Professional Documents
Culture Documents
Luca Vitagliano
Copyright © 2018 Luca Vitagliano
Licensed under the Creative Commons Attribution-NonCommercial 3.0 Unported License (the
“License”). You may not use this file except in compliance with the License. You may obtain a
copy of the License at http://creativecommons.org/licenses/by-nc/3.0. Unless required
by applicable law or agreed to in writing, software distributed under the License is distributed on an
“AS IS ” BASIS , WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and limitations under the License.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Notation and Conventions 8
3 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1 Tangent Spaces 55
3.2 Tangent Maps 60
3.3 The Tangent Bundle 68
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Introduction
Differential Geometry is one of the big branches of modern Geometry. It studies geometric problems
with tools from Calculus, Linear Algebra and Commutative Algebra. Standard topics in Differential
Geometry are smooth curves and surfaces in the standard Euclidean space Rn . However, nowadays
those notions are encompassed by the abstract definition of a smooth manifold. Smooth manifolds
are the main objects in Differential Geometry, and these notes provide a short introduction to them
and to calculus on them. Historically the definition of a smooth manifold has two main motivations:
Defining a coordinate invariant calculus. In undergraduate Analysis courses we learn calculus
on Rn and its open subsets. The standard Euclidean space is equipped with canonical coordinates
t 1 , . . . ,t n , and we traditionally build calculus in terms of those coordinates. However, it is often
useful to change coordinates. For instance, polar coordinates in R3 are better suited to study the
motion of a particle moving under the action of a central potential with center located at the origin.
We are thus led to the question of how to translate the main constructions in calculus (derivatives,
integrals, etc.) from standard coordinates to polar coordinates, or, more generally, any coordinate
system. There is an even more fundamental question: does calculus have any coordinate invariant
meaning, i.e. is there anything in calculus which is independent of the choice of coordinates? Notice
that this question is particularly relevant in view of the fact that coordinates are actually a human
artifact: there are no coordinates in nature! For instance, in the physical 3-dimensional space (or,
better to say, 4-dimentional space-time) where we live, there are no distinguished coordinates, and
using Cartesian coordinates, as we often do, is an arbitrary human choice, with some advantages
in some cases, but still a human choice. In some sense, Differential Geometry is what remains of
calculus after removing all the dependences on the arbitrary choice of a coordinate system.
Defining a calculus on more general spaces. As we already stressed, in undergraduate
Analysis courses, we learn calculus on Rn and its open subsets. However there are more general
spaces than that where it is useful to perform calculus. For instance, when describing the motion
of a particle on the surface of the 2-dimensional sphere, it would be useful to be able to compute
derivatives of curves and functions on the sphere, independently on how the sphere embeds into R3 .
Even more generally, there are physical systems whose configuration space (space of positions)
cannot be embedded in the standard Euclidean space in any physically meaningful way. The typical
8
example is the rigid body. Let B be a rigid body moving in our 3-dimensional physical Euclidean
space E. We choose Cartesian coordinates on E and think of it as R3 . In particular, it is equipped
with a standard orthonormal frame Rcan . Now let G be the center of mass of B, and let RB be an
orthonormal reference frame attached to B and defining the same orientation as Rcan . Then, the
positions of B in E are parameterized by the position of G, given by its coordinates, together with
the position of RB with respect to Rcan , given by a special orthogonal matrix. Summarizing, the
configuration space of B is (isomorphic to) R3 × SO(3), and a natural question is: is it possible to
define a calculus on such space (so to study the motion of B)? Actually R3 × SO(3) is an example
of a smooth manifold, and calculus on smooth manifolds is a part of Differential Geometry.
So Differential Geometry was born to provide a coordinate invariant calculus on one side, and
to define a calculus on more general spaces on the other side. Later, mostly for its deep conceptual
contents, it became an important field of research with connections to several other branches of
Mathematics including: Algebraic Topology, Algebraic Geometry, Mathematical Physics, Partial
Differential Equations, etc.
covers A.
A countable set is a set which is either finite or in bijection with integers.
We denote by R the field of real numbers. Let n be a non-negative integer and let Rn be
the standard n-dimensional Euclidean space. We will usually denote by t 1 , . . . ,t n the standard
coordinates on Rn interpreting them as real valued functions on Rn . As such, they can be seen as
the components of the identity map idRn : Rn → Rn . The standard coordinates t 1 , . . . ,t n restrict to
any open subset U ⊆ Rn , and we denote with the same symbols t 1 , . . . ,t n the restrictions. When
n = 1, we usually denote by t (or x) the only coordinate. When n = 2 (resp. 3), we usually write x, y
(resp. x, y, z) instead of t 1 ,t 2 (resp. t 1 ,t 2 ,t 3 ). Occasionally, we will deviate from these conventions.
Let n, m be non-negative integers, and let U ⊆ Rn and V ⊆ Rm be open subsets. By definition,
a real function f : U → R is C∞ or smooth (at some point P ∈ U) if it possesses partial derivatives
of arbitrarily high order (in P). Given smooth functions f , g : U → R, the sum
f g : U → R, P 7→ ( f g)(P) := f (P)g(P)
∂ ∂ ∂f
| 1 n f (t 1 , . . . ,t n ) or simply |P f or (P).
∂t i (t ,...,t )=P ∂t i ∂t i
The partial derivative functions will be denoted
∂f 1 ∂f
(t , . . . ,t n ) or simply .
∂t i ∂t i
3 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . 55
3.1 Tangent Spaces
3.2 Tangent Maps
3.3 The Tangent Bundle
■ Example 1.2 Example 1.1 can be slightly generalized considering any n-dimensional real vector
space V and any frame R = (e1 , . . . , en ) of V . Recall that the coordinate map associated to R is
the vector space isomorphism
mapping a vector v to its components (x1 (v), . . . , xn (v)) in the frame R. In their turn, (x1 (v), . . . , xn (v))
are implicitly given by v = ∑ni=1 xi (v)ei . It is clear that (V, ϕR ) is an n-dimensional chart on V
14 Chapter 1. Charts, Atlases and Smooth Manifolds
M U
p
P
"
!
ℝ$
■ Example 1.3 — The Stereographic Charts on the Sphere. Consider the (n + 1)-dimensional
standard Euclidean space Rn+1 with standard coordinates t 1 , . . . ,t n+1 . The n-dimensional sphere is
the subset Sn ⊆ Rn+1 defined by
Sn = P ∈ Rn+1 : ∥P∥ = 1 ,
ℝ'!(
P+
)'
P
j! (#) 0
on the sphere, from Example 1.3. Show that they are indeed charts, and that the stereographic
projections are explicitly given by
Pi Pi
X+i (P) = and X−i (P) = , for P = (P1 , . . . , Pn+1 ) ∈ Sn .
1 − Pn+1 1 + Pn+1
■
■Example 1.4 — More Charts on the Sphere. In this example, we define 2(n + 1) more charts
on Sn . We begin considering the standard n-dimensional open disk, i.e. the open subset Dn ⊆ Rn
defined by
Dn := {P ∈ Rn : ∥P∥ < 1}.
Now, for all i = 1, . . . , n + 1, let Ui,± ⊆ Sn be the subset defined by
Ui,± := (P1 , . . . , Pn+1 ) ∈ Sn : ±Pi > 0 .
tn+1
ℝ!"#
P+
0
p!"# (%)
(!
■ Example 1.5 — Affine Charts on the Projective Space. On Rn+1 ∖ {0} with standard coordi-
nates t 0 , . . . ,t n , consider the equivalence relation ∼ given by
P ∼ Q ⇔ there exists a non-zero real number λ such that P = λ Q.
The n-dimensional (real) projective space is the quotient
RPn := Rn+1 ∖ {0}/ ∼
of Rn+1 ∖ {0} under the equivalence relation ∼. Equivalently, RPn is the space of lines through
the origin in Rn+1 . The equivalence class [P] of the point P = (P0 , . . . , Pn ) ∈ Rn+1 ∖ {0} will be
denoted by
[P0 : · · · : Pn ],
and the real numbers P0 , . . . , Pn are called the homogeneous coordinates of [P] ∈ RPn . In this
example we define n-dimensional charts on RPn . First of all, for every i = 0, . . . , n, let Ui ⊆ RPn be
the subset defined by
Ui := [P0 : · · · : Pn ] : Pi ̸= 0 .
The map
!
n 0 n P0 Pi
c Pn
ϕi : Ui → R , [P] = [P : · · · : P ] 7→ ϕi ([P]) := , . . . , , . . . , ,
Pi Pi Pi
Hence, the (Ui , ϕi ) are all n-dimensional charts on the n-dimensional projective space. They are
called the affine charts on RPn . The i-th affine chart (Ui , ϕi ) is centered at
[0 : · · · : |{z}
1 : · · · : 0].
i-th place
■
1.1 Charts and Atlases 17
Given an n-dimensional chart (U, ϕ) on a set M, we can use the coordinate map ϕ to identify
the coordinate domain U with U b := ϕ(U). In turn, we can use this identification to transfer notions
in calculus (smoothness, derivatives, etc.) from U
b to U. For instance, we could declare that a real
function f : U → R is smooth at a point p ∈ U if the composition f ◦ ϕ −1 : U b → R is smooth at
ϕ(p). If we want to do the same for the whole M, we need to have, at least, a family of charts
A = {(U, ϕ)} covering M in the sense that M = (U,ϕ)∈A U. Suppose we have such a family A ,
S
let f : M → R be a (real valued) function on M, and let p ∈ M. Then, it is natural to declare that f
is smooth at p if the composition f ◦ ϕ −1 : U → R is smooth, for some chart (U, ϕ) ∈ A around
p. However, the latter condition may depend on (U, ϕ), unless A satisfies suitable additional
conditions.
The above discussion motivates the following two definitions.
Definition 1.1.2 — Compatible Charts. Two n-dimensional charts (U, ϕ = (x1 , . . . , xn )), (V, ψ =
(y1 , . . . , yn )) on a set M are compatible if either
✓ U ∩V = ∅, or
✓ U ∩V ̸= ∅ and, additionally, the following two conditions are satisfied:
(1) ϕ(U ∩V ) ⊆ U b and ψ(U ∩V ) ⊆ Vb are (non-empty) open subsets, and
(2) ψ ◦ ϕ : ϕ(U ∩V ) → ψ(U ∩V ) is a diffeomorphism (hence ϕ ◦ ψ −1 : ψ(U ∩V ) →
−1
M
p V
U
j y
P y ∘ j"#
%
$
()
ℝ'
ℝ'
Notice that the compatibility between charts is a reflexive and symmetric relation, but it’s not
transitive in general (can you provide a counter-example?).
Definition 1.1.3 — Atlas. An n-dimensional smooth atlas (or, simply, an atlas) on a set M is a
family A = {(U, ϕ)} of n-dimensional charts such that
✓ A covers M, i.e. M = (U,ϕ)∈A U, and
S
Before providing examples of atlases we want to stress that there are various ways how to
18 Chapter 1. Charts, Atlases and Smooth Manifolds
■ Example 1.7 Let M be a set and let (U, ϕ) be an n-dimensional chart on M. Choose any
diffeomorphism Ψ : U b →W b ⊆ Rn . The composition Ψ ◦ ϕ : U → W
b onto an(other) open subset W b
is bijective, hence (U, Ψ ◦ ϕ) is a chart. Additionally, the transition map between (U, ϕ) and
(U, Ψ ◦ ϕ) is precisely Ψ : U
b →W b . Hence (U, ϕ) and (U, Ψ ◦ ϕ) are compatible. ■
Exercise 1.2 Let M be a set, let p ∈ M be a point, and let (U, ϕ) be a chart on M around p.
Show that there exists a chart centered at p and compatible with (U, ϕ). ■
■ Example 1.9 Let V be an n-dimensional real vector space and let R be a frame of V . The chart
■ Example 1.10 — Two Atlases on the Sphere. Consider the n-dimensional sphere Sn ⊆ Rn+1 .
The stereographic projections (U+ , ϕ+ ) (U− , ϕ− ) (Example 1.3) cover Sn and are compatible (see
Exercise 1.3 below). Hence they form an atlas on the sphere, called the stereographic atlas.
Similarly, the charts (Ui,± , πi ) from Example 1.4 cover Sn and are pairwise compatible. Hence
they form another atlas {(Ui,± , πi )}i=1,...,n+1 on the sphere. Finally, notice that the charts (Ui,± , πi )
are also compatible with the stereographic projections (Exercise 1.3). ■
■ Example 1.11 — The Affine Atlas of the Projective Space. Consider the n-dimensional
projective space RPn . The affine charts (Ui , ϕi ) (Example 1.5) cover RPn and are compatible (see
Exercise 1.4 below). Hence they form an atlas, called the affine atlas of the projective space. ■
Exercise 1.4 Consider the n-dimensional projective space RPn . Show that the affine charts
(Example 1.5) form an n-dimensional atlas on it. ■
We will show in the next chapter that the notion of smoothness of a function f : M → R (at a
point p ∈ M) is independent of the choice of a chart (around p) in a fixed atlas. However, it may
still depend on the atlas, unless we restrict ourselves to a class of compatible atlases.
Definition 1.1.4 — Compatible Atlases. Two atlases A , B on a set M are compatible if the
charts of A are compatible with the charts of B, or, equivalently, if A ∪ B is an atlas as well.
■ Example 1.12 Let V be an n-dimensional real vector space and let R, S be frames of V . The
charts (V, ϕR ) and (V, ϕS ) are compatible. Indeed the transition map is
−1
ϕS ◦ ϕR : Rn → Rn
1.1 Charts and Atlases 19
which is a vector space isomorphism, hence a diffeomorphism. It follows that the atlases
{(V, ϕR )} and {(V, ϕS )}
are compatible. ■
■ Example 1.13 According to Exercise 1.3, the stereographic atlas and the atlas {(Ui,± , πi )}i=1,...,n+1
consisting of the orthogonal projections are compatible atlases on the n-dimensional sphere. ■
Proof. The compatibility between atlases is clearly reflexive and symmetric. It remains to prove
that it is transitive. So, let A , B, C be atlases on a set M and suppose that A is compatible with B,
and B is compatible with C . We want to show that A is compatible with C . To do this, consider a
chart (U, ϕ) in A and a chart (W, χ) in C and show that they are compatible. If U ∩W = ∅ there
is nothing to prove. So let U ∩W ̸= ∅ and look at the transition map:
χ ◦ ϕ −1 : ϕ(U ∩W ) → χ(U ∩W ). (1.1)
We remark, preliminarily, that (1.1) is invertible. Now, ϕ(U ∩W ) ⊆ Ub = ϕ(U) is an open subset.
