structure.
Corollary 4.3.3. Let N ⊂ M be a subset of a differentiable manifold. Then
there exists at most one submanifold structure on N .
proof. Let L, N be submanifolds of M , and assume that L = N as sets. The
claim is that then their smooth structures agree. By definition the inclusion
map L → M is smooth, hence it follows from Corollary 4.3.2 that the identity
map L → N is smooth. Likewise it is smooth in the opposite direction. Hence
the two structures agree.
Being a submanifold is thus really just a property of the subset N ⊂ M .
The following lemma shows that this property is ‘local’.
Lemma 4.3. Let M be a differentiable manifold and N ⊂ M a subset.
Then N is a submanifold if and only if for each p ∈ N there exists an open
neighborhood W of p in M such that W ∩ N is a submanifold of W .
Proof. It follows directly from Definition 4.2 that if N is a submanifold of
M , then N ∩ W is a submanifold of W for all open sets W ⊂ M . Conversely,
assume that N satisfies the local condition. By applying the first part of
Theorem 4.3 to the submanifold N ∩W of W we obtain charts of the specified
Submanifolds 55
type within W , for each of the sets W . Since W is open in M , a chart on W

is also a chart on M , and since all the sets W cover N , the second part of
Theorem 4.3 implies that N is a submanifold.
Notice that it follows from Lemma 2.9 that if M has a countable atlas,
then so has every submanifold N , since it inherits from M the property of
having a countable base for the topology.
4.4 Level sets

The following theorem is an efficient tool for the construction of manifolds.
It represents the analog for abstract manifolds of Theorem 1.6. Let M ,
N be differentiable manifolds of dimension m and n, respectively, and let
f : M → N be a smooth map. We say that f is submersive at p ∈ M if the
differential dfp : Tp M → Tq N is surjective.
Theorem 4.4. Let q ∈ N be given. If it is not empty, then the set
L := {p ∈ f −1 (q) | f is submersive at p}
is a submanifold of M of dimension m − n. Furthermore, the tangent space
Tp L is the null space for the linear map dfp : Tp M → Tq N for each p ∈ L.
Notice that if L is not empty, then n ≤ m, since a linear map into a higher
dimensional space cannot be surjective.
Proof. According to Lemma 4.3, it suffices to consider a neighborhood of
every point in L. Let p ∈ L be given.
Let σ: U → M and τ : V → N be charts about p and q, say with σ(x0 ) = p
and τ (y0 ) = q, and assume that f (σ(U )) ⊂ τ (V ). Put W = σ(U ), then
σ: U → W is a diffeomorphism.
The differential at x0 of τ −1 ◦f ◦σ: U → V (see (3.18)) is surjective. Hence
x0 belongs to the set
S := {x ∈ U | τ −1 ◦ f ◦ σ(x) = y0 , rank D(τ −1 ◦ f ◦ σ)(x) = n},
which by Theorem 1.6 is an m − n-dimensional manifold in Rm .
Notice that σ(S) = L ∩ W . Since σ is a diffeomorphism, it follows that
L ∩ W is an m − n-dimensional submanifold of W .
The tangent space Tp L is the image by dσx0 of the tangent space Tx0 S,
and the latter space is identified in Example 3.2.2 as the null space of D(τ −1 ◦
f ◦ σ)(x0 ) in Rm . By the chain rule, dσx0 maps this null space onto the null
space of dfp .
Definition 4.4. An element q ∈ N is called a regular value for f if f is

submersive at p for every p ∈ f −1 (q). A critical value is an element q ∈ N
which is not regular.
Notice that by this definition all elements q ∈
/ f (M ) are regular values.
56 Chapter 4
Corollary 4.4. If q ∈ f (M ) is a regular value, then the level set f −1 (q) is

a submanifold of M of dimension m − n.
Example 4.4 Let M ⊂ R3 be the unit sphere, and let h: M → R denote

the ‘height’ function h(x, y, z) = z. The map h: M → R is smooth, since it
is the restriction of the map H: R3 → R, also given by H(x, y, z) = z. The
differential of h at p ∈ M is the restriction to Tp M of dHp : R3 → R, hence
dhp (v) = v · (0, 0, 1). It follows that dhp is surjective Tp M → R if and only
Tp is not perpendicular to the z-axis. This happens exactly when p is not
one of the poles ±(0, 0, 1). We conclude that the critical values y ∈ R are
y = ±1. Away from these, level sets are submanifolds of M :
z
h 1 critical
h−1 (y) y regular
y 0
x
−1 critical
The height function on the sphere, see Example 4.4
4.5 Lie subgroups

The notion of a submanifold gives rise to a notion of Lie subgroups.
Definition 4.5. Let G be a Lie group. An (embedded) Lie subgroup of G
is a subgroup which is also a submanifold.
As mentioned there exist different notions of submanifolds. Recall that
we use the strong notion of embedded submanifolds from Definition 4.2. See
Definition 4.9.3 for a notion of Lie subgroups with weaker properties.
Theorem 4.5. Let H ⊂ G be a Lie subgroup of a Lie group G. Then H is
a Lie group, and it is a closed subgroup of G.
Proof. To see that H is a Lie group we need to show that the multiplication
is smooth from H × H to H, and that inversion is smooth H → H. Since the
inclusion of H in G is smooth, hence also the inclusion of H × H in G × G,
it is clear that multiplication is smooth H × H → G. Likewise the inversion
is smooth H → G. Now Corollary 4.3.3 implies that these maps are actually
smooth into H.
To see that H is closed, we first note that the closure H̄ of H is stable under
multiplication and inversion (and thus, it is a subgroup of G). This follows
Submanifolds 57
already from the fact that these operations are continuous. Next we choose
a neighborhood W of e in G according to Corollary 4.3.1. It follows from the
continuity of the map π: W → H in that corollary, that the equality π(h) = h
holds also for all h ∈ W ∩ H̄. Hence W ∩ H̄ ⊂ H. Let g ∈ H̄ be given. Then
since multiplication is continuous, {y ∈ G | g −1 y ∈ W } is a neighborhood of
g and hence contains an element h ∈ H. Then g −1 h ∈ W ∩ H̄ ⊂ H, and we
conclude that g = h(g −1 h)−1 ∈ H.
By a remarkable theorem of Cartan, every closed subgroup of a Lie group
is an (embedded) Lie subgroup. The proof is too complicated to be given
here.
Example 4.5 If G = GL(n, R) then the first statement in Theorem 4.5
follows from Theorem 2.8.
4.6 The orthogonal group

As an example of the application of Theorem 4.4 we will show that the
orthogonal group O(n) is a Lie group, and we will determine its dimension.
Recall that by definition O(n) is the group of all n × n real matrices A, which
are orthogonal, that is, which satisfy
AAt = I.
Theorem 4.6. O(n) is a Lie group of dimension n(n − 1)/2.

Proof. It follows from Theorem 4.5 that we only have to establish that O(n)
is a submanifold of M(n, R), of the mentioned dimension.
Let Sym(n) denote the set of all symmetric n × n real matrices, this is
a vector space of dimension n(n + 1)/2. Hence Sym(n) is a manifold of
dimension n(n + 1)/2.
Furthermore, let f : M(n, R) → Sym(n) be the map given by f (A) = AAt ,
clearly this is a smooth map. Then O(n) = f −1 (I), where I ∈ Sym(n) is
the identity matrix. It will be shown below that I is a regular value for f ,
hence it follows from Theorem 4.4 that O(n) is a submanifold of M(n, R) of
dimension n2 − n(n + 1)/2 = n(n − 1)/2.
In order to show that I is a regular value, we must determine the differen-
tial dfA of f at each A ∈ O(n), and show that it is surjective from TA M(n, R)
to Tf (A) Sym(n).
Since M(n, R) is an n2 -dimensional vector space, its tangent space at each
point is identified with M(n, R). The tangent vector corresponding to a
matrix X ∈ M(n, R) is the the derivative at s = 0 of the curve γ(s) = A+sX.
By Theorem 3.8, the differential maps this tangent vector to the derivative
at 0 of the curve f ◦ γ. Since
f (A + sX) = (A + sX)(A + sX)t = AAt + s(AX t + XAt ) + s2 XX t

58 Chapter 4
we conclude that dfA (X) = AX t + XAt ∈ Sym(n).

The claim has thus been reduced to the claim that if A is orthogonal,
then the linear map X → AX t + XAt is surjective of M(n, R) onto Sym(n).
This is easily seen, for if B ∈ Sym(n) is given, then B = AX t + XAt where
X = 12 BA.
Notice that it follows from Theorem 4.4 and the proof above that the
tangent space of O(n) at A is
TA O(n) = {X ∈ M(n, R) | AX t + XAt = 0}.
In particular, the tangent space at the identity matrix I is the space
TI O(n) = {X ∈ M(n, R) | X t = −X}
of antisymmetric matrices.
4.7 Domains with smooth boundary

As mentioned in Example 4.2.3, an open subset of an m-dimensional dif-
ferentiable manifold is an m-dimensional submanifold. In this section we will
discuss some particularly well behaved open subsets.
Definition 4.7. Let M be an m-dimensional differentiable manifold. A
non-empty open subset D ⊂ M is said to be a domain with smooth boundary
if for each p ∈ ∂D there exists an open neighborhood W of p in M , and a
smooth function f : W → R such that 0 is a regular value, f (p) = 0 and
W ∩ D = {x ∈ W | f (x) > 0}.
In most examples D is given by a single function f as above, with W = M .

Example 4.7.1 Let M denote the cylinder
{(x, y, z) ∈ R3 | x2 + y 2 = 1}
in R3 . The half-cylinder
D = {(x, y, z) ∈ R3 | x2 + y 2 = 1, z > 0}
is an open submanifold of M . The function f : M → R defined by f (x, y, z) =
z has the required property, hence D is a domain with smooth boundary
in M .
z
∂D
y
x
Submanifolds 59
Example 4.7.2 A particularly simple case is that where M = Rm . For

example, the half-space
D = Hm = {x ∈ Rm | xm > 0}
is a domain with smooth boundary. The boundary is ∂D = Rm−1 , standardly

embedded as {x | xm = 0}. The unit ball
D = {x ∈ Rm | x2 < 1}
is another example. The boundary is the unit sphere ∂D = S m−1 .

Lemma 4.7. Let D ⊂ M be a domain with smooth boundary, and let p, W
and f be as in Definition 4.7. Then
W ∩ ∂D = {x ∈ W | f (x) = 0}.
Proof. Let x ∈ W ∩ ∂D be given. If f (x) > 0 then x belongs to D, which is

open, and hence x ∈ / ∂D. If f (x) < 0 then it follows from the continuity of f
that an open neighborhood of x is disjoint from D, and again we conclude
that x ∈/ ∂D. Thus f (x) = 0.
For the opposite inclusion we use the condition of regularity in Definition
4.7. This condition implies that if x ∈ W is a point with f (x) = 0 then
dfx (v) = 0 for some tangent vector v ∈ Tx W = Tx M . We may assume
dfx (v) > 0. If γ is a curve on M with γ(0) = x, representing v, then it
follows that (f ◦ γ) (0) > 0. Hence f (γ(t)) > 0 for t positive and sufficiently
close to 0, which shows that x is a limit of points in W ∩D. Hence x ∈ ∂D.
Corollary 4.7. Let D ⊂ M be a domain with smooth boundary. If the
boundary ∂D is not empty, it is an m − 1-dimensional submanifold of M .
Proof. We apply Lemma 4.3. Let p ∈ ∂D, and let W and f be as above. It
follows from Lemma 4.7 and Theorem 4.4 that ∂D∩W is an m−1-dimensional
submanifold of W .
Theorem 4.7. Let D ⊂ M be open and non-empty. Then D is a domain
with smooth boundary if and only if for each p ∈ ∂D there exists a chart
σ: U → M on M around p = σ(0) such that
D ∩ σ(U ) = σ(U + ) where U + = {x ∈ U | xm > 0}.
If this is the case and σ is such a chart, then in addition
∂D ∩ σ(U ) = σ(U ◦ ) where U ◦ = {x ∈ U | xm = 0},
and the map x → σ(x, 0), defined for all x = (x1 , . . . , xm−1 ) ∈ Rm−1 with
(x, 0) = (x1 , . . . , xm−1 , 0) ∈ U ◦ , is a chart on ∂D.
60 Chapter 4
M
U+ σ
D
σ(U + ) = D ∩ σ(U )
◦
U
U−
Proof. Assume the condition of existence of the charts σ. It is easily seen

that 0 is a regular value for the smooth function f defined by f (σ(x)) = xm
on the open set W = σ(U ). Hence D is a domain with smooth boundary,
and Lemma 4.7 shows that ∂D ∩ σ(U ) = σ(U ◦ ). It follows from the last
statement in Theorem 4.3 that x → σ(x, 0) is a chart on ∂D.
Conversely, assume that D is a domain with smooth boundary, and let
p ∈ ∂D. It follows from Corollary 4.7 and Theorem 4.3 that there exists a
chart σ: ] − , [m → M with σ(0) = p, such that σ(x) ∈ ∂D if and only if
xm = 0. Let y, z ∈ U be two points, for which ym and zm are either both
positive or both negative. Then xm = 0 on the line segment from y to z, and
hence the curve γ(t) = σ(y + t(z − y)), where t ∈ [0, 1], does not meet ∂D.
If one, but not both, of the points σ(y) and σ(z) belongs to D, we reach a
contradiction, since in that case some point along γ would belong to ∂D (if
σ(y) ∈ D and σ(z) ∈ / D then γ(t) ∈ ∂D where t0 = sup{t | γ(t) ∈ D}). Thus
the two points both belong to D or they both belong to the complement. It
follows that σ(U ) ∩ D equals either σ(U + ) or σ(U − ). In the latter case we
change sign in the last coordinate of σ, and thus in any case we reach that
σ(U ) ∩ D = σ(U + ).
4.8 Orientation of the boundary

Let D be a domain with smooth boundary in M . If an orientation of
M is given, we can establish an orientation of ∂D in the following manner.
Let p ∈ ∂D, then we have to decide whether a given basis v1 , . . . , vm−1 for
Tp ∂D is positive or negative. We choose a neighborhood W and a function
f ∈ C ∞ (W ) as in Definition 4.7. Since 0 is a regular value, there exists a
vector v0 ∈ Tp M with dfp (v0 ) < 0. The inequality means that the tangent
vector v0 points out from D. We now say that v1 , . . . , vm−1 is a positive basis
for Tp ∂D if and only if v0 , v1 , . . . , vm−1 is a positive basis for Tp M .
We have to verify that what we have introduced is in fact an orientation
of the tangent space Tp ∂D, that is, if v1 , . . . , vm−1

is another basis with the

same property that v0 , v1 , . . . , vm−1 is positive, then the transition matrix
from v1 , . . . , vm−1 to v1 , . . . , vm−1

has positive determinant. At the same
Submanifolds 61
time we can verify that the definition is independent of the choices of W , f

and v0 . The details are omitted, as they are completely straightforward.
After that we also have to verify that what we have introduced is in fact an
orientation of ∂D, that is, there exists an atlas which induces the above on all
tangent spaces. Let σ be a chart on M as in Theorem 4.7. We can arrange,
by changing the sign on x1 if necessary, that the basis −em , e1 , . . . , em−1 for
Rm is mapped to a positive basis for Tp M by dσp , for each p ∈ U . It then
follows from the definition above that the corresponding chart on ∂D, given
by (x1 , . . . , xm−1 ) → σ(x1 , . . . , xm−1 , 0), induces a positive basis for Tp ∂D.
These charts constitute an atlas on ∂D, as desired.
Theorem 4.8. If M is orientable then so is ∂D, for any domain D ⊂ M
with smooth boundary.
Proof. See the discussion above.
Example 4.8.1. Assume that D is a domain with smooth boundary in

M = R2 . Then D is open in R2 and the boundary C = ∂D is a curve in
R2 . For example, D could be the ball {(x, y) | x2 + y 2 < 1} and C the circle
{(x, y) | x2 +y 2 = 1}. When R2 is oriented in the standard fashion, a vector v
in Tp C has positive direction if and only if the normal vector v̂ points into D.
Example 4.8.2. Assume similarly that D is a domain with smooth bound-
ary in M = R3 (for example, the unit ball). Then the boundary S = ∂D is
a surface in R3 (in the example, the unit sphere). With the standard orien-
tation of R3 , a basis v1 , v2 for Tp S is positive if and only if v1 × v2 points
outward of D.
Example 4.8.3. Assume that M is a curve in Rn given by a parametrization
γ: I → Rn , where I is an open interval. The parametrization determines an
orientation of M . Let D = γ(]a, b[) where [a, b] ⊂ I, then D is a domain with
smooth boundary. The boundary is ∂D = {γ(a)} ∪ {γ(b)}. By definition, an
orientation of a point set is a choice of a sign ± at each point. The definition
above is interpreted so that the sign is + in the case where v0 , the tangent
vector of M that points out of D, is positively oriented in M . This means
that the orientation of ∂D is + at γ(b) and − at γ(a).
Example 4.8.4. Let M be the cylinder {(x, y, z) | x2 + y 2 = 1} in R3 ,
oriented with outward positive normal, and let D be a band around the
waist, say where 0 < z < 1. The boundary ∂D consists of two circles. The
upper circle at z = 1 is oriented clockwise, and the lower circle at z = 0 is
oriented oppositely.
Example 4.8.5. Let N be an arbitrary oriented manifold, and let M be
the product manifold M = N × R. The subset D = N ×]0, 1[ is a domain
with smooth boundary ∂D = (N × {0}) ∪ (N × {1}). The two copies of N
are oppositely oriented, as in the preceding example.
62 Chapter 4
4.9 Immersed submanifolds

The other notion of submanifolds, which was mentioned first in the chap-
ter, is briefly mentioned. For clarity, in this subsection we call a submanifold
(according to Definition 4.2) an embedded submanifold.
Definition 4.9.1. Let M be a differentiable manifold. An immersed
submanifold is a subset N ⊂ M equipped as a topological space (but not
necessarily with the topology induced from M ) and a smooth structure such
that the inclusion map i: N → M is smooth with a differential dip : Tp N →
Tp M which is injective for each p ∈ N .
This is a weaker notion, because an embedded submanifold always carries
the induced topology, whereas here we are just requiring some topology. It is
required that the inclusion map is smooth, in particular continuous. Hence
every relatively open subset in N is open for this topology, but not vice-versa.
For example, the figure ∞ in R2 can be obtained as an immersed submanifold
(see Example 1.2.2), but it is not a submanifold of R2 .
Definition 4.9.2. Let M and N be differentiable manifolds, and let f : N →
M be a map. Then f is called an embedding/immersion if its image f (N )
can be given the structure of an embedded/immersed submanifold of M onto
which f is a diffeomorphism.
The notion of immersed submanifolds is important in the theory of Lie
groups, where the following definition turns out to be useful.
Definition 4.9.3. Let G be a Lie group. An (immersed) Lie subgroup of
G is a subgroup which is also an immersed submanifold, and which is a Lie
group on its own with this structure.
It should be emphasized that this notion of Lie subgroups is more general
than the one in Definition 4.5. It is in fact customary to reserve the term
‘Lie subgroup’ (without the specification in front) to Definition 4.9.3, and
use the term ‘embedded Lie subgroup’ for the other.
4.10 Exercises
1 Verify the statement below Example 4.2.4, about transitivity of the prop-
erty of being a submanifold.
2 Let M, N be differentiable manifolds and fix an element y0 ∈ N . Prove
that the subset {(x, y0 ) | x ∈ M } is a submanifold of M × N .
3 Let M be a differentiable manifold. Prove that the diagonal {(x, x) | x ∈
M } is a submanifold of M × M .
4 Let f : M → N be a diffeomorphism of manifolds, and let A ⊂ M be a
submanifold. Show that f (A) is a submanifold in N .
Submanifolds 63
5 Let L, M be differentiable manifolds, and let N ⊂ M be a submanifold.

Prove that a map f : N → L is smooth if and only if, for each p ∈ N there
exists an open neighborhood W of p in M , and a smooth map F : W → L
such that the functions f and F agree on W ∩ N .
6 Verify that the image by π of the equator of S 2 is a submanifold in RP 2 ,
which is diffeomorphic to RP 1 and to S 1 (see also Exercise 2.7).
7 Let N ⊂ M be a submanifold of a differentiable manifold. Show that N
is locally closed in M : Every p ∈ N has an open neighborhood W in M
such that W ∩ N is closed in W . Show by an example that N need not
be closed.
8 Prove the following converse to Theorem 4.4. Let L be an l-dimensional
submanifold of an m-dimensional differentiable manifold M . Prove that
for each p ∈ L there exists an open neighborhood W of p in M , a smooth
map f : W → Rm−l such that y = f (p) ∈ Rm−l is a regular value and
L ∩ W = f −1 (y) (hint: Apply Theorem 4.3).
9 Show that S 3 can be given
a structure as a Lie group. Hint: Consider the
group SU(2) = { −β̄ ᾱ |α, β ∈ C, |α|2 + |β|2 = 1}.
α β
10 Show that the special linear group
SL(n, R) := {A ∈ GL(n, R) | det A = 1}
is a Lie group. Hint: the differential d(det)I at I of the determinant map

det: M(n, R) → R is the trace M(n, R) → R.
11 Let M be a differentiable manifold, and let p ∈ M .
a. Let f be a diffeomorphism of M to itself with p as fixed point, f (p) = p.
Show that if M is oriented, then f preserves the orientation at p if and
only if det dfp > 0.
b. Let σ and σ̃ be charts on M around p = σ(x) = σ̃(x̃). Show that σ
and σ̃ induce the same orientation at p if and only if the Jacobian matrix
D(σ̃ −1 ◦ σ)(x) has positive determinant.
12 Verify that the unit sphere S n is orientable. Let α: S n → S n be the
antipodal map. Verify that it is a diffeomorphism, and determine for each
n whether it is orientation preserving or -reversing.
13 Let A, B ⊂ S 2 denote, respectively, the cap {(x, y, z) | z > 12 } and the
band {(x, y, z) | |z| < 12 }, and put D = π(A), E = π(B) ⊂ RP 2 . Prove
that D and E are domains with a common smooth boundary C = ∂D =
∂E, which is diffeomorphic to a circle, and that RP 2 = D ∪ C ∪ E (in
this fashion, RP 2 can be thought of as a disc and a Möbius band, sewn
together along the edge of each).
64 Chapter 4
14 Let F : M → N be a diffeomorphism. If D ⊂ M is a domain with smooth

boundary, verify that so is F (D), and ∂(F (D)) = F (∂D). Furthermore,
if M and N are oriented, and F is orientation preserving/reversing, then
so is its restriction ∂D → ∂(F (D)). Prove finally that the rotation of
S n−1 by an element A ∈ O(n) is orientation preserving if det A = 1 and
-reversing if det A = −1.
15 Prove that if M and N are orientable manifolds, then so is their product
M × N.
Chapter 5
Topological properties of manifolds
In this chapter we investigate the consequences for a manifold, when cer-

tain topological properties are assumed. Furthermore, we develop an impor-
tant analytical tool, called partition of unity.
5.1 Compactness
Recall that in a metric space X, a subset K is said to be compact, if
every sequence from K has a subsequence which converges to a point in
K. Recall also that every compact set is closed and bounded, and that the
converse statement is valid for X = Rn with the standard metric, that is,
the compact subsets of Rn are precisely the closed and bounded ones.
The generalization of compactness to an arbitrary topological space X
does not invoke sequences. It originates from another important property
of compact sets in a metric space, called the Heine-Borel property, which
concerns coverings of K.
Let X be a Hausdorff topological space, and let K ⊂ X.
Definition 5.1.1. A covering of K is a collection of sets Ui ⊂ X, where
i ∈ I, whose union ∪i Ui contains K. A subcovering is a subcollection (Uj )j∈J ,
where J ⊂ I, which is again a covering of K. An open covering is a covering
by open sets Ui ⊂ X.
Definition 5.1.2. The set K is said to be compact if every open covering

has a finite subcovering.
It is a theorem that for a metric space the property in Definition 5.1.2 is
equivalent with the property that K is compact (according to the definition
with sequences), hence there is no conflict of notations. The space X itself
is called compact, if it satisfies the above definition with K = X.
The following properties of compact sets are well known and easy to prove.
Let X and Y be Hausdorff topological spaces.
Lemma 5.1.1. Let f : X → Y be a continuous map. If K ⊂ X is compact,
then so is the image f (K) ⊂ Y .
Example 5.7.1 The canonical map S 2 → RP2 is continuous. Hence RP2 is
compact.
66 Chapter 5
Lemma 5.1.2. Let K ⊂ X be compact. Then K is closed, and every closed

subset of K is compact.
Lemma 5.1.3. Assume that X is compact, and let f : X → Y be a continu-
ous bijection. Then f is a homeomorphism.
Proof. We have to show that f −1 is continuous, or equivalently, that f car-
ries open sets to open sets. By taking complements, we see that it is also
equivalent to show that f carries closed sets to closed sets. This follows from
Lemma 5.1.1, in view of Lemma 5.1.2.
5.2 Countable exhaustion by compact sets

The following property of a manifold with a countable atlas will be used
in the next section.
Theorem 5.2. Let M be a differentiable manifold. The following conditions
are equivalent.
(i) There exists a countable atlas for M .
(ii) There exists a sequence K1 , K2 , . . . of compact sets with union M .
(iii) There exists a sequence D1 , D2 , . . . of open sets with union M , such
that each Dn has compact closure D̄n .
(iv) There exists a sequence K1 , K2 , . . . of compact sets with union M
such that
K1 ⊂ K2◦ ⊂ · · · ⊂ Kn ⊂ Kn+1 ◦
⊂ ....
A topological space M satisfying (ii) is said to be σ-compact.