Indeed, let P ∈ ϕ(U ∩W ), and let p = ϕ (P). As B is an atlas, there is a chart (V, ψ) in B around
−1
Given an atlas A on a set M, it is often useful to consider all charts on M that are compatible
with all charts in A . With this in mind, we give the following
20 Chapter 1. Charts, Atlases and Smooth Manifolds
V
M
p W
U
j c
y
P c ∘ y'( "
"
! y ∘ j'( %
#$
ℝ* ℝ*
ℝ*
Exercise 1.5 Prove Proposition 1.1.2. (Hint: for the first part define
For instance, the charts on a real vector space determined by frames are all in the same maximal
atlas. Similarly, the stereographic atlas on the sphere and the atlas of orthogonal projections are
contained in the same maximal atlas.
Exercise 1.6 — The standard smooth structure on Rn . Consider the standard one-chart atlas
{id : Rn → Rn } on Rn . According to Proposition 1.1.2, it is contained in a unique maximal
atlas Acan , called the standard smooth structure on Rn . Show that Acan consists of charts (U, ϕ),
where U ⊆ Rn is any open subset, and ϕ : U → U b is any diffeomorphism between open subsets
n
of R . ■
1.2 Topological Spaces 21
Proposition 1.1.3 A subset U ⊆ M is A -open if and only if, for every chart (V, ψ) ∈ A ,
ψ(U ∩V ) ⊆ Vb is an open subset.
Proof. The “if” part of the statement is obvious. Let’s prove the “only if” part. So let U ⊆ M
be an A -open subset, and let (V, ψ) be a chart in A . We want to show that ψ(U ∩ V ) ⊆ Vb is
an open subset. If U ∩ V = ∅, then ψ(U ∩ V ) = ∅ ⊆ Vb is an open subset. On the other hand,
let U ∩V ̸= ∅, pick a point P ∈ ψ(U ∩V ), and let p = ψ −1 (P). As U is open, there is a chart
(U, ϕ) ∈ A around p such that ϕ(U ∩U) is open. As (V, ψ) and (U, ϕ) are in the same atlas, they
are compatible, hence the transition map
ψ ◦ ϕ −1 : ϕ(U ∩V ) → ψ(U ∩V )
In the following, we denote by τA the family of A -open subsets in M. The main properties of
τA are summarized in the following
Exercise 1.7 Prove Proposition 1.1.4. (Hint: use Proposition 1.1.3. For Properties (1), (2), (3),
use also the analogous properties enjoyed by the open subsets of Rn ) ■
Properties (1), (2), (3) in Proposition 1.1.4 can be axiomatized to get what is called a topology on
a set. A set equipped with a topology is called a topological space, and the branch of Mathematics
studying topological spaces is Topology. In the next section we briefly recall those fundamentals of
Topology that will be useful in the rest of these notes.
■ Example 1.14 — Atlas Topology. Let (M, A ) be a set equipped with an atlas. The family τA
of A -open subsets in M is a topology, called the atlas topology, which only depends on the smooth
structure Amax containing A : τA = τAmax . Accordingly, M is a topological space, whose open
subsets are A -open subsets. In what follows we will call them just open subsets if there is no risk
of confusion. ■
■ Example 1.15 The open subsets in Rn form a topology denoted τcan and called the standard
topology of Rn . On the other hand the standard atlas A = {id : Rn → Rn } induces its own atlas
topology τA . It is easy to see that, actually, τA = τcan . Indeed a subset U ⊆ Rn is A -open if
id(U ∩ Rn ) is open, i.e. U ∈ τcan . In the following, we will always interpret Rn as a topological
space with the standard topology. ■
■ Example 1.16 On any set X there are at least two canonical topologies:
✓ the smallest possible one, consisting of ∅ and X only, called the concrete topology, and
denoted τcon , and
✓ the largest possible one, consisting of all subsets in X, called the discrete topology, and
denoted τdis .
A topological space (X, τ) is concrete if τ = τcon , and it is discrete if τ = τdis . Notice that a
topological space (X, τ) is discrete if and only if every singleton {p} ⊆ X, with p ∈ X, is an open
subset. It immediately follows from this remark that a 0-dimensional atlas induces the discrete
topology. ■
R Let (X, τ) be a topological space, and let σ be the family of closed subsets of X. It is easy to
see that
(1) ∅, X ∈ σ ,
(2) for every subfamily G ⊆ σ , the intersection C∈G C is in σ ,
T
τY := {Y ∩ U : U ⊆ X is an open subset} .
Then τY is a topology on Y , called the subspace topology. The subset Y ⊆ X equipped with the
subspace topology is called a subspace. Notice that
✓ if Y itself is open in X, then its subspace topology consists of open subsets of X contained in
Y , and, similarly,
✓ if Y is closed, then closed subsets with respect to the subspace topology are closed subsets of
X contained in Y .
■ Example 1.17 Zn ⊆ Rn is a closed and discrete subspace. Every finite subspace in Rn is closed
and discrete. ■
1.2 Topological Spaces 23
■ Example 1.18 Being a subset in Rn+1 , the n-dimensional sphere Sn is a subspace. Denote by τSn
the subspace topology. On the other hand, the stereographic atlas A , or, equivalently, the atlas of
orthogonal projections, induces its own atlas topology τA . We will show in the next section (see
Example 1.21) that, actually, τSn = τA . ■
The notion of topology, allows us to talk about continuity. Before giving the main definition, we
recall that, given open subsets U ⊆ Rm and V ⊆ Rn , the notion of continuous map F : U → V can
be expressed purely in terms of the (subspace) topologies of U and V . Namely, a map F : U → V
is continuous if and only if, for every open subset V ⊆ V , the preimage F −1 (V ) ⊆ U is an open
subset. This suggests the following
Definition 1.2.2 — Continuous Map. A map F : X → Y between topological spaces is contin-
uous if, for every open subset V ⊆ Y , the preimage F −1 (V ) ⊆ X is an open subset in X.
Let (X, τX ) and (Y, τY ) be topological spaces and let F : X → Y be a map. We write F :
(X, τX ) → Y , or F : X → (Y, τY ), or F : (X, τX ) → (Y, τY ), if we want to insist that X and Y are
topological spaces with the topologies τX and τY respectively (and not different ones).
For instance, every map F : X → (Y, τcon ) to a concrete space is continuous. Every map
F : (X, τdis ) → Y from a discrete space is continuous. The inclusion iA : A ,→ X of a subspace
A ⊆ X is also continuous. Indeed, for any subset B ⊆ X, the preimage iY−1 (B) is simply A ∩ B, so, if
B is open, iY−1 (B) is open (with respect to the subspace topology).
For any topological space (X, τ), the identity map id : (X, τ) → (X, τ) is continuous. However,
if the domain and the codomain are equipped with different topologies, say τ and τ ′ , then id :
(X, τ) → (X, τ ′ ) needs not to be continuous. For instance, if X possesses more than one element,
then the concrete topology τcon and the discrete topology τdis on X are different, and, while the
map id : (X, τdis ) → (X, τcon ) is continuous, the map id : (X, τcon ) → (X, τdis ) is not. Finally, it is
easy to see that the composition of continuous maps is continuous. It follows that the restriction
F : A → Y of a continuous map F : X → Y to a subspace A ⊆ X is also continuous. Indeed, it is the
composition of the continuous map iA : A ,→ X followed by F : X → Y .
■ Example 1.19 Let G : X → Y be a (not necessarily continuous) map between topological spaces,
let B ⊆ Y be a subspace, and suppose that G takes values in B. In particular, we can consider the
restriction G : X → B of G : X → Y to B in the codomain, and G : X → B is continuous if and
only if so is G : X → Y . Indeed, let G : X → B be continuous, then G : X → Y is continuous as
well, because it is the composition of the continuous map G : X → B followed by the inclusion
iB : B ,→ Y . Conversely, let G : X → Y be continuous, and let U ⊆ B be an open subset. To avoid
confusion, denote GB : X → B the restriction. By definition of subspace topology, there is an open
subset U ⊆ Y such that U = U ∩ B = i−1 B (U ). Hence
G−1 −1 −1 −1 −1
B (U) = GB (iB (U )) = (iB ◦ GB ) (U ) = G (U )
which is open because G : X → Y is continuous. ■
Notice that, as the example provided by the maps id : (X, τdis ) → (X, τcon ) and id : (X, τcon ) →
(X, τdis ) shows, a continuous, bijective map needs not to be a homeomorphism, hence we cannot
remove the last condition from the definition. On the other hand, the inverse of a homeomorphism
is clearly always a homeomorphism. The identity id : (X, τ) → (X, τ) is a homeomorphism and the
composition Ψ ◦ Φ : X → Z of homeomorphisms Φ : X → Y and Ψ : Y → Z is a homeomorphism
whose inverse homeomorphism is Φ−1 ◦ Ψ−1 : Z → X. Finally, diffeomorphisms between open
subsets of Rn are, in particular, homeomorphisms.
24 Chapter 1. Charts, Atlases and Smooth Manifolds
A Hausdorff space enjoys the nice property that its one-point subspaces are closed. Indeed
let X be a Hausdorff space, and let p ∈ X be a point. We want to show that X ∖ {p} is open. To
do this take q ∈ X ∖ {p}, i.e. q ̸= p. From the Hausdorff property, there is, in particular, an open
neighborhood Vq of q not containing p, i.e. Vq ⊆ X ∖ {p}. Hence
[
X ∖ {p} = Vq
q∈X∖{p}
is open.
Not all topological spaces are Hausdorff. For instance, a concrete space with more than one
element is not Hausdorff.
We now come to II-countability. First we need to discuss the new notion of basis for a topology.
Let (X, τ) be a topological space. A basis for the topology τ is a subfamily B ⊆ τ such that every
open subset can be written as a union of elements of B. For instance, the open disks form a basis
for the standard topology of Rn . The open disks with rational centers and radii do also form a basis.
Finally, multi-rectangles
I1 × · · · × In ∈ Rn ,
with I1 , . . . , In ⊆ R open intervals, do also form a basis (and we can even restrict to multi-rectangles
with rational vertices). Notice that any basis for a discrete topology should contain all one-point
subsets.
Let (X, τ) be a topological space, and let B be a basis for τ. It immediately follows from the
definition, that B is an open cover of X, i.e. it consists of open subsets covering X: X = ∪U ∈B U .
Additionally, for all U , V ∈ B, the intersection U ∩ V is a union of open subsets in B. Conversely,
if X is a set, and B is a family of subsets covering X and such that for all U , V ∈ B, the intersection
U ∩ V is a union of subsets in B, then B is a basis for a, necessarily unique, topology τB .
Specifically, τB consists of subsets of the form
U
[
U ∈F
for some subfamily F ⊆ B. The proof of this fact is easy and it is left as Exercise 1.9.
1.2 Topological Spaces 25
Exercise 1.9 Let X be a set, and let B be a family of subsets covering X and such that for all
U , V ∈ B, the intersection U ∩ V is a union of subsets in B. Show that
τB := { U ∈F U : F ⊆ B}
S
Given a topological space (X, τ), it is often useful to have a basis B for the topology τ. For
instance a map F : Y → X from another topological space Y is continuous provided only F −1 (B) ⊆ Y
is open for any B ∈ B (show it as an exercise!).
A topological space (X, τ) is II-countable if there is a countable basis for the topology τ. For
instance, Rn is II-countable (open disks with rational centers and radii are countably many), and
any concrete space is II-countable. Every subspace of a II-countable space is II-countable (see
Exercise 1.10), so the n-dimensional sphere, with the subspace topology, is II-countable.
Exercise 1.10 Show that every subspace of a II-countable space is II-countable. (Hint: let (X, τ)
be a topological space and let Y ⊆ X be a subspace. Denote by τY the subspace topology of Y .
Show, preliminarily, that, given a basis B for τ, the family of subsets BY := {B ∩Y : B ∈ B} is
a basis for τY . Now, choose B to be countable.) ■
Not all topological spaces are II-countable. For instance, a discrete space with uncountably
many points is not II-countable.
We conclude this section discussing products of topological spaces. Let X1 , . . . , Xk be topologi-
cal spaces. In the Cartesian product X1 × · · · × Xk consider the following family of subsets
It is easy to see that B enjoys the properties of the basis, i.e. the properties stated in Exercise 1.9
(show it as an exercise). It follows that B is a basis for a topology on X1 × · · · × Xk called the
product topology. The Cartesian product X1 × · · · × Xk equipped with the product topology is called
the product space. Notice that, if, for all i, Bi is a basis for the topology of Xi , then
pr−1 U × · · · × Xk
i (U ) = X1 × · · · × |{z}
i-th place
is a homeomorphism. Indeed, first remember that the open multi-rectangles form a basis for the
topology of Rk (for all k). Hence products of open multi-rectangles form a basis for the topology
of Rn1 × · · · × Rnk . Now, the preimage with respect to Φ of an open multi-rectangle in Rn is the
26 Chapter 1. Charts, Atlases and Smooth Manifolds
The product of Hausdorff spaces is Hausdorff, and the product of II-countable spaces is II-
countable. Indeed, let X1 , . . . , Xk be Hausdorff spaces. Take two different points p = (p1 , . . . , pk )
and q = (q1 , . . . , qk ) in the product space X1 × · · · × Xk . This means that there exists i ∈ {1, . . . , k}
such that pi ̸= qi . As Xi is Hausdorff, pi , qi are separated by open subsets Ui ,Vi . But then p, q are
separated by the open subsets pr−1 −1
i (U1 ), pri (Vi ). For the II-countability, it is enough to notice
that, if B1 , . . . , Bk are countable bases of X1 , . . . , Xk then B × (see (1.2)) is a countable basis of
X1 × · · · × Xk .
Proposition 1.3.1 — Properties of the Atlas Topology. Let (M, A ) be a set equipped with an
n-dimensional atlas. The atlas topology τA induced on M by A is the unique topology such
that, for every chart (U, ϕ) ∈ A ,
(1) the coordinate domain U ⊆ M is open, and
(2) the coordinate map ϕ : U → U b ⊆ Rn is a homeomorphism.
Here both U and U b are equipped with the subspace topology.
Proof. First of all, we show that the atlas topology τA possesses both Properties (1) and (2) in
the statement. Property (1) is clear. For Property (2), we have to show that both ϕ : U → U b and
−1
ϕ :U b → U are continuous. Equivalently, we have to show that for every open subsets V ⊆ U
and V ⊆ U,
b b both ϕ −1 (Vb ) and ϕ(V ) = (ϕ −1 )−1 (V ) are open. So let Vb ⊆ U
b be an open subset, and
−1 −1 −1
consider ϕ (Vb ) ⊆ U. We have ϕ(ϕ (Vb ) ∩U) = ϕ(ϕ (Vb )) = Vb , which is open. This shows
that ϕ −1 (Vb ) ∈ τA . So it is an open subset of M contained in the open subspace U, hence it is open
in the subspace U. Now, let V ⊆ U be an open subset. Being an open subset in an open subspace, it
belongs to τA . In particular, ϕ(V ∩U) = ϕ(V ) ⊆ U b is an open subset.