Example 5.2 Each of the properties (i)-(iv) is easily verified for M = Rn .
Proof. The implications (iv)⇒(iii)⇒(ii) are easily seen. The implication
(ii)⇒(i) follows from the fact, seen from Definition 5.1.2, that every com-
pact set K ⊂ M can be covered by finitely many charts.
We establish (i)⇒(iii). For each chart σ: U → M the collection of closed
balls B̄(x, r) in U with rational center and rational radius is countable, and
the corresponding open balls cover U (see Example 2.9). Since σ is a home-
omorphism, the collection of all the images σ(B(x, r)) of these balls, for all
charts in a countable atlas, is a countable collection of open sets with the
desired property.
Finally we prove that (iii)⇒(iv). Let D1 , D2 , . . . be as in (iii). Put K1 =
D̄1 . By the compactness of K1 we have
K1 ⊂ D1 ∪ D2 ∪ · · · ∪ Di1
for some number i1 . Put
K2 = D̄1 ∪ D̄2 ∪ · · · ∪ D̄i1 ,

Topological properties of manifolds 67
then K2 is compact and K1 ⊂ K2◦ . Again by compactness we have
K2 ⊂ D1 ∪ D2 ∪ · · · ∪ Di2
for some number i2 > i1 . Put
K3 = D̄1 ∪ D̄2 ∪ · · · ∪ D̄i2 ,
then K3 is compact and K2 ⊂ K3◦ . Proceeding inductively in this fashion we

obtain the desired sequence.
5.3 Locally finite atlas

It follows from Definition 5.1.2 that a compact manifold has a finite atlas.
We shall define a much more general property.
Definition 5.3. A collection of subsets of a topological space X is said to
be locally finite, if for each element x ∈ X there exists a neighborhood which
intersects non-trivially with only finitely many of the subsets.
An atlas of a differentiable manifold M is said to be locally finite if the
collection of images σ(U ) is locally finite in M .
In the following lemma we give a useful criterion for the existence of a
locally finite atlas.
Lemma 5.3.1. Let M be a differentiable manifold. There exists a locally
finite atlas for M if and only if the following criterion holds.
There exists a covering M = ∪α∈A Kα of M by compact sets Kα , and a
locally finite covering M = ∪α∈A Wα by open sets Wα (with the same set A
of indices), such that Kα ⊂ Wα for each α ∈ A.
Before proving the lemma, we shall verify that the criterion holds in case
of a manifold with a countable atlas.
Lemma 5.3.2. Let M be a differentiable manifold for which there exists a
countable atlas. Then the criterion in Lemma 5.3.1 holds for M .
Proof. Let L1 , L2 , . . . be an increasing sequence of compact sets in M as in
Theorem 5.2(iv). Put K1 = L1 and Kn = Ln \ L◦n−1 for n > 1, and put
W1 = L◦2 , W2 = L◦3 and Wn = L◦n+1 \ Ln−2 for n > 2. It is easily seen that
the criterion is satisfied by these collections of sets.
Proof of Lemma 5.3.1. Assume that M has a locally finite atlas. For each
chart σ in this atlas, we can apply the preceding lemma to the manifold σ(U )
(which has an atlas of a single chart). It follows that there exist collections of
sets as described, which cover σ(U ). The combined collection of these sets,
over all charts in the atlas, satisfies the desired criterion for M .
68 Chapter 5
Assume conversely that the criterion holds for M . For each α ∈ A we

can cover the compact set Kα by finitely many charts σ, each of which maps
into the open set Wα . Let us call this finite collection of charts Aα . The
combined collection A = ∪Aα of all these charts for all α is an atlas because
∪α Kα = M . Since the collection of the Wα ’s is locally finite, each p ∈ M
has a neighborhood V which is disjoint from all Wα except finitely many, say
Wα1 , . . . , Wαn . This neighborhood V is then also disjoint from all the charts
from A except those from Aα1 , . . . , Aαn . Hence A is locally finite.
Corollary 5.3. Every differentiable manifold with a countable atlas has a
locally finite atlas.
Proof. Follows immediately from the two lemmas (in fact, by going through
the proofs one verifies that there exists a countable locally finite atlas).
5.4 Bump functions

The following lemma will be used several times in the future, when we
need to pass between local and global objects on a manifold.
Lemma 5.4. Let σ: U → M be a chart on a differentiable manifold, and
let B(x0 , r) ⊂ B̄(x0 , s) ⊂ U be concentric balls in U with 0 < r < s. There
exists a smooth function g ∈ C ∞ (M ), which takes values in [0, 1], such that
g(q) = 1 for q ∈ σ(B(x0 , r)) and g(q) = 0 for q ∈ / σ(B(x0 , s)).
The function g is called a bump function around p = σ(x0 ), because of the
resemblance with speed bumps used to reduce traffic. The following figure
shows the graph of a bump function around p = 0 in R.
x
r s
Proof. It suffices to prove that there exists a function ϕ ∈ C ∞ (Rm ) with

values in [0, 1] such that ϕ(x) = 1 for |x| ≤ r and ϕ(x) = 0 for |x| ≥ s,
because then g: M → R defined by

ϕ(y) if q = σ(y) ∈ σ(U )
g(q) =
g(q) = 0 otherwise,
has the required properties. In particular g is smooth, because it is smooth
on both σ(B(x0 , s)) (where it equals ϕ ◦ σ −1 ) and M \ σ(B̄(x0 , r)) (where it
is 0), hence on a neighborhood of every point in M .
In order to construct ϕ, we recall that the function ψ on R, defined by

ψ(t) = e−1/t for t > 0 and ψ(t) = 0 for t ≤ 0, is smooth (see the graph
below). The function
ψ(s − t)
h(t) =
ψ(s − t) + ψ(t − r)
is smooth, and it takes the value 1 for t ≤ r and 0 for t ≥ s. The function
ϕ(x) = h(|x|) has the required property.
y y
x x
y = ψ(x) y = h(x)
5.5 Partition of unity

Recall that the support, supp f , of a function f : M → R is the closure of
the set where f is non-zero.
Definition 5.5. Let M be a differentiable manifold. A partition of unity
for M is a collection (fα )α∈A of functions fα ∈ C ∞ (M ) such that:
(1) 0 ≤ fα ≤ 1,
the collection of the supports supp fα , α ∈ A, is locally finite in M ,
(2)
(3) α∈A fα (x) = 1 for all x ∈ M .
Notice that because of condition (2) the (possibly infinite) sum in (3) has
only finitely many non-zero terms for each x (but the non-zero terms are not
necessarily the same for all x).
It is clear that there exist a partition of unity. For example the trivial
collection consisting only of the constant function 1 is a partition of unity.
However, to obtain a useful collection, one needs a restriction on the support
of each member function fα .
Theorem 5.5. Let M be a differentiable manifold for which there exists a
locally finite atlas, and let M = ∪α∈A Ωα be an arbitrary open cover of M .
Then there exists a partition of unity (fα )α∈A for M (with the same set of
indices), such that fα has support inside Ωα for each α.
Proof. Let M = ∪β∈B Kβ and M = ∪β∈B Wβ be coverings as in Lemma 5.3.1,
with Kβ ⊂ Wβ . Let β ∈ B be arbitrary. For each p ∈ Kβ we choose α ∈ A
70 Chapter 5
such that p ∈ Ωα , and we choose a chart σ: U → M (not necessarily from the

given atlas) with σ(x) = p and with image σ(U ) ⊂ Ωα ∩ Wβ . Furthermore,
in U we choose a pair of concentric open balls B(x, r) B(x, s) around x,
such that the closure of the larger ball B(x, s) is contained in U . Since Kβ is
compact a finite collection of the images of the smaller balls σ(B(x, r)) cover
it. We choose such a finite collection of pairs of concentric balls for each β.
Since the sets Kβ have union M , the combined collection of all the images
of inner balls, over all β, covers M . Furthermore, for each p ∈ M there exists
a neighborhood which meets only finitely many of the images of the outer
balls, since the Wβ are locally finite. That is, the collection of the images of
the outer balls is locally finite.
The rest of the proof is based on Lemma 5.4. For each of the above men-
tioned pairs of concentric balls we choose a smooth function g ∈ C ∞ (M )
with the properties mentioned in this lemma. Let (gi )i∈I denote the collec-
tion of all these functions. By the remarks in the preceding paragraph, we
see that for each p ∈ M there exists some gi with gi (p) = 1, and only finitely
many of the functions gi are non-zero at p.
Hence the sum g = i gi has only finitely many terms in a neighborhood
of each point p. It follows that the sum makes sense and defines a positive
smooth function. Let fi = gi /g, then this is a partition of unity. Moreover,
for each i there exists an α ∈ A such that gi , hence also fi , is supported
inside Ωα .
We need to fix the index set for the gi so that it is A. For each i choose αi
such that fi has support inside Ωαi . For each α ∈ A let fα denote the sum
of those fi for which αi = α, if there are any. Let fα = 0 otherwise. The
result follows easily.
Corollary 5.5. Let M be a differentiable manifold with a locally finite atlas,
and let C0 , C1 be closed, disjoint subsets. There exists a smooth function
f ∈ C ∞ (M ) with values in [0, 1], which is 1 on C0 and 0 on C1 .
Proof. Apply the theorem to the covering of M by the complements of C1
and C0 .
The corollary is a smooth analogue of Urysohn’s lemma for normal topo-
logical spaces.
5.6 Embedding in Euclidean space

As an illustration of the use of partitions of unity, we shall prove the fol-
lowing theorem, which is a weak version of Whitney’s Theorem 2.10. Notice
that the theorem applies to all compact manifolds. Recall that by definition
an embedding is the same as a diffeomorphism onto a submanifold.
Theorem 5.6. Let M be a differentiable manifold, for which there exists a
finite atlas. Then there exists a number N ∈ N and an embedding of M into
RN .
Proof. Let σi : Ui → M , where i = 1, . . . , n be a collection of charts on

M , which comprise an atlas. We will prove the theorem with the value
N = n(m + 1) where m = dim M .
Let f1 , . . . , fn be a partition of unity for M such that supp fi ⊂ σi (Ui )
for each i. Its existence follows from Theorem 5.5. Then fi ∈ C ∞ (M ) and
f1 + · · · + fn = 1.
For each i = 1, . . . , n we define a function hi : M → Rm by

i fi (p)x if p = σi (x) ∈ σi (Ui )
h (p) =
0 if p ∈ M \ supp fi .
Then hi is smooth, since it restricts to a smooth function on the two open

sets σi (Ui ) and M \ supp fi covering M .
We now define a smooth map F : M → RN = Rm × · · · × Rm × Rn by
F (p) = (h1 (p), . . . , hn (p), f1 (p), . . . , fn (p)).
We first show that F is injective. Let p ∈ M and choose i such that fi (p) > 0.
Then p ∈ supp fi ⊂ σi (Ui ) and hi (p) = fi (p)x with x = σ −1 (p) ∈ Ui . Hence
p = σi (x) = σi (hi (p)/fi (p)) (5.1)
is uniquely determined from the coordinates of F (p).

We define S := F (M ) and aim to prove that this is a manifold in RN . For
each i = 1, . . . , n let
xi
Wi = {(x1 , . . . , xn , y1 , . . . , yn ) ∈ RN | yi > 0, ∈ Ui },
yi
where x1 , . . . , xn ∈ Rm and y1 , . . . , yn ∈ R. Then Wi is open in RN . We

claim that if p ∈ M then
F (p) ∈ Wi ⇔ fi (p) > 0. (5.2)
The implication ⇒ is obvious. Conversely if fi (p) > 0 then p ∈ σi (Ui ) and it

follows from (5.1) that hi (p)/fi (p) = x ∈ Ui . Hence F (p) ∈ Wi follows from
the definition of Wi . In particular, (5.2) implies that S ⊂ W1 ∪ · · · ∪ Wn .
Let
Vi = {x ∈ Ui | fi (σi (x)) > 0} ⊂ Rm ,
then Vi is open. We define τi = F ◦ σi : Vi → S. Clearly, τi is smooth and
it follows from (5.2) that it maps onto S ∩ Wi . Furthermore it follows from
(5.1) that
(x1 , . . . , xn , y1 , . . . , yn ) → xi /yi , Wi → Ui
72 Chapter 5
provides a continuous inverse to τi . Hence τi is an embedded m-dimensional

parametrized manifold in RN , whose image is S ∩ Wi . It follows that S is a
manifold in RN .
The map F is smooth and bijective M → S, and it follows from the
analysis above that it is a diffeomorphism onto its image.
Example 5.6 In Section 2.4 we equipped the projective space M = RP2
with an atlas consisting of n = 3 charts. Hence it follows from the proof just
given that there exists a diffeomorphism of it into R9 (in fact, as mentioned
below Theorem 2.10, there exists an embedding into R4 ).
5.7 Connectedness
In this section two different notions of connectedness for subsets of a topo-
logical space are introduced and discussed. Let X be a non-empty topological
space.
Definition 5.7. (1) X is said to be connected if it cannot be separated in
two disjoint non-empty open subsets, that is, if X = A1 ∪ A2 with A1 , A2
open and disjoint, then A1 or A2 is empty (and A2 or A1 equals X).
(2) X is called pathwise connected if for each pair of points a, b ∈ X there
exists real numbers α ≤ β and a continuous map γ: [α, β] → X such that
γ(α) = a and γ(β) = b (in which case we say that a and b can be joined by
a continuous path in X).
(3) A non-empty subset E ⊂ X is called connected or pathwise connected
if it has this property as a topological space with the induced topology.
The above definition of ”connected” is standard in the theory of topolog-
ical spaces. However, the notion of ”pathwise connected” is unfortunately
sometimes also referred to as ”connected”. The precise relation between the
two notions will be explained in this section and the following. The empty
set was excluded in the definition, let us agree to call it both connected and
pathwise connected.
Example 5.7.1 A singleton E = {x} ⊂ X is clearly both connected and
pathwise connected.
Example 5.7.2 A convex subset E ⊂ Rn is pathwise connected, since by
definition any two points from E can be joined by a straight line, hence a
continuous curve, inside E. It follows from Theorem 5.7.3 below that such a
subset is also connected.
Example 5.7.3 It is a well-known fact, called the intermediate value prop-
erty, that a continuous real function carries intervals to intervals. It follows
from this fact that a subset E ⊂ R is pathwise connected if and only if it is
an interval. We shall see below in Theorem 5.7.1 that likewise E is connected
if and only if it is an interval. Thus for subsets of R the two definitions agree.
Example 5.7.4 For a set X to be pathwise connected, it is sufficient that

there is some point x0 in X to which every other x ∈ X can be joined by
a continuous path, because every pair of points a, b ∈ X can then be joined
by the conjunction of the paths that join a and b to x0 . For example, we see
that the 2-sphere S 2 is pathwise connected, since every point x can be joined
to a pole by a segment of a great circle.
Lemma 5.7. Let (Ei )i∈I be a collection of subsets of X with non-empty
intersection ∩i Ei . If all the sets Ei are connected, respectively pathwise con-
nected, then so is their union E = ∪i Ei .
Proof. Let x0 ∈ ∩i Ei .
Assume first that all the Ei are connected, and that E = A1 ∪ A2 with
A1 , A2 disjoint and relatively open in E. Assume that x0 ∈ A1 . For each i
the two sets Ei ∩ A1 and Ei ∩ A2 are disjoint and relatively open in Ei . Since
Ei is connected, one of these sets is empty. Since x0 ∈ Ei ∩ A1 we conclude
that Ei ∩ A2 = ∅ for all i. Hence A2 = ∅, and E is connected
Assume next that the Ei are pathwise connected. Then each point x ∈ E
belongs to some Ei , hence can be joined to x0 by a continuous path within
that set. It follows that E is pathwise connected (see Example 5.7.4).
Example 5.7.5 Let H ⊂ R3 be the surface
{(x, y, z) | x2 + y 2 − z 2 = 1}
called the one-sheeted hyperboloid. We will prove that it is pathwise con-
nected. Let
H0 = {(x, 0, z) | x2 − z 2 = 1, x > 0} ⊂ H
and for each t ∈ R,
Ct = {(x, y, t) | x2 + y 2 = 1 + t2 } ⊂ H.
z
Ct
y
x
H0
Then every point (x, y, t) ∈ H can be joined to e1 = (1, 0, 0) by a segment

of Ct composed with a segment of H0 . Hence H is pathwise connected (see
Example 5.7.4).
74 Chapter 5
Theorem 5.7.1. Let E ⊂ R be non-empty. Then E is connected if and only

if it is an interval.
Proof. Assume that E is connected. Let a = inf E and b = sup E. For each
element c ∈ R with a < c < b the sets E∩] − ∞, c[ and E∩]c, ∞[ are disjoint
and open in E, and it follows from the definitions of a and b that they are
both non-empty. Since E is connected, their union cannot be E, hence we
conclude that c ∈ E. Hence E is an interval with endpoints a and b.
Assume conversely that E is an interval, and that E = A ∪ B where A
and B are open, disjoint and non-empty. Since they have open complements,
A and B are also closed in E. Choose a ∈ A and b ∈ B, and assume for
example that a < b. Then [a, b] ⊂ E, since E is an interval. The set [a, b] ∩ A
is non-empty, since it contains a, let c be its supremum. Since A is closed it
contains c, hence c < b and ]c, b] ⊂ B. Since B is closed it also contains c,
contradicting that A and B are disjoint.
One of the most fundamental properties of connected sets is expressed in

the following theorem, which generalizes the intermediate value property for
real functions on R (see Example 5.7.3).
Theorem 5.7.2. Let f : X → Y be a continuous map between topological
spaces. If E ⊂ X is connected, then so is the image f (E) ⊂ Y . Likewise, if
E is pathwise connected then so is f (E).
Proof. We may assume E = X (otherwise we replace X by E).
1) Assume f (X) = B1 ∪ B2 with B1 , B2 open and disjoint, and let Ai =
−1
f (Bi ). Then A1 , A2 are open, disjoint and with union X. Hence if X is
connected then A1 or A2 is empty, and hence B1 or B2 is empty.
2) If a, b ∈ X can be joined by a continuous path γ, then f (a) and f (b)
are joined by the continuous path f ◦ γ.
Example 5.7.6 The 1-sphere S 1 is both connected and pathwise connected,

since θ → (cos θ, sin θ) is continuous. Since the n-sphere S n is the image of
[− π2 , π2 ] × S n−1 → S n by the map
(θ, x) → (cos(θ)x, sin(θ)),
it follows by induction that S n is also connected and pathwise connected.

Theorem 5.7.3. Every pathwise connected topological space is connected.
Proof. Suppose X were pathwise connected but not connected. Then X =
A ∪ B with A, B open, disjoint and nonempty. Let a ∈ A, b ∈ B, then there
exists a continuous path γ: [α, β] → X joining a to b. The image C = γ([α, β])
is the disjoint union of C ∩ A and C ∩ B. These sets are open subsets of
the topological space C, and they are nonempty since they contain a and
b, respectively. Hence C is not connected. On the other hand, since the
interval [α, β] is connected, it follows from Theorem 5.7.2 that C = γ([α, β])
is connected. We have reached a contradiction.
The converse statement is false. There exists subsets of, for example Rn
(n ≥ 2), which are connected but not pathwise connected (an example in
R2 is given below). However, for open subsets of Rn the two notions of
connectedness agree. This will be proved in the following section.
Example 5.7.7 The graph of the function

sin(1/x) x = 0
f (x) =
0 x=0
is connected but not pathwise connected.
5.8 Connected manifolds

In this section we explore the converse to Theorem 5.7.3, with the case in
mind that the topological space X is a differentiable manifold.
Definition 5.8. A topological space X is said to be locally pathwise con-
nected if it has the following property. For each point x ∈ X and each
neighborhood V there exists an open pathwise connected set U such that
x∈U ⊂ V.
Example 5.8 The space X = Rn is locally pathwise connected, since all
open balls are pathwise connected. The same is valid for a differentiable
manifold, since each point has a neighborhood, which is the image by a chart
of a ball in Rm , hence pathwise connected.
Lemma 5.8. In a locally pathwise connected topological space, every open
connected set E is pathwise connected.
Proof. For a pair a, b ∈ E we write a ∼ b if a and b can be joined by a
continuous path in E. It is easily seen that this is an equivalence relation.
Since E is open there exists for each a ∈ E an open neighborhood V ⊂ E,
hence an open pathwise connected set U with a ∈ U ⊂ E. For all points x
in U we thus have a ∼ x. It follows that the equivalence classes for ∼ are
open. Let A be an arbitrary (non-empty) equivalence class, and let B denote
the union of all other equivalence classes. Then A and B are open, disjoint
and have union E. Since E is connected and A was non-empty, B must be
empty. Hence A = E, that is, a ∼ b for all a, b ∈ E, which means that E is
pathwise connected.
Theorem 5.8. Let M be a differentiable manifold. Each open connected
subset E of M is also pathwise connected.
Proof. Follows immediately from Lemma 5.8, in view of Example 5.8.
In particular, every open connected subset of Rn is pathwise connected.
76 Chapter 5
5.9 Components
Let X be topological space. We shall determine a decomposition of X as
a disjoint union of connected subsets. For example, the set R× = R \ {0} is
the disjoint union of the connected subsets ] − ∞, 0[ and ]0, ∞[.
Definition 5.9. A component (or connected component) of X is a subset
E ⊂ X which is maximal connected, that is, it is connected and not properly
contained in any other connected subset.
Example 5.9.1 The components of Q (with the standard metric from R)
are the point sets {q}. This is easily seen.
In the example mentioned before Definition 5.9, the components ] − ∞, 0[
and ]0, ∞[ are both open and closed in R× . In general, components need
not be open (see Example 5.9.1), but they are always closed. This is a
consequence of the following lemma.
Lemma 5.9. The closure of a connected subset E ⊂ X is connected.
Proof. Assume that Ē is separated in a disjoint union Ē = A1 ∪ A2 where
A1 , A2 are relatively open in Ē. Then Ai = Wi ∩ Ē for i = 1, 2, where W1
and W2 are open in X. Hence each set Wi ∩ E = Ai ∩ E is relatively open in
E, and these two sets separate E in a disjoint union. Since E is connected,
one of the two sets is empty. If for example W1 ∩ E = ∅, then E is contained
in the complement of W1 , which is closed in X. Hence also Ē is contained
in this complement, and we conclude that A1 = Ē ∩ W1 is empty.
Theorem 5.9. X is the disjoint union of its components. If X is locally
pathwise connected, for example if it is a differentiable manifold, then the
components are open and pathwise connected.
Proof. Components are disjoint, because if two components overlapped, their
union would be connected by Lemma 5.7, and none of them would be max-
imal. Let x ∈ X be arbitrary. It follows from Lemma 5.7 that the union
of all the connected sets in X that contain x, is connected. Clearly this
union is maximal connected, hence a component. Hence X is the union of
its components.
Assume that X is locally pathwise connected, and let E ⊂ X be a compo-
nent of X. For each x ∈ E there exists a pathwise connected open neighbor-
hood of x in X. This neighborhood must be contained in E (by maximality
of E). Hence E is open. Now Lemma 5.8 implies that E is pathwise con-
nected.
Example 5.9.2 Let H ⊂ R3 be the surface {(x, y, z) | x2 + y 2 − z 2 = −1},

called the two-sheeted hyperboloid. We claim it has the two components
H + = {(x, y, z) ∈ H | z > 0}, H − {(x, y, z) ∈ H | z < 0}.

It is easily seen that |z| ≥ 1 for all (x, y, z) ∈ H, hence H = H + ∪ H − , a

disjoint union. The verification that H + and H − are pathwise connected is
similar to that of Example 5.7.5.
z
H+
x y
H−
Corollary 5.9. Let M be a differentiable manifold. The components of M

are submanifolds of M of the same dimension as M . If an atlas is given
for each component, then the combined collection of the charts comprises an
atlas for M .
Proof. Follows from the fact that the components are open and cover M .
In particular, since there is no overlap between the components, M is
orientable if and only if each of its components is orientable.
5.10 The identity component of a Lie group

Let G be a Lie group. The connected component G0 of G, which contains
the neutral element e, is called the identity component.
Lemma 5.10. The identity component is a Lie subgroup of G.
Proof. It suffices to prove that G0 is a subgroup, since Corollary 5.9 ensures
it is a submanifold. Since the inversion g → g −1 is continuous, it follows
from Theorem 5.7.2 that G−1 0 := {g
−1
| g ∈ G0 } is connected. Since e ∈ G−1
0
it follows that G−1
0 ⊂ G 0 by definition of G0 . Likewise, by continuity of
−1
multiplication each coset xG0 is connected. If x ∈ G0 , then e = xx ∈ xG0 ,
and again we conclude that xG0 ⊂ G0 . Thus G0 is a subgroup.
Recall that O(n) denotes the Lie group of orthogonal n × n-matrices. The
elements g ∈ O(n) have determinant +1 or −1. The subgroup
SO(n) := {g ∈ O(n) | det g = 1}
is called the special orthogonal group.