′
Now we prove uniqueness. Let τ, τ be topologies on M with both Properties (1) and (2), and
prove that τ = τ ′ . First, for any chart (U, ϕ) ∈ A , denote by τU the subspace topology on U
induced by τ, and by τU′ the subspace topology induced by τ ′ . Take an element U ∈ τ. From
Property (1), U ∈ τ for every (U, ϕ) ∈ A , hence U ∩ U ∈ τ. So U ∩ U is an open subset of
(M, τ) contained in the open subspace U. It follows that U ∩ U ∈ τU . As ϕ : (U, τU ) → U b is a
homeomorphism, ϕ(U ∩ U ) ⊆ U is an open subset. But ϕ : (U, τU ) → U is a homeomorphism as
b ′ b
well, hence U ∩ U ∈ τU′ ⊆ τ ′ . Finally,
U = U ∩U
[
(U,ϕ)∈A
is the open union of open subsets of (M, τ ′ ). This shows that τ ⊆ τ ′ . Exchanging the role of τ and
τ ′ we see that τ ′ ⊆ τ, and this concludes the proof. ■
■ Example 1.21 — The Atlas Topology on the Sphere is the Subspace Topology. Consider the
n-dimensional sphere Sn ⊆ Rn+1 , and its stereographic atlas A . In this example, we use Proposition
1.3.1 to show that the subspace topology τSn and the atlas topology τA coincide. According to
Proposition 1.3.1, it is enough to show that τSn possesses Properties (1) and (2) in the statement. So,
consider the stereographic charts (U+ , ϕ+ ), (U− , ϕ− ). The coordinate domain U+ is
U+ = Sn ∖ {P+ } = Sn ∩ (Rn ∖ {P+ }).
1.3 Smooth Manifolds 27
As the second factor is open in Rn , it follows that U+ is open in the subspace topology of Sn , and
similarly for U− . For Property (2), notice that ϕ+ : U+ → Rn is the restriction to the subspace U+
of the continuous map:
where
Pi
F i (P) = , i = 1, . . . , n.
1 − Pn+1
Hence, ϕ+ is continuous with respect to the subspace topology. It is easy to see that its inverse
ϕ+−1 : Rn → U+ is the restriction to the subspace U+ ⊆ Rn+1 in the codomain of the continuous
map
where
2Qi
Gi (Q) = , for i = 1, . . . , n,
∥Q∥2 + 1
and
∥Q∥2 − 1
Gn+1 (Q) = .
∥Q∥2 + 1
Hence, from Example 1.19, ϕ+−1 : Rn → U+ is also continuous with respect to the subspace
topology. We conclude that ϕ+ : U+ → Rn is a homeomorphism with respect to the subspace
topology. Similarly for ϕ− . Summarizing, τSn shares both Properties (1) and (2) in the statement
of Proposition 1.3.1 and coincides with τA . It immediately follows that (Sn , τA ) is a Hausdorff,
II-countable topological space. ■
Let (M, A ) be a set equipped with a smooth structure. From now on, by a chart (resp. an atlas)
on M we will always mean a chart in A (resp. an atlas contained in A ) unless otherwise stated.
R For every open subset U ⊆ M and every p ∈ U , there is a chart (U, ϕ) around p such that
U ⊆ U . In other words, the coordinate domains of a maximal atlas A form a basis for the
topology τA . Indeed, let (U ′ , ϕ) be any chart around p, and put U := U ′ ∩ U . Then U is
an open subset contained in U ′ and (U, ϕ) is a subchart such that U ⊆ U . But, as maximal
atlases contain all subcharts of its charts, we have (U, ϕ) ∈ A .
There are simple criteria to check whether (M, A ) is a smooth manifold or not, i.e. whether
(M, τA ) is Hausdorff and II-countable or not.
28 Chapter 1. Charts, Atlases and Smooth Manifolds
Proposition 1.3.2 Let (M, A ) be a set equipped with a smooth structure. Then
✓ (M, τA ) is Hausdorff if and only if for any two points p, q ∈ M either p, q belong to the
same coordinate domain, or they belong to disjoint coordinated domains;
✓ (M, τA ) is II-countable if and only if it possesses a countable atlas.
Proof. Let p, q be distinct points of M. If M is Hausdorff, then p, q are separated by open subsets
U , V . As coordinate domains form a basis for the topology of M, then p, q are also separated by
coordinate domains. Conversely, let p, q belong to disjoint coordinate domains U,V , then U,V ⊆ M
are open subsets separating them. On the other hand, if p, q belong to the coordinate domain U of
the same chart (U, ϕ), then ϕ(p), ϕ(q) ∈ U b are distinct point. As Ub is Hausdorff, there are open
−1 −1
subsets V,W ⊆ U separating them. Hence ϕ (V ), ϕ (W ) are open subsets of U (hence of M)
b
separating p, q.
For the second part of the statement, let B ⊆ τA be a countable basis. Denote by BA ⊆ B the
subfamily consisting of open subsets B ∈ B with the property that there exists (U, ϕ) ∈ A such
that B ⊆ U. As the coordinate domains cover M, and every coordinate domain is covered by open
subsets in BA , it follows that BA covers M. Additionally, being an open subset in a coordinate
domain, every B ∈ BA can be seen itself as the coordinate domain of a (sub)chart (B, ϕ). Choose
just one such chart (B, ϕB ) for every B and notice that {(B, ϕB ) : B ∈ BA } is a countable atlas on
M. Conversely, let A0 ⊆ A be a countable atlas. For every (U, ϕ) ∈ A0 , let FU be the family
of open subsets of U of the form ϕ −1 (D), where D ⊆ U b is an open disk with rational center and
radius. It is clear that
FU
[
(U,ϕ)∈A0
■ Example 1.23 Let V be an n-dimensional real vector space. We know that all the charts (V, ϕR )
determined by frames R of V belong to the same maximal atlas Amax . Then (V, Amax ) is an
n-dimensional smooth manifold. ■
■ Example 1.24 — Real Projective Spaces are Manifolds. The n-dimensional real projective
space RPn equipped with its affine atlas is an n-dimensional smooth manifold. Indeed, let [P̄] =
[P̄0 : · · · : P̄n ], [Q̄] = [Q̄0 : · · · : Q̄n ] be distinct points of RPn . Then either they belong to the same
coordinated domain or they don’t belong simultaneously to any of the affine coordinate domain, in
other words, whenever Pi ̸= 0 we must have Qi = 0 and vice-versa. In this second case, without
loosing much generality, we assume that P̄0 , . . . , P̄k−1 ̸= 0, P̄k = · · · = P̄n = 0 for some 0 < k ≤ n.
Then we have Q̄0 = · · · = Q̄k−1 = 0 and Q̄k , . . . , Q̄n ̸= 0, and we can even choose P̄0 = Q̄k = 1.
Now, let δ , ε be positive real numbers, and, in Rn , consider the following open subsets:
and
(B1 , . . . , |{z}
1 , . . . , Bn ) with |B1 | < ε. (1.4)
k-th place
1 1
− 0 ̅ 𝑘−1
𝜀 𝑃 𝜀
[ ] ℝ
Finally, we show that products of smooth manifolds are smooth manifolds, and open subsets in
a smooth manifold are smooth manifolds in a natural way. This will enrich significantly our list of
examples.
Let M1 , . . . , Mk be smooth manifolds of dimension n1 , . . . , nk and let A1 , . . . , Ak be their smooth
structures. Put n := n1 + · · · + nk . First of all, we can construct an n-dimensional atlas on the
product M1 × · · · × Mk as follows. For every i = 1, . . . , k, let (Ui , ϕi ) ∈ Ai be a chart on Mi . Consider
the product U1 × · · · ×Uk ⊆ M1 × · · · × Mk and the map
ϕ1 × · · · × ϕk : U1 × · · · ×Uk → U bk ⊆ Rn ,
b1 × · · · × U (p1 , . . . , pk ) 7→ (ϕ1 (p1 ), . . . , ϕk (pk )).
is a homeomorphism between the product spaces. Finally, use this fact and Proposition 1.3.1 to
prove the claim.) ■
1
T n := S · · × S}1
| × ·{z
n times
We now come to open subsets in a manifold. So, let M be a manifold, let A be its smooth
structure, and let U ⊆ M be an open subset. Every chart (U, ϕ) ∈ A such that U ⊆ U is clearly a
chart on U . Additionally, the coordinate domains of such charts cover U . In other words
AU := {(U, ϕ) ∈ A : U ⊆ U }
is an atlas on U . It is almost immediate to see that the atlas topology induced by AU coincides
with the subspace topology (do you see it?). Hence with the atlas topology, U is Hausdorff and
II-countable. We conclude that U is a manifold of the same dimension as M. Every such manifold
is called an open submanifold of M.
■Example 1.26 — The General Linear Group. Consider the vector space M(n, R) of n × n real
matrices. The general linear group GL(n, R) consists, by definition, of invertible matrices, or,
equivalently, non-zero determinant matrices. As the determinant map
det : M(n, R) → R
is continuous (even polynomial), and R ∖ {0} ⊆ R is an open subset, it follows that GL(n, R) =
det−1 (R ∖ {0}) ⊆ M(n, R) is an open subset, hence an open submanifold of M(n, R). ■
2. Smooth Maps and Submanifolds
In this chapter, we show how the smooth structure on a smooth manifold allows us to define the
notion of smoothness. In particular, we will define smooth real valued functions on a manifold,
smooth maps between manifolds, and smooth subsets of a manifold, called submanifolds. These
notions are the first building blocks of a differential calculus on manifolds, in particular a coordinate-
free calculus.
Proposition 2.1.1 A function M → R is smooth if and only if all its coordinate representations
are smooth.
Proof. The “if part” of the statement is obvious. Let’s prove the “only if part”. So let f : M → R
be a smooth function, and let (V, ψ) be any chart on M. We want to show that the coordinate
representation fbV = f ◦ ψ −1 : Vb → R is smooth. It is enough to show that fbV is smooth around every
point. To do this, take Q ∈ Vb , and let p = ψ −1 (Q). As f is smooth, there is a chart (U, ϕ) around p
such that the coordinate representation fb = f ◦ ϕ −1 : Ub → R is smooth. Hence, so is the restriction
fb : ϕ(U ∩V ) → R of fb to the open subset ϕ(U ∩V ). Consider the open subset ψ(U ∩V ) ⊆ Vb and
32 Chapter 2. Smooth Maps and Submanifolds
notice that the restriction fbV : ψ(V ∩U) → R is given by the following composition (see Figure
2.1)
fbV = fb◦ (ϕ ◦ ψ −1 ).
As both the transition map ϕ ◦ ψ −1 and the coordinate representation fb are smooth, so is fbV :
ψ(V ∩U) → R. From the arbitrariness of Q, the coordinate representation fbV : Vb → R is smooth
as well. ■
M
p V
U
j y
j ∘ y"#
%
$
Q ()
f +,-
ℝ'
+, ℝ'
ℝ
0
Exercise 2.1 Let Q(t 0 , . . . ,t n ), R(t 0 , . . . ,t n ) be homogeneous polynomials of the same degree
in the real undeterminates t 0 , . . . ,t n . Assume, additionally, that R(t 0 , . . . ,t n ) does only vanish at
the origin (e.g., this is the case if R(t 0 , . . . ,t n ) = (t 0 )2 + · · · + (t n )2 ). Prove that the real valued
function
Q(P0 , . . . , Pn )
f : RPn → R, [P] = [P0 : · · · : Pn ] 7→ f ([P]) :=
R(P0 , . . . , Pn )
It is clear that a function f : M → R is smooth if and only if it is locally smooth, i.e., for every
point p ∈ M, there exists an open neighborhood U ⊆ M of p such that the restriction f |U : U → R
of f to U is smooth, where we interpret U as a manifold with the open submanifold structure. The
claim easily follows from the fact that the charts in M whose coordinate domain is contained in U
form an atlas on U (check the details as an exercise!). Otherwise stated, it holds the following
2.1 Smooth Functions 33
Lemma 2.1.2 — Gluing Lemma. Let M be a manifold, and let C = {U } be an open cover
of M (i.e. every elements U ∈ C is an open subset U ⊆ M, and M = U ). For every
S
U ∈C
U ∈ C , let fU : U → R be a smooth function. If
fV |U ∩V = fU |U ∩V
for every U , V ∈ C , then there exists a unique smooth function f : M → R such that
f |U = fU
for all U ∈ C .
There are a lot of smooth functions on a manifold, as the following theorem shows.
Theorem 2.1.3 — Bump Function Theorem. Let M be a smooth manifold. For every point
p ∈ M, and every open neighborhood U ⊆ M of p, there exists a smooth function β : M → R on
M such that β = 1 around p (i.e. there is an open neighborhood V ⊆ M of p such that β |V = 1),
and the support of β is contained in U .
The statement of Theorem 2.1.3 needs some explanations. First of all, let X be a topological
space, and let A ⊆ X be a subset. The closure of A in X is the smallest closed subset A ⊆ X
containing A. Equivalently, A is the (closed) intersection of all closed subsets of X containing A.
Given a real valued function f : X → R, the support of f is the closure
Notice that, if f is a continuous function, then the subset {p ∈ X : f (p) ̸= 0} is open (while the
support of f is always closed by definition).
We omit a full proof of Theorem 2.1.3. However, we stress that the Hausdorff property of M
is a key ingredient of the proof (and the theorem does not hold true in general for a set M with a
smooth structure A such that (M, τA ) is not Hausdorff). A smooth function β like in the statement
of Theorem 2.1.3 is called a bump function (relative to the data (p, U )). Notice that, as a real
valued function on M vanishes outside its support, given a bump function relative to the data (p, U ),
in particular we have f |M∖U = 0. As already mentioned, we will not prove the Bump Function
Theorem in full generality. However, in Example 2.2 we show that it is true when M = Rn . The
general case can be proved from this one (via charts, the Hausdorff property of manifolds, and
some little more Topology). Before doing this we present some algebraic structures on the space of
smooth functions on a manifold. We begin defining real algebras.
Definition 2.1.2 — Algebra. A real, associative, commutative algebra with unit, or simply
an R-algebra, or just an algebra, is a real vector space (A, +, ·) equipped with an additional
interior operation ⋆ : A × A → A such that
✓ ⋆ is R-bilinear,
✓ ⋆ is associative.
In particular, (A, +, ⋆) is a ring. Additionally, we require that the ring (A, +, ⋆) is commutative
and possesses a unit. A subalgebra of an algebra A is a vector subspace B ⊆ A which is
additionally a subring containing 1.
Notice that a subalgebra B of an algebra A is also an algebra with the restricted operations.
As for rings and vector spaces, the symbols ⋆, · are usually omitted in products, and we write aα,
and αβ instead of a · α and α ⋆ β , for a ∈ R, and α, β elements in an algebra. As a first trivial
example, notice that R is an algebra in the obvious way. Complex numbers do also form a real
algebra. More generally, for any algebra A, the map i : R → A, a 7→ a · 1, is an injection identifying
34 Chapter 2. Smooth Maps and Submanifolds
R with a subalgebra of A. In the following, we will often identify R with its image i(R) under i.
This shows that an algebra A can be also seen as a ring extension R ⊆ A.