78 Chapter 5
Example 5.10. Let

cos θ − sin θ cos θ sin θ
kθ = and lθ = ,
sin θ cos θ sin θ − cos θ
for θ ∈ R. Then O(2) = {kθ } ∪ {lθ }, and these are the connected compnents.
The special orthogonal group SO(2) = {kθ } is the identity component.
The preceding example can be generalized to O(n) for all n ≥ 1.
Theorem 5.10. The special orthogonal group SO(n) is connected. The com-
ponents of O(n) are SO(n) and {g ∈ O(n) | det g = −1}.
Proof. The proof is by induction on n. The cases n = 1 and n = 2 are easy.
Before we begin the induction, we need some notation.
Let G = SO(n) and H = SO(n − 1). We consider the embedding of H in
G as the subgroup of (n − 1) × (n − 1)-blocks in the lower right corner:

1 0
K = {kh := ∈ G | h ∈ H}.
0 h
Notice that K can be characterized as the subgroup {g ∈ G | ge1 = e1 } of

matrices for which e1 is a fixed vector. Indeed, ge1 = e1 means that the first
column of g is e1 (viewed as a column), and since g is orthogonal this forces
the first row also to be e1 (viewed as a row).
Let T = {tu | u ∈ R} with
⎛ ⎞
cos u − sin u 0
tu = ⎝ sin u cos u 0 ⎠ ∈ G
0 0 I
where I denotes the (n − 2) × (n − 2) identity matrix. Then T is connected.

We observe that g → ge1 is surjective SO(n) → S n−1 . In fact, already
the map φ: (k, t) → kte1 is surjective K × T → S n−1 , so that g → ge1 is
surjective from the subset of products kt in G. The surjectivity of φ can be
established by induction: Every element in x ∈ S n−1 can be written in the
form x = (cos u, ω sin u) with u ∈ R and ω ∈ S n−2 , and by induction we have
ω = he1 in Rn−1 for some h ∈ SO(n − 1). Then x = kh tu e1 .
We now claim that every element g ∈ G can be written as a triple product
g = k1 tk2 with k1 , k2 ∈ K and t ∈ T . Let g ∈ G be given and let x = ge1 .
We write x = φ(k, t) = kte1 as above and observe that then t−1 k −1 ge1 = e1 .
Hence k2 := t−1 k −1 g ∈ K and we have g = ktk2 as claimed.
Finally we can prove by induction that G = SO(n) is connected. Assuming
that H = SO(n − 1) is connected we have that K ⊂ G is connected as a
continuous image of H. Now the product map K × T × K → G is continuous
and surjective, hence G is connected (see Exervise 8).
Fix an arbitrary element g0 ∈ O(n) with determinant −1. Then the set
{g ∈ O(n) | det g = −1} is the image of SO(n) by g → g0 g, and hence it
is connected. Hence O(n) is the disjoint union of two connected subsets. If
these were not components of O(n), then O(n) would be connected. But
this is impossible, since {±1} is the image of O(n) by the determinant map,
which is continuous.
5.11 The Jordan-Brouwer theorem

The proof of the following theorem is too difficult to be given here.
Theorem 5.11. Let S ⊂ Rn be an n − 1–dimensional compact connected
manifold in Rn . The complement of S in Rn consists of precisely two com-
ponents, of which one, called the outside is unbounded, and the other, called
the inside is bounded. Each of the two components is a domain with smooth
boundary S.
The Jordan curve theorem for smooth plane curves is obtained in the
special case n = 2.
Example 5.11 Let S = S n−1 denote the unit sphere in Rn . Its inside is
the open n-ball {x | x < 1}, and the outside is the set {x | x > 1}.
Corollary 5.11. Let S be an n − 1–dimensional compact manifold in Rn .
Then S is orientable.
Proof. It suffices to prove that each component of S is orientable, so we
may as well assume that S is connected. The inside of S is orientable,
being an open subset of Rn . Hence it follows from Theorem 4.8 that S is
orientable.
In particular, all compact surfaces in R3 are orientable.
5.12 Exercises
1 Prove Lemmas 5.1.1 and 5.1.2.

2 Let X, Y be a Hausdorff topological spaces, and let K ⊂ X be compact.
a. Prove that K × {y} is compact for each y ∈ Y .
b. Prove that if K × {y} ⊂ W for some open set W in X × Y , then there
exist open sets U ⊂ X and V ⊂ Y such that K × {y} ⊂ U × V ⊂ W .
3 Let X, Y be a Hausdorff topological spaces, and let K ⊂ X, L ⊂ Y be
compact. Prove that K × L is compact. (Hint: If an open cover of K × L
is given, apply (a) above to obtain, for each y, a finite subcover of K ×{y}.
Then apply (b) above to show that finitely many of these subcovers will
cover the entire set K × L.)
80 Chapter 5
4 Prove that O(n) is compact.

5 Let M and N be differentiable manifolds, and let (fα )α∈A and (gβ )β∈B
be partitions of unity of them. Show that the collection of functions on
M × N given by all products fα (x)gβ (y) is a partition of unity on M × N .
6 Let f : X → Y be a continuous map between topological spaces. Prove
that if X is connected, then the graph {(x, f (x) | x ∈ X} is connected in
X × Y . Prove the same statement for pathwise connectedness.
7 Prove that the graph in Example 5.7.7 is connected [Hint: Apply 5.9 with
E equal to the part of the graph with x > 0]. Verify that the function f
is not continuous.
8 Prove that if X and Y are pathwise connected topological spaces, then so
is the product X × Y . Prove the corresponding statement for connected
topological spaces [Hint: Apply Lemma 5.7].
9 Let N ⊂ M be a submanifold of a differentiable manifold, and let f ∈
C ∞ (N ). Assume that there exists a locally finite atlas for M .
a. Prove that there exists an open set W ⊂ M with N ⊂ W , and a
function F ∈ C ∞ (W ) such that F (x) = f (x) for all x ∈ N [Hint: use
Exercise 4.5 and partition of unity].
b. Assume now in addition that N is closed in M . Prove that the above
can be accomplished with W = M .
c. Give an example which shows the extra condition in part (b) is neces-
sary [Hint: 1/x near x = 0]
10 Let M and N be oriented and connected differentiable manifolds, and
let f : M → N be a diffeomorphism. Prove that either f is orientation
preserving at every p ∈ M , or it is orientation reversing at avery p ∈ M
[Hint: Prove that the set where it is orientation preserving is open. For
this, apply continuity of the determinant of the matrix dfp , with respect
to standard bases for given charts].
11 Show that RP n is connected for all n. Show that it is not orientable when
n is even [Hint: Consider the composition of π with the antipodal map α
of the sphere. Apply Exercise 4.12 together with Exercise 9 above].
12 Let (Ei )i∈I be a collection of subsets of X, and let E0 ⊂ X be a subset
with the property that Ei ∩ E0 = ∅ for all i. Prove that if both E0 and
all the sets Ei are connected, respectively pathwise connected, then so is
their union E = E0 ∪ (∪i Ei ).
Chapter 6
Vector fields and Lie algebras
By definition, a vector field on a manifold M assigns a tangent vector

Y (p) ∈ Tp M to each point p ∈ M . One can visualize the vector field as a
collection of arrows, where Y (p) is placed as an arrow tangent to the manifold
and with its basis at p.
6.1 Smooth vector fields

We would like the vector field Y to vary smoothly with p. Since Y : p →
Y (p) maps into a space that varies with p, the precise formulation of smooth-
ness requires some care. For vector fields on a manifold S in Rn , we can
express the desired smoothness easily by means of the ambient space Rn .
Definition 6.1.1. A smooth vector field on a manifold S ⊂ Rn is a smooth
map Y : S → Rn such that Y (p) ∈ Tp S for all p ∈ S.
The following result gives a reformulation, which paves the way for the
abstract case.
Lemma 6.1.1. Let S ⊂ Rn be a manifold and let Y : S → Rn be such that
Y (p) ∈ Tp S for all p ∈ S. The following conditions are equivalent
(a) Y is smooth.
(b) For each chart σ: U → S we have

m
Y (σ(u)) = ai (u)σu i (u) (6.1)
i=1
for all u ∈ U , with coefficient functions a1 , . . . , am in C ∞ (U ).

(c) The same as (b), but only for σ in a given atlas.
82 Chapter 6
Proof. By Theorem 2.6 the first condition is equivalent to Y ◦σ being smooth

U → Rn for all charts σ in a given atlas. It follows immediately from (6.1)
that this is the case if the functions a1 , . . . , am are smooth. Thus (c) implies
(a).
The implication (b)⇒(c) is clear. Finally we will derive (b) from (a). Since
the elements σu 1 (u), . . . , σu m (u) form a basis for Tσ(u) S, there exist for each
u ∈ U some unique real numbers a1 (u), . . . , am (u) such that (6.1) holds. The
claim is that (a) implies these numbers depend smoothly on u.
Let p = σ(u0 ) ∈ σ(U ) be given. It follows from Theorem 1.7 that σ −1 has
a smooth extension ϕ to a neighborhood W of p in Rn . Then ϕ ◦ σ is the
identity map in a neighborhood of u0 . The entries in the Jacobian matrix
Dϕ(x) of ϕ depend smoothly on x ∈ W . According to the chain rule we have
Dϕ(σ(u))Dσ(u) = I for all u ∈ U . Let a(u) ∈ Rm denote the column whose
elements are the numbers a1 (u), . . . , am (u). The vectors σu i (u) in (6.1) are
the columns of Dσ(u), and the expression on the right of (6.1) is Dσ(u)a(u).
Hence it follows from (6.1) that
Dϕ(σ(u))Y (σ(u)) = Dϕ(σ(u))Dσ(u)a(u) = a(u)
and hence a(u) depends smoothly on u.

Let now M be an abstract manifold. We will use a generalized version
of the conditions (b)-(c) in our definition of a smooth vector field on M .
Recall from Section 3.9, that we introduced the standard basis vectors for
the tangent space Tp M at p of M ,
dσu (e1 ), . . . , dσu (em ) (6.2)
where σ(u) = p, and where ei ∈ Rm are the canonical basis vectors. These
are the analogs of the vectors σu i (u) in (6.1).
Definition 6.1.2. A vector field on M is an assignment of a tangent vector
Y (p) ∈ Tp M to each p ∈ M . It is called a smooth vector field if, for each
chart σ: U → M in a given atlas of M , there exist a1 , . . . , am in C ∞ (U ) such
that
m
Y (σ(u)) = ai (u)dσu (ei ) (6.3)
i=1
for all u ∈ U . The vector space of smooth vector fields on M is denoted

X(M ).
It is a consequence of Lemma 6.2.2, to be proved below, that the definition
is unchanged if the atlas is replaced by a compatible one. The concept
introduced thus only depends on the smooth structure of M .
Notice that each chart σ in the given atlas for M is a diffeomorphism
U → σ(U ), and the coefficients a1 (u), . . . , an (u) in (6.3) are the components
Vector fields and Lie algebras 83
of the vector (dσu )−1 (Y (p)) ∈ Rm , where p = σ(u). Hence the condition of
smoothness can also be phrased as the condition that
u → (dσu )−1 (Y (σ(u))) (6.4)
is smooth U → Rm .
Example 6.1 Let M ⊂ Rm be an open subset, regarded as a manifold with
the chart of the identity map Rm → Rm . We know that at each point p ∈ M
the tangent space Tp M can be identified with Rm , and a smooth vector field
on M is nothing but a smooth map Y : M → Rm .
If we regard tangent vectors to Rm as directional differentiation operators
(see Section 3.6), the standard basis vectors are the partial derivative oper-
∂
ators ∂x i
, for i = 1, . . . , m. The operator corresponding to Y is the partial
differential operator
m
∂
ai (x) ,
i=1
∂xi
where the coefficients a1 , . . . , am ∈ C ∞ (M ) are the components of Y . In

other words, we can think of a smooth vector field on Rm as a first order
partial differential operator with smooth coefficients. In the following section
we shall see that a similar interpretation is possible for vector fields on general
abstract manifolds.
Definition 6.1.3. Let f ∈ C ∞ (M ) and Y ∈ X(M ). The product f Y ∈
X(M ) is defined by (f Y )(p) = f (p)Y (p) for all p ∈ M .
The fact that f Y is smooth is easily seen from Definition 6.1.2.
6.2 An equivalent formulation of smoothness

Let Y be a vector field on a differentiable manifold M . In analogy with
Example 6.1, we shall interpret Y as a first order differential operator. Recall
from Definition 3.7 that a tangent vector X ∈ Tp M can be brought to act
on a smooth function f by the directional derivative Dp,X . For f ∈ C ∞ (Ω),
where Ω ⊂ M is open, it thus makes sense to apply Y (p) to f for each p ∈ Ω.
We denote by Y f = Y (f ) the resulting function p → Y (p)f = Dp,Y (p) f .
It is convenient to introduce the coordinate functions ξ1 , . . . , ξm associated
with a chart σ: U → M . They map σ(U ) → U and are defined by
ξi (σ(u)) = ui , u ∈ U.
Lemma 6.2.1. The coordinate functions ξi belong to C ∞ (σ(U )). For each
X ∈ Tp M the coordinates of X with respect to the standard basis (6.2) are
Dp,X (ξ1 ), . . . , Dp,X (ξm ).
84 Chapter 6
Proof. The function ξi is equal to σ −1 composed with projection on the i-th

coordinate of Rm . Hence it is smooth (see Example 2.7.1). Recall
∂
Dp,dσu (ei ) f = (f ◦ σ)(u) (6.5)
∂ui

(see (3.21)). If X = i ai dσu (ei ) ∈ Tp M we derive
∂ ∂
Dp,X (ξj ) = ai (ξj ◦ σ)(u) = ai (uj ) = aj .
i
∂ui i
∂ui
Notice that it follows from the preceding lemma, that a tangent vector
X ∈ Tp M is uniquely determined by its action on functions. If we know
Dp,X f for all smooth functions f defined on arbitrary neighborhoods of p,
then we can determine X by taking f = ξ1 , . . . , f = ξm with respect to some
chart. In particular, it follows that a vector field is uniquely determined by
its action on functions. Because of this, it is quite customary to identify a
smooth vector field Y on M with its action on smooth functions, and thus
regard the operator f → Y f as being the vector field itself. The following
lemma supports this point of view.
Lemma 6.2.2. Let Y be a vector field on an abstract manifold M . The
following conditions are equivalent:
(i) Y is smooth,
(ii) Y f ∈ C ∞ (M ) for all f ∈ C ∞ (M ),
(iii) Y f ∈ C ∞ (Ω) for all open sets Ω ⊂ M and all f ∈ C ∞ (Ω).
Proof. Let σ be a chart on M , and let (6.3) be the associated expression of
Y . It follows from (6.5) that

m
m
∂
Y f (σ(u)) = ai (u)Dσ(u),dσu (ei ) f = ai (u) (f ◦ σ)(u). (6.6)
i=1 i=1
∂ui
The implication of (i)⇒(ii) will be proved from this equation.

Assume (i) and let f ∈ C ∞ (M ). Assume that σ belongs to the given atlas.
Then the ai are smooth functions of u, and it follows from (6.6) that Y f ◦ σ
is smooth. Since σ was arbitrary within an atlas, Y f is smooth. This proves
(ii).
We prove (ii)⇒(iii). Let Ω ⊂ M be open, and let f ∈ C ∞ (Ω). Let p ∈ Ω
be given, and chose a chart σ on M with p ∈ σ(U ) ⊂ Ω. Let g ∈ C ∞ (M ) be
a smooth ‘bump’ function around p with support inside σ(U ), as in Lemma
5.4. Then the function h on M defined by h = gf on Ω and h = 0 otherwise,
belongs to C ∞ (M ). Furthermore h = f in a neighborhood of p, and hence
Y h = Y f in that neighborhood. Since Y h is smooth by assumption (ii), it
follows that Y f is smooth in a neighborhood of p. Since p was arbitrary in

Ω, (iii) follows,
For the last implication, (iii)⇒(i), we apply Lemma 6.2.1. Let σ be an
arbitrary chart on M , and let ξi be a coordinate function. It follows from
(iii) with Ω = σ(U ) that Y ξi is smooth on this set, and it follows from the
last statement in Lemma 6.2.1 that (6.3) holds with ai = Y ξi . Hence Y is
smooth.
Notice that we have the following linearity,
Y (f + g) = Y f + Y g, Y (λf ) = λY f
for f, g ∈ C ∞ (Ω) and λ ∈ R.
6.3 Derivations
Smooth vector fields obey a generalized Leibniz rule.
Definition 6.3. A linear map Y: C ∞ (M ) → C ∞ (M ) satisfying
Y(f g) = f Y(g) + gY(f )
for all f, g ∈ C ∞ (M ) is called a derivation.

Lemma 6.3. Every smooth vector field Y ∈ X(M ) is a derivation.
Proof. Let σ be a chart on M , and let (6.3) be the associated expression of
Y . It follows from (6.5) that on σ(U ) we have
∂
Y (f g) = ai [(f g) ◦ σ].
i
∂ui
The lemma now follows by applying the usual Leibniz rule to the differenti-
ation of the product [(f g) ◦ σ] = (f ◦ σ)(g ◦ σ).
In fact, the following converse is valid. The proof will not be given here.
Theorem 6.3. Every derivation Y of C ∞ (M ) is given by a unique vector
field Y ∈ X(M ).
6.4 The tangent bundle

In Definition 6.1.2 we defined a vector field on a differentiable manifold M
as an assignment of a tangent vector Y (p) to each point p ∈ M . Thus Y is a
map from M into the set of all tangent vectors at all points of M . It turns
out that this set of tangent vectors can be given a differential structure of its
own, by means of which the smoothness of Y can be elegantly expressed.
86 Chapter 6
Definition 6.4. The tangent bundle of M is the union T M = ∪p∈M Tp M

of all tangent vectors at all points.
For a given element X ∈ T M the point p ∈ M for which X ∈ Tp M is
called the base point. The map π: T M → M which assigns p to X, is called
the projection.
Notice that the union we take is disjoint, that is, there is no overlap
between Tp1 M and Tp2 M if p1 = p2 . Formally, an element in T M is actually
a pair (p, X) where p ∈ M and X ∈ Tp M , and π maps (p, X) to the first
member of the pair. Notationally it is too cumbersome to denote elements
in this fashion, and hence the base point p is suppressed.
If M = S, a manifold in Rn , then we are accustomed to viewing tangent
vectors as elements in the ambient space, and of course the same vector
v ∈ Rn can easily be tangent vector to S at several points. For example, two
antipodal points on a sphere have identical tangent spaces. However, in the
tangent bundle T M it is important that we make distinction between copies
of v which are attached to different base points (otherwise, for example, the
projection would not be well-defined). Thus the tangent bundle of a manifold
in Rn is not to be conceived as a subset of Rn .
Example 6.4 Let U ⊂ Rm be an open set, which we view as an m-
dimensional manifold. At each point x ∈ U the tangent space Tx U is a
copy of Rm , hence we can identify the tangent bundle of U as T U = U × Rm .
The projection π is given by π(x, y) = x.
Let f : M → N be a smooth map between manifolds, then its differential
dfp maps Tp M into Tf (p) N for each p ∈ M . The collection of these maps,
for all p ∈ M , is a map from T M to T N , which we denote by df .
In the special case where f : U → Rn maps an open set U ⊂ Rm smoothly
into Rn , the differential df : U ×Rm → Rn ×Rn is the map given by df (p, v) =
(f (p), Df (p)v). Observe that this is a smooth map.
Theorem 6.4. Let M be a differentiable manifold. The collection consisting
of the maps
dσ: U × Rm → T M,
for all charts σ in an atlas on M , is an atlas for a structure of an abstract
manifold on the tangent bundle T M .
With this structure the projection π: T M → M is smooth, and a vector
field Y on M is smooth if and only if it is smooth as a map from M to T M .
Proof. Before we can prove that the chosen collection is an atlas on T M , we
need to give T M a Hausdorff topology. We declare a subset W of T M to be
open if and only if, for each σ in the atlas of M , the preimage dσ −1 (W ) is
open in U × Rm . The conditions for a topology are easily verified. Moreover,
with this definition it is clear that each map dσ is a continuous bijection
onto its image. To show that the inverse is continuous we need to verify that
dσ(A) ⊂ T M is open when A ⊂ U × Rm is open. By the definition we just

gave, we must verify that dτ −1 (dσ(A)) is open in Rm × Rm for all τ in the
atlas of M . Since dτ −1 (dσ(A)) = (dσ −1 ◦ dτ )−1 (A), this follows from the
continuity of d(σ −1 ◦ τ ). Hence dσ is a homeomorphism onto its image. It is
easily seen that π: T M → M is continuous.
We need to verify the Hausdorff axiom. Let two distinct elements X1 , X2 ∈
T M be given, and let p1 , p2 ∈ M be their base points. If p1 and p2 are
distinct, there exist disjoint open sets V1 , V2 ⊂ M around them (since M
is Hausdorff), and then the sets π −1 (V1 ), π −1 (V2 ) are open, disjoint neigh-
borhoods of X1 , X2 . On the other hand, if p1 = p2 there exists a chart
σ around this point, and there exist distinct elements w1 , w2 ∈ Rm such
that X1 = dσp (w1 ) and X2 = dσp (w2 ). Then (since Rm is Hausdorff)
there exist disjoint open sets W1 , W2 in Rm around w1 , w2 , and the sets
dσ(U × W1 ), dσ(U × W2 ) are open disjoint neighborhoods of X1 and X2 .
We have to show that our charts on T M overlap smoothly, that is, we have
to verify that if σ1 : U1 → M and σ2 : U2 → M are charts on M , then dσ1−1 ◦dσ2
is smooth. It follows from the chain rule that dσ1−1 ◦ dσ2 = d(σ1−1 ◦ σ2 ),
and thus the assertion follows from the smoothness of σ1−1 ◦ σ2 (see the
observation before the theorem). This completes the verification that T M is
a differentiable manifold. It is easily seen that π is smooth for this structure.
Let Y be a vector field on M , viewed as a map M → T M . Let p ∈ M
be given, and let σ be a chart around it. Then dσ is a chart on T M around
Y (p), and hence the map Y : M → T M is smooth at p = σ(u) if and only if
the coordinate expression dσ −1 ◦ Y ◦ σ is smooth at u. As remarked earlier
(see (6.4)), this is exactly the condition in Definition 6.1.2.
Let f : M → N be a smooth map between manifolds. It follows from the

chain rule that if σ is a chart on M and τ a chart on N , then dτ −1 ◦ df ◦ dσ =
d(τ −1 ◦ f ◦ σ). Using the atlas on T M as above, and the corresponding one
for T N , we now see that df is a smooth map between T M and T N .
6.5 The Lie bracket
Let X, Y ∈ X(M ) be smooth vector fields on M , and let Ω ⊂ M be open.

For a given function f ∈ C ∞ (Ω) we obtain a new function Y f ∈ C ∞ (Ω) by
application of Y (see Lemma 6.2.2). Applying X to this function we obtain
a function X(Y f ) ∈ C ∞ (Ω), which we denote XY f . The two vector fields
thus give rise to a linear operator XY : C ∞ (Ω) → C ∞ (Ω).
2
∂ f
For example, the mixed second derivative ∂x∂y of a function f ∈ C ∞ (R2 )
∂ ∂
is obtained in this fashion with X = ∂x and Y = ∂y . In this example
it is known that the two derivatives commute, that is, XY f = Y Xf for
all f ∈ C ∞ (R2 ), and hence the operators XY and Y X agree. The same
commutation does not occur in general. The non-commutativity is expressed
88 Chapter 6
by the difference XY − Y X of the two operators, which turns out to be a

fundamental object.
Definition 6.5. Let X, Y ∈ X(M ) be smooth vector fields on M . The

Lie bracket [X, Y ] is the linear operator C ∞ (Ω) → C ∞ (Ω) defined by the
equation
[X, Y ]f = XY f − Y Xf
for f ∈ C ∞ (Ω).
Example 6.5 Let X be the vector field dx d

on R and let Y be the vec-
d2
tor field x dx d
on R. Then XY f = dx d d
(x dx f ) = x dx d
2 f + dx f and Y Xf =
2
d d d
x dx ( dx f ) = x dx 2 f , and hence the Lie bracket is the first order operator
given by [X, Y ]f = XY f − Y Xf = dx d
f.
None of the operators XY and Y X are vector fields – they are second
order operators, whereas vector fields are first order operators. However, it
turns out that their difference [X, Y ] is again a vector field, the essential
reason being that the second order terms cancel with each other. This is
established in the following theorem.
Theorem 6.5. Let X, Y ∈ X(M ) be smooth vector fields on a differentiable

manifold. The Lie bracket [X, Y ] is again a smooth vector field on M , that
is, there exist a unique element Z ∈ X(M ) such that [X, Y ]f = Zf for all
f ∈ C ∞ (Ω) and all open sets Ω ⊂ M .
Proof. Let p ∈ M . We will show that there exists a tangent vector Z(p) ∈
Tp M such that Z(p)f = ([X, Y ]f )(p) for all f ∈ C ∞ (Ω) where p ∈ Ω (as
remarked below Lemma 6.2.1 such a tangent vector, if it exists, is unique).
Choose a chart σ: U → M around p, and let

m
m
X(σ(u)) = ai (u)dσu (ei ), Y (σ(u)) = bj (u)dσu (ej )
i=1 j=1
be the expressions for the smooth vector fields X and Y , as in Definition

6.1.2. By applying (6.6) separately for X and Y , in both orders, we obtain

m
∂ ∂ ∂ ∂
([X, Y ]f ) ◦ σ = ai bj (f ◦ σ) − bj ai (f ◦ σ)
i,j=1
∂ui ∂uj ∂uj ∂ui
m
∂bj ∂ ∂ai ∂
= ai (f ◦ σ) − bj (f ◦ σ).
∂ui ∂uj ∂uj ∂ui
i,j=1
In the last step we used the Leibniz rule and cancellation. Let

m
∂bj ∂ai
Z(p) = ai dσu (ej ) − bj dσu (ei )
∂ui ∂uj
i,j=1
∂bi ∂ai
= aj − bj dσu (ei ) ∈ Tp M,
i j
∂uj ∂uj
where the coefficient functions are evaluated at u ∈ U with σ(u) = p. This

tangent vector Z(p) ∈ Tp M has the desired property.
Since p was arbitrary, we have obtained a vector field Z on M , such that
Z(p)f = [X, Y ]f (p) for all f ∈ C ∞ (Ω). It follows from Lemma 6.2.2 that
this vector field is smooth, since the Lie bracket [X, Y ] has the properties (ii)
and (iii) in that lemma.
6.6 Properties of the Lie bracket

Because of the result in Theorem 6.5, we regard the Lie bracket as a map
X(M ) × X(M ) → X(M ). It has the following elementary properties, which
are easily verified.
Lemma 6.6.1. The Lie bracket satisfies
[aX + bY, Z] = a[X, Z] + b[Y, Z], [X, aY + bZ] = a[X, Y ] + b[X, Z],
[X, Y ] = −[Y, X],
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0,
for all X, Y, Z ∈ X(M ), and a, b ∈ R.
The latter equation, which is a replacement for associativity, is known as
the Jacobi identity. The following property of the Lie bracket is verified by
means of Lemma 6.3.
Lemma 6.6.2. Let X, Y ∈ X(M ) and f, g ∈ C ∞ (M ). Then
[f X, gY ] = f g[X, Y ] + f (Xg)Y − g(Y f )X.
The following lemma essentially shows that Lie brackets transform nicely
by smooth maps. Let ψ: M → N be a smooth map between differentiable
manifolds, and let X ∈ X(M ) and V ∈ X(N ). We shall say that dψ trans-
forms X into V if
dψp (X(p)) = V (ψ(p))
for all p ∈ M . Since vector fields are uniquely determined by their action on
smooth functions, an equivalent condition is obtained from Lemma 3.8.3,
X(ϕ ◦ ψ) = (V ϕ) ◦ ψ
for all ϕ ∈ C ∞ (N ).
90 Chapter 6
Lemma 6.6.3. Let ψ: M → N be smooth, and let X, Y ∈ X(M ) and V, W ∈

X(N ) be given. If dψ transforms X into V and Y into W , then it also
transforms [X, Y ] into [V, W ].
Proof. As mentioned the assumptions amount to X(ϕ ◦ ψ) = (V ϕ) ◦ ψ and
Y (ϕ ◦ φ) = (W ϕ) ◦ ψ for all ϕ ∈ C ∞ (N ), and likewise the desired conclusion
amounts to [X, Y ](ϕ ◦ ψ) = ([V, W ]ϕ) ◦ φ, for all ϕ ∈ C ∞ (N ).
The proof is now a straightforward computation:
[X, Y ](ϕ ◦ ψ) = X(Y (ϕ ◦ ψ)) − Y (X(ϕ ◦ ψ))

= X((W ϕ) ◦ ψ) − Y ((V ϕ) ◦ ψ)
= (V (W ϕ)) ◦ ψ − (W (V ϕ)) ◦ ψ = ([V, W ]ϕ) ◦ ψ.
Corollary 6.6. Let M ⊂ N be a submanifold.

(1) Let V ∈ X(N ). If V (p) ∈ Tp M for all p ∈ M , then V |M ∈ X(M ).
(2) Let V, W ∈ X(N ) and assume that V (p), W (p) ∈ Tp M for all p ∈ M .
Then
[V |M , W |M ] = [V, W ]|M .
Proof. For (1), see Exercise 5. For (2) we apply Lemma 6.6.3. It is easily
seen that the inclusion map transforms X ∈ X(M ) into V ∈ X(N ) if and
only if X = V |M .
6.7 The Lie algebra of a Lie group

In the remainder of this chapter we describe an important application of
Lie brackets, in the theory of Lie groups. The multiplication in a group
can be quite difficult to handle algebraically, one reason being that it is not
commutative in general. In a Lie group G we have the extra structure of
a manifold, which allows the application of differential calculus. By means
of this tool we shall develop an algebraic object related to G, called the
Lie algebra of G. The Lie algebra is more easily handled, since it is linear
in nature. In spite of the fact that it is a simpler object, the Lie algebra
contains a wealth of information about the Lie group.
Let G be a Lie group and let g ∈ G. The map x → g (x) = gx of G into
itself is called left translation by g. Obviously, this is a smooth map.
Definition 6.7.1. A vector field X on G is said to be left invariant if
d(g )x (X(x)) = X(gx)
for all elements g and x in G.

A shorter notation for the condition above is dg ◦ X = X ◦ g . It is easily
seen that a linear combination of left invariant vector fields is left invariant.
Example 6.7.1 Let G = Rn with addition. The left translation by an

element y ∈ Rn is the map x → y + x of Rn to itself. The differential of
this map is the identity map of Rn to itself, and hence the condition that
a vector field X: Rn → Rn is left invariant is that X(x) = X(y + x) for all
x ∈ Rn , or in other words, that X is a constant map. When X is viewed
as a differentiation operator,
the ∂condition is that it should have constant
coefficients, that is X = i=1 ai ∂xi where ai ∈ R is constant.
Lemma 6.7. If X, Y are left invariant smooth vector fields on G, then their
Lie bracket [X, Y ] is again left invariant.
Proof. Lemma 6.6.3 is used with φ = g and V = X, W = Y .
Example 6.7.2 Again, let G = Rn with addition. The Lie bracket of two
vector fields with constant coefficients,
∂ ∂
[X, Y ] = [ ai , bj ]=0
i
∂xi j ∂xj
is zero because the partial derivatives commute.
Definition 6.7.2. A Lie algebra is a vector space g over R, equipped with

a map [·, ·]: g × g → g satisfying the properties
[aX + bY, Z] = a[X, Z] + b[Y, Z], [X, aY + bZ] = a[X, Y ] + b[X, Z],
[X, Y ] = −[Y, X],
[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0,
for all X, Y, Z ∈ g and a, b ∈ R.
Example 6.7.3 Let M be a differentiable manifold. Then X(M ) with the

Lie bracket is a Lie algebra.
Example 6.7.4 Let gl(n, R) denote the vector space M(n, R) of all real
n × n matrices, equipped with the commutator bracket [A, B] = AB − BA,
where AB and BA are determined by ordinary matrix multiplication. Then
gl(n, R) is a Lie algebra (the verification is by straightforward computations).
By definition, a Lie subalgebra of a Lie algebra g is a linear subspace h
such that [X, Y ] ∈ h for all X, Y ∈ h.
Corollary 6.7. Equipped with the Lie bracket the space g of left invariant
smooth vector fields on a Lie group G is a algebra.
Proof. It is a Lie subalgebra of X(G).
The set g defined in this corollary is called the Lie algebra of G. The
choice, that the elements in the Lie algebra of G are left invariant vector
92 Chapter 6
fields, is not canonical. It can be shown that the use of right translation
instead of left translation would lead to an isomorphic Lie algebra (but in
general, it would be realized by a different set of vector fields).
Example 6.7.5 Let G = Rn with addition. It follows from Example 6.7.2
that g = Rn , and that the Lie bracket is trivial [X, Y ] = 0 for all X, Y ∈ g.
In the preceding example the Lie group G is commutative, and all the Lie
brackets [X, Y ] are trivial. This is not a coincidence, in fact it can be shown
that a connected Lie group is commutative if and only if its Lie algebra has
trivial brackets (we shall not prove this here). This is a simple instance of
the fact mentioned above, that a lot of information about the Lie group can
be retrieved from its Lie algebra.
6.8 The tangent space at the identity

Let G be a Lie group with Lie algebra g. We shall give a relation, which
describes g and its vector space structure directly in terms of the structure
of G as a differentiable manifold. First, we need a lemma.
Lemma 6.8.1. Every left invariant vector field is smooth.
Proof. We apply Lemma 6.8.2 below. Assume X is a left invariant vector
field on G, then according to Lemma 6.2.2 we must prove that Xf is smooth
for all f ∈ C ∞ (G). Since X is left invariant we obtain with Lemma 3.8.3,
Xf (g) = X(g)(f ) = d(g )e (X(e))(f ) = X(e)(f ◦ g ).
The map F (x, y) = f (xy) is smooth on G × G, and the lemma shows that
g → X(e)(F (g, ·)) is smooth on G. This is exactly the desired conclusion.
Lemma 6.8.2. Let M, N be differentiable manifolds, and let F ∈ C ∞ (M ×
N ). For each x ∈ M , denote by F (x, ·) the smooth function y → F (x, y)
on N . Let q ∈ N and Y ∈ Tq N be fixed, then x → Y (F (x, ·)) is a smooth
function on M .
Proof. Let σ: U → M be a chart, and let γ: I → N be a parametrized curve
through q, which represents Y . The map (u, t) → F (σ(u), γ(t)) belongs to
C ∞ (U × I), hence so does its derivative with respect to t. As a function of
u, this derivative is Y (F (σ(u), ·)).
Theorem 6.8. Let G be a Lie group with Lie algebra g. The evaluation map
X → X(e) is a linear isomorphism of g onto the tangent space Te G.
Proof. The map is clearly linear, and it is injective, for if X(e) = 0 then
X(g) = (dg )e (X(e)) = 0 for all g ∈ G, hence X = 0. Finally, we show that
it is surjective. Let Y ∈ Te G be given, and define X(x) = d(x )e (Y ) for all
x ∈ G. Then X is a left invariant vector field on G, since by the chain rule
d(g )x (X(x)) = d(g )x (d(x )e (Y )) = d(g ◦ x )e (Y ) = d(gx )e (Y ) = X(gx),

where we used the multiplicative property gx = g ◦ x of left multiplication.

It now follows from Lemma 6.8.1 that X is smooth.
Corollary 6.8. The Lie algebra g is finite dimensional, with dimension equal
to that of the manifold G.
6.9 Homomorphisms of Lie groups
Definition 6.9. Let H, G be Lie groups. A map φ: H → G which is smooth

and a homomorphism in the group theoretical sense, that is, which satisfies
φ(xy) = φ(x)φ(y) ∀x, y ∈ H,
is called a homomorphism of Lie groups.
For example, the determinant is a Lie homomorphism from GL(n, R)
to R× . The natural map from SO(2) to GL(2, R) is another example.
Lemma 6.9. Let φ: H → G be a homomorphism of Lie groups, and let h, g
denote the corresponding Lie algebras. For each X ∈ h there exists a unique
element V ∈ g such that V (φ(h)) = dφh (X(h)) for all h ∈ H.
Proof. Let X ∈ h. It follows from Theorem 6.8 that there is a unique left in-
variant vector field V ∈ g such that V (e) = dφe (X(e)). The homomorphism
property of φ implies that φ(h) ◦ φ = φ ◦ h . Hence with the chain rule
d(φ(h) )e (V (e)) = dφh (d(h )e (X(e)).
It follows by left invariance that V (φ(h)) = dφh (X(h)) for all h ∈ H.

We denote the map X → V from h to g by dφ and call it the induced map.
Theorem 6.9. Let φ: H → G be a homomorphism of Lie groups. The
induced map dφ is a homomorphism of Lie algebras, that is, it is linear and
satisfies
dφ([X1 , X2 ]) = [dφ(X1 ), dφ(X2 )] (6.7)
for all X1 , X2 ∈ h.
Proof. The induced map is linear since dφh is linear for each h. Let X ∈ h
and V = dφ(X) ∈ g. Then V (φ(h)) = dφh (X(h)) and φ transforms X to V .
Now (6.7) follows from Lemma 6.6.3.
Assume that H is a Lie subgroup of a Lie group G. Obviously the inclusion
map ι: H → G is a homomorphism of Lie groups. As usual we regard the
tangent space Th H of H as a subspace of Th G, for each h ∈ H, with the
inclusion map dιh . From the theory above we now obtain
Corollary 6.9. If H ⊂ G is a Lie subgroup, then the Lie algebra h of H
admits a natural inclusion as a Lie subalgebra of the Lie algebra g of G.
94 Chapter 6
6.10 The Lie algebra of GL(n, R)

In this section we determine the Lie algebra of the Lie group G = GL(n, R)
(see Example 2.8.4). Recall that it is an n2 -dimensional Lie group with
2
the manifold structure as an open subset of M(n, R) Rn . The tangent
space Tx M(n, R) of M(n, R) at a matrix x is conveniently also identified
with M(n, R), such that the tangent vector in Tx GL(n, R) which corresponds
to a given matrix A ∈ M(n, R) is represented by the curve t → x + tA
in GL(n, R) ⊂ M(n, R). Accordingly, the linear isomorphism g → Te G in
Theorem 6.8 is thus in this case a linear isomorphism g → M(n, R). In
order to describe the Lie algebra of GL(n, R), we need to determine the Lie
bracket, that is, we need to describe the isomorphic image in M(n, R) of
[X, Y ] in terms of the images of X and Y .
Recall from Example 6.7.4 that we equipped the vector space M(n, R),
with the commutator bracket AB − BA. We claim that it is this structure
on M(n, R), which represents the Lie bracket of GL(n, R), that is, if X and
Y are left invariant vector fields on GL(n, R), then [X, Y ](e) = X(e)Y (e) −
Y (e)X(e). In other words, we claim the following.
Theorem 6.10. The Lie algebra of GL(n, R) is gl(n, R).
Proof. Let G = GL(n, R). We first investigate the left invariant vector fields
on G. Let C ∈ M(n, R) and consider the curve t → x + tC, the class of which
represents C as a tangent vector at x ∈ G. For g ∈ G, the left multiplication
g : x → gx transforms our curve into the curve t → gx + tgC, of which the
derivative at t = 0 is gC. We conclude that d(g )x (C) = gC. For a left
invariant vector field X on G it thus follows that X(gx) = gX(x) for all
g, x ∈ G, when the values X(x) are regarded as elements in M(n, R). In
particular, X(g) = gX(e), and it follows that the action of X on a function
f ∈ C ∞ (G) is given by
d
Xf (g) = f (g + tgA)|t=0 (6.8)
dt
when A = X(e).
We now turn to the actual proof. Let X, Y be left invariant vector fields,
and let A = X(e) and B = Y (e). Let T denote the left invariant vector field
for which T (e) = AB − BA. The claim is then that [X, Y ] = T , that is, we
need to show
[X, Y ]f = T f (6.9)
for all f ∈ C ∞ (G).

Let ξ: M(n, R) → R be one of the n2 maps, which takes a matrix to a
given entry. These are the coordinate functions for our chart on GL(n, R).
In Lemma 6.2.1 it was seen that a tangent vector is uniquely determined
by its action on the coordinate functions, hence it follows that it suffices to

prove (6.9) for f = ξ.
We insert f = ξ in (6.8), and observe that ξ is a linear function of the
matrices. Hence,
d
Xξ(g) = (ξ(g) + tξ(gA))|t=0 = ξ(gA). (6.10)
dt
Likewise Y ξ(g) = ξ(gB) and T ξ(g) = ξ(g(AB − BA)).

Next we insert f = Y ξ in (6.8) and obtain
d d
XY ξ(g) = Y ξ(g + tgA)|t=0 = ξ((g + tgA)B)|t=0 = ξ(gAB).
dt dt
Likewise Y Xξ(g) = ξ(gBA), and thus [X, Y ]ξ(g) = ξ(g(AB − BA)). Now
(6.9) follows.
Example 6.10 Let H = O(n). In Section 4.5 we determined the tangent

space TI O(n) at the identity as the set of n × n antisymmetric matrices. We
denote this set by o(n),
o(n) = {A ∈ M(n, R) | At = −A}.
Since O(n) is a Lie subgroup of GL(n, R), it follows from Corollary 6.9 that
o(n) is a Lie subalgebra of the Lie algebra of GL(n, R), and then it follows
from Theorem 6.10 that the Lie bracket of o(n) is the commutator bracket
[A, B] = AB − BA.
6.11 Exercises
1 Let Y denote the vector field Y (x, y) = (x, −y) on R2 . Determine its
coordinates with respect to the chart σ(u, v) = (u + v, u − v).
2 Prove that Y (x, y, z) = (xz, yz, z 2 − 1) is a smooth vector field on the
sphere S 2 .
3 Consider M = R2 . Compute the Lie bracket [X, Y ] when
a) X(x, y) = (x, 0), Y (x, y) = (0, y)
b) X(x, y) = (0, x), Y (x, y) = (y, 0).
4 Consider M = R. Determine the Lie bracket of two arbitrary smooth
vector fields X(x) = f (x) and Y (x) = g(x).
5 Let N ⊂ M be a submanifold of a differentiable manifold, and let V ∈
X(M ). Assume that V (p) ∈ Tp N for all p ∈ N . Prove that the restriction
V |N is a smooth vector field on N .
96 Chapter 6
6 Let N ⊂ M be as above, and let Y ∈ X(N ).

a. Prove that there exists an open set W ⊂ M with N ⊂ W , and a
smooth vector field Ỹ on W whose restriction to N is Y . Prove that if N
is closed, this can be obtained with W = M .
b. Prove that if a function f ∈ C ∞ (N ) is extended to a smooth function
F ∈ C ∞ (M ) (see Exercise 5.9), then Y f is the restriction of Ỹ F .
7 Prove Lemma 6.6.2.
8 Verify the Jacobi identity for the commutator bracket on gl(n, R) (Exam-
ple 6.7.4).
9 a Define [(x, y), (v, w)] = (0, xw − yv) for (x, y) and (v, w) in R2 . Show
that in this fashion R2 is a Lie algebra [Hint: Rather than verifying x ythe

Jacobi identity directly, you can consider 2×2 matrices of the form 0 0 ].
b. Let g be a 2-dimensional Lie algebra for which there are basis vectors
e, f such that [e, f ] = f . Prove that it is isomorphic with the mentioned
Lie algebra on R2 .
c. Prove finally that every 2-dimensional Lie algebra with a non-trivial
bracket (that is, [X, Y ] is not always zero) has such a basis. Conclude that
up to isomorphism there are only two different Lie algebras of dimension 2.
10 Let G be a Lie group. For two subsets A, B ⊂ G let
AB = {xy | x ∈ A, y ∈ B}.
Show that AB is open if A or B is open, and that it is compact if A and

B are compact.
11 Let G be a Lie group, and let H be an open subgroup. Show that H
is closed [Hint: prove that the complement H c satisfies H c = HH c ].
Conclude that a connected Lie group has no open subgroups (besides
itself).
12 Let G be a connected Lie group. Choose a neighborhood W of e, and put
Dn = W n = {x1 · · · xn | x1 , . . . , xn ∈ W }.
Show that G = ∪n Dn . Prove that G has a countable atlas [Hint: Choose

W with compact closure. Now apply Theorem 5.2].
13 Let G be a Lie group, and let G0 be the identity component of G (see
Section 5.10) which contains the neutral element e. Show that G0 is an
open subgroup, and that it is contained in every open subgroup of G.
14 Let G be a commutative Lie group. Prove that its Lie algebra is abelian,
that is, all Lie brackets [X, Y ] are 0 [Hint: g → g −1 is a Lie homomor-
phism].
Chapter 7
Tensors
In this chapter some fundamental constructions for a real vector space

V are introduced: The dual space V ∗ , the tensor spaces T k (V ) and the
alternating tensor spaces Ak (V ). The presentation is based purely on linear
algebra, and it is independent of all the preceding chapters. The bridge to
geometry will be built in the following chapter, where we shall apply the
theory of the present chapter to the study of manifolds. The linear space V
will then be the tangent space Tp M at a given point.
Let V be a vector space over R. For our purposes only finite dimensional
spaces are needed, so we shall assume dim V < ∞ whenever it is convenient.
7.1 The dual space

We recall that the dual space of a vector space is defined as follows.
Definition 7.1. The dual space V ∗ is the space of linear maps ξ: V → R.
A linear map ξ ∈ V ∗ is often called a linear form. Equipped with the
natural algebraic operations of addition and scalar multiplication of maps,
V ∗ becomes a vector space on its own.
The basic theorem about V ∗ , for V finite dimensional, is the following.
Theorem 7.1.1. Assume dim V = n ∈ N, and let e1 , . . . , en be a basis.
(i) For each i = 1, . . . , n, an element ξi ∈ V ∗ is defined by
ξi (a1 e1 + · · · + an en ) = ai , a1 , . . . , an ∈ R.
(ii) The elements ξ1 , . . . , ξn form a basis for V ∗ (called the dual basis).
Proof. (i) is easy. For (ii), notice first that two linear forms on a vector space
are equal, if they agree on each element of a basis. Notice also that it follows
from the definition of ξi that ξi (ej ) = δij . Let ξ ∈ V ∗ , then

n
ξ= ξ(ei )ξi , (7.1)
i=1
because the two sides agree on each ej . This shows that

the vectors ξ1 , . . . , ξn
∗
span V . They are also linearly independent, for if i bi ξi = 0 then bj =

i bi ξi (ej ) = 0 for all j.
98 Chapter 7
Corollary 7.1. If dim V = n then dim V ∗ = n.
If a linear form ξ ∈ V ∗ satisfies that ξ(v) = 0 for all v ∈ V , then by

definition ξ = 0. The similar result for elements v ∈ V needs a proof:
Lemma 7.1.1. Let v ∈ V . If v = 0 then ξ(v) = 0 for some ξ ∈ V ∗ .
Proof. For simplicity we assume dim V < ∞ (although the result is true in
general). Assume v = 0. Then there exists a basis e1 , . . . , en for V with
e1 = v. The element ξ1 of the dual basis satisfies ξ1 (v) = 1.
The dual space is useful for example in the study of subspaces of V . This
can be seen from the following theorem, which shows that the elements of V ∗
can be used to detect whether a given vector belongs to a given subspace.
Lemma 7.1.2. Let U ⊂ V be a linear subspace, and let v ∈ V . If v ∈

/ U
∗
then ξ(v) = 0 for some ξ ∈ V with ξ|U = 0.
Proof. The proof is similar to the previous one, which corresponds to the
special case U = {0}. As before we assume dim V < ∞. Let e1 , . . . , em be
a basis for U , let em+1 = v, and extend to a basis e1 , . . . , en for V . The
element ξm+1 of the dual basis satisfies ξm+1 |U = 0 and ξm+1 (v) = 1.
The term ‘dual’ suggests some kind of symmetry between V and V ∗ . The
following theorem indicates such a symmetry for the finite dimensional case.
We define V ∗∗ = (V ∗ )∗ as the dual of the dual space.
Theorem 7.1.2(duality theorem). The map Φ : V → V ∗∗ defined by
Φ(v)(ξ) = ξ(v)
for v ∈ V , ξ ∈ V ∗ is linear and injective. If dim V < ∞ it is an isomorphism.
Notice that Φ is defined without reference to a particular basis for V .
Proof. It is easily seen that Φ maps into V ∗∗ , and that it is linear. If Φ(v) = 0
then ξ(v) = 0 for all ξ, hence v = 0 by Lemma 7.1.1. Thus Φ is injective.
If dim V < ∞ then dim V ∗∗ = dim V ∗ = dim V , and hence Φ is also surjec-
tive.
7.2 The dual of a linear map
Let V, W be vector spaces over R, and let T : V → W be linear. By

definition, the dual map T ∗ : W ∗ → V ∗ takes a linear form η ∈ W ∗ to its
pull-back by T , that is, T ∗ (η) = η ◦ T . It is easily seen that T ∗ is linear.
Tensors 99
Lemma 7.2.1. Assume V and W are finite dimensional, and let T : V → W

be linear. If T is represented by a matrix (aij ) with respect to some given
bases, then T ∗ is represented by the transposed matrix (aji ) with respect to
the dual bases.
Proof. Let e1 , . . . , en and f1 , . . . , fm denote the given bases for V and W .
The fact that T is represented by aij is expressed in the equality

T ej = aij fi .
i
Let ξ1 , . . . , ξn and η1 , . . . , ηm denote the dual bases for V ∗ and W ∗ . Then

aij = ηi (T ej ).
We now obtain from (7.1) with ξ = T ∗ ηk that

T ∗ ηk = T ∗ ηk (ej )ξj = ηk (T ej )ξj = akj ξj .
j j j
Lemma 7.2.2. (i) T is surjective if and only if T ∗ is injective.