■ Example 2.1 — Algebra of Real Valued Functions. Let X be a set. The space F (X, R) of real
valued functions f : X → R is an algebra with the following point-wise operations
where f , g ∈ F (X, R), and a ∈ R. The zero in F (X, R) is the constant function whose constant
value is 0, the 1 in F (X, R) is the constant function whose constant value is 1, and the inclusion
i : R ,→ F (X, R) identifies a real number a with the constant function, also denoted a, whose
constant value is a itself. ■
Exercise 2.2 Prove Proposition 2.1.4. (Hint: show that, given smooth functions f , g, a real
number a, and a chart (U, ϕ), we have the following relations between coordinate representa-
tions:
f + g = fb+ gb,
[
f · g = fb· gb,
d
acf = a fb.
Finally, use the fact that sum, products, and products by real scalars of smooth functions on U
b
are smooth functions.) ■
By studying the derivatives of f at the origin, one can prove that f is a smooth function. Second,
for every two real numbers r, R such that r < R, we define a smooth function h : R → R by putting
f (R − t)
h(t) = , t ∈ R.
f (R − t) + f (t − r)
Clearly, h(t) = 1 for all t ≤ r, and h(t) = 0 for all t ≥ R. Additionally, 0 < h(t) < 1 for r < t < R.
The function h is called a cut-off function. Finally, let P0 ∈ Rn , and let U ⊆ Rn be any open
neighborhood of P0 . There are positive real numbers r, R such that r < R, and, additionally, the
2.2 Smooth Maps 35
1
"
0
ℝ
1
ℎ
0
ℝ
# $
closed disks Dr , DR centered at P0 with radii r, R are contained in U . It follows that the smooth
function
β : Rn → R, P 7→ β (P) := h (∥P − P0 ∥)
is a bump function relative to the data (P0 , U ). Indeed β |Dr = 1, in particular, f = 1 in an open
neighborhood of P0 , and supp β = DR ⊆ U . ■
$
&'
%
ℝ"
U
M F N
p V
((,)
U
j
y
%&
P (&
"
! (& (*)
ℝ$ ℝ'
M and (V, ψ = (y1 , . . . , yn )) a chart on N such that F(U) ⊆ V . In this situation, we will sometimes
consider the vector valued map ψ ◦ F : U → Vb . Its components will be denoted F 1 , . . . , F n :
ψ ◦ F = (F 1 , . . . , F n ) and we will write F a = F a (x1 , . . . , xm ), a = 1, . . . , n, to express the fact that
the F a can be seen as functions of the coordinates x1 , . . . , xm . Notice that F a = ya ◦ F : U → R for
all a = 1, . . . , n, and their coordinate representations Fba with respect to the chart (U, ϕ) are exactly
the components of the coordinate representation F: b Fb = (Fb1 , . . . , Fbn ). Accordingly, F is smooth if
and only if it has coordinate representations FbU,V around every point and, additionally, the F a are
smooth for every (U, ϕ), (V, ψ) as above.
We now provide a list of examples of smooth maps. A map F : U → V between open subsets
of standard Euclidean spaces U ⊆ Rm and V ⊆ Rn is smooth in the sense of Definition 2.2.1 if
and only if it is smooth in the standard sense. All constant maps const : M → N have constant
coordinate representations, hence they are smooth. A map f : M → R is smooth if and only if it
is a smooth function in the sense of Definition 2.1.1. Finally, the identity id : M → M is a smooth
map. More generally, the inclusion iU : U ,→ M of an open submanifold U ⊆ M has coordinate
representations, and they can be chosen to be simply the identity (choose in U and in M the same
chart). Hence iU is a smooth map.
■ Example 2.3 Consider the n-dimensional projective space RPn , and let
be the canonical projection. Then π is a smooth map. To see this, fix a point P0 = (P00 , . . . , P0n ) ∈
Rn+1 ∖ {0}. Then there exists i ∈ {0, . . . , n} such that P0i ̸= 0. Denote Vi := {t i ̸= 0} ⊆ Rn+1 ∖ {0}.
It is an open subset and (Vi , id) is a subchart of the chart (Rn+1 ∖ {0}, id) on Rn+1 ∖ {0} around P0 .
Additionally, π(Vi ) ⊆ Ui , the coordinate domain of the i-th affine chart (Ui , ϕi ) on RPn . This shows
that π has coordinate representations around every point P0 . Now, we compute
πb = ϕi ◦ π ◦ id−1 = ϕi ◦ π : Vbi = Vi → Rn .
So πb is a smooth map for all i (it has smooth components). We conclude that π is a smooth map. ■
Exercise 2.4 Prove that the inclusion iSn : Sn ,→ Rn+1 of the n-dimensional sphere in the
(n + 1)-dimensional standard Euclidean space is a smooth map. ■
■ Example 2.4 — Projections onto Factors are Smooth. Let M1 , . . . , Mk be smooth manifolds of
dimensions n1 , . . . , nk respectively. Consider the product manifold M1 × · · · × Mk . The projection
onto the i-th factor:
is a smooth map for every i = {1, . . . , k}. Indeed, let ( p̄1 , . . . , p̄k ) ∈ M1 × · · · × Mk , and, for every
j = {1, . . . , k}, let (U j , ϕ j ) be a chart on M j around p̄ j . Consider the product chart
(U1 × · · · ×Uk , ϕ1 × · · · × ϕk )
38 Chapter 2. Smooth Maps and Submanifolds
Clearly pri (U1 × · · · ×Uk ) ⊆ Ui , showing that pri has coordinate representation
pr b1 × · · · × U
bi :U bk → U
bi .
n
Additionally, denote by (t(1j) , . . . ,t( j)j ) the standard coordinates on U
b j . Then pr
b i is given by
1 n1 1 nk 1 ni
b i (t(1)
pr , . . . ,t(1) , . . . ,t(k) , . . . ,t(k) ) = (t(i) , . . . ,t(i) ),
which is smooth. ■
Lemma 2.2.2 — Gluing Lemma for Smooth Maps. Let M, N be manifolds, and let C = {U }
be an open cover of M. For every U let FU : U → N be a smooth map. If
FU |U ∩V = FV |U ∩V
for every U , V ∈ C , then there exists a unique smooth map F : M → N such that
F|U = FU
for all U ∈ C .
Let M, N be manifolds, and let F : M → N be a map. In order to check whether or not F is
smooth we have to check, first of all, whether or not F has coordinate representations. Notice
that continuous maps F : M → N do always have coordinate representations around every point.
Indeed, let p ∈ M, and let (V, ψ) be a chart on N around F(p). The preimage F −1 (V ) ⊆ M is
an open neighborhood of p. Hence there is a chart (U, ϕ) around p such that U ⊆ F −1 (V ) (the
coordinate domains of charts in the maximal atlas form a basis for the topology). It follows that
F(U) ⊆ V . Actually, a map F : M → N is smooth if and only if it is continuous, and its coordinate
representations (they exist by continuity) are smooth. The only non-obvious thing in this claim is
that a smooth map is continuous. This is the content of the next
Proof. Let F : M → N be a smooth map between manifolds, and let V ⊆ N be an open subset. We
want to show that the preimage F −1 (V ) ⊆ M is an open subset. So, let p ∈ F −1 (V ) be a point,
and let (U, ϕ) be a chart on M around p, and (V, ψ) a chart on N around F(p) such that F(U) ⊆ V
(they exist because F is smooth). Consider also the coordinate representation
Fb = ψ ◦ F ◦ ϕ −1 : U
b → Vb .
Exercise 2.6 Complete the proof of Proposition 2.2.3 showing that for any smooth map
F : M → N, any chart (U, ϕ) on M, any chart (V, ψ) on N such that F(U) ⊆ V , and any open
subset V (intersecting V ) we have
M N
F -1(V ) F V
p
V ((,)
U
j
y
%&
P (&
"
! (& (*)
ℝ$ ℝ'
Proof. Let F : M → N, and G : N → Q be smooth maps between manifolds. We want to show that
G ◦ F : M → Q is a smooth map. So take a point p ∈ M. As G is smooth, there is a chart (V, ψ) on
N around F(p), and a chart (W, χ) on Q around G(F(p)) such that G(V ) ⊆ W and the coordinate
representation G b is smooth. Since F is continuous, F −1 (V ) ⊆ M is an open neighborhood
b : Vb → W
of p. Hence there is a chart (U, ϕ) on M around p such that U ⊆ F −1 (V ), i.e. F(U) ⊆ V . Denote
b → Vb the coordinate representation of F with respect to (U, ϕ) and (V, ψ). All coordinate
by Fb : U
representations of a smooth map are smooth, so Fb is smooth. Collecting all previous remarks we get
M N Q
𝐹 −1 (𝑉) F G
p
U 𝐹(𝑝) V 𝐺(𝐹(𝑝)) W
ℝ𝑚 ̂
𝑈 ℝ𝑛 𝑉 ℝ𝑞 ̂
𝑊
b ◦ Fb : U
that G b →W
b is a smooth coordinate representation of G ◦ F around p. From the arbitrariness
40 Chapter 2. Smooth Maps and Submanifolds
F : M → M1 × · · · × Mk .
Φ : Rn1 × · · · × Rnk → Rn
R : Sn → Sn , P 7→ −P
■ Example 2.7 Let M be a smooth manifold, and let k be a positive integer. Consider the product
M ×k := M
| × ·{z
· · × M}.
k times
Prove that it is a well-defined diffeomorphism. (Hint: consider the affine charts (U0 , ϕ0 ), (U1 , ϕ1 )
on the projective line RP1 and the stereographic charts (U+ , ϕ+ ), (U− , ϕ− ) on the circle. Prove
that Φ(U0 ) ⊆ U+ and Φ(U1 ) ⊆ U− , and, more precisely,
Proposition 2.2.5 Let M be an n-dimensional manifold and let (U, ϕ) be a chart on M. Then
the coordinate map
ϕ :U →U
b
H(αβ ) = H(α)H(β ).
H(1) = 1.
Notice that an algebra homomorphism is, in particular, a ring homomorphism. The identity
id : A → A is an algebra homomorphism. More generally, the inclusion of a subalgebra is an algebra
homomorphism. The composition of algebra homomorphisms is an algebra homomorphism and the
inverse of an algebra isomorphism is an algebra isomorphism. Finally, the composition of algebra
isomorphisms is an algebra isomorphism.
42 Chapter 2. Smooth Maps and Submanifolds
■ Example 2.8 — Pull-Back of Real Valued Functions. Let X,Y be sets and let F : X → Y be a
map. Consider the algebras of real-valued functions F (X, R), F (Y, R). It is easy to see that the
map
is an algebra homomorphism, called the pull-back along F. It is clear that the pull-back along the
identity id : X → X is the identity id : F (X, R) → F (X, R). If Z is another set and G : Y → Z is
another map. Then
(G ◦ F)∗ = F ∗ ◦ G∗ .
We leave to the reader to check all the claims in this example as Exercise 2.10. ■
Now, let F : M → N be a smooth map between smooth manifolds. It follows from Proposition
2.2.4 that the pull-back F ∗ : F (N, R) → F (M, R) takes smooth functions to smooth functions:
F ∗ (C∞ (N)) ⊆ C∞ (M).
The restriction F ∗ : C∞ (N) → C∞ (M) is also an algebra homomorphism, and we also call it
the pull-back along F. Analogous remarks as for maps of sets hold in the case of smooth maps of
manifolds: the pull-back along the identity id : M → M is the identity id : C∞ (M) → C∞ (M). If Q is
another smooth manifold and G : N → Q is another smooth map, then (G ◦ F)∗ = F ∗ ◦ G∗ . Finally,
the pull-back Φ∗ : C∞ (N) → C∞ (M) along a diffeomorphism Φ : M → N is an algebra isomorphism
with inverse given by the pull-back (Φ−1 )∗ : C∞ (M) → C∞ (N) along the inverse diffeomorphism
Φ−1 : N → M.
R We actually have a much stronger result that we state without proof in the following
Theorem 2.2.6 Let M, N be smooth manifolds.
(1) A map F : M → N is smooth if and only if the pull-back F ∗ takes smooth functions
to smooth functions.
(2) Every algebra homomorphism H : C∞ (N) → C∞ (M) is the pull-back along a unique
smooth map F : M → N, i.e. for every algebra homomorphism H : C∞ (N) → C∞ (M)
there is a unique smooth map F : M → N such that H = F ∗ .
In other words, the correspondence F 7→ F ∗ is a(n identity preserving and composition
reversing) bijection between smooth maps M → N and algebra homomorphisms C∞ (N) →
C∞ (M).
Proof. Omitted. ■
2.3 Submanifolds
In this section we present submanifolds. They are particularly nice subsets of a manifold M. They
are so nice that they inherit a smooth structure from that of M. We will make a systematic use of
2.3 Submanifolds 43
the Implicit Function Theorem. Therefore, we begin recalling the appropriate version of it. First of
all, we define regular systems. So, let d, n be non-negative integers such that d ≤ n. Consider the
standard Euclidean space Rn with standard coordinates (t 1 , . . . ,t n ), let U ⊆ Rn be an open subset,
and let
F = (F 1 , . . . , F n−d ) : U → Rn−d
Z(F) := {P ∈ U : F(P) = 0} ⊆ U .
Denote by M(n − d, n; R) the real vector space of real (n − d) × n matrices and recall that the
Jacobian matrix of F is the matrix-valued smooth map
a=1,...,n−d
∂ Fa
JF : U → M(n − d, n; R), P 7→ JF (P) := (P) .
∂t i i=1,...,n
Here the index a ranges over the rows, and the index i ranges over the columns. We say that 0 is a
regular value of the smooth map F, or that the system SF is regular, if
■ Example 2.9 Let F : Rn → Rn−d be a surjective affine map. Then F is of the form
F(P) = A · P + P0 , P ∈ Rn ,
for some P0 ∈ Rn−d , and some (n − d) × n matrix A such that rank A = n − d. Clearly, JF = A
constantly. In particular,
SF : {A · P + P0 = 0
is a regular system. This shows that every reduced linear system is regular. ■
So
d+1
t
− f 1 (t 1 , . . . ,t d ) = 0
SF : .. ,
.
n
t −f n−d 1 d
(t , . . . ,t ) = 0
whose rank is n − d at all points of U × Rn−d , in particular all points of Z(F). Hence SF is a regular
system. ■
is regular. ■
■ Example 2.12 On R2 with coordinates (x, y) consider the smooth map F : R2 → R given by
F(x, y) := xy.