(ii) T ∗ is surjective if and only if T is injective.
Proof. Assume for simplicity that V and W are finite dimensional. It is
known from linear algebra that a map represented by a matrix is surjective if
and only if the rank of the matrix equals the number of rows, and injective if
and only if the rank equals the number of columns. The lemma follows from
the preceding, since the rank of (aij ) is equal to the rank of its transpose.
7.3 Tensors
We now proceed to define tensors. Let k ∈ N. Given a collection of
vector spaces V1 , . . . , Vk one can define a vector space V1 ⊗ · · · ⊗ Vk , called
their tensor product. The elements of this vector space are called tensors.
However, we do not need to work in this generality, and we shall be content
with the situation where the vector spaces V1 , . . . , Vk are all equal to the
same space. In fact, the tensor space T k V we define below corresponds to
V ∗ ⊗ · · · ⊗ V ∗ in the general notation.
Definition 7.3.1. Let V k = V × · · · × V be the Cartesian product of k
copies of V . A map ϕ from V k to a vector space U is called multilinear if it
is linear in each variable separately (i.e. with the other variables held fixed).
Definition 7.3.2. A (covariant) k-tensor on V is a multilinear map T : V k →

R. The set of k-tensors on V is denoted T k (V ).
In particular, a 1-tensor is a linear form, T 1 (V ) = V ∗ . It is convenient to
add the convention that T 0 (V ) = R. The set T k (V ) is called tensor space, it
is a vector space because sums and scalar products of multilinear maps are
again multilinear.
100 Chapter 7
Lemma 7.3. Let η1 , . . . , ηk ∈ V ∗ . The map
(v1 , . . . , vk ) → η1 (v1 ) · · · ηk (vk ) (7.2)
is a k-tensor.
Proof. This is clear.
The map (7.2) is denoted η1 ⊗ · · · ⊗ ηk . More generally, if S ∈ T k (V ) and

T ∈ T l (V ) are tensors, we define the tensor product S ⊗ T ∈ T k+l (V ) by
S ⊗ T (v1 , . . . , vk , vk+1 , . . . , vk+l ) = S(v1 , . . . , vk )T (vk+1 , . . . , vk+l ).
It is easily seen that S ⊗ T is a k + l-tensor, and that the algebraic operation

⊗ between tensors is associative and satisfies the proper distributive laws of
a product. The associativity is expressed through
(R ⊗ S) ⊗ T = R ⊗ (S ⊗ T )
and the content of the distributive law is that S ⊗ T depends linearly on

S and T . Because of the associativity no brackets are needed in the tensor
product of three or more tensors. This explains the notation η1 ⊗ · · · ⊗ ηk .
Notice that for k = 0 the tensor S ∈ T k (V ) is a number. In this case (and
similarly when l = 0) the tensor product S ⊗ T is defined by multiplication
of T with that number.
Theorem 7.3. Assume dim V = n with e1 , . . . , en a basis. Let ξ1 , . . . , ξn ∈

V ∗ denote the dual basis. The elements ξi1 ⊗ · · · ⊗ ξik , where I = (i1 , . . . , ik )
is an arbitrary sequence of k numbers in {1, . . . , n}, form a basis for T k (V ).
Proof. Let TI = ξi1 ⊗ · · · ⊗ ξik . Notice that if J = (j1 , . . . , jk ) is another

sequence of k integers, and we denote by eJ the element (ej1 , . . . , ejk ) ∈ V k ,
then
TI (eJ ) = δIJ ,
that is, TI (eJ ) = 1 if J = I and 0 otherwise. It follows

that the TI are
linearly independent, for if a linear combination T = I aI TI is zero, then
aJ = T (eJ ) = 0.
It follows from the multilinearity that a k-tensor is uniquely determined
by its values on all elements inV k of the form eJ . For any given k-tensor
T we have that the k-tensor I T (eI )TI agrees with T on all eJ , hence

T = I T (eI )TI and we conclude that the TI span T k (V ).
Tensors 101
Corollary 7.3. If dim V = n then dim T k (V ) = nk .

It follows from Theorem 7.3 that every element of T k (V ) is a sum of tensor
products of the form η1 ⊗ · · · ⊗ ηk , where η1 , . . . , ηk ∈ V ∗ . For this reason it
is customary to denote T k (V ) = V ∗ ⊗ · · · ⊗ V ∗ .
Remark (especially for physicists). The coefficients of a vector v ∈ V with
respect to a basis v1 , . . .
, vn are often numbered with superscripts instead of
subscripts, so that v = ai vi . Furthermore, the elements of the dual basis
are numbered with superscripts, ξ 1 , . . . , ξ n , and the coefficients of a vector
ξ ∈ V∗ with respect to this basis are then numbered with subscripts so that
ξ= bi ξ i . Then
ξ(v) = a i bi .
These rules are adapted to the so-called ”Einstein convention” which is to
sum over an index which appears both above and below.
The coefficients of a covariant tensor are numbered with subscripts, so
that the expression by means of the basis in Theorem 7.3 becomes

T = ai1 ,...,ik ξ i1 ⊗ · · · ⊗ ξ ik .
There is a similar construction of so-called contravariant tensors, which have
superscript indices, and also mixed tensors which carry both types of indices,
but we do not need them here. In the physics literature, a tensor T is often
identified with the array of coefficients aji11,...,i
,...,jl
k
with respect to a particular
basis. The array is then required to obey a transformation rule for the passage
between different bases.
7.4 Alternating tensors

Let V be a real vector space. In the preceding section the tensor spaces
k
T (V ) were defined, together with the tensor product
(S, T ) → S ⊗ T, T k (V ) × T l (V ) → T k+l (V ).
There is an important construction of vector spaces which resemble tensor
powers of V , but for which there is a more refined structure. These are the
so-called exterior powers of V , which play an important role in differential
geometry because the theory of differential forms is built on them. They are
also of importance in algebraic topology and many other fields.
A multilinear map
ϕ: V k = V × · · · × V → U
where k > 1, is said to be alternating if for all v1 , . . . , vk ∈ V the value
ϕ(v1 , . . . , vk ) changes sign whenever two of the vectors v1 , . . . , vk are inter-
changed, that is
ϕ(v1 , . . . , vi , . . . , vj , . . . , vk ) = −ϕ(v1 , . . . , vj , . . . , vi , . . . , vk ). (7.3)
102 Chapter 7
Since every permutation of the numbers 1, . . . , k can be decomposed into

transpositions, it follows that
ϕ(vσ(1) , . . . , vσ(k) ) = sgn σ ϕ(v1 , . . . , vk ) (7.4)
for all permutations σ ∈ Sk of the numbers 1, . . . , k.

Examples 7.4.1. 1. Let V = R3 . The vector product (v1 , v2 ) → v1 × v2 ∈ V
is alternating from V × V to V .
2. Let V = Rn . The n × n determinant is multilinear and alternating in
its columns, hence it can be viewed as an alternating map (Rn )n → R.
Lemma 7.4.1. Let ϕ: V k → U be multilinear. The following conditions are
equivalent:
(a) ϕ is alternating,
(b) ϕ(v1 , . . . , vk ) = 0 whenever two of the vectors v1 , . . . , vk coincide,
(c) ϕ(v1 , . . . , vk ) = 0 whenever the vectors v1 , . . . , vk are linearly depen-
dent.
Proof. (a)⇒(b) If vi = vj the interchange of vi and vj does not change the
value of ϕ(v1 , . . . , vk ), so (7.3) implies ϕ(v1 , . . . , vk ) = 0.
(b)⇒(a) Consider for example the interchange of v1 and v2 . By linearity
0 = ϕ(v1 + v2 , v1 + v2 , . . . )
= ϕ(v1 , v1 , . . . ) + ϕ(v1 , v2 , . . . ) + ϕ(v2 , v1 , . . . ) + ϕ(v2 , v2 , . . . )
= ϕ(v1 , v2 , . . . ) + ϕ(v2 , v1 , . . . ).
It follows that ϕ(v2 , v1 , . . . ) = −ϕ(v1 , v2 , . . . ).

(b)⇒(c) If the vectors v1 , . . . , vk are linearly dependent then one of them
can be written as a linear combination of the others. It then follows that
ϕ(v1 , . . . , vk ) is a linear combination of terms in each of which some vi appears
twice.
(c)⇒(b) Obvious.
In particular, if k > dim V then every set of k vectors is linearly dependent,
and hence ϕ = 0 is the only alternating map V k → U .
Definition 7.4. An alternating k-form is an alternating k-tensor V k → R.
The space of these is denoted Ak (V ), it is a linear subspace of T k (V ).
We define A1 (V ) = V ∗ and A0 (V ) = R.
Example 7.4.2. Let η, ζ ∈ V ∗ . The 2-tensor η ⊗ ζ − ζ ⊗ η is alternating.
The following lemma exhibits a standard procedure to construct alternat-

ing forms.
Tensors 103
Lemma 7.4.2. Let k > 1. For each T ∈ T k (V ) the element Alt(T ) ∈ T k (V )

defined by
1
Alt(T )(v1 , . . . , vk ) = sgn σ T (vσ(1) , . . . , vσ(k) ) (7.5)
k!
σ∈Sk
is alternating. Moreover, if T is already alternating, then Alt(T ) = T .

Proof. Let τ ∈ Sk be the transposition corresponding to an interchange of
two vectors among v1 , . . . , vk . We have
1
Alt(T )(vτ (1) , . . . , vτ (k) ) = sgn σ T (vτ ◦σ(1) , . . . , vτ ◦σ(k) ).
k!
σ∈Sk
Since σ → τ ◦σ is a bijection of Sk we can substitute σ for τ ◦σ. Using sgn(τ ◦

σ) = − sgn(σ), we obtain the desired equality with − Alt(T )(v1 , . . . , vk ).
If T is already alternating, all summands of (7.5) are equal to T (v1 , . . . , vk ).
Since |Sk | = k! we conclude that Alt(T ) = T .
The definition in Lemma 7.4.2 for k > 1 is supplemented with the following
convention
Alt(T ) = T, T ∈ A0 (V ) = R or T ∈ A1 (V ) = V ∗ .
Notice that for η1 , . . . , ηk ∈ V ∗ and v1 , . . . , vk ∈ V we obtain

1
Alt(η1 ⊗ · · · ⊗ ηk )(v1 , . . . , vk ) = sgn σ η1 (vσ(1) ) · · · ηk (vσ(k) )
k!
σ∈Sk
and hence
1
Alt(η1 ⊗ · · · ⊗ ηk )(v1 , . . . , vk ) = det[(ηi (vj ))ij ]. (7.6)
k!
Let e1 , . . . , en be a basis for V , and ξ1 , . . . , ξn the dual basis for V ∗ . We
saw in Theorem 7.3 that the elements ξi1 ⊗ · · · ⊗ ξik form a basis for T k (V ).
We will now exhibit a similar basis for Ak (V ). We have seen already that
Ak (V ) = 0 if k > n.
Theorem 7.4. Assume k ≤ n. For each subset I ⊂ {1, . . . , n} with k
elements, let 1 ≤ i1 < · · · < ik ≤ n be its elements, and let
ξI = Alt(ξi1 ⊗ · · · ⊗ ξik ) ∈ Ak (V ). (7.7)
These elements ξI form a basis for Ak (V ). In particular,
n!
dim Ak (V ) = .
k!(n − k)!
104 Chapter 7
Proof. It follows from the last statement in Lemma 7.4.2 that Alt: T k (V ) →
Ak (V ) is surjective. Applying Alt to all the basis elements ξi1 ⊗ · · · ⊗ ξik for
T k (V ), we therefore obtain a spanning set for Ak (V ). It follows from (7.6)
that Alt(ξi1 ⊗ · · · ⊗ ξik ) = 0 if there are repetitions among the i1 , . . . , ik .
Moreover, if we rearrange the order of the numbers i1 , . . . , ik , the element
Alt(ξi1 ⊗ · · · ⊗ ξik ) is unchanged, apart from a possible change of the sign.
Therefore Ak (V ) is spanned by the elements ξI in (7.7).
Consider an arbitrary linear combination T = I aI ξI with coefficients
aI ∈ R. Let J = (j1 , . . . , jk ) where 1 ≤ j1 < · · · < jk ≤ n, then it follows
from (7.6) that

1/k! if I = J
ξI (ej1 , . . . , ejk ) =
0 otherwise
It follows that T (ej1 , . . . , ejk ) = aJ /k! for J = (j1 , . . . , jk ). Therefore, if

T = 0 we conclude aJ = 0 for all the coefficients. Thus the elements ξI are
independent.
Notice the special case k = n in the preceding theorem. It follows that
An (V ) is one-dimensional, and spanned by the map
(v1 , . . . , vn ) → det[(ξi (vj ))ij ]
for an arbitrary (dual) basis for V ∗ (thus demonstrating the uniqueness of

the determinant as an alternating n-multilinear map).
7.5 The wedge product

In analogy with the tensor product (S, T ) → S ⊗ T , from T k (V ) × T l (V )
to T k+l (V ), there is a construction of a product Ak (V ) × Al (V ) → Ak+l (V ).
Since tensor products of alternating tensors are not alternating, it does not
suffice just to take S ⊗ T .
Definition 7.5. Let S ∈ Ak (V ) and T ∈ Al (V ). The wedge product S ∧ T ∈
Ak+l (V ) is defined by
S ∧ T = Alt(S ⊗ T ). (7.8)
Notice that in the case k = 0, where Ak (V ) = R, the wedge product is

just scalar multiplication.
In these notes we use the simple normalization (7.8) for the wedge product.
Another convention is also common, according to which the right hand side
of (7.8) is replaced by
(k + l)!
Alt(S ⊗ T ).
k! l!
With the unwieldy factor in front at this place some later formulas would
become simpler (most prominently, the factorial k! would disappear in (7.9)).
Tensors 105
Example 7.5 Let η1 , η2 ∈ A1 (V ) = V ∗ . Then by definition
1
η1 ∧ η2 = (η1 ⊗ η2 − η2 ⊗ η1 ).
2
Since the operator Alt is linear, the wedge product depends linearly on
the factors S and T . It is more cumbersome to verify the associative rule for
∧. In order to do this we need the following lemma.
Lemma 7.5.1. Let S ∈ T k (V ) and T ∈ T l (V ). Then
Alt(Alt(S) ⊗ T ) = Alt(S ⊗ Alt(T )) = Alt(S ⊗ T ).
Proof. We will only verify
Alt(Alt(S) ⊗ T ) = Alt(S ⊗ T ).
The proof for the other expression is similar.

Let G = Sk+l and let H ⊂ G denote the subgroup of permutations leaving
each of the last elements k+1, . . . , k+l fixed. Then H is naturally isomorphic
to Sk . Now
Alt(Alt(S) ⊗ T )(v1 , . . . , vk+l )

1
= (k+l)! sgn σ Alt(S)(vσ(1) , . . . , vσ(k) ) T (vσ(k+1) , . . . , vσ(k+l) )
σ∈Sk+l

1 1
= (k+l)! sgn σ k! sgn τ S(vσ(τ (1)) , . . . , vσ(τ (k)) )
σ∈Sk+l τ ∈Sk
T (vσ(k+1) , . . . , vσ(k+l) )

1
= (k+l)!k! sgn(σ ◦ τ ) S(vσ(τ (1)) , . . . , vσ(τ (k)) )
σ∈G τ ∈H
T (vσ(τ (k+1)) , . . . , vσ(τ (k+l)) ).
Since σ → σ ◦ τ is a bijection of G we can substitute σ for σ ◦ τ , and we

obtain the desired expression, since there are k! elements in H.
Lemma 7.5.2. Let R ∈ Ak (V ), S ∈ Al (V ) and T ∈ Am (V ). Then
(R ∧ S) ∧ T = R ∧ (S ∧ T ) = Alt(R ⊗ S ⊗ T ).
Proof. It follows from the preceding lemma that
(R ∧ S) ∧ T = Alt(Alt(R ⊗ S) ⊗ T ) = Alt(R ⊗ S ⊗ T )
106 Chapter 7
and
R ∧ (S ∧ T ) = Alt(R ⊗ Alt(S ⊗ T )) = Alt(R ⊗ S ⊗ T ).
Since the wedge product is associative, we can write any product T1 ∧· · ·∧

Tr of tensors Ti ∈ Aki (V ) without specifying brackets. In fact, it follows by
induction from Lemma 7.5.1 that
T1 ∧ · · · ∧ Tr = Alt(T1 ⊗ · · · ⊗ Tr )
regardless of how brackets are inserted in the wedge product.

In particular, it follows from (7.6) that
1
η1 ∧ · · · ∧ ηk (v1 , . . . , vk ) = det[(ηi (vj ))ij ] (7.9)
k!
for all v1 , . . . , vk ∈ V and η1 , . . . , ηk ∈ V ∗ , where the vectors in V ∗ are viewed

as 1-forms. The basis elements ξI in Theorem 7.4 are written in this fashion
as
ξI = ξi1 ∧ · · · ∧ ξik
where I = (i1 , . . . , ik ) is an increasing sequence from 1, . . . , n. This will be

our notation for ξI from now on.
The wedge product is not commutative. Instead, it satisfies the following
relation for the interchange of factors.
Lemma 7.5.3. Let η, ζ ∈ V ∗ , then
ζ ∧ η = −η ∧ ζ. (7.10)
More generally, if S ∈ Ak (V ) and T ∈ Al (V ) then
T ∧ S = (−1)kl S ∧ T (7.11)
Proof. The identity (7.10) follows immediately from the fact that η ∧ ζ =
1
2 (η ⊗ ζ − ζ ⊗ η).
Since Ak (V ) is spanned by elements of the type S = η1 ∧ · · · ∧ ηk , and
A (V ) by elements of the type T = ζ1 ∧ · · · ∧ ζl , where ηi , ζj ∈ V ∗ , it suffices
l
to prove (7.11) for these forms. In order to rewrite T ∧ S as S ∧ T we must

let each of the k elements ηi pass the l elements ζj . The total number of sign
changes is therefore kl.
Tensors 107
7.6 The exterior algebra

Let
A(V ) = A0 (V ) ⊕ A1 (V ) ⊕ · · · ⊕ An (V )
be the direct sum of all the spaces of alternating forms on V . Notice that the
general elements in A(V ) are formal sums of k-forms with different values
of k for the components, and it is only the individual components that are
multilinear, alternating maps.
We extend the wedge product ∧ to A(V ) by linearity, and in this fashion
we obtain an algebra, called the exterior algebra of V . The element 1 ∈ A0 (V )
is a unit for this algebra. The exterior algebra has a linear basis consisting of
all the elements ξI , where I is an arbitrary subset of {1, . . . , n}. Its dimension
is
dim A(V ) = 2n .
7.7 Exercises
1 Let V = R2 , let e1 , e2 be the canonical basis vectors and ξ1 , ξ2 the dual

basis for V ∗ . Another basis for V is given by the vectors v1 = (1, 0),
v2 = (1, −1). Determine the dual basis of that, in terms of ξ1 and ξ2
2 Let V be finite dimensional. Prove that every basis for V ∗ is dual of some
basis for V .
3 Let V be a 2-dimensional vector space with basis e1 , e2 and dual basis
ξ1 , ξ2 .
a. Define a tensor S ∈ T 2 (V ) by
S = ξ1 ⊗ ξ1 + ξ1 ⊗ ξ2 − ξ2 ⊗ ξ1 − ξ2 ⊗ ξ2
Determine η1 , η2 in V ∗ such that S = η1 ⊗ η2 .

b. A 2-tensor S ∈ T 2 (V ) for which it is possible to find η1 , η2 in V ∗ such
that S = η1 ⊗ η2 is said to be pure. Show that if
S = aξ1 ⊗ ξ1 + bξ1 ⊗ ξ2 + cξ2 ⊗ ξ1 + dξ2 ⊗ ξ2
is pure, then ad − bc = 0.
c. Find a 2-tensor which is not pure. Is the set of pure 2-tensors a linear
subspace in T 2 (V )?
4 For two vector spaces V, W , we denote by Hom(V, W ) the space of lin-
ear maps from V to W . For example, V ∗ = Hom(V, R). Determine an
isomorphism between T k (V ) and Hom(V, T k−1 (V )).
108 Chapter 7
5 If a multilinear map F : (W ∗ )k → V is given, show that there exists a

unique element T ∈ Hom(T k (W ), V ) such that
T (η1 ⊗ · · · ⊗ ηk ) = F (η1 , ..., ηk )
for all sets of k vectors η1 , ..., ηk in W ∗ .
In the following exercises V is a finite dimensional vector space with basis

e1 , . . . , en , and ξ1 , . . . , ξn is the dual basis for V ∗ .
6 Verify that (ξ1 +2ξ2 −ξ4 )∧(3ξ2 +ξ3 ) = (2ξ1 +ξ2 −ξ3 −2ξ4 )∧(−ξ1 +ξ2 +ξ3 +ξ4 )
7 Write ξ1 ∧ ξ2 + ξ2 ∧ ξ3 + ξ3 ∧ ξ1 in the form η ∧ ζ where η, ζ ∈ V ∗ .
8 Assume dim V = 3. Show that every element in A2 (V ) can be written in
the form η ∧ ζ where η, ζ ∈ V ∗ .
9 Show that the map (η1 , . . . , ηk ) → η1 ∧ · · · ∧ ηk is multilinear and alter-
nating. from (V ∗ )k to Ak (V ).
10 Let η1 , . . . , ηk ∈ V ∗ . Show that these vectors are linearly independent if
and only if their wedge product η1 ∧ · · · ∧ ηk is non-zero [Hint: For one
direction, apply the previous exercise. For the other, arrange that the
vectors are part of a basis, and apply Theorem 7.4]
11 Let f : V ∗ → V ∗ be a linear map, and let A denote its matrix with respect
to the basis ξ1 , . . . , ξn . Show that
f (ξ1 ) ∧ · · · ∧ f (ξn ) = det A ξ1 ∧ · · · ∧ ξn
Suppose ξ˜1 . . . , ξñ is a second basis for V ∗ . Show that the bases ξ1 , . . . , ξn
and ξ˜1 . . . , ξ̃n are equally oriented if and only if ξ˜1 ∧ · · · ∧ ξñ = cξ1 ∧ · · ·∧ ξn
with c > 0.
Chapter 8
Differential forms
customary to write the integral of a real function f : R → R in the

m
It is
form f (x) dx1 . . . dxm . The quantity dx1 . . . dxn in the formula is regarded
just as formal notation, which reminds us of the fact that in the definition of
the Riemann integral (for functions of one variable), a limit is taken where
the increments Δx tend to zero. Thus dx is regarded as an ‘infinitesimal’
version of Δx. In the theory of differential forms the infinitesimal quantity
is replaced by an object dx1 ∧ · · · ∧ dxm which we shall see has a precise
meaning as a k-form, acting on elements of the tangent space.
The main motivation is to develop theories of differentiation and integra-
tion which are valid for smooth manifolds without reference to any particular
charts.
8.1 The cotangent space

Let M be an m-dimensional abstract manifold, and let p ∈ M .
Definition 8.1. The dual space (Tp M )∗ of Tp M is denoted Tp∗ M and called
the cotangent space of M at p. Its elements are called cotangent vectors or
just covectors.
By now we have become accustomed to the notion of tangent vectors, and
we draw them as arrows. A cotangent vector associates a number in R with
each tangent vector. Thus one can think of a covector as an ”animal, which
feeds on arrows”. By lack of better visualization, we draw it as follows:
a covector
Example 8.1 Consider a function f ∈ C ∞ (M ). Its differential dfp at p ∈ M

is a linear map from Tp M into R (see Example 3.8.2). Hence
dfp ∈ Tp∗ M.
We recall that dfp (X) = Dp,X f = (f ◦ γ) (t0 ) for X ∈ Tp M , if γ is a

parametrized curve through γ(t0 ) = p, representing X.
110 Chapter 8
Given a chart σ: U → M where U ⊂ Rm is open, we can now give dxi a

precise meaning as follows. Recall from Lemma 6.2.1 that we defined the i-th
coordinate function ξi : σ(U ) → R as the smooth map which takes p = σ(u)
to the i-th coordinate ui of u ∈ U . At this point it is convenient to change
notation, and denote the coordinate function simply by xi . Thus
σ −1 (p) = (x1 (p), . . . , xm (p)) ∈ U.
The change of notation is motivated by the desire to give dxi the precise
meaning, which it now obtains as the differential of xi at p. In order to
avoid the double subscript of d(xi )p , we denote this differential by dxi (p).
According to Example 8.1 it is a covector,
dxi (p) ∈ Tp∗ M,
for each p ∈ σ(U ).

Lemma 8.1. Let σ: U → M be a chart on M , and let p = σ(u) ∈ σ(U ).
The tangent vectors
dσu (e1 ), . . . , dσu (em ) (8.1)
form a basis for Tp M . The dual basis for Tp∗ M is
dx1 (p), . . . , dxm (p). (8.2)
Thus, dxi (p) is the linear form on Tp M , which carries a vector to its i-th
coordinate in the basis (8.1).
Proof. The first statement is just a repetition of the fact that the standard
basis is a basis (see Section 3.9). For the second statement, we notice that
by the chain rule
∂ui
dxi (p)(dσu (ej )) = d(xi ◦ σ)u (ej ) = = δij , (8.3)
∂uj
since xi ◦ σ(u) = ui .
8.2 Covector fields

Recall that a vector field Y on M associates a tangent vector Y (p) ∈ Tp M
to each point p ∈ M . Similarly, a covector field associates a covector to each
point. If we visualize a vector field as a collection of tangent arrows, one
based in each point on M , then the corresponding picture of a covector field
is that it is a collection of arrow-eating animals, one in each point of M .
In this section we will define what it means for a covector field to be
smooth. The definition is analogous to Definition 6.1.2.
Differential forms 111
a field of covectors
Definition 8.2. A covector field ξ on M is an assignment of a covector
ξ(p) ∈ Tp∗ M
to each p ∈ M . It is called smooth if, for each chart σ: U → M in a given

atlas of M , there exist smooth functions a1 . . . , am ∈ C ∞ (σ(U )) such that

ξ(p) = ai (p)dxi (p) (8.4)
i
for all p ∈ σ(U ). The space of smooth covector fields is denoted X∗ (M ).

It follows from the next lemma that the notion of smoothness is indepen-
dent of the atlas.
Notice that a covector ξ(p) can be applied to tangent vectors in Tp M . If
Ω ⊂ M is open and Y a vector field on Ω, it thus makes sense to apply a
covector field ξ to Y . The result is a function, ξ(Y ) on Ω, given by
ξ(Y )(p) = ξ(p)(Y (p)) ∈ R.
Lemma 8.2.1. Let ξ be a covector field on M . Then ξ is smooth if and

only if for each open set Ω ⊂ M and each smooth vector field Y ∈ X(Ω), the
function ξ(Y ) belongs to C ∞ (Ω).
Proof. Assume first that ξ is smooth. Let Ω be open, and let σ be a chart
(of the given atlas). For each Y ∈ X(Ω) we can write Y (p) = bj (p)dσu (ej )
for p ∈ Ω ∩ σ(U ), with coefficients bj (p) that are smooth functions of p. If
we write ξ according to (8.4), it follows from (8.3) that

ξ(Y )(p) = ai (p)dxi (p)( bj (p)dσu (ej )) = ai (p)bi (p),
i j i
and hence ξ(Y ) is smooth.

Conversely, assume that ξ(Y ) ∈ C ∞ (Ω) for all Ω and all Y ∈ X(Ω).
In particular, we can take Ω = σ(U ) and Y = dσ(ei ) for a given chart,
112 Chapter 8
and conclude that the function ai defined by ai (p) = ξ(p)(dσu(ei )) depends

smoothly on p. By applying (7.1) to the covector ξ(p) and the dual bases
(8.1)-(8.2) in Lemma 8.1 we see that (8.4) holds. Hence ξ is smooth.
Notice that if ξ is a smooth covector field on M , and ϕ ∈ C ∞ (M ) a smooth
function, then the covector field ϕξ defined by the pointwise multiplication
(ϕξ)(p) = ϕ(p)ξ(p) for each p, is again smooth. This follows from Definition
8.2, since the coefficients of ϕξ in (8.4) are just those of ξ, multiplied by ϕ.
If f ∈ C ∞ (M ) then the differential df is a covector field, as it is explained
in Example 8.1.
Lemma 8.2.2. If f ∈ C ∞ (M ) then df is a smooth covector field on M . For
each chart σ on M , the expression for df by means of the basis (8.2) is
m
∂(f ◦ σ)
df = dxi . (8.5)
i=1
∂ui
The differentials satisfy the rule d(f g) = g df + f dg for all f, g ∈ C ∞ (M ).

Proof. The smoothness of df follows from (8.5), because the coefficient func-
tions are smooth. The rule for the differential of f g also follows from (8.5),
by means of the Leibniz rule for the partial differentiation of a product. Thus
we only need to prove this equality.
◦σ)
In order to do that we recall that dfp (dσu (ej )) = ∂(f
∂uj (u) (see (3.21)).
By applying (7.1) to the covector dfp and the dual bases (8.1)-(8.2) we obtain
the following expression from which (8.5) then follows

dfp = dfp (dσu (ei ))dxi (p).
i
Example 8.2 Let M = R2 with coordinates (x1 , x2 ), and let f ∈ C ∞ (R2 ).

Then the differential of f is given by (8.5) as
∂f ∂f
df = dx1 + dx2 ,
∂x1 ∂x2
A smooth covector field ξ = a1 dx1 + a2 dx2 with a1 , a2 ∈ C ∞ (R2 ) is said to

be exact if it has the form df for some function f . It follows immediately
from the identity
∂ 2f ∂ 2f
= (8.6)
∂x1 ∂x2 ∂x2 ∂x1
that an exact covector field is necessarily closed, that is, it satisfies
∂a1 ∂a2
= . (8.7)
∂x2 ∂x1
Conversely, let a closed form ξ = a1 dx1 + a2 dx2 on R2 be given. We

claim that ξ is exact. Choose a function f1 ∈ C ∞ (R2 ) with ∂x ∂f1
= a1 (for
x1 1
example, f1 (x1 , x2 ) = 0 a1 (t, x2 ) dt). It follows from (8.7) and (8.6) that
2
∂a2
∂x1
= ∂x∂1 ∂x
f1
2
. We deduce that a2 − ∂x ∂f1
2
does not depend on x1 , and choose
a one-variable function f2 ∈ C (R) with f2 = a2 − ∂x
∞ ∂f1
2
, as a function of x2 .
∂f
It is easily seen that f (x1 , x2 ) = f1 (x1 , x2 ) + f2 (x2 ) satisfies both a1 = ∂x 1
∂f
and a2 = ∂x 2
. Hence ξ is exact as claimed.
The analogous result is not valid in an arbitrary 2-dimensional manifold
(for example, it fails on M = R2 \ {(0, 0)}).
8.3 Differential forms

The notion of a covector field can be generalized to a notion of fields of
k-forms. Recall from Definition 7.4 that we denote by Ak (V ) the space of
alternating k-tensors on the vector space V . Recall also that A0 (V ) = R and
A1 (V ) = V ∗ , and if ξ1 , . . . , ξm is a basis for V ∗ , then a basis for Ak (V ) is
obtained from the set of elements
ξi1 ∧ · · · ∧ ξik
where 1 ≤ i1 < · · · < ik ≤ m (see Theorem 7.4).

Definition 8.3.1. A k-form ω on M is an assignment of an element
ω(p) ∈ Ak (Tp M )
to each p ∈ M .
In particular, given a chart σ: U → M , the elements dxi1 ∧ · · · ∧ dxik ,
where 1 ≤ i1 < · · · < ik ≤ m, are k-forms on the open subset σ(U ) of M .
For each p ∈ σ(U ), a basis for Ak (Tp M ) is obtained from these elements.
Therefore, every k-form ω on M has a unique expression on σ(U ),

ω= aI dxi1 ∧ · · · ∧ dxik
I={i1 ,...,ik }
where aI : σ(U ) → R.
Definition 8.3.2. We call ω smooth if all the functions aI are smooth, for
each chart σ in an atlas of M . A smooth k-form is also called a differential
k-form. The space of differential k-forms on M is denoted Ak (M ).
In particular, since A0 (Tp M ) = R, we have A0 (M ) = C ∞ (M ), that is, a
differential 0-form on M is just a smooth function ϕ ∈ C ∞ (M ). Likewise, a
differential 1-form is nothing but a smooth covector field.
114 Chapter 8
It can be verified, analogously to Lemma 8.2.1, that a k-form ω is smooth if

and only if ω(X1 , . . . , Xk ) ∈ C ∞ (Ω) for all open sets Ω and all X1 , . . . , Xk ∈
X(Ω). In particular, it follows that the notion of smoothness of a k-form is
independent of the chosen atlas for M .
Example 8.3 Let S be an oriented surface in R3 . The volume form on S is
the 2-form defined by ωp (X1 , X2 ) = 12 (X1 × X2 ) · Np for X, Y ∈ Tp S, where
Np ∈ R3 is the positive unit normal of S at p. In the local coordinates of a
chart, it is given by
ω = (EG − F 2 )1/2 dx1 ∧ dx2
where E, F and G are the coefficients of the first fundamental form. The
volume form is smooth, since these coefficients are smooth functions.
Lemma 8.3. Let ω be a k-form on M and ϕ a real function on M , and
define the product ϕω pointwise by
ϕω(p) = ϕ(p)ω(p).
Then ϕω is a smooth k-form if ϕ and ω are smooth.
Let θ be an l-form on M , and define the wedge product θ ∧ ω pointwise by
(θ ∧ ω)(p) = θ(p) ∧ ω(p)
for each p. Then θ ∧ ω is a smooth k + l-form if ω and θ are smooth.
Proof. Easy.
8.4 Pull back

An important property of differential forms is that they can be pulled back
by a map from one manifold to another. The construction is analogous to
that of the dual map in Section 7.2.
Definition 8.4. Let f : M → N be a smooth map of manifolds, and let
ω ∈ Ak (N ). We define f ∗ ω ∈ Ak (M ), called the pull back of ω, by
f ∗ ω(p)(v1 , . . . , vk ) = ω(f (p))(dfp (v1 ), . . . , dfp (vk ))
for all v1 , . . . , vk ∈ Tp M and p ∈ M .
It is easily seen that for each p ∈ M the operator f ∗ ω(p) so defined is
multilinear and alternating, so that it belongs to Ak (Tp M ). We have to
verify that it depends smoothly on p. This will be done in the following
lemma. It is clear that f ∗ ω depends linearly on ω, this will be used in the
proof of the lemma.
Let
ω= aI dyi1 ∧ · · · ∧ dyik
I={i1 ,...,ik }
be the expression for ω by means of the coordinates of a chart τ : V → N
on N . Here yi is the i-th coordinate of τ −1 for i = 1, . . . , n, and I is an
increasing sequence of k numbers among 1, . . . , n.
Lemma 8.4.1. With notation from above,

f ∗ω = (aI ◦ f ) d(yi1 ◦ f ) ∧ · · · ∧ d(yik ◦ f ) (8.8)
I
on f −1 (τ (V )). In particular, f ∗ ω is smooth.

Proof. In view of Lemma 8.3, the last statement follows from (8.8), since all
the ingredients in this expression are smooth, and since M is covered by the
open subsets f −1 (τ (V )), for all τ in an atlas of N .
In order to verify (8.8), it suffices to prove the following rules for pull
backs:
f ∗ (ϕω) = (ϕ ◦ f )f ∗ ω, (8.9)
f ∗ (θ ∧ ω) = f ∗ (θ) ∧ f ∗ (ω), (8.10)
∗
f (dϕ) = d(ϕ ◦ f ). (8.11)
These rules follow easily from the definitions.

In particular, Definition 8.4 allows us to define the restriction of a k-form
on a manifold N to a submanifold M . Let i: M → N be the inclusion map,
then we define ω|M = i∗ ω. It is given at p ∈ M simply by
ω|M (p)(v1 , . . . , vk ) = ω(p)(v1 , . . . , vk )
for v1 , . . . , vk ∈ Tp M ⊂ Tp N .
The following result will be important later.
Lemma 8.4.2. Let f : M → N be a smooth map between two manifolds of
the same dimension k. Let x1 , . . . , xk denote the coordinate functions of a
chart σ on M , and let y1 , . . . , yk denote the coordinate functions of a chart
τ on N . Then
f ∗ (dy1 ∧ · · · ∧ dyk ) = det(df ) dx1 ∧ · · · ∧ dxk (8.12)
where det(df ) in p is the determinant of the matrix for dfp with respect to
the standard bases for Tp M and Tf (p) N given by the charts.
Proof. The element f ∗ (dy1 ∧ · · · ∧ dyk ) is a k-form on M , hence a constant
multiple of dx1 ∧ · · · ∧ dxk (recall that dim Ak (V ) = 1 when dim V = k).
We determine the constant by applying f ∗ (dy1 ∧ · · · ∧ dyk ) to the vector
(dσp (e1 ), . . . , dσp (ek )) in (Tp M )k . We obtain, see (7.6),
f ∗ (dy1 ∧ · · · ∧ dyk )(dσp (e1 ), . . . , dσp (ek ))

= (dy1 ∧ · · · ∧ dyk )(dfp (dσp (e1 )), . . . , dfp (dσp (ek )))
1
= det(dyi (dfp (dσp (ej )))).
k!
116 Chapter 8
On the other hand, applying dx1 ∧ · · · ∧ dxk to the same vector in (Tp M )k ,
we obtain, again by (7.6),
1 1
(dx1 ∧ · · · ∧ dxk )(dσp (e1 ), . . . , dσp (ek )) = det(δij ) = ,
k! k!
since dx1 , . . . , dxk is the dual basis. It follows that the constant we are
seeking is the determinant det(dyi (dfp (dσp (ej ))))
Notice that dyi (dfp (dσp (ej ))) are exactly the matrix elements for dfp . The
formula (8.12) now follows.
8.5 Exterior differentiation

In this section a differentiation operator d is defined for differential forms
on a smooth manifold M . It is a generalization of the operator which maps
a smooth function to its differential, in general it will map k-forms to k + 1-
forms. The operator simply consists of applying d to the coefficients aI in
the coordinate expression for the k-form.
Definition 8.5. Fix a chart σ: U → M , and let ω ∈ Ak (M ). On its image
σ(U ) ⊂ M we can write

ω= aI dxi1 ∧ · · · ∧ dxik (8.13)
I
where I = (i1 , . . . , ik ) with 1 ≤ i1 < · · · < ik ≤ m = dim M , and where

aI ∈ C ∞ (σ(U )) for each I (see below Definition 8.3.1). On the image σ(U )
we then define
dω = daI ∧ dxi1 ∧ · · · ∧ dxik , (8.14)
I
where daI is the differential of aI . Then dω ∈ Ak+1 (σ(U )).

The operator d is called exterior differentiation. It is defined above by
reference to a fixed chart σ, but we shall see in the following theorem that
it can be extended to a globally defined operator, which is independent of
any choice of chart or atlas. The theorem also mentions the most important
properties of d.
Theorem 8.5. There exists for each k ∈ N a unique map d: Ak (M ) →
Ak+1 (M ) such that (8.14) holds on σ(U ) for each chart σ in a given atlas
for M . This map d is independent of the choice of the atlas. Furthermore,
these maps have the following properties:
(a) if k = 0 then d agrees with the differential on functions,
(b) d: Ak (M ) → Ak+1 (M ) is linear,
(c) d(ϕω) = dϕ ∧ ω + ϕ dω for ϕ ∈ C ∞ (M ), ω ∈ Ak (M ),
(d) d(ω1 ∧ ω2 ) = dω1 ∧ ω2 + (−1)k ω1 ∧ dω2 for ω1 ∈ Ak (M ), ω2 ∈ Al (M ),

(e) d(df1 ∧ · · · ∧ dfk ) = 0 for all f1 , . . . , fk ∈ C ∞ (M ),
(f) d(dω) = 0 for all ω ∈ Ak (M ).
Proof. Suppose an atlas is given. The maps d are unique because (8.14)
determines dω uniquely on σ(U ), for each chart (σ, U ) in the atlas. However,
we have to verify the existence, the independency on the atlas, and the listed
properties.
For every chart (σ, U ) on M (not necessarily from the given atlas) and
each k we define a σ-dependent operator dσ : Ak (σ(U )) → Ak+1 (σ(U )) by
(8.13)-(8.14). We want to define d: Ak (M ) → Ak+1 (M ) such that
(dω)|σ(U) = dσ (ω|σ(U) ) (8.15)
for ω ∈ Ak (M ) and for all σ. We need to verify that the expressions (8.14)
agree on overlaps of charts. Before we commence on this we shall prove that
dσ satisfies (a)-(f) above, with M replaced by σ(U ) everywhere.
(a) Let k = 0 and ω = f ∈ C ∞ (σ(U )). The only summand in the
expression (8.13) corresponds to I = ∅, and the coefficient a∅ is equal to f
itself. The expression (8.14) then reads dσ ω = df .
(b) The coefficients aI in (8.13) depend linearly on ω, and the expression
(8.14) is linear with respect to these coefficients, hence dσ ω depends linearly
on ω.
(c) Let ϕ ∈ C ∞ (σ(U )), ω ∈ Ak (σ(U )), and write ω according to (8.13). In
the corresponding expression for the product ϕω each coefficient aI is then
multiplied by ϕ, and hence definition (8.14) for this product reads

dσ (ϕω) = d(ϕaI ) ∧ dxi1 ∧ · · · ∧ dxik
I

= (aI dϕ + ϕ daI ) ∧ dxi1 ∧ · · · ∧ dxik
I

= dϕ ∧ aI dxi1 ∧ · · · ∧ dxik + ϕ daI ∧ dxi1 ∧ · · · ∧ dxik
I I
= dϕ ∧ ω + ϕ dσ ω
where the rule in Lemma 8.2.2 was applied in the second step.
(d) Notice first that if I = (i1 , . . . , ik ) with 1 ≤ i1 < · · · < ik ≤ m then by
definition
dσ (a dxi1 ∧ · · · ∧ dxik ) = da ∧ dxi1 ∧ · · · ∧ dxik (8.16)
for a ∈ C ∞ (σ(U )). The expression (8.16) holds also without the assumption
that the indices i1 , . . . , ik are increasingly ordered, for the dxi1 , . . . , dxik can
118 Chapter 8
be reordered at the cost of the same change of sign on both sides. The
expression also holds if the indices are not distinct, because in that case both
sides are 0.
Let ω1 ∈ Ak (σ(U )), ω2 ∈ Al (σ(U )), and write both of these forms as
follows, according to (8.13):

ω1 = aI dxi1 ∧ · · · ∧ dxik , ω2 = bJ dxj1 ∧ · · · ∧ dxjl ,
I J
where I = (i1 , . . . , ik ) and J = (j1 , . . . , jl ). Then

ω1 ∧ ω2 = aI bJ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
I,J
and hence according to the remarks just made

dσ (ω1 ∧ ω2 ) = d(aI bJ ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
I,J

= (bJ daI + aI dbJ ) ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
I,J

= bJ daI ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
I,J

+ aI dbJ ∧ dxi1 ∧ · · · ∧ dxik ∧ dxj1 ∧ · · · ∧ dxjl
I,J
= d ω1 ∧ ω2 + (−1)k ω1 ∧ dσ ω2
σ
where the sign (−1)k in the last step comes from passing dbJ to the right
past the elements dxi1 up to dxik .
(e) is proved first for k = 1. In this case f ∈ C ∞ (σ(U )) and
m
∂(f ◦ σ)
σ
d f = df = dxi .
i=1
∂xi
Hence
m
∂(f ◦ σ) m m
∂ 2 (f ◦ σ)
σ σ
d (d f ) = d ∧ dxi = dxj ∧ dxi .
i=1
∂x i i=1 j=1
∂x j ∂x i
2 2
The latter expression vanishes because the coefficient ∂∂x(fj ∂x◦σ)
i
= ∂∂x(fi ∂x
◦σ)
j
oc-
curs twice with opposite signs.
The general case is now obtained by induction on k. It follows from
property (d) that dσ (df1 ∧ · · · ∧ dfk ) can be written as a sum of two terms
each of which involve a factor where d is applied to a wedge product with

fewer factors. Applying the induction hypothesis, we infer the statement.
(f) Let ω be given by (8.13). Then

dσ (dσ ω) = d( daI ∧ dxi1 ∧ · · · ∧ dxik ) = 0
I
by property (e).
Having established these properties for the operators dσ on σ(U ), we pro-
ceed with the proof that for two different charts σ, σ , the corresponding

operators dσ and dσ agree on the overlap σ(U ) ∩ σ (U ).
Assume first that U ⊂ U and σ = σ|U . In the expression (8.13) for
ω on σ (U ) we then have the restriction of each aI to this set, and since

d(aI |σ (U ) ) = daI |σ (U ) it follows that the expression (8.14) for dσ ω on

σ (U ) is just the restriction of the same expression for dσ ω. Hence dσ ω =
dσ ω on σ (U ) as claimed.
For the general case the observation we just made implies that we can
replace U by the open subset σ −1 (σ (U ) ∩ σ(U )) and σ by its restriction to
this subset, without affecting dσ ω on σ (U ) ∩ σ(U ). Likewise, we can replace
U by σ −1 (σ (U ) ∩ σ(U )). The result is that we may assume σ(U ) = σ (U ).
We now apply the rules (a)-(f) for dσ to the expression

ω= aI dxi1 ∧ · · · ∧ dxik
I
for ω with respect to the chart σ . It follows from (b), (c) and (e) that

dσ ω = dσ (aI dxi1 ∧ · · · ∧ dxik )
I

= daI ∧ dxi1 ∧ · · · ∧ dxik ,
I

and hence dσ ω = dσ ω as claimed.
Finally we can define the global operator d. Let p ∈ M be given and
choose a chart σ around p. We define dω(p) for each ω ∈ Ak (M ) by the
expression in Definition 8.5. It follows from what we just proved, that dω(p)
is independent of the choice of chart (in particular, it is also independent of
the atlas).
It is now clear that dω is a well defined smooth k + 1-form satisfying
(8.15). The properties (a)-(f) for d follow immediately from (8.15) and the
corresponding properties for dσ .
As in Section 7.6 we define
A(M ) = A0 (M ) ⊕ A1 (M ) ⊕ · · · ⊕ Am (M )
120 Chapter 8
where m = dim M . The elements of A(M ) are called differential forms. Thus
a differential form is a map which associates to each point p ∈ M a member
of the exterior algebra A(Tp M ), in a smooth manner. The operators d of
Theorem 8.5 can then be combined in a single linear operator, also denoted
d: A(M ) → A(M ).
8.6 Some examples

In the following we treat M = Rn for various values of n. In each case we
denote by σ: Rn → Rn the standard chart on Rn , the identity map.
Example 8.6.1 Let M = R. Since dim M = 1, we have A(R) = A0 (R) ⊕
A1 (R). Furthermore, A0 (R) = C ∞ (R) and A1 (R) = {a dx | a ∈ C ∞ (R)}.
The exterior differentiation is given by df = f dx for f ∈ A0 (R) and df = 0
for f ∈ A1 (R).
Example 8.6.2 Let M = R2 . Since dim M = 2, we have
A(R2 ) = A0 (R2 ) ⊕ A1 (R2 ) ⊕ A2 (R2 ).
Furthermore, A0 (R2 ) = C ∞ (R2 ) and
A1 (R2 ) = {a1 dx1 + a2 dx2 | a1 , a2 ∈ C ∞ (R2 )}
A2 (R2 ) = {a dx1 ∧ dx2 | a ∈ C ∞ (R2 )}
The exterior differentiation is given by
∂f ∂f
df = dx1 + dx2
∂x1 ∂x2
(see Example 8.2), and
d(a1 dx1 +a2 dx2 )
= da1 ∧ dx1 + da2 ∧ dx2

∂a1 ∂a1 ∂a2 ∂a2
= dx1 + dx2 ∧ dx1 + dx1 + dx2 ∧ dx2
∂x1 ∂x2 ∂x1 ∂x2

∂a1 ∂a2
= − + dx1 ∧ dx2
∂x2 ∂x1
Example 8.6.3 Let M = R3 . Then

A(R3 ) = A0 (R3 ) ⊕ A1 (R3 ) ⊕ A2 (R3 ) ⊕ A3 (R3 ).
Furthermore, A0 (R3 ) = C ∞ (R3 ) and
A1 (R3 ) = {a1 dx1 + a2 dx2 + a3 dx3 | a1 , a2 , a3 ∈ C ∞ (R3 )}
A2 (R3 ) = {b1 dx2 ∧ dx3 + b2 dx3 ∧ dx1 + b3 dx1 ∧ dx2 | b1 , b2 , b3 ∈ C ∞ (R3 )}
A3 (R3 ) = {a dx1 ∧ dx2 ∧ dx3 | a ∈ C ∞ (R3 )}.
The exterior differentiation is given by

∂f ∂f ∂f
df = dx1 + dx2 + dx3
∂x1 ∂x2 ∂x3
and
d(a1 dx1 +a2 dx2 + a3 dx3 )
= da1 ∧ dx1 + da2 ∧ dx2 + da3 ∧ dx3

∂a1 ∂a1 ∂a1
= dx1 + dx2 + dx3 ∧ dx1
∂x1 ∂x2 ∂x3

∂a2 ∂a2 ∂a2
+ dx1 + dx2 + dx3 ∧ dx2
∂x1 ∂x2 ∂x3

∂a3 ∂a3 ∂a3
+ dx1 + dx2 + dx3 ∧ dx3
∂x1 ∂x2 ∂x3

∂a3 ∂a2
= − dx2 ∧ dx3
∂x2 ∂x3

∂a1 ∂a3
+ − dx3 ∧ dx1
∂x3 ∂x1

∂a2 ∂a1
+ − dx1 ∧ dx2
∂x1 ∂x2
Furthermore,
d (b1 dx2 ∧ dx3 + b2 dx3 ∧ dx1 + b3 dx1 ∧ dx2 )
= db1 ∧ dx2 ∧ dx3 + db2 ∧ dx3 ∧ dx1 + db3 ∧ dx1 ∧ dx2

∂b1 ∂b2 ∂b3
= + + dx1 ∧ dx2 ∧ dx3
∂x1 ∂x2 ∂x3
and
d(a dx1 ∧ dx2 ∧ dx3 ) = 0.
The expressions that occur as coefficients of dω in the three degrees are
familiar from vector calculus. The vector field

∂f ∂f ∂f
grad f = , ,
∂x1 ∂x2 ∂x3
is the gradient of f , and the vector field

∂a3 ∂a2 ∂a1 ∂a3 ∂a2 ∂a1
curl a = − , − , −
∂x2 ∂x3 ∂x3 ∂x1 ∂x1 ∂x2
is the curl (or rotation) of a. Finally, the function
∂b1 ∂b2 ∂b3
div b = + +
∂x1 ∂x2 ∂x3
is the divergence of b.
122 Chapter 8
8.7 Exterior differentiation and pull back

The following property of exterior differentiation plays an important role
in the next chapter.
Theorem 8.7. Let g: M → N be a smooth map of manifolds, and let
g ∗ : Ak (N ) → Ak (M )
denote the pull-back, for each k (see Definition 8.4). Then
dM ◦ g ∗ = g ∗ ◦ dN ,
where dM and dN denotes the exterior differentiation for the two manifolds.
Proof. By Lemma 8.4.1, and properties (b), (c) and (e) above,

d(g ∗ ω) = d( (aI ◦ g) d(yi1 ◦ g) ∧ · · · ∧ d(yik ◦ g))
I

= d(aI ◦ g) ∧ d(yi1 ◦ g) ∧ · · · ∧ d(yik ◦ g).
I
On the other hand, by the properties of g ∗ listed in the proof of Lemma 8.4.1,

g ∗ (dω) = g ∗ ( daI ∧ dyi1 ∧ · · · ∧ dyik )
I

= d(aI ◦ g) ∧ d(yi1 ◦ g) ∧ · · · ∧ d(yik ◦ g).
I
Corollary 8.7. Let M be a submanifold of the abstract manifold N , and let

ω be a differential form on N . Then the restriction of ω to M is a differential
form on M , and it satisfies d(ω|M ) = dω|M , that is, exterior differentiation
and restriction commute.
Proof. The restriction ω|M is exactly the pull back i∗ ω by the inclusion map
i: M → N . The result is an immediate consequence of Theorem 8.7.
8.8 Exercises
1 Let ξ denote the covector field
2xy dx + (x2 + y) dy
on R2 . Verify that ξ is closed, and find f ∈ C ∞ (R2 ) such that df = ξ.