The Jacobian matrix of F is the row
JF = (y, x)
whose rank is 0 at the origin and 1 elsewhere. However, the origin is in Z(F), so the system
SF : {xy = 0
is not regular. ■
Notice that, if rank JF (P0 ) = n − d for some P0 ∈ U , then there exists an (n − d) × (n − d) minor
M of JF which is non-zero at P0 . Then, by continuity, M is non-zero in a full open neighborhood
V ⊆ U of P0 (in particular, rank JF (P) = n − d for all P ∈ V ). It follows from this discussion that,
if SF is a regular system, then, for every P0 ∈ Z(F) there exists an (n − d) × (n − d) minor M of JF
which is non zero in an open neighborhood V ⊆ U of P0 . Let M be the minor identified by (all the
rows and) the columns
i1 , . . . , in−d
of JF . In this situation, we denote by (v1 , . . . , vn−d ) the coordinates (t i1 , . . . ,t in−d ) respectively, and
by (u1 , . . . , ud ) the remaining coordinates, in their order. So, the n-tuple (u1 , . . . , ud , v1 , . . . , vn−d )
differs from (t 1 , . . . ,t n ) only by a permutation. Denote it σ . In the following, we also denote
u = (u1 , . . . , ud ), and v = (v1 , . . . , vn−d ). Given open subsets V ⊆ Rn−d and U ⊆ Rd , we denote by
U ×σ V ⊆ Rn the open subset defined by
U ×σ V := {P ∈ Rn : u(P) ∈ U, and v(P) ∈ V } .
2.3 Submanifolds 45
Exercise 2.11 Prove Proposition 2.3.1 (Hint: First of all, let W ⊆ Rl be an open subset and
let G : W → U be a smooth map. Use the chain rule for the derivatives of a composition of
smooth maps to show that:
for all Q ∈ W (see also Chapter 4). Now apply (2.2) to the case F = Φ−1 and G = Φ to show
that the Jacobian matrix JΦ of Φ is invertible (the inverse being JΦ−1 ◦ Φ). Finally, apply (2.2)
to the case G = Φ.) ■
Theorem 2.3.2 — Implicit Function Theorem. Let d, n be non-negative integers such that
d ≤ n, let U ⊆ Rn be an open subset, and let
F : U → Rn−d
be a smooth map. If 0 is a regular value of F, then, locally, around every point, Z(F) is the graph
of a smooth map, i.e., for every point P0 ∈ Z(F), there are
(1) an open neighborhood U ⊆ Rd of u(P0 ),
(2) an open neighborhood V ⊆ Rn−d of v(P0 ), and
(3) a smooth map f = ( f 1 , . . . , f n−d ) : U → V ,
such that U ×σ V ⊆ U , and, in U ×σ V , the system SF is equivalent to the system
1 1 1 d
v = f (u , . . . , u )
.. , (2.3)
.
n−d
v = f n−d 1 d
(u , . . . , u )
i.e. Z(F) ∩ U ×σ V is exactly the space of solutions of (2.3) (here (u, v) are the tuples of
coordinates defined in the discussion preceding Exercise 2.11).
R In the situation described in the statement of the Implicit Function Theorem we also say
that, locally, around the point P0 ∈ Z(F), we can solve the system SF with respect to the
variables (v1 , . . . , vn−d ). This can be done even globally, if F is, for instance, an affine map
(see Example 2.9), and, in this case, the proof of the Implicit Function Theorem boils down
to the standard Gauss elimination method.
#$ , … , # '
( = /(#)
($ , … , ( ")'
*(,)
01
ℝ"
and
Yet in other words, (locally around P0 ) we were able to solve the equation
(t 1 )2 + · · · + (t n+1 )2 − 1 = 0
with respect to t i . The case P0i < 0 can be discussed in a similar (and obvious) way (do you see it?).
■
Exercise 2.12 Given the open subset U ⊆ Rn and the smooth map F : U → Rn−d , prove that
0 is a regular value of F, and, for every point P0 ∈ Z(F), find U,V, f as in the statement of the
Implicit Function Theorem, in the following two cases:
✓ in R3 with coordinates (x, y, z), U = R3 ∖ {x = y = 0}, and F : U → R given by
p 2
F(x, y, z) = x 2 + y2 − R + z2 − r 2 ,
so that
w2 − x 2 − 1 = 0
SF : .
y2 + z2 − 1 = 0
■
We finally come to the main definition of this section. Let M be an n-dimensional manifold.
Definition 2.3.1 — Submanifold. A d-dimensional submanifold of M is a subset S ⊆ M such
that, for every point p0 ∈ S, there is a chart (U, ϕ) on M around p0 such that ϕ(S ∩U) is the
b → Rn−d such that
solution space of a regular system, i.e. there is a smooth map F : U
(1) 0 is a regular value of F, and
(2) ϕ(S ∩U) = Z(F).
A submanifold is a smooth subset in the sense that, in local coordinates, it looks like the solution
space of a regular system, hence, by the Implicit Function Theorem, as the graph of a smooth map.
Proposition 2.3.3 Let S ⊆ M be a submanifold. Then, for every point p0 ∈ S, and every chart
(V, ψ) around p0 , there is a subchart (V0 , ψ) around p0 , such that ψ(S ∩V0 ) is the solution space
of a regular system.
Proof. Let (U, ϕ) be a chart around p0 such that ϕ(S ∩U) is the solution space of a regular system.
Put U0 = U ∩ V . Then ϕ(S ∩ U0 ) is the solution space of a regular system (defined on the open
subset ϕ(U0 ) ⊆ U).
b We want to show that ψ(S ∩U0 ) is also the solution space of a regular system.
To do this, notice that
(1) ϕ(U0 ) and ψ(U0 ) are related by the transition map ϕ ◦ ψ −1 : ψ(U0 ) → ϕ(U0 ), and
(2) the transition map ϕ ◦ ψ −1 identifies ψ(S ∩U0 ) and ϕ(S ∩U0 ).
Finally, use Proposition 2.3.1. ■
■Example 2.15 — Graph of a Smooth Map. Let F : M → N be a smooth map between manifolds,
dim M = n. By definition, the graph of F is the subset graph F ⊆ M × N in the product manifold
defined by
Then graph F ⊆ M × N is an n-dimensional submanifold. Indeed, pick a point p ∈ M, and let (U, ϕ)
be a chart on M around p, and (V, ψ) a chart on N around F(p) such that F(U) ⊆ V . Denote by
(U ×V, ϕ × ψ) the product chart on M × N. It is easy to see that ϕ × ψ(graph F) is the graph of
b → Vb , hence it is the solution space of a regular system as in
the coordinate representation Fb : U
Example 2.10. ■
■ Example 2.16 — Spheres are Submanifolds. Example 2.11 shows that the n-dimensional
sphere is an n-dimensional submanifold of Rn+1 . ■
■ Example 2.17 In the 3-dimensional projective space RP3 consider the subset
S := [P0 : P1 : P2 : P3 ] ∈ RP3 : P0 P3 − P1 P2 = 0 .
48 Chapter 2. Smooth Maps and Submanifolds
We want to show that S ⊆ RP3 is a 2-dimensional submanifold. To do this, consider the affine
charts (Ui , ϕi ) on RP3 , i = 0, . . . , 3, and compute ϕi (S ∩Ui ) ⊆ R3 . For instance, if i = 0,
S ∩U0 = [1 : Q1 : Q2 : Q3 ] : Q3 − Q1 Q2 = 0
S : t 3 − t 1t 2 = 0 .
Similarly for i = 1, 2, 3. ■
There are two useful characterizations of submanifolds. We summarize them in the following
Proposition 2.3.4 Let M be an n-dimensional manifold and let S ⊆ M be a subset. The following
three conditions are equivalent:
(1) S is a d-dimensional submanifold;
(2) for every P0 ∈ S there is a chart (W, ψ) around P0 such that W b = U ×V , with U ⊆ Rd and
V ⊆ Rn−d open subsets, and ψ(S ∩W ) is the graph of a smooth map f : U → V ;
(3) for every P0 ∈ S there is a chart (W0 , χ = (x1 , . . . , xd , y1 , . . . , yn−d )) around P0 such that
W b0 = U0 ×V0 , with U0 ⊆ Rd and V0 ⊆ Rn−d open subsets, and S ∩W0 is the zero locus of
the ya , i.e. S ∩W0 = {p ∈ U0 : y1 (p) = · · · = yn−d (p) = 0} (see Figure 2.10).
Proof.
(1) ⇒ (2) It follows immediately from the Implicit Function Theorem, restricting to suitable
subcharts.
(2) ⇒ (3) First of all, let U ⊆ Rd be an open subset, and let f : U → Rn−d be a smooth map.
The map
is a diffeomorphism with inverse Φ− f . Additionally, Φ f maps the graph of f to the graph of the
zero map 0 : U → Rn−d . Now, let S, P0 , and (W, ψ) be as in the statement. Then, composing
with Φ f we get a new chart (W, χ := Φ f ◦ ψ) such that χ(W ∩ S) is the graph of the zero section.
We can always restrict to a subchart (W0 , χ) around P0 whose coordinate domain is of the form
Wb0 = U0 ×V0 (see Figure 2.9), and this concludes the proof.
(3) ⇒ (1) χ(S ∩W0 ) is given by the vanishing of the last n − d coordinates, and this condition
is a regular system. ■
ℝ)*( ℝ)*(
%
$ % )
Φ" Φ" ($
graph f
V P
ℝ( %,
$ Φ" (+)
ℝ(
U
Our next aim is to show that submanifolds in a manifold M inherit from M a manifold structure.
This is explained in the next theorem and its proof.
Theorem 2.3.5 — Submanifold Chart Theorem. Let M be an n-dimensional manifold and let
S ⊆ M be a d-dimensional submanifold in M. There exists a unique smooth structure AS on S
such that
(1) (S, AS ) is a d-dimensional manifold,
(2) the inclusion iS : S ,→ M is a smooth map, and
(3) given a map between manifolds F : N → M taking values in S, i.e. F(N) ⊆ S, we have that
F : N → M is smooth if and only if F : N → S is smooth.
Proof. We begin with the existence. First of all, we consider two auxiliary smooth maps:
and
ϕS := prd ◦ ϕ : US → U
bS
is inverted by
ϕ −1 ◦ ind : U
bS → US .
Hence (US , ϕS ) is a chart on S around p. We will refer to any such chart as a submanifold chart.
50 Chapter 2. Smooth Maps and Submanifolds
M
U p j ℝ%&$ ℝ%
"
!
!- P
S ℝ$
j-
in$ pr$
'(
ℝ$
"-
!
Submanifold charts cover S. The next step is proving that they actually form an atlas. We
have to prove that any two submanifold charts are compatible. So let (U, ϕ), (V, ψ) be adapted
charts on M around p, and let (US , ϕS ), (VS , ψS ) be the induced submanifold charts. First of all
ϕS (US ∩VS ) ⊆ U
bS is an open subset, indeed ind maps it to
ϕ(U ∩V ) ∩ U bS × {0}
bS × {0} ⊆ U.
which is open in the subspace U b Similarly, ψS (US ∩VS ) ⊆ VbS is an open subset. The
transition map
ψS ◦ ϕS−1 : ϕS (US ∩VS ) → ψS (US ∩VS )
can be written as the following composition of smooth maps:
ψS ◦ ϕS−1 = prd ◦ (ψ ◦ ϕ −1 ) ◦ ind
(see Figure 2.11), hence it is smooth. Exchanging the role of (U, ϕ) and (V, ψ) shows that the
inverse ϕS ◦ψS−1 is also smooth, so that the transition map is a diffeomorphism and (US , ϕS ), (VS , ψS )
are compatible charts.
The atlas {(US , ϕS )} on S consisting of submanifold charts will be referred to as the submanifold
atlas, and it is contained in a unique smooth structure denoted AS . In order to prove that (S, AS ) is
a smooth manifold, we have to show that the atlas topology τAS induced by the maximal atlas AS ,
or, equivalently, the submanifold atlas, is Hausdorff and II-countable. Notice that it is enough to
prove that τAS coincides with the subspace topology τS . To do this we use Proposition 1.3.1. We
have to show that for every submanifold chart (US , ϕS ), US ⊆ S is an open subset with respect to
the subspace topology, which is obvious because US = S ∩U and U ⊆ M is an open subset, and,
additionally, that ϕS : US → U bS is a homeomorphism when US ⊆ M is equipped with the subspace
topology. For the continuity of ϕS , write ϕS as the following continuous composition of continuous
maps:
ϕS = prd ◦ ϕ ◦ iS : US → U
bS .
U M
V
S
ℝ%&$ j ℝ%&$
.
"
!
*+
. ∘ j&0
ℝ$ ℝ$
in$ pr$
ℝ$ ℝ$
")
! *+)
.) ∘ j)&0
hence ϕS−1 is continuous with respect to the subspace topology (see Example 1.19). We con-
clude that τAS = τS , in particular (S, τAS ) is Hausdorff and II-countable, hence it is a smooth
(d-dimensional) manifold.
Now we have to show that the manifold (S, AS ) enjoys Properties (2) and (3) in the statement.
For the smoothness of iS : S ,→ M notice that, if (US , ϕS ) is a submanifold chart induced by an
adapted chart (U, ϕ), then iS (US ) ⊆ U showing that iS has coordinate representations. Additionally,
the coordinate representation biS : U bS → U b with respect to the charts (US , ϕS ) and (U, ϕ) is simply
ind which is smooth. So iS is smooth. Finally, take a map between manifolds F : N → M such
that F(N) ⊆ S. For the sake of clarity, we denote by FS : N → S the restriction of F to S in the
codomain. We have to show that F is smooth if and only if so is FS . If FS is smooth, then F = iS ◦ FS
is the smooth composition of smooth maps, hence it is smooth. Conversely, let F be smooth.
Take a point q ∈ N, let (U, ϕ) be an adapted chart around F(q), and let (US , ϕS ) be the induced
submanifold chart. As F is a continuous map, F −1 (U) ⊆ N is an open subset, hence there is a
chart (V, ψ) on N around q such that V ⊆ F −1 (U), i.e. F(V ) ⊆ U. But F takes values in S, hence
FS (V ) = F(V ) ⊆ U ∩ S = US , showing that FS has coordinate representations. It remains to show
that the coordinate representation FbS : Vb → U bS with respect to the charts (V, ψ), (US , ϕS ) is smooth.
It is easy to see that, actually, FbS = prd ◦ F,
b in particular, it is a smooth composition of smooth maps.
This concludes the proof of Property (3) for the smooth structure AS .
We conclude the proof of the theorem showing that AS is the only smooth structure with
Properties (1), (2) and (3) in the statement. So, suppose A ′ , A ′′ are smooth structures on S with
those properties. We denote by S′ the manifold (S, A ′ ), and by S′′ the manifold (S, A ′′ ). We want
to show that S′ = S′′ , i.e. A ′ = A ′′ . To do this, it is enough to prove that the identity map id : S → S
induces a diffeomorphism id : S′ → S′′ (do you see it?). So, denote by iS′ : S′ ,→ M the inclusion.
As S′ has Property (2), iS′ is smooth. Additionally, it takes values in S. As S′′ has Property (3), it
follows that the restriction of iS′ to S′′ in the codomain is also smooth. But the latter restriction is
52 Chapter 2. Smooth Maps and Submanifolds
U
M
0(2)
N F S
",
0 '4
(") - ℝ&'% ℝ&
0,
V
#
"
q
0/ ℝ%
!
-,
in% pr%
./
0/,
ℝ%
#,
"
exactly id : S′ → S′′ . Exchanging the role of S′ and S′′ we see that the inverse map id : S′′ → S′ is
also smooth, hence id : S′ → S′′ is a diffeomorphism as claimed, and this concludes the proof. ■
Definition 2.3.3 — Submanifold Atlas. A chart (US , ϕS ) on S as in the proof of Theorem 2.3.5
is called a submanifold chart, and the atlas {(US , ϕS )} is a submanifold atlas.
■ Example 2.20 In Rn+1 consider the n-dimensional sphere Sn ⊆ Rn+1 . We already explained
that Sn is an n-dimensional submanifold. So, it inherits a smooth structure ASn from Rn+1 , and
(Sn , ASn ) is an n-dimensional manifold. We want to show that ASn is not a new smooth structure on
Sn but it coincides with the smooth structure A containing the stereographic atlas (or, equivalently,
the atlas consisting of orthogonal projections). We do this via Theorem 2.3.5. We already know
that iSn : (Sn , A ) → Rn+1 is smooth (Exercise 2.4). It remains to show that (Sn , A ) has Property
(3) in the statement of Theorem 2.3.5. So, let F : N → Rn+1 be a smooth map taking values in Sn .
We want to show that the restriction FSn : N → Sn of F to Sn in the codomain is smooth. We use a
trick: we interpret F as a map with values in Rn+1 ∖ {0}:
F : N → Rn+1 ∖ {0}.
(show that π is actually smooth as an exercise). Conclude noticing that FSn = π ◦ F, i.e. it is the
smooth composition of smooth maps. ■
2.3 Submanifolds 53
f |S = i∗S ( f ) = f ◦ iS :
the pull-back of f along the smooth map iS . However, in general, not all smooth functions g ∈ C∞ (S)
are restrictions of smooth functions on M as the following counter-example shows.
■ Example 2.21 In M = R with coordinate t, consider the open submanifold S = R ∖ {0}. The
function
1
:S→R
t
is a smooth function, but it is not the restriction of a smooth function on R. ■
Lemma 2.3.6 — Local Extension Lemma. Let M be a manifold, and let S ⊆ M be a submani-
fold. For every smooth function g ∈ C∞ (S) and every point p0 ∈ S, there exist
(1) a smooth function ge ∈ C∞ (M), and
(2) an open neighborhood V ⊆ M of p0 ,
such that ge and g agree on S ∩V , i.e. g|S∩V = ge|S∩V .
Proof. The statement is a corollary of the Bump Function Theorem (Theorem 2.1.3). Choose an
bS → R
adapted chart (U, ϕ) around p0 , and let (US , ϕS ) be the induced submanifold chart. Let gb : U
be the coordinate representation of g with respect to the chart (US , ϕS ). The composition
ge′ := gb ◦ prd ◦ ϕ : U → R
is a smooth function on U, i.e. ge′ ∈ C∞ (U). It is easy to see that ge′ |US = g|US , however, in general,
ge′ is not the restriction to U of a smooth function on M. We can cure this problem locally, using
a bump function. Se let β ∈ C∞ (M) be a smooth function such that supp β ⊆ U and β |V = 1 for
some open neighborhood V ⊆ U of p. Consider the function ge : M → R defined by
g ′ (p) if p ∈ U
β (p)e
ge(p) := .
0 if p ∈ M ∖ supp β
Clearly, ge is well-defined, and it is smooth by the Gluing Lemma. Finally, for every p in S ∩V , we
have
g ′ (p) = g(p),
ge(p) = β (p)e
as claimed. ■
3. Tangent Vectors
Given a curve γ : I → M, we will usually denote by t the standard coordinate on I and interpret
it as the “time”. The next step is defining the velocity of γ at the time t0 ∈ I. However, this is not
as easy as in the case when M is a submanifold in Rn , because we don’t know (yet) how to take
derivatives of a map with values in a manifold. Notice that, however defined, the velocity of γ
56 Chapter 3. Tangent Vectors
ℝ"
Figure 3.1: Velocity of a curve in a submanifold of Rn .
should measure how fast does the point γ (t) ∈ M changes when t changes in I. In order to go back
to a more familiar situation, we pick a “test function” f ∈ C∞ (M) and, instead of measuring directly
how fast does γ (t) change at the time t0 , we measure first how fast does the real number f (γ (t))
change at the time t0 . This can be easily done taking the derivative at t0 of the real valued function
f ◦ γ : I → R:
d
|t f ◦ γ.
dt 0
In order to get rid of the dependence on the test function f , we repeat this for every function
f ∈ C∞ (M). In this way we get a map
d
γ̇ (t0 ) : C∞ (M) → R, f 7→ γ̇ (t0 )( f ) := |t f ◦ γ. (3.1)
dt 0
This map measures well enough how fast does γ (t) change. Hence we call it the velocity of γ at the
time t0 and denote it also by
d d dγ
|t=t γ (t) or |t γ or (t0 ).
dt 0 dt 0 dt
The main properties of γ̇ (t0 ) are summarized in the next
Exercise 3.1 Show that γ̇ (t0 ) : C∞ (M) → R is R-linear and satisfies the following identity: for
all f , g ∈ C∞ (M)
We interpret γ̇ (t0 ) as a tangent vector to M at the point γ (t0 ). Even more, given a manifold
M, and a point p ∈ M, we axiomatize the properties of γ̇ (t0 ) stated in Exercise 3.1 and give the
following
3.1 Tangent Spaces 57
Definition 3.1.2 — Tangent Vector. A tangent vector to M at the point p is a linear map
v : C∞ (M) → R
satisfying the following Leibniz rule at the point p: for all f , g ∈ C∞ (M),
The set of all tangent vectors to M at p is called the tangent space to M at p and denoted by
Tp M.
R Given a tangent vector v to M at the point p, we will always assume p as part of the defining
data of v. From this point of view, a tangent vector should be better defined as a pair (p, v)
where p ∈ M is a point, and v is as in Definition 3.1.2. Doing this we avoid anomalies like the
zero map 0 : C∞ (M) → R being a tangent vector at all points. We will explicitly understand
tangent vectors as pairs (p, v) when defining the tangent bundle later on in this chapter.
Later on in this chapter, we will actually show that every tangent vector in the sense of Definition
3.1.2 is the velocity of some curve at some time. We now give another example of a tangent vector
besides the velocity of a curve.
■ Example 3.1 — Coordinate Vectors. Let M be an n-dimensional manifold and let p ∈ M be a
point. Fix a chart (U, ϕ = (x1 , . . . , xn )) on M around p, and, for each i = 1, . . . , n, define the map
∂ ∂ ∂ fb
| p : C∞ (M) → R, f 7→ | p f := i (ϕ(p)),
∂ xi ∂x i ∂t
where, as usual, fb = f ◦ ϕ −1 : U
b → R is the coordinate representation of f . The map ∂ i | p is called
∂x
the i-th coordinate vector and it is indeed a tangent vector to M at p. The proof is left as Exercise
3.2. ■
Exercise 3.2 Show that the i-th coordinate vector ∂∂xi | p is a tangent vector to M at p. ■
In the next lemma, we present two properties of tangent vectors following from the Leibniz
rule (and the linearity).
where, in the second step, we used the Leibniz rule. Now, let c ∈ C∞ (M) be a constant function.
Then c identifies with a real number, also denoted by c ∈ R, and c = c · 1, so that
For the second part of the statement, let f , g ∈ C∞ (M) be functions that agree around p,
i.e. f |U = g|U for some open neighborhood U ⊆ M of p, then h := f − g vanishes around p,
i.e. h|U = 0. If we prove that v vanishes on h, we have done by linearity. Indeed, if v(h) = 0, then
v( f ) − v(g) = v( f − g) = v(h) = 0.
So it remains to prove that v vanishes on every function h such that h|U = 0. To do this, consider a
bump function β ∈ C∞ (M) relative to the data (p, U ), i.e. β = 1 around p, and supp β ⊆ U , and
denote η = 1 − β . It follows that h = η · h (do you see it?), so that
We conclude this section remarking that the tangent space Tp M to a manifold M at a point
p ∈ M is a vector space. First notice that, by definition of tangent vectors, Tp M is a subset in the
space HomR (C∞ (M), R) of R-linear maps C∞ (M) → R (the dual space of the vector space C∞ (M)):
An easy check, that we leave as Exercise 3.3, reveals that Tp M is actually a vector subspace. In
particular, it is a vector space.
Exercise 3.3 Show that the tangent space to M at the point p is a vector subspace in the space
HomR (C∞ (M), R) dual to the vector space C∞ (M). ■
Our intuition suggests that the tangent space Tp M should share the same dimension as M. This
is indeed true, according to the following
form a frame in Tp M (called the coordinate frame). In particular, Tp M is finitely generated and
dim Tp M = dim M = n.
We postpone a detailed proof of the Tangent Vector Theorem to the next section. For now, we
only prove it in the case M = Rn . The proof relies on the following
Lemma 3.1.3 — Hadamard Lemma. Let f ∈ C∞ (Rn ) be a smooth function on Rn , and let
P0 = (P01 , . . . , P0n ) ∈ Rn be a point. Then there exist smooth functions g1 , . . . , gn ∈ C∞ (Rn ) such
that
n
f = f (P0 ) + ∑ (t i − P0i ) · gi , (3.3)
i=1
and, additionally,
∂f
gi (P0 ) = (P0 ), i = 1, . . . , n. (3.4)
∂t i
So, when P = P0 , P(ε) = P0 constantly, otherwise, it is the generic point on the line through P and
P0 . In any case, P(0) = P0 , and P(1) = P. Compute
and (3.3) follows from the arbitrariness of P. Finally, we check (3.4). Remember that when P = P0 ,
we have P(ε) = P0 constantly, so, for all i ∈ {1, . . . , n},
Z 1 Z 1
∂f ∂f ∂f
gi (P0 ) = (P ) dε =
i 0
(P0 ) dε = (P0 ).
0 ∂t ∂t i 0 ∂t i
■
We are now ready to prove the Tangent Vector Theorem in the case M = Rn . In other words we
are proving the following
Lemma 3.1.4 — Tangent Vector Theorem for Rn . Consider the n-dimensional standard
Euclidean space Rn with standard coordinates (t 1 , . . . ,t n ), and let P0 ∈ Rn be a point. Then the
coordinate vectors
∂ ∂
|P , . . . , n |P0 (3.5)
∂t 1 0 ∂t
Proof. Notice, preliminarly, that the coordinate vectors (3.5) are exactly the partial derivatives (of
smooth functions Rn → R) at P0 . Now, let a1 , . . . , an ∈ R be such that the linear combination
n
∂
v := ∑ ai |P
i=1 ∂t i 0
for all j, where δij is the Kronecker delta. This shows that the ∂t∂ i |P0 are independent. Finally, we
use the Hadamard Lemma (Lemma 3.1.3) to show that they generate TP0 Rn . So let v ∈ TP0 Rn be any
60 Chapter 3. Tangent Vectors
tangent vector, and let f ∈ C∞ (Rn ) be an arbitrary smooth function. We compute v( f ) by writing f
in the Hadamard form (3.3):
!
n
v( f ) = v f (P0 ) + ∑ (t i − P0i ) · gi (Hadamard Lemma)
i=1
n
= v ( f (P0 )) + ∑ v (t i − P0i ) · gi )
(v is R-linear)
i=1
n
= ∑ v (t i − P0i ) · gi )
(v vanishes on constants)
i=1
n
= ∑ v(t i − P0i ) · gi (P0 ) + (P0i − P0i ) · v(gi )
(Leibniz rule at the point P0 )
i=1
n
∂
= ∑ v(t i ) |P f , (Equation 3.4)
i=1 ∂t i 0
We leave as Exercise 3.4 to prove that the composition of the pull-back F ∗ : C∞ (N) → C∞ (M)
followed by a tangent vector v : C∞ (M) → R to M at the point p is indeed a tangent vector to N at
the point F(p), and that d p F is indeed a linear map.
Exercise 3.4 Prove that the tangent map d p F in (3.6) is well-defined (i.e. d p F(v) ∈ TF(p) N for
every v ∈ Tp M) and R-linear. ■
■ Example 3.2 Let M be a manifold, let γ : I → M be a smooth curve, and let t0 ∈ I. It immediately
follows from the definition of velocity γ̇(t0 ) of γ at the time t0 (3.1), that γ̇(t0 ) is the image of the
coordinate tangent vector dtd |t0 ∈ Tt0 I under the tangent map dt0 γ :
d
γ̇(t0 ) = dt0 γ |t
dt 0
Proposition 3.2.1 — Chain Rule. Let M, N, Q be smooth manifolds, and let F : M → N and
G : N → Q be smooth maps. For every point p ∈ M,
(1) the tangent map to the identity idM : M → M is the identity:
d p idM = idTp M ;
(2) the tangent map to the composition G ◦ F : M → Q is the composition of the tangent maps:
d p (G ◦ F) = dF(p) G ◦ d p F, (3.7)
R It follows from Proposition 3.2.1 and the Tangent Vector Theorem (Theorem 3.1.2, not
proved yet in full generality) that two diffeomorphic manifolds M, N share the same dimen-
sion: dim M = dim N. In particular, there are no diffeomorphisms between open subsets in
Euclidean spaces of different dimensions.
d p iS : Tp S → Tp M.
It is enough to show that ker d p iS = {0}. So let v ∈ Tp S be such that d p iS (v) = 0. We have to show
that v(g) = 0 for all smooth functions g on S. So, take g ∈ C∞ (S) and use the Local Extension
Lemma (Lemma 2.3.6) to pick a smooth function ge ∈ C∞ (M) such that g and ge agree on an open
neighborhood of p in S. As tangent vectors are local operators (Lemma 3.1.1, Point (2)), we have
g|S ) = v(i∗S (e
v(g) = v(e g)) = d p iS (v)(e
g) = 0.
This concludes the proof of the first part of the statement. For the second part, let S = U ⊆ M
be an open submanifold. We have to show that, for any v ∈ Tp M, there exists w ∈ Tp U such that
v = d p iU (w). We define w : C∞ (U ) → R as follows. Let g ∈ C∞ (U ), and let ge ∈ C∞ (M) be a
62 Chapter 3. Tangent Vectors
smooth function on M that coincides with g around p (it exists in view of the Local Extension
Lemma). Put
g).
w(g) := v(e
First of all, as tangent vectors are local operators, v(eg) does only depend on the values of ge around
p, hence w(g) is well-defined. Secondly, if f , g are functions on S, and fe, ge ∈ C∞ (M) coincide with
them around p, then a fe+ be g, fege coincide with a f + bg, f g around p for all a, b ∈ R. So
and
w( f g) = v( fege) = fe(p)v(e
g) + ge(p)v( fe) = f (p)w(g) + g(p)w( f ).
This shows that w is a tangent vector. Finally, let h ∈ C∞ (M) and compute
In the following, we will use d p iS to identify Tp S with its image (unless otherwise stated).