∂aj
2 A covector field ξ = i ai dxi on Rn is said to be closed if ∂ai
∂xj = ∂xi for
all pairs of indices i, j. Consider the covector field
(y 2 + 2xz + 2x)dx + (z 2 + 2xy + 2y)dy + (x2 + 2yz + 2z)dz
on R3 . Verify that it is closed, and find f ∈ C ∞ (R3 ) such that df = ξ.

3 Let M be an abstract manifold, and let f ∈ C ∞ (M ). Assume that df = 0.
Prove that each level set {p ∈ M | f (p) = c} is both closed and open.
Conclude that f is constant on each component of M , and show that
the dimension of the kernel of d in C ∞ (M ) is the number of components
(which is either a finite number or infinite).
4 Let M = R2 \ {(x, 0) | x ≤ 0}, and let (r, ϕ) denote the polar coordinates
of (x, y), that is, (x, y) = (r cos ϕ, r sin ϕ). Here ϕ ∈] − π, π[.
a. Determine dx and dy in terms of r, ϕ, dr and dϕ.
b. Determine dr and dϕ in terms of x, y, dx and dy.
c. Show that the covector field
−y x
ξ= dx +
x2 + y 2 x2 + y 2
is closed on R2 \{(0, 0)}, and that it is exact on M but not on R2 \{(0, 0)}.
5 Let ξ be a covector field on an abstract manifold M . Prove that ξ is
smooth if ξ(Y ) ∈ C ∞ (M ) for all Y ∈ X(M ) (thus improving the ‘if’ of
Lemma 8.2.1).
6 Prove Lemma 8.3.
7 Let g: R2 → R3 be given by g(u, v) = (cos u cos v, cos u sin v, sin v). Deter-
mine g ∗ (xdz) and g ∗ (zdx ∧ dy).
8 Let M = R3 . Let ω1 = adx and ω2 = adx ∧ dy, where a ∈ C ∞ (M ).
Determine dω1 and dω2 . Compute the following differential forms
d(xz 2 dx), d(xz 2 dx ∧ dy), d(xz 2 dx) ∧ dy.
9 Verify Theorem 8.7 by direct computation in the case where g is the map
in exercise 7, and the operators are applied to the differential form yzdx.
124 Chapter 8
Chapter 9
Integration
The purpose of this chapter is to define integration on smooth manifolds,
and establish its relation with the differentiation operator d of the previous
chapter.
9.1 Null sets

Recall from Geometry 1 that a compact set D ⊂ R2 is said to be a null
set, if for each > 0 it can be covered by a finite union of rectangles with
total area < . We will generalize this concept to subsets D of an abstract
manifold M .
We first treat the case of sets in Rn . The notion of a rectangle is generalized
as a box, by which we mean a product set D = I1 × · · · × In ⊂ Rn , where
each Ii is a closed interval [ai , bi ] with −∞ < ai ≤ bi < ∞. The volume of
D is the product of the lengths of the intervals. The box is called a cube if
the intervals have equal length.
Definition 9.1.1. A set A ⊂ Rn is said to be a null set if for each > 0
it can be covered by a countable collection of boxes, of which the sum of
volumes is < .
It is clear that any subset of a null set is again a null set.
Example 9.1.1 Clearly a singleton A = {x} is a null set. More generally
every countable subset of Rn is a null set, since we can cover it by countably
many boxes, each of volume 0.
Example 9.1.2 Let m < n and consider a set in Rn of the form
A = {(x1 , . . . , xm , 0, . . . , 0) | x ∈ Rm }.
By writing Rm as a countable union of boxes Bi we can arrange that A is
covered by the boxes Bi × {0}, all of volume 0.
In fact for our present purpose we need only consider compact sets A. In
this case, we can replace the word countable by finite in the above definition.
Indeed, suppose a compact set A is covered by countably many boxes with
volume sum < . By increasing each box slightly, say the volume of the i-th
box is increased at most by 2−i , we can arrange that A is contained in a
countable union of open boxes with volume sum < 2. By Definition 5.1.2,
A is then covered by a finite subcollection, which of course also has volume
sum < 2.
126 Chapter 9
Lemma 9.1. Let A ⊂ Rn be a null set, and let f : A → Rn be a smooth map.

Then f (A) is a null set.
Proof. We only give the proof for the case that A is compact. By Definition
2.6.1, f has an extension F to a smooth map defined on an open set W
containing A. Let D ⊂ W be compact, then there exists an upper bound C
on the absolute values of the first order derivatives of F over D. Using the
mean value theorem, one then obtains an estimate |F (x) − F (y)| ≤ C|x − y|
for all x, y ∈ D. It follows from this estimate that the image by F of a ball
inside D of radius r is contained in a ball of radius Cr.
Let > 0 be given and choose a finite collection of boxes, which cover A
and whose volumes add to < . By possibly increasing some of the boxes
slightly, we can arrange that all the boxes have positive and rational side
lengths, still with volumes adding to < . We can then write each box as a
union of cubes of any size which divides the lengths of all sides of the box,
and which only intersect along their boundary. We thus obtain a covering
of A by finitely many cubes, whose volumes add to < . The circumscribed
ball of a cube has a volume which is proportional to that of the cube, with
a factor c that depends only on n. It follows that we can cover A by finitely
many arbitrarily small closed balls, whose volumes add to < c. In particular,
we may assume that each of these balls is contained in W . The discussion
in the first part of the proof (with D equal to the union of the balls) then
implies that f (A) is covered by finitely many balls of total volume < C n c.
By circumscribing these balls with cubes, we obtain a covering of F (A) by
boxes as desired.
Definition 9.1.2. Let M be an abstract manifold of dimension m. A

subset A ⊂ M is called a null set in M , if there exists a countable collection
of charts σi : Ui → M and null sets Ai ⊂ Ui such that A = ∪i σi (Ai ).
In order for this to be a reasonable definition, we should prove that a
diffeomorphism of open sets in Rm carries a null set to a null set, so that
the condition on Ai is unchanged if σi is reparametrized. This follows from
Lemma 9.1.
Example 9.1.3 Let N be a submanifold of M , with dim N < dim M , and
assume that N has a countable atlas. Then N is a null set in M . It follows
from Theorem 4.3 that for each p ∈ N there exists a chart σ on M around
p, such that N ∩ σ(U ) is the image of a null set in U (see Example 9.1.2). A
countable collection of these charts will already cover N , since N is second
countable (see Lemma 2.9).
9.2 Integration on Rn
Recall from Geometry 1, Chapter 3, that a non-empty subset D ⊂ R2 is
called an elementary domain if it is compact and if its boundary is a finite
Integration 127
union of smooth curves [a, b] → R2 . If D is an elementary domain, U ⊂ R2

an open set and f : U → R a continuous function, the plane integral D f dA
was given a sense in Geometry 1. The crucial property of D in this definition
is that D is compact and that its boundary is a null set. In particular, if
D = [a, b] × [c, d] is a rectangle,
b d d b
f dA = f (u, v) dv du = f (u, v) du dv.
D a c c a
These definitions can be generalized to Rn in analogous fashion. A non-

empty compact subset D ⊂ Rn is called a domain of integration if its bound-
ary is a null set in Rn . Assume that D is a domain of integration and that
D ⊂ U , where U is open. In this case, if f : U → R is continuous, the in-
tegral D f dV can be given a sense. The details are omitted since they are
analogous to those in Geometry 1.
In particular, if D = [a1 , b1 ] × · · · × [an , bn ] is a generalized rectangle, then
b1 bn
f dV = ··· f (x1 , . . . , xn ) dxn . . . dx1 ,
D a1 an
and the order of the consecutive integrals over x1 , . . . , xn can be interchanged

freely. Of course, for n = 1 the concept coincides with the ordinary integral
of a continuous function if D is an interval. In the trivial case n = 0, where
Rn = {0}, we interprete the definition so that the integral over D = {0} of
f is the evaluation of f in 0.
Example 9.2.1 If Ω ⊂ Rn is bounded and is a domain with smooth bound-
ary, then the closure D = Ω̄ is a domain of integration. The boundary ∂Ω is
an n − 1-dimensional manifold, hence a null set according to Example 9.1.3.

The integral D f dV has the following properties, also in analogy with
the two-dimensional case:

f + g dA = f dA + g dA
D

D D
cf dA = c f dA
D
D
| f dA | ≤ |f | dA
D
D

f dA = f dA + f dA,
D1 ∪D2 D1 D2
where in the last line D1 and D2 are domains of integration with disjoint
interiors.
There is an important rule for change of variables, which reads as follows.
It will not be proved here.
128 Chapter 9
Theorem 9.2. Let φ: U1 → U2 be a diffeomorphism of open sets, and let

D ⊂ U1 be a domain of integration. Then φ(D) ⊂ U2 is a domain of inte-
gration and
f dV = (f ◦ φ) |det(Dφ)| dV
φ(D) D
for all continuous f : U2 → R.
9.3 Integration on a chart

Let S ⊂ R3 be a surface, and let σ: U → S be a chart on it. If D ⊂ U
is a domain of integration, the area of the set R = σ(D) ⊂ S is defined
as the integral D (EG − F 2 )1/2 dA, where E, F , G are the coefficients of
the first fundamental form. It was shown in Geometry 1 that the area is
independent of the chosen parametrization σ, thanks to the presence of the
function (EG − F 2 )1/2 . The Jacobian determinant that arises in a change of
coordinates, is exactly compensated by the change of this square root when
E, F and G are replaced by their counterparts for the new coordinates.
Similarly, if f : S → R is a continuous function, its integral over R is defined
to be
f dA = (f ◦ σ) (EG − F 2 )1/2 dA.
R D
This integral is independent of the chart, for the same reason as before.
These considerations can be generalized to an m-dimensional smooth sur-
face in Rk . The factor (EG − F 2 )1/2 in the integral over D is generalized as
the square root of the determinant det(σu i · σu j ), i, j = 1, . . . , m, for a given
chart σ. The dot is the scalar product from Rn .
When trying to generalize further to an abstract m-dimensional manifold
M we encounter the problem that in general there is no dot product on Tp M ,
hence there is no way to generalize E, F and G. As a consequence, it is not
possible to define the integral of a function on M in a way that is invariant
under changes of of charts. One way out is to introduce the presence of an
inner product on tangent spaces as an extra axiom in the definition of the
concept of a manifold - this leads to so-called Riemannian geometry, which is
the proper abstract framework for the theory of the first fundamental form.
Another solution is to replace the integrand including (EG − F 2 )1/2 by a
differential form of the highest degree. This approach, which we shall follow
here, leads to a theory of integration for m-forms, but not for functions.
In particular, it does not lead to a definition of area, since that would be
obtained from the integration of the constant 1, which is a function, not a
form. However, we can still view the theory as a generalization to the theory
for surfaces, by defining the integral of a function f over a surface as the
integral of f times the volume form (see Example 8.3).
Integration 129
For some applications it is important to be able to integrate forms which

are continuous, but not smooth. The definition of continuity of forms is
analogous to Definition 8.3.2: A continuous k-form on an abstract manifold
M is a k-form ω for which all the functions aI are continuous, for each chart
in an atlas of M . In particular, an m-form on an m-dimensional manifold
thus has the form
ω = φ dx1 ∧ · · · ∧ dxm (9.1)
on each chart σ, with φ a continuous function on σ(U ).
Definition 9.3. Let M be an m-dimensional oriented abstract manifold, and
let σ: U → M be a positive chart on M . Let ω ∈ Am (M ) be a continuous
m-form on M . Let D ⊂ U be a domain of integration, and let R = σ(D).
Then we define
ω= φ ◦ σ dV (9.2)
R D
where φ is the function determined in (9.1).

We shall now see that the definition is independent of the choice of chart.
The point is that when coordinates are changed, the function φ that rep-
resents ω changes by a factor that exactly compensates for the Jacobian
determinant resulting from the change. The Jacobian determinant has sim-
ply been built into the object that we integrate. This is the central idea
behind the theory.

Theorem 9.3. The integral R ω is independent of the choice of positive
chart σ. More precisely, if R is contained in the image of some other chart,
then the integral defined in (9.2) has the same value for the two charts.
Proof. The statement is the special case of the identity mapping M → M in
the following more general result.
Lemma 9.3. Let M , M̃ be oriented abstract manifolds, and let g: M̃ → M
be an orientation preserving diffeomorphism. Let σ: U → M and σ̃: Ũ → M̃
be charts, and let D ⊂ U and D̃ ⊂ Ũ be domains of integration. Assume that
g(R̃) = R, where R = σ(D) and R̃ = σ̃(D̃).
Furthermore, let ω be a continuous m-form on M and let ω̃ = g ∗ ω be its
pull back by g (defined as in Definition 8.4). Then

ω̃ = ω.
R̃ R
Proof. The open subset σ̃(Ũ ) ∩ g −1 (σ(U )) of σ̃(Ũ ) contains R̃. Hence we can
replace σ̃ by its restriction to the preimage in Ũ of this set with no effect
on the integral over R̃. That is, we may assume that σ̃(Ũ ) ⊂ g −1 (σ(U )), or
equivalently, g(σ̃(Ũ )) ⊂ σ(U ).
130 Chapter 9
Similarly, since g(σ̃(Ũ )) contains R, we can replace σ by its restriction to

the preimage in U of this set with no effect on the integral over R. In other
words, we may in fact assume that g(σ̃(Ũ )) = σ(U ).
Let Φ denote the coordinate expression for g,
Φ = σ −1 ◦ g ◦ σ̃: Ũ → U.
It is a diffeomorphism of Ũ onto U . The claim is that

φ̃ ◦ σ̃ dV = φ ◦ σ dV
D̃ D
where ω = φ dx1 ∧ · · · ∧ xm and ω̃ = φ̃ dx̃1 ∧ · · · ∧ x̃m are the expressions for

ω and ω̃.
It follows from Lemma 8.4.2 that
g ∗ (dx1 ∧ · · · ∧ dxm ) = det(dg) dx̃1 ∧ · · · ∧ dx̃m , (9.3)
where det(dgp ) is the determinant of the matrix for dgp with respect to the
standard bases for the tangent spaces. In other words, det(dgp ) is the deter-
minant of the Jacobian matrix DΦ of the coordinate expression Φ. Applying
(8.9) and (9.3) we conclude that
ω̃ = g ∗ ω = g ∗ (φ dx1 ∧ · · · ∧ dxm ) = (φ ◦ g) det(DΦ) dx̃1 ∧ · · · ∧ dx̃m .
It follows that
φ̃ = det(DΦ) (φ ◦ g). (9.4)
Since D = Φ(D̃) we obtain from Theorem 9.2

φ ◦ σ dV = φ ◦ σ ◦ Φ |det(DΦ)| dV. (9.5)
D D̃
Observe that σ ◦ Φ = g ◦ σ̃, and that det(DΦ) is positive since g preserves
orientation. It now follows from (9.4) that the expression (9.5) equals

φ̃ ◦ σ̃ dV,
D
and the lemma is proved.
Example 9.3 Let γ: I → R2 be an embedded parametrized curve, and let
ω be a continuous 1-form on R2 , say ω = f1 (x1 , x2 ) dx1 + f2 (x1 , x2 ) dx2 . Let
[a, b] ⊂ I and R = γ([a, b]). Then
b
ω= F (γ(u)) · γ (u) du (9.6)
R a
where F (x1 , x2 ) = (f1 (x1 , x2 ), f2 (x1 , x2 )). This is seen as follows. The basis
for the 1-dimensional tangent space of γ at p = γ(u) is γ (u), and the dual
basis vector du(p) is defined by du(p)(γ (u)) = 1. On the other hand, ωp (v) =
F (p) · v for v ∈ R2 , and in particular ωp (γ (u)) = F (p) · γ (u). Hence ω|γ(I) =
φ du with φ(p) = F (p) · γ (u). Now (9.6) follows from Definition 9.3. The
expression (9.6) is called the line integral of ω along R.
Integration 131
9.4 Integration on a manifold

We shall generalize the definition of R
ω to the case where R is not nec-
essarily contained inside a single chart.
Definition 9.4.1. A compact subset R of an abstract manifold M is said
to be a domain of integration, if the boundary ∂R is a null set.
For example, if Ω ⊂ M is a domain with smooth boundary, such that the
closure R = Ω̄ is compact, then this closure is a domain of integration, by
Corollary 4.7 and Example 9.1.3.
It is easily seen that the union R1 ∪ R2 , the intersection R1 ∩ R2 and the
set difference R1 \ R2◦ , of two domains of integration R1 , R2 ⊂ M is again a
domain of integration (in all three cases, the boundary is contained in the
union of the boundaries of R1 and R2 ).
In particular, if R ⊂ M is a domain of integration, and if σ: U → M is a
chart with another domain of integration D ⊂ U , then σ(D) ∩ R is a domain
of integration inside σ(U ), to which we can apply the preceding definition of
the integral. We shall define the integral over R as a sum of such integrals,
for a collection of charts that cover R. In order to combine these integrals
defined by different charts, we employ a partition of unity, see Section 5.4.
We therefore assume that M has a locally finite atlas.
We choose around each point p ∈ M a chart σ: U → M and a domain
of integration D ⊂ U such that p is contained in the image σ(D◦ ) of the
interior. The collection of all these charts is denoted σα : Uα → M , where A
is the index set. The open sets σα (Dα◦ ) for all α ∈ A cover M . We choose
a partition of unity (ϕα )α∈A such that each function ϕα has support inside
the corresponding set σ(Dα◦ ) (see Theorem 5.5). Notice that the collection
of the supports of the functions ϕα is locally finite, hence only finitely many
of these functions are non-zero on the compact set R.
Definition 9.4.2. Let ω and be a continuous m-form on M , and let R ⊂ M
be a domain of integration. We define

ω= ϕα ω,
R α∈A R∩σα (Dα )
where the integrals on the right are defined with respect to the charts σα , as
in Definition 9.3. The sum is finite, since only finitely many ϕα are non-zero
on R.
Theorem 9.4. The definition given above is independent of the choice of
the charts σα and the subsequent choice of a partition of unity.
Proof. Indeed, if ϕ̃α̃ , where α̃ ∈ Ã, is a different partition of unity relative

132 Chapter 9
to a different atlas, then

ϕα ω = ( ϕ̃α̃ )ϕα ω
α∈A R∩σα (Dα ) α∈A R∩σα (Dα )
α̃∈Ã

= ϕα ϕ̃α̃ ω
R∩σα (Dα )
α̃∈Ã α∈A
and since ϕα ϕ̃α̃ ω is supported inside the intersection R ∩ σα (Dα ) ∩ σα̃ (Dα̃ ),
the latter expression equals

ϕα ϕ̃α̃ ω
R∩σα (Dα )∩σ̃α̃ (Dα̃ )
α̃∈Ã α∈A
The integral over R ∩ σα (Dα ) ∩ σ̃α̃ (Dα̃ ) has the same value for the two charts
by Theorem 9.3. Hence the last expression above is symmetric with respect
to the partitions indexed by A and Ã, and hence the original sum has the
same value if the partition is replaced by the other one.