In other words, we will interpret Tp S as a subspace in Tp M, and write Tp S ⊆ Tp M. In particular,
if U ⊆ M is an open submanifold, then Tp U = Tp M. So a tangent vector v ∈ Tp M can be
safely applied to any smooth function f defined only locally around p, i.e. any smooth function
f ∈ C∞ (U ) on some open neighborhood U ⊆ M of p. In the following, we will do this without
further comments. We are finally ready to prove the Tangent Vector Theorem (Theorem 3.1.2).
Proof of the Tangent Vector Theorem 3.1.2. Let M be an n-dimensional manifold, let p ∈ M be a
point, and let (U, ϕ = (x1 , . . . , xn )) be a chart on M around p. The claim is that the coordinate
vectors
∂ ∂
|p, . . . , n |p (3.8)
∂ x1 ∂x
form a frame of Tp M. To prove this, denote by P the image ϕ(p), and consider the chain of
isomorphisms
d p iU dpϕ d p iUb
Tp M o TpU / TPU
b / TP Rn . (3.9)
It easily follows from the definition of coordinate vectors that (3.9) identify
∂ ∂ ∂ ∂
| p , . . . , | p and |P , . . . , |P
∂ x1 ∂ xn ∂t 1 ∂t n
(do you see it?). As the latter is a frame of TP Rn (Lemma 3.1.4), the former must be a frame of
Tp M as claimed. ■
for some real numbers vi . Notice that, from the definition of coordinate vector,
∂
| p x j = δij , i, j = 1, . . . , n. (3.11)
∂ xi
It follows that the components v1 , . . . , vn of a tangent vector v ∈ Tp M in the coordinate frame are
given by the action of v on the coordinates x1 , . . . , xn , i.e.
v j = v(x j ), j = 1, . . . , n. (3.12)
Indeed
n n
∂
v(x j ) = ∑ vi
∂ xi
| p x j
= ∑ vi δij = v j ,
i=1 i=1
∂
v = vi |p,
∂ xi
instead of (3.10).
■ Example 3.3 — Coordinate Expression for the Velocity of a Curve. let γ : I → M be a
curve, and let t0 ∈ I. Denote p0 = γ(t0 ). As γ is smooth we can choose an open subinterval J ⊆ I
containing t0 , and a chart (U, ϕ = (x1 , . . . , xn )) on M around p0 , such that γ(J) ⊆ U. Denote by
γ i = γ ∗ (xi ) : J → R the components of (the coordinate representation of) γ with respect to the chart
(U, ϕ). Then, for every t ∈ J,
∂
γ̇(t) = vi | ,
∂ xi γ(t)
d i dγ i
γ̇(t)(xi ) = x (γ(t)) = (t).
dt dt
Summarizing
dγ i ∂
γ̇(t) = (t) i |γ(t) . (3.13)
dt ∂x
■
Clearly, if we change coordinates around p, the coordinate frame changes. We want to find
out how. So let (Ũ, ϕ̃ = (x̃1 , . . . x̃n )) be another chart on M around p. We denote by t 1 , . . . ,t n the
standard coordinates on ϕ(U), and, to avoid confusion, by t˜1 , . . . , t˜n the standard coordinates on
ϕ̃(Ũ). We also put P = ϕ(p). The transition matrix
is given by
∂ ∂
aij = the j-th component of i | p in the frame R̃ = i | p x̃ j ,
∂x ∂x
and, using the definition of coordinate vector, we find
∂ j ∂ j −1 ∂ j −1 −1 ∂ j −1 ∂ (ϕ̃ ◦ ϕ −1 ) j
| p x̃ = |P x̃ ◦ ϕ = |P x̃ ◦ ϕ̃ ◦ ϕ̃ ◦ ϕ = |P ˜
t ◦ (ϕ̃ ◦ ϕ ) = (P).
∂ xi ∂t i ∂t i ∂t i ∂t i
This shows that MR,R̃ is exactly the Jacobian matrix at the point P = ϕ(p) of the transition map
Next we compute the representative matrix of the tangent map with respect to two coordinate
frames. Namely, let M, N be manifolds with dim M = m, dim N = n, let F : M → N be a smooth
map, and let p ∈ M be a point. Choose a chart (U, ϕ = (x1 , . . . , xm )) on M around p and a chart
(V, ψ = (y1 , . . . , yn )) on N around F(p) such that F(U) ⊆ V . Put P = ϕ(p). The restriction
F : U → V is completely determined by functions F a := ya ◦ F = F ∗ (ya ) ∈ C∞ (U), a = 1, . . . , n.
The representative matrix
d F
MSp,R = (bai )a=1,...,n
i=1,...,m ∈ M(n, m; R)
∂ ∂ −1 ∂ Fba
bai = | p F a
= |P F a
◦ ϕ = (P),
∂ xi ∂t i ∂t i
where the Fba are the components of the coordinate representation Fb : U
b → Vb of F. This shows that
dpF
MS ,R is exactly the Jacobian matrix at the point P = ϕ(p) of F.
b
We conclude this section discussing two further applications of the Tangent Vector Theorem.
We begin showing that every tangent vector is the velocity of a suitable (non-unique) curve.
Proposition 3.2.3 Let M be a manifold, and let p ∈ M. For every tangent vector v ∈ Tp M there
exists a curve γ : I → M, such that 0 ∈ I, γ (0) = p, and γ̇(0) = v.
Proof. Let n = dim M, choose a chart (U, ϕ = (x1 , . . . , xn )) around p, and put P = ϕ(p). Then
∂
v = vi |p
∂ xi
for some n-tuple V := (v1 , . . . , vn ) ∈ Rn . Consider the following curve in Rn
γb : R → Rn , t 7→ γb(t) := P + tV.
3.2 Tangent Maps 65
U
M
v
p
j
P V "
!
%
%&
0 ℝ$
I
The pre-image γb−1 (U) b is an open neighborhood of 0 in R. In particular, there is an open interval
I ⊆ R containing 0, such that γb(I) ⊆ U. b Denote again by γb : I → U b the restriction, and let
1 n
(γ , . . . , γ ) be its components. The curve γb is the coordinate representation of a unique curve
γ = ϕ −1 ◦ γb : I → M (taking values in U). The curve γ is the one we are looking for. To show this,
we compute the velocity γ̇ (0). From (3.13),
dγ i ∂ ∂
γ̇ (0) = (0) i |γ(0) = vi i | p = v.
dt ∂x ∂x
This concludes the proof. ■
Exercise 3.6 Let F : M → N be a smooth map between manifolds, and let γ : I → M be a curve
in M. Show that, for every t0 ∈ I, the tangent map to F transforms the velocity of a curve γ into
the velocity of the transformed curve F ◦ γ, i.e.
dγ d(F ◦ γ )
dγ (t0 ) F (t0 ) = (t0 ), (3.14)
dt dt
■
Notice that, in view of Proposition 3.2.3, Equation (3.14) can be understood as an alternative
definition of the tangent map.
Our second application of the Tangent Vector Theorem is a description of the tangent spaces to
a vector space. So, let V be an n-dimensional real vector space. In particular, V is an n-dimensional
manifold. We begin noticing that there are two (a priori different) ways to understand the velocity
of a curve γ : I → V in V at time t0 ∈ I. First of all, as the vector γ̇ (t0 ) in the tangent space Tγ (t0 )V .
Secondly, as a vector in V , denoted γ ′ (t0 ), as follows. Choose a frame R = (e1 , . . . , en ) in V . Then,
for all t ∈ I, γ (t) = vi (t)ei , where the vi : I → R are smooth functions. Put
dvi
γ ′ (t0 ) = (t0 )ei ∈ V.
dt
It is easy to see that γ ′ (t0 ) is actually independent of the choice of R (do you see it?).
66 Chapter 3. Tangent Vectors
Exercise 3.7 Let V be a finite dimensional real vector space, let v1 , . . . , vk be (non-necessarily
independent) vectors in V , and let f1 , . . . fk : I → R be smooth functions on some open interval
I ⊆ R. Show that the map
Proposition 3.2.4 — Tangent Space to a Vector Space. For any v0 ∈ V , there is a canonical
vector space isomorphism V ∼
= Tv0 V given by
d
V → Tv0 V, v 7→ v↑ := |t=0 (v0 + tv). (3.15)
dt
Additionally, for any curve γ : I → V , and any t0 ∈ I, we have
γ ′ (t0 )↑ = γ̇ (t0 ),
Probably, the definition in (3.15) requires some explanations. For any v0 , v ∈ V , we are
considering the smooth curve Γ : R → V , t 7→ Γ(t) := v0 + tv in V (do you see that Γ is smooth?),
and v↑ ∈ Tv0 V is the velocity of Γ at time 0.
Proof of Proposition 3.2.4. We have to show that the map V → Tv0 V , v 7→ v↑ is linear and bijective.
We take care of both things at once. Choose a frame R = (e1 , . . . , en ) in V , and let
(V, ϕR = (x1 , . . . , xn ))
be the associated chart on V . Recall that, for any v ∈ V , xi (v) is, by definition, the i-th component
vi of v in the frame R. Let
∂ ∂
RT = |v , . . . , n |v0
∂ x1 0 ∂x
be the coordinate frame in Tv0 V induced by the chart (V, ϕR ). Finally, let ϕRT : Tv0 V → Rn be the
coordinate isomorphism induced by RT . We want to compute ϕRT (v↑ ). To do this, remember that
the i-th component of a velocity vector in the coordinate frame is the time derivative of the i-th
component of the coordinate representation of the curve (see (3.13)). As v↑ is the velocity at time 0
of the curve Γ, we have, in particular, that the coordinate vector of v↑ in the coordinate frame RT is
dΓ1 dΓn
↑
ϕRT (v ) = (0), . . . , (0)
dt dt
Γi = vi0 + tvi ,
3.2 Tangent Maps 67
hence
dΓi
(0) = vi
dt
for all i = {1, . . . , n}. We conclude that
This shows that the map V → Tv0 V , v 7→ v↑ , is given by the composition of the vector space
−1
isomorphism ϕR : V → Rn , followed by the vector space isomorphism ϕR T
: Rn → Tv0 V :
v↑ = (ϕR
−1
T
◦ ϕR )(v), for all v ∈ V .
!" + (!
! V
!
0V
!"
#$% &
For the second part of the proposition, let γ : I → V , t 7→ γ (t) = vi (t)ei be a curve in V , and let
t0 ∈ I. By definition of γ ′ (t0 ), we have
dv1 dvn
′
ϕR (γ (t0 )) = (t0 ), . . . , (t0 ) .
dt dt
On the other hand, the coordinate representation of γ with respect to the chart (V, ϕR ) is ϕR ◦ γ =
(v1 , . . . , vn ) : I → Rn , and, from (3.13) again,
dvi ∂
γ̇ (t0 ) = (t0 ) i |γ(t0 )
dt ∂x
so that
dv1 dvn
ϕRT (γ̇ (t0 )) = (t0 ), . . . , (t0 ) = ϕR (γ ′ (t0 ))
dt dt
In the following, given a curve γ : I → V in a finite dimensional real vector space V , we will
always interpret its velocity γ̇ (t0 ) at time t0 ∈ I as a vector in V , identifying γ̇ (t0 ) and γ ′ (t0 ) via the
isomorphism V ∼ = Tv0 V , unless otherwise stated.
68 Chapter 3. Tangent Vectors
T M :=
∏ T M = {(p, v) : p ∈ M and v ∈ T M} .
p p
p∈M
When it is clear what is p, a point (p, v) in T M will be also denoted simply by v (see also the remark
following Definition 3.1.2). The set T M comes with a natural surjection τ : T M → M, (p, v) 7→ p,
and the pair (T M, τ) is called the tangent bundle to M. Sometimes we understand τ and call T M
itself the tangent bundle. The smooth structure on M induces an atlas on T M as we now show.
Begin with a chart (U, ϕ = (x1 , . . . , xn )) on M, and define a chart (TU, T ϕ) on T M as follows.
First of all, put
TU := τ −1 (U) =
∏ T M = {(p, v) ∈ T M : p ∈ U} .
p
p∈U
As the tangent spaces to U identify canonically with the tangent spaces to M at points of U, TU is
exactly the tangent bundle to U. Next define a map:
b × Rn ,
T ϕ : TU → U (p, v) 7→ T ϕ(p, v) := (ϕ(p); v(x1 ), . . . , v(xn )).
Notice that the last n entries of T ϕ(p, v) are nothing but the components of v in the coordinate
frame
∂ ∂
Rp = |p, . . . , n |p
∂ x1 ∂x
Proposition 3.3.1 — Smooth Structure on the Tangent Bundle. Standard charts on T M form
an atlas. With the associated smooth structure, T M is a smooth 2n-dimensional manifold, and
τ : T M → M is a smooth map.
Proof. As the charts on M cover M, the standard charts on T M cover T M. Next we show that any
two standard charts are compatible. So, take two charts (U, ϕ = (x1 , . . . , xn )), (Ũ, ϕ̃ = (x̃1 , . . . , x̃n )),
and the associated standard charts (TU, T ϕ), (T Ũ, T ϕ̃). If U ∩ Ũ = ∅, then TU ∩ T Ũ = ∅, and
(TU, T ϕ), (T Ũ, T ϕ̃) are compatible. If U ∩ Ũ ̸= ∅, then TU ∩ T Ũ = T (U ∩ Ũ) and we have to
show that the transition map
is indeed an open subset, and similarly for T ϕ̃(TU ∩ T Ũ). Now, check the smoothness of T ϕ̃ ◦
(T ϕ)−1 : take a point (P; v1 , . . . , vn ) ∈ ϕ(U ∩ Ũ) × Rn and compute
(T ϕ̃ ◦ (T ϕ)−1 )(P; v1 , . . . , vn )
−1 i ∂
= T ϕ̃ ϕ (P); v i |ϕ −1 (P)
∂x
∂ ∂
= (ϕ̃ ◦ ϕ −1 )(P); vi i |ϕ −1 (P) x̃1 , . . . , vi i |ϕ −1 (P) x̃n
∂x ∂x
−1 1 −1 n
−1 i ∂ (ϕ̃ ◦ ϕ ) i ∂ (ϕ̃ ◦ ϕ )
= (ϕ̃ ◦ ϕ )(P); v (P), . . . , v (P) .
∂t i ∂t i
As (U, ϕ), (Ũ, ϕ̃) are compatible, the first n entries depend smoothly on P. The last n entries depend
smoothly on P, and linearly, hence smoothly, on v1 , . . . , vn . This shows that T ϕ̃ ◦ (T ϕ)−1 is smooth.
Changing the roles of (U, ϕ) and (Ũ, ϕ̃) reveals that T ϕ ◦ (T ϕ̃)−1 is smooth as well. We conclude
that both transition maps T ϕ̃ ◦ (T ϕ)−1 and T ϕ ◦ (T ϕ̃)−1 are diffeomorphisms, and this concludes
the proof of the first part of the statement. The atlas on T M consisting of standard charts will be
called the standard atlas.