It is easily seen that M ω depends linearly on ω. Moreover, in analogy
with Lemma 9.3:
Lemma 9.4. Let g: M̃ → M be an orientation preserving diffeomorphism,
and R̃ ⊂ M̃ a domain of integration. Then R = g(R̃) ⊂ M is a domain
of integration. Furthermore, let ω be a continuous m-form with pull back
ω̃ = g ∗ ω. Then
ω̃ = ω.
R̃ R
Proof. Let (ϕα )α∈A be a partition of unity on M , as in Definition 9.4.2, and

put ψα = ϕα ◦ g ∈ C ∞ (M̃ ). Since g is a diffeomorphism, this is a partition
of unity on M̃ . It follows from (8.9), that g ∗ (ϕα ω) = ψα ω̃. The proof is now
straightforward from Definition 9.4.2 and Lemma 9.3.
The case of a 0-dimensional manifold needs special interpretation. Recall
that a 0-dimensional abstract manifold is a discrete set M . By definition, an
orientation of M assigns + or − to each point in M , it can thus be viewed as
a function o: M → {+, −}. A subset R is compact if and only if it is finite.
A 0-form on M is a function f : M → R, and the interpretation of its integral
over a finite set D is that it is the signed sum

o(p)f (p).
p∈D
This interpretation may seem peculiar, but it is extremely convenient (for

example, in Example 9.7.1 below).
Integration 133
9.5 A useful formula

The definition of R ω by means of a partition of unity is almost impossible
to apply, when it comes to explicit computations, because it is very difficult
to write down explicit partitions of unity. The following theorem, of which
the proof is omitted, can be used instead. The idea is to chop up R in finitely
many pieces and treat each of them separately. The pieces are chosen such
that they are convenient for computations.
Theorem 9.5. Let M be an abstract oriented manifold, let ω be a continuous
m-form on M , and let R ⊂ M be a domain of integration.
a) Assume that R = R1 ∪ · · · ∪ Rn where each set Ri ⊂ M is a domain of
integration, such that all the interiors Ri◦ are disjoint from each other. Then
n

ω= ω.
R i=1 Ri
b) Let i = 1, . . . , n and assume that there exist an open set Ui ⊂ Rm ,

a smooth map gi : Ui → M and a domain of integration Di ⊂ Ui such that
Ri = gi (Di ), and such that the restriction gi |Di◦ is a positive chart. Then

ω= gi∗ ω.
Ri Di
Let φi : Ui → R be the function determined by gi∗ ω = φi du1 ∧ · · · ∧ dum ,

with u1 , . . . , um the coordinates of Rm . Then the last formula above reads

ω= φi dV.
Ri Di
Notice that the requirement on the map gi is just that its restriction to the
interior of Di is a chart. If gi itself is a chart, σ = gi , then the present
function
φi is identical with the function φ ◦ σ in (9.2), and the formula for
Ri
ω by means of gi∗ ω is identical with the one in Definition 9.3.
For example, the unit sphere S 2 is covered in this fashion by a single map
g: D → S 2 of spherical coordinates
g(u, v) = (cos u cos v, cos u sin v, sin u)
with D = [−π/2, π/2] × [−π, π], and thus we can compute the integral of a
2-form over S 2 by means of its pull-back by spherical coordinates, in spite of
the fact that this is only a chart on a part of the sphere (the point being, of
course, that the remaining part is a null set).
134 Chapter 9
9.6 Stokes’ theorem

Stokes’ theorem is the central result in the theory of integration on man-
ifolds. It gives the relation between exterior differentiation and integration,
and it generalizes the fundamental theorem of calculus.
Let M be an m-dimensional oriented abstract manifold, and let Ω ⊂ M be
a domain with smooth boundary. The boundary ∂Ω (which may be empty)
is equipped with the induced orientation, see Section 4.8. For example, Ω
could be the inside of a compact connected manifold in Rn , as in the Jordan-
Brouwer theorem, Theorem 5.11.
As mentioned below Definition 9.4.1,
if Ω̄ is compact, then Ω̄ is a domain of
integration, and hence the integral Ω̄ ω is defined for all continuous m-forms
on M . The requirement that Ω̄ is compact can be relaxed as follows. Observe
that if R is a domain of integration, then so is the intersection Ω̄ ∩ R, since
the boundary of each set is a null set. Let now ω be a continuous m-form.
The support supp ω of ω is the closure of the set {p ∈ M | ω(p) = 0}. Assume
that Ω̄∩supp ω is compact, and choose a domain of integration R ⊂ M which
contains this set. We can then define the integral of ω over Ω (or Ω̄) by

ω= ω,
Ω Ω̄∩R
the point being that the right hand side is independent of the choice of R.
Theorem 9.6. Stokes’ theorem. Let ω be a differential m − 1-form on
M , and assume that Ω̄ ∩ supp ω is compact. Then

dω = ω. (9.7)
Ω ∂Ω
The right-hand side is interpreted as 0 if ∂Ω is empty.

The result is remarkable, because it shows that by knowing ω only on the
boundary of Ω, we can predict a property of it on the interior, namely the
total integral of its derivative.
Proof. Let R ⊂ M be a domain of integration containing Ω̄ ∩ supp ω. Then
this set also contains Ω̄ ∩ supp dω. Hence by definition

dω = dω and ω= ω.
Ω Ω̄∩R ∂Ω ∂Ω∩R
Choose around each point p ∈ M a chart σ: U → M and an m-dimensional

rectangle D ⊂ U such that p ∈ σ(D◦ ), where D◦ is the interior of D. In
addition we assume that these charts are chosen so that if p ∈ ∂Ω then
σ(U ) ∩ Ω = σ(U + ) where U + = {x ∈ U | um > 0} (see Theorem 4.7), if
p ∈ Ω then σ(U ) ⊂ Ω and if p ∈ M \ Ω̄ then σ(U ) ⊂ M \ Ω̄.
Integration 135
We may now assume that ω is supported inside σ(D◦ ) for one of these
charts. For if the result has been established in this generality, we can apply
it to ϕα ω for each element ϕα in a partition of unity as in Definition 9.4.2.
Since the set R is compact, only finitely many ϕα ’s are non-zero on it, and
as both sides of (9.7) depend linearly on ω, the general result then follows.
The equation (9.7) clearly holds if σ(U ) ⊂ M \ Ω̄, since then both sides
are 0. This leaves the other two cases, σ(U ) ∩ Ω = σ(U + ) and σ(U ) ⊂ Ω, to
be checked.
We may also assume that
ω = f dx1 ∧ · · · ∧ dxj−1 ∧ dxj+1 ∧ · · · ∧ dxm
for some j between 1 and m, because in general ω will be a sum of such

forms. m ∂f
Since df = i=1 ∂xi
dxi (see Lemma 8.2.2), it then follows that
∂f
dω = (−1)j−1 dx1 ∧ · · · ∧ dxm ,
∂xj
where the sign appears because dxj has been moved past the 1-forms from
dx1 up to dxj−1 .
Assume first that j < m. We will prove that then

dω = ω = 0.
Ω ∂Ω
By definition

∂f ∂(f ◦ σ)
dω = (−1) j−1
dx1 ∧ · · · ∧ dxm = (−1)j−1 dV,
Ω Ω ∂xj ∂uj
where the last integration takes place over the set {x ∈ D | σ(x) ∈ Ω̄}, that
is, over D ∩ U + or D, in the two cases.
In the integral over the rectangle D ∩ U + or D, we can freely interchange
the order of integrations over u1 , . . . , um . Let us take the integral over uj
first (innermost), and let us denote its limits by a and b. Notice that since
j < m these are the limits for the uj variable both in D ∩ U + and D. Now
by the fundamental theorem of calculus
b
∂(f ◦ σ)
duj = f (σ(u1 , . . . , uj−1 , b, uj+1 , . . . , um ))
a ∂uj
− f (σ(u1 , . . . , uj−1 , a, uj+1 , . . . , um )).
However, since ω is supported in σ(D◦ ), it follows that these values are zero
for all u1 , . . . uj−1 , uj+1 , . . . um , and hence Ω dω = 0 as claimed.
136 Chapter 9
On the other hand, if σ(U )∩Ω = σ(U + ), then xm = 0 along the boundary,
and it follows immediately
that dxm , and hence also ω, restricts to zero on
∂Ω. Therefore also ∂Ω ω = 0. The same conclusion holds trivially in the
other case, σ(U ) ⊂ Ω.
Next we assume j = m. Again we obtain 0 on both sides of (9.7) in the
case σ(U ) ⊂ Ω, and we therefore assume the other case. We will prove that
then

m
dω = ω = (−1) f (σ(u1 , . . . , um−1 , 0)) du1 . . . dum−1 ,
Ω ∂Ω
where the integral runs over the set of (u1 , . . . , um−1 ) ∈ Rm−1 for which
σ(u1 , . . . , um−1 , 0) ∈ ∂Ω
Following the preceding computation of Ω dω, we take the integral over
uj = um first. This time, however, the lower limit a is replaced by the value
0 of xm on the boundary, and the previous conclusion fails, that f vanishes
here. Instead we obtain the value f (σ(u1 , . . . , um−1 , 0)), with a minus in
front because it is the lower limit in the integral. Recall that there was a
factor (−1)j−1 = (−1)m−1 in front. Performing the integral over the other
variables as well, we thus obtain the desired integral expression

m
(−1) f (σ(u1 , . . . , um−1 , 0)) du1 . . . dum−1

for dω.
Ω
On the other hand, the integral ∂Ω ω can be computed by means of the
restricted chart σ|U∩Rm−1 on ∂Ω. However, we have to keep track of the
orientation of this chart. By definition (see Section 4.8), the orientation of
the basis
dσ(e1 ), . . . , dσ(em−1 )
for Tp ∂M is the same as the orientation of the basis
−dσ(em ), dσ(e1 ), . . . , dσ(em−1 )
for Tp M . The orientation of the latter is (−1)m , because σ is a positive

chart. It follows that

m
ω = (−1) f (σ(u1 , . . . , um−1 , 0)) du1 . . . dum−1 ,
∂Ω
as claimed.
Integration 137
9.7 Examples from vector calculus

In the following examples we always use the standard coordinates on the
Euclidean space Rn .
Example 9.7.1. Let M = R, then m = 1 and ω is a 0-form, that is, a
function f ∈ C ∞ (R). The differential of f is the 1-form df = f (x)dx (see
Example 8.6.1). Let Ω =]a, b[, then the boundary ∂Ω consists of the two
points a and b, oriented by −1 and +1, respectively (see Example 4.8.3).
Hence b b
dω = df = f (x) dx
Ω a a
and
ω = f (b) − f (a)
∂Ω
Thus we see that in this case Stokes’ theorem reduces to the fundamental
theorem of calculus.
Example 9.7.2. Let M = R2 , then m = 2 and ω is a 1-form. The boundary
of the open set Ω is a union of smooth curves, we assume for simplicity it
is a single simple closed curve. Write ω = f (x, y) dx + g(x, y) dy, then (see
Example 8.6.2)
∂f ∂g
dω = (− + ) dx ∧ dy
∂y ∂x
and hence
∂f ∂g
dω = − + dA.
Ω Ω ∂y ∂x

On the other hand, the integral ∂Ω ω over the boundary can be computed
as follows. Assume γ: [0, T ] → ∂Ω is the boundary curve, with end points
γ(0) = γ(T ) (and no other self intersections). Theorem 9.5 can be applied
with D1 = [0, T ] and g1 = γ. Write γ(t) = (x(t), y(t)), then by definition
g1∗ ω = (f ◦ γ) d(x ◦ γ) + (g ◦ γ) d(y ◦ γ),

and ∂Ω
ω is the line integral

ω= f (x, y)dx + g(x, y)dy.
∂Ω γ
We thus see that in this case Stokes’ theorem reduces to the classical Green’s
theorem (which is equivalent with the divergence theorem for the plane):

∂f ∂g
− + dA = f (x, y)dx + g(x, y)dy.
Ω ∂y ∂x γ
138 Chapter 9
The orientation of γ is determined as follows. If n and t are the outward

normal vector and the positive unit tangent vector, respectively, in a given
point of γ, then (n, t) should be positively oriented, that is, t = n̂, which is
exactly the standard counter clockwise orientation of a closed curve in R2 .
Example 9.7.3. Let M = R3 , and let Ω be a domain with smooth bound-
ary. The boundary ∂Ω is then a surface S in R3 . Let ω be a 2-form on
M:
ω = f (x, y, z) dy ∧ dz + g(x, y, z) dz ∧ dx + h(x, y, z) dx ∧ dy.
Then (see Example 8.6.3)
∂f ∂g ∂h
dω = ( + + ) dx ∧ dy ∧ dz
∂x ∂y ∂z
and hence
dω = div(f, g, h) dV,
Ω Ω
where div(f, g, h) = ∂f ∂g ∂h
∂x + ∂y + ∂z is the divergence of the vector field (f, g, h).

On the other hand, the integral ∂Ω ω over the boundary ∂Ω = S can be
computed as follows. SupposeD1 , . . .
, Dn and gi : Di → S are as in Theorem
n
9.5 for the manifold S. Then S ω = i=1 Di gi∗ ω. Let σ(u, v) = gi (u, v) be
one of the functions gi . Then
σ ∗ ω = (f ◦σ) d(y◦σ)∧d(z◦σ)+(g◦σ) d(z◦σ)∧d(x◦σ)+(h◦σ) d(x◦σ)∧d(y◦σ).
Furthermore
∂σ2 ∂σ2 ∂σ3 ∂σ3
d(y ◦ σ) ∧ d(z ◦ σ) = ( du + dv) ∧ ( du + dv)
∂u ∂v ∂u ∂v
∂σ2 ∂σ3 ∂σ2 ∂σ3
=( − ) du ∧ dv
∂u ∂v ∂v ∂u
and similarly
∂σ3 ∂σ1 ∂σ1 ∂σ3
d(z ◦ σ) ∧ d(x ◦ σ) = ( − ) du ∧ dv
∂u ∂v ∂v ∂u
∂σ1 ∂σ2 ∂σ2 ∂σ1
d(x ◦ σ) ∧ d(y ◦ σ) = ( − ) du ∧ dv.
∂u ∂v ∂v ∂u
Notice that the three expressions in front of du∧dv are exactly the coordinates
of the normal vector σu × σv . Thus we see that
σ ∗ ω = (f ◦ σ, g ◦ σ, h ◦ σ) · (σu × σv ) du ∧ dv.
Let N(u, v) denote the outward unit normal vector in σ(u, v), then σu ×σv =
σu × σv N and we see that D σ ∗ ω is the surface integral of the function
Integration 139
(f, g, h) · N over the image of σ = gi (recall that the surface integral of

a function includes a factor σu × σv = (EG − F 2 )1/2 ). Summing over
i = 1, . . . , n we finally obtain the surface integral

ω = (f, g, h) · N dA.
S S
We conclude that in this case Stokes’ theorem reduces to the divergence

theorem (sometimes also called the Gauss theorem):

div(f, g, h) dV = (f, g, h) · N dA.
Ω ∂Ω
Example 9.7.4. Let M = S be an oriented smooth surface in R3 , and let

Ω ⊂ M be a domain with smooth boundary. Let ω be a 1-form on M , and
assume ω is the restriction of a smooth 1-form
f (x, y, z) dx + g(x, y, z) dy + h(x, y, z) dz
defined on an open neighborhood of M in R3 . A similar analysis as in the
previous example shows that in this case Stokes’ theorem reduces to the
original theorem of Stokes, which is

(curl F · N) dA = f dx + g dy + h dz.
Ω ∂Ω
where
∂h ∂g ∂f ∂h ∂g ∂f
curl F = ( − , − , − ).
∂y ∂z ∂z ∂x ∂x ∂y
The preceding examples show that in addition to the fundamental theo-

rem of calculus, Stokes’ theorem also generalizes the three main theorems of
classical vector calculus.
9.8 Exercises
1 Let ω = f (x, y) dx + g(x, y) dy be a 1-form defined on an open set Ω ⊂ R2 ,

and let γ(t) = (x(t), y(t)): I → Ω be a parametrized smooth curve, where
I is an open interval. Let a, b ∈ I with a < b. Prove the formula
b
ω= f (x(t), y(t))x(t) + g(x(t), y(t))y (t) dt.
γ([a,b]) a
Compute the integral of y dx + x dy,

a. along the line segment from (0, 0) til (1, 1),
b. over the unit circle with counterclockwise orientation.
Repeat for y dx − x dy.
140 Chapter 9
2 Compute the integral of y dx + x dy, along the curve γ(t) = (sin7 t, sin17 t)
from t = 0 to t = π/2.
3 Let M = {(x, y, v, w)|x2 + y 2 = v 2 + w2 = 1} in R4 , and choose the
orientation as in Example 3.8.2. Let ω be the 2-form adx ∧ dv on R4 , with
a ∈ C ∞ (R4 ). Determine the integral of ω over M as an expression with
a, and compute it in each of the cases y 2 and a = yw.
4 On R2 , Stokes’ theorem is Green’s theorem (see Example 9.7.2).
a. Use the theorem to verify the following formulas for the area of Ω

1
A(Ω) = x dy = − y dx = x dy − y dx.
γ γ 2 γ
b. Show that

−y x 0 if(0, 0) ∈
/ Ω̄
2 2
dx + 2 dy =
γ x +y x + y2 2π if(0, 0) ∈ Ω
for all closed and regular smooth curves γ, where Ω is the interior (hint:
if (0, 0) ∈ Ω, use the theorem to replace γ by the unit circle).
5 On R3 , Stokes’ theorem is the divergence theorem (see Example 9.7.3)
a. Use (f, g, h) = (x, y, z) on the unit ball. What does the theorem tell?
b. Show that Green’s theorem can be reformulated as a 2-dimensional
version of the divergence theorem.
c. Find formulas for the volume of Ω, which are analogous to those of 4a
above.
d. Let
(x, y, z)
(f, g, h) = 2
(x + y 2 + z 2 )3/2
for which the divergence is zero. State and prove a result analogous the
that of 4b.
Index
algebraic variety, 11 diffeomorphism, 23, 27
alternating, 101 differentiable manifold, 20
form, 102 differential, 43,
ambient space, 10 form, 113
atlas, 8, 19 dimension of manifold, 1, 10, 20
compatible, 20 directional derivative, 39, 41, 40
countable, 29 discrete
base, for topology, 29 set, 10
basis, standard, 45 topology, 18
boundary, 18 divergence, div, 121
orientation, 60 theorem, 139
bump function, 68 domain
chain rule, 42, 43 of integration, 127, 131
chart, 8 with smooth boundary, 58
closed dual
set, 18 basis, 97
vector field, 112 map, 98
closure, 18 vector space, 97
compact, 65 Einstein convention, 101
compatible, 20 embedded
component, 76 parametrized manifold, 3
connected, 72 submanifold, 62
component, 76 embedding, 62
continuous, 18 exact vector field, 112
contravariant, 101 extension, local 23
coordinate exterior
expression, 34, 36 algebra, 107
function, 83, 110 differentiation, 116
map, 11 Gauss, 139
cotangent space, 109 GL(n, R), 28
countable, gl(n, R), 91, 94
atlas, 29 gradient, grad, 121
second, 30 Green, 137
covariant, 99 group, Lie, 28
covector, 109 orthogonal, 57
field , 111 Hausdorff topology, 19
cover, 8 Heine-Borel property, 65
covering, 65 homomorphism, Lie, 93
open, 65 homeomorphism, 3, 19
critical value, 55 immersed submanifold, 62
curl, 121 immersion, 62
curve, 5 induced topology, 19
142 Index
integral, 127, 129, 131 orthogonal group, O(n), 57

interior, 18 parametrized curve on, 34, 36
intermediate value property, 72 partition of unity, 69
isolated point, 10 pathwise connected, 72
Jacobi identity, 89 locally, 75
Jordan-Brouwer theorem, 79 product
Jordan curve theorem, 79 manifold, 23
k-form, 113 topology, 22
left translation, 90 projection, 86
left invariant, 90 pull back, 114
level set, 55 regular, 1
Lie value, 55
algebra, 91 relatively open, 3, 19
bracket, 88 reparametrization, 2
group, 28 restriction, of k-form, 115
homomorphism, 93 second countable, 30
locally finite, 67 smooth,
manifold, atlas, 19
abstract, 20 boundary, 58
differentiable, 20 covector field, 111
embedded parametrized, 3 differential form, 113
in Rn , 10 extension, 23
parametrized, 1 map, 23, 25, 27
product, 23 structure, 20
map, vector field, 81, 82
continuous, 18 space,
smooth, 23, 25, 27 ambient, 10
mixed tensor, 101 Hausdorff, 19
Möbius band, 46 projective, 21
multilinear, 99 tangent, 34, 33, 38
neighborhood, 18 tensor, 99
null set, 125, 126 topological, 17
O(n), 57, 95 standard basis, 45
o(n), 95 Stokes’ theorem, 134, 139
open, 17 structure, smooth, 20
covering, 65 subcovering, 65
relatively, 3, 19 submanifold, 49, 50
orientable, 46, 79 structure, 54
orientation, 46 embedded, 62
of boundary,60 immersed, 62
preserving, 46 support, 69
oriented, 45, surface, 7
Index 143
tangent
bundle, 86
map, 41
space, 34, 33, 38
tangential, 37
tensor, 99
contravariant, 101
covariant, 99
mixed, 101
product, 100
space, 99
topological space, 17
topology,
discrete, 18
Hausdorff, 19
trivial, 18
value,
critical, 55
regular, 55
variety, 11
vector,
cotangent, 109
field, 81
tangent, 38
volume form, 114
wedge product, 104
Whitney’s theorem, 30, 70

Differentiable Manifolds: Lecture Notes For Geometry 2

Uploaded by

Differentiable Manifolds: Lecture Notes For Geometry 2

Uploaded by

Diﬀerentiable manifolds

Lecture Notes for Geometry 2

4.7 Orientation of the boundary . . . . . . . . . . . . . . . 60

8.6 Exterior diﬀerentiation and pull back . . . . . . . . . . 122

Manifolds in Euclidean space

In Geometry 1 we have dealt with parametrized curves and surfaces in

1.1 Parametrized manifolds

Clearly, a parametrized manifold with m = 2 and n = 3 is the same

1.2 Embedded parametrizations

σ(V ) = σ(U ) ∩ W. (1.1)

The importance of this condition is illustrated in Example 1.2.2 below.

The restriction γ|I is not a homeomorphism from I to C. The problem occurs

Deﬁnition 1.3. A curve in R2 is a non-empty set C ⊂ R2 satisfying the

Example 1.3.1. The image C = γ(I) of an embedded parametrized curve

some other parametrization could satisfy the requirement in Deﬁnition 1.3.

The deﬁnition of a curve allows the following useful reformulation.

fx 1 (p) = · · · = fx n (p) = 0.

Theorem 1.3. Let f : Ω → R be a smooth function, where Ω ⊂ R2 is open,

C = {p ∈ Ω | f (p) = c, p is not critical }

Example 1.3.5. The set C = {(x, y) | x2 + y 2 = c} is a curve in R2 for each

Example 1.4.1. The image S = σ(U ) of an embedded parametrized surface

Theorem 1.4. Let f : Ω → R be a smooth function, where Ω ⊂ R3 is open,

S = {p ∈ Ω | f (p) = c, p is not critical } (1.4)

1.5 Chart and atlas

σ(u, v) = (cos v, sin v, u), u, v ∈ R

is regular and covers the cylinder S = {(x, y, z) | x2 + y 2 = 1}, but it is not

U1 = {(u, v) ∈ R2 | −π < v < π}, U2 = {(u, v) ∈ R2 | 0 < v < 2π},

and let σi denote the restriction of σ to Ui for i = 1, 2. Then σ1 and σ2 are

Example 1.5.3. The spherical coordinate map

σ(u, v) = (cos u cos v, cos u sin v, sin u), − π2 < u < π

and its variation

σ̃(u, v) = (cos u cos v, sin u, cos u sin v), − π2 < u < π

Theorem 1.5. Let S be a surface. There exists an atlas of it.

S = {p ∈ Ω | f (p) = c, rank Df (p) = k}

A manifold S in Rn which is constructed as in Theorem 1.6 as the set of

is an m-dimensional manifold in Rm+1 .

Example 1.6.6 The set

S = S 1 × S 1 = {x ∈ R4 | x21 + x22 = x23 + x24 = 1}

is a 2-dimensional manifold in R4 . Let

Deﬁnition 1.6.2. Let S be an m-dimensional manifold in Rn . A chart on

As in Theorem 1.5 it is seen that every manifold in Rn possesses an atlas.

1.7 The coordinate map of a chart

In Deﬁnition 1.6.2 we require that σ is injective, but we do not require

Deﬁnition 1.7. Let S be a manifold in Rn , and let σ: U → S be a chart.

Theorem 1.7. Let σ: U → Rn be a chart on a manifold S ⊂ Rn , and let

and a smooth map ϕ: W → V such that

ϕ(σ(q)) = q, σ(ϕ(x)) = x (1.6)

for all q ∈ V and all x ∈ S ∩ W .

Proof. Let W ⊂ Rn be an open neighborhood of p0 such that S ∩ W is

σ̃(x1 , . . . , xm ) = (x, h(x)) ∈ Rn ,

where Ũ ⊂ Rm is open and h: Ũ → Rn−m smooth, then S ∩ W = σ̃(Ũ ). Let

for each point p = (x, h(x)) in S ∩ W .

Let W̃ = W ∩ π −1 (Ṽ ) ⊂ Rn . This is an open set, and it satisﬁes

S ∩ W̃ = S ∩ W ∩ π −1 (Ṽ ) = σ̃(Ũ ) ∩ π −1 (Ṽ ) = σ̃(Ṽ ) = σ(V ).

The map ϕ = (π ◦ σ)−1 ◦ π: W̃ → V is smooth and satisﬁes (1.6).

1.8 Transition maps

f −1 (σ(U )) = {x ∈ Ω | f (x) ∈ σ(U )} ⊂ Rk ,

is called the coordinate expression for f with respect to σ.

Vi = σi−1 (σ1 (U1 ) ∩ σ2 (U2 )) ⊂ Ui

for i = 1, 2. These are open sets and the transition map

Proof. Immediate from Lemma 1.8.

1 Consider the parametrized curve

with t ∈] − ∞, −1[ ∪ ] − 1, ∞[. Verify that γ is a bijection onto the set

2 Let γ be as above, with t ∈ I where −1 ∈ / I ⊂ R. In each of the following

12 Let C ⊂ R2 be a curve in R2 , and put S = C × R = {(x, y, z) | (x, y) ∈ C}.

Show that P = (0, −1, 0) is isolated in M (that is, there is a neighborhood

Prove that this is a surface, and describe it in words.

The notion of a manifold S deﬁned in the preceding chapter assumes S to

2.1 Topological spaces

Example 2.1.3 Let X be an arbitrary set. If we equip X with the collection

Proof. The conditions in Deﬁnition 2.1.1 are easily veriﬁed.

2.2 Abstract manifolds

Example 2.2 Let S ⊂ Rn be an m-dimensional manifold in Rn (see Deﬁni-

In the preceding example S was equipped with an atlas as in Deﬁnition

is a linear bijection Rm → M . We equip M with the distance function

is compatible with the previous atlas. The transition maps σ −1 ◦ τ and

(x, y) → xy, x → x−1 ,