For the second part of the statement, we have to show that the atlas topology induced by the
standard atlas is Hausdorff and II-countable. We use the criterion provided by Proposition 1.3.2. For
the Hausdorff property, take (p, v), (q, w) ∈ T M. If p = q then p, q belong to the same coordinate
domain U, hence (p, v), (q, w) belong to the same standard coordinate domain TU. If p ̸= q, we
can find disjoint coordinate domains U,V such that p ∈ U and q ∈ V . It follows that TU, TV are
disjoint standard coordinate domains such that (p, v) ∈ TU and (q, w) ∈ TV . For II-countability, we
cover M by countably many charts (U, ϕ). Hence T M is covered by the standard charts (TU, T ϕ),
and they are countably many. We leave it to the reader to prove the very last part of the statement
(about the smoothness of τ : T M → M) as Exercise 3.8. ■
Exercise 3.8 Complete the proof of Proposition 3.3.1 showing that the natural projection
τ : T M → M is smooth. ■
We now come to a special class of maps M → T M that will play an important role in the next
chapter: sections of T M.
Definition 3.3.1 — Section of the Tangent Bundle. A section of T M is a map s : M → T M
that inverts τ : T M → M on the right, i.e. τ ◦ s = idM . In other words, a section is the assignment
of a tangent vector s(p) to M at the point p, for every point p ∈ M.
There is a natural algebraic structure on the space of sections of T M: they form a module over
the algebra C∞ (M). More generally, let A be a real algebra.
Definition 3.3.2 — Module. A module over A, or simply an A-module, is a set M equipped
with two operations +, · such that
(1) + : M × M → M is an interior operation and (M , +) is an abelian group,
(2) · : A × M → M is an exterior operation and
✓ α · (β · µ) = (α · β ) · µ,
✓ (α + β ) · µ = α · µ + β · µ,
✓ α · (µ + ν) = α · µ + α · ν,
✓ 1 · µ = µ,
for all α, β ∈ A, and µ, ν ∈ M .
A submodule of an A-module M is a subset N ⊆ M which is stable under both operations in
M.
70 Chapter 3. Tangent Vectors
%& '
%'
"(') "(!)
( "
! '
In other words, a module satisfies all the axioms of a vector space except that the ring of scalar
is now a generic algebra, not necessarily a field. We usually write α µ, instead of α · µ, for α ∈ A
and µ ∈ M . As a first trivial example, notice that any real vector space is an R-module. Conversely,
every module over a real algebra is also a real vector space. A submodule N of a module M is
also a module with the restricted operations.
In many respects, the theory of modules is similar to the theory of vector spaces. For instance,
linearly independent systems and systems of generators are defined exactly as for vector spaces,
except that now linear combinations are taken with coefficients in the algebra rather than just
in the field. A basis of a module is a system of independent generators. The main difference
between modules and vector spaces is that a module may not possess a basis. Linear maps between
modules over the same algebra A are defined exactly as for vector spaces, except that now they are
compatible with the multiplication by a scalar in the algebra rather than just in the field. In order
to distinguish explicitly this kind of linearity from R-linearity, we call it A-linearity, and A-linear
maps between A-modules are also called A-module homomorphisms. An invertible A-linear map is
called an A-module isomorphism, and its inverse is also an A-module isomorphism.
The following example is our main example of a module.
■ Example 3.4 — Module of Sections of T M. Let M be a manifold. The space of sections of the
tangent bundle T M is a C∞ (M)-module with the following point-wise operations:
where f ∈ C∞ (M), and s, s1 , s2 are sections. The zero section is the section mapping a point p to
the zero tangent vector 0 ∈ Tp M, for all p. ■
We are mainly interested in smooth sections. Discussing the smoothness of a section is easier
than discussing the smoothness of a generic map. Indeed, let M be a manifold, and let s be a section
of the tangent bundle T M. For every chart (U, ϕ) on M, the section s maps U to TU. This shows
that all sections have coordinate representations. Now, look at the restriction
s : U → TU.
3.3 The Tangent Bundle 71
As already remarked in the discussion following Definition 2.2.1, s is smooth provided only the
pull-backs s∗ (xi ), s∗ (ẋi ) of the coordinates on TU are smooth, i = 1, . . . , n. Now, by definition of
section, for every p ∈ U,
which is automatically smooth in its argument p. We conclude that the section s is smooth provided
only the functions s∗ (ẋi ) on U are smooth for some family of charts (U, ϕ) covering M. In the
following, we denote si := s∗ (ẋi ) and call them the components of the section s in the chart (U, ϕ).
We can then rephrase the preceding discussion saying that a section s of T M is smooth if and only
if its components si are smooth in every chart of an atlas.
Proposition 3.3.2 Smooth sections of T M form a submodule in the module of all sections.
Exercise 3.9 Prove Proposition 3.3.2. (Hint: prove that, for every three sections s, s1 , s2 , any
smooth function f , and any chart (U, ϕ) on M the following identities hold:
for all i ∈ {1, . . . , n}. Conclude that, if s, s1 , s2 are smooth, then so are s1 + s2 and f s.) ■
We denote by Γ(T M) the C∞ (M)-module of smooth sections of the tangent bundle T M. In the
following, by “section” we will always mean “smooth section”, unless otherwise stated.
We conclude this chapter discussing the tangent bundle to a submanifold. So let M be an
n-dimensional manifold, and let S ⊆ M be a d-dimensional submanifold. As Tp S ⊆ Tp M for all
p ∈ S, the tangent bundle to S can be seen as a subset in the tangent bundle to T M: T S ⊆ T M.
Proof. Let p0 ∈ S, and let (U, ϕ = (y1 , . . . , yd ; z1 , . . . , zn−d )) be an adapted chart on M around p0 .
This means that
Consider also the submanifold chart (US , ϕS = (y1S , . . . , ydS )). As we know, the coordinate represen-
tation of the inclusion iS : S ,→ M with respect to the charts (US , ϕS ) and (U, ϕ) is
bS → U,
biS = ind : U b (t 1 , . . . ,t d ) 7→ ind (t 1 , . . . ,t d ) = (t 1 , . . . ,t d , 0, . . . , 0).
∂
v = vi | p ∈ Tp S
∂ yiS
∂
d p iS (v) = vi | p ∈ Tp M.
∂ yi
72 Chapter 3. Tangent Vectors
∂ ∂
w = vi | p + wa a | p ∈ Tp M, p ∈ US ,
∂ yi ∂z
is in Tp S if and only if
In other words,
showing that (TU, T ϕ) is an adapted chart to T S (up to a permutation of the coordinates). The last
part of the statement follows from the easy remark that, for every chart (U, ϕ) on M adapted to
S, the submanifold chart induced on T S by (TU, T ϕ), agrees with (TUS , T ϕS ): the standard chart
induced by the submanifold chart (US , ϕS ) (do you see it?). So T S can be covered by charts that are
simultaneously in the submanifold atlas and the standard atlas. ■
4. Full Rank Smooth Maps
By definition, the rank of a smooth map F : M → N between manifolds is the rank of the tangent
map. So the rank is a non-negative integer valued function on M, bounded from the above by
the minimum of the dimensions of M and N. Different smooth maps F, G : M → N may look
very different, even locally, depending on how does their rank vary on M. However, all constant
rank smooth maps, with fixed rank, say r, look locally the same. This is the main content of the
Rank Theorem, which is a nice and important example of a “local normal form theorem” or “local
classification theorem” in Differential Geometry. In this chapter, we begin defining the simplest
possible examples of constant rank maps: immersions, submersions, and local diffeomorphisms.
Thereafter, we state the Rank Theorem, and discuss a couple of consequences.
✓ F is an immersion if m ≤ n, and rank p F = m for all p ∈ M (in other words, the tangent
map d p F is injective for all p ∈ M).
✓ F is a submersion if m ≥ n, and rank p F = n for all p ∈ M (in other words, the tangent
map d p F is surjective for all p ∈ M).
✓ F is a local diffeomorphism if m = n, and rank p F = m = n for all p ∈ M (in other words,
the tangent map d p F is an isomorphism for all p ∈ M).
is exactly the representative matrix of the tangent map d p F : Tp M → TF(p) N in the coordinate
frames
∂ ∂ ∂ ∂
|p, . . . , m |p and | , . . . , n |F(p) .
∂ x1 ∂x ∂ y1 F(p) ∂y
It follows that
We conclude that
0 if P = (0, 0)
rankP F = rank JF (P) = .
2 if P ̸= (0, 0)
■ Example 4.3 Consider the standard Euclidean space R2 with coordinates (x, y). The curve
γ : R → R2 , t 7→ (sint, cost)
4.1 Full Rank Maps 75
Jγ = (cost, − sint) .
Whose rank is constant and equal to 1. Alternatively, we could argue as follows. For any t0 ∈ R, the
tangent space Tt0 R is spanned by dtd |t0 , whose image under the tangent map dt0 γ : Tt0 R → Tγ(t0 ) R2 is
d ∂ ∂
dt0 γ |t0 = γ̇(t0 ) = cost0 |γ(t0 ) − sint0 |γ(t0 ) ̸= 0.
dt ∂x ∂y
It follows that dt0 γ is injective for all t0 ∈ R, hence γ is an immersion. More generally, a curve
γ : I → M in a manifold M is an immersion if and only if the velocity γ̇(t0 ) is non-zero for all t0 ∈ I.
■
0 ··· 1 0 ··· 0
■ Example 4.6 Consider the standard Euclidean space Rn+1 with coordinates (t 0 , . . . ,t n ). We
already know that the canonical projection
is smooth (see Example 2.3). We want to show that it is a submersion. To do this, pick a point
P = (P0 , . . . , Pn ) ∈ Rn+1 ∖ {0}, and let i ∈ {0, . . . , n} be such that Pi ̸= 0. So (Vi := {t i ̸= 0}, id) is
a chart around P, and π(Vi ) ⊆ Ui , where, as usual, we denote by (Ui , ϕi ) the i-th affine chart on
RPn . Recall from Example 2.3 that the coordinate representation of π with respect to these charts is
!
n 0 n 0 n t0 tbi tn
πb : Vi → R , (t , . . . ,t ) 7→ πb(t , . . . ,t ) = ,..., i ,..., i .
ti t t
whose rank is constant and equal to n. In particular, rankP π = n. So, from the arbitrariness of P, π
is a submersion. ■
76 Chapter 4. Full Rank Smooth Maps
■ Example 4.7 Let U ⊆ Rn be an open subset, and let F : U → Rm be a smooth map, with m ≤ n.
Assume that 0 is a regular value of F. We already remarked that, in this case, rank JF = m in an
open neighborhood of any point P ∈ Z(F) (see the discussion following Example 2.12). Hence
rank F = m in an open neighborhood of V ⊆ U of Z(F), and F|V : V → Rm is a submersion.
Conversely, if there exists an open neighborhood V ⊆ U of Z(F) such that F|V : V → Rm is a
submersion, then 0 is a regular value of F (do you see it?). ■
where we used the chain rule (3.7). As dt γ is injective, dt γ ′ is injective as well, and, from
the arbitrariness of t, γ ′ is also an immersion. Finally, dim R = dim S1 = 1, hence γ ′ is a local
diffeomorphism. ■
■ Example 4.10 Consider the canonical projection π : Rn+1 ∖ {0} → RPn , and restrict it to the
sphere Sn ⊆ Rn+1 ∖ {0} to get a new smooth map π ′ : Sn → RPn . We want to show that π ′ is a
local diffeomorphism. As dim Sn = dim RPn = n, it is enough to check that π ′ is a submersion. To
do this, consider the projection
P
Π : Rn+1 ∖ {0} → Sn , P 7→ Π(P) := .
∥P∥
We already mentioned in Example 2.20 that Π is smooth. Additionally, π is the composition of Π
followed by π ′ (do you see it?). Hence, for any P ∈ Rn+1 ∖ {0},
dP π = dΠ(P) π ′ ◦ dP Π.
As dP π is surjective, dΠ(P) π ′ is surjective as well. Finally, from the surjectivity of Π, every point in
Sn is of the form Π(P), and π ′ is a submersion (hence a local diffeomorphism). ■
In particular,
✓ if m ≤ n, then F is an immersion if and only if, around every point of M, it has a coordinate
representation Fb such that
b 1 , . . . ,t m ) = (t 1 , . . . ,t m , 0, . . . , 0) = inm (t 1 , . . . ,t m );
F(t
✓ if m ≥ n, then F is a submersion if and only if, around every point of M, it has a coordinate
representation Fb such that
b 1 , . . . ,t m ) = (t 1 , . . . ,t n ) = prn (t 1 , . . . ,t m );
F(t
✓ if m = n, then F is a local diffeomorphism if and only if, around every point of M, it has a
coordinate representation Fb such that
b 1 , . . . ,t m ) = (t 1 , . . . ,t m ) = id(t 1 , . . . ,t m ).
F(t
Proof.
Exercise 4.2 Prove the implication (2) ⇒ (1) in the statement of the Rank Theorem (Theorem
4.2.1). ■
In practice, the Rank Theorem says that the local structure of a smooth map F : M → N of
constant rank r does only depend on dim M, dim N and r. Yet in other words, all smooth maps F of
constant rank r between an m-dimensional and an n-dimensional manifold do locally look the same.
For this reason we also say that the Rank Theorem is a local normal form theorem. We now discuss
some consequences of the rank theorem.
Proof. Let p ∈ M, and let (U, ϕ) and (V, ψ) be charts as in the statement of the Rank Theorem. As
prr maps open subsets to open subsets, and inr is a homeomorphism onto its image,
W := (prr ◦ F)(
b U) b ⊆ Rr
b = prr (U)
is an open neighborhood of 0 in Rr , and there exist open subsets W ⊆ W ⊆ Rr and W ′ ⊆ Rn−r , such
that 0 ∈ W ×W ′ ⊆ Vb . Consider the subcharts (V0 := ψ −1 (W ×W ′ ), ψ), and (U0 := F −1 (V0 )∩U, ϕ).
We have
Hence (V0 , ψ) is an adapted chart to F(U), around F(p). We conclude that F(U) is an r-
dimensional submanifold. Finally, the adapted chart (V0 , ψ) induces a submanifold chart (W0 , ψ0 :=
78 Chapter 4. Full Rank Smooth Maps
b 1 , . . . ,t m ) = (t 1 , . . . ,t r ) = prr (t 1 , . . . ,t m ),
F(t
ℝ7'%
U 5
M p 2
1
19 ℝ%
N 4(6)
"9 ,′
ℝ%
"#
in% pr%
,9
V !-(/)
ℝ%
,
For the second item in the statement, let q ∈ N be a point in the image of F, and let p ∈ Sq .
This simply means that F(p) = q. Take charts (U, ϕ) and (V, ψ) as in the statement of the Rank
Theorem. Then
But Z(F)
b is the solution space of the following regular system or r equations
1
t =0
.. .
.
r
t =0
ℝ2'%
U 0
M p
)
(
ℝ%
N ℝ%
1 "#
But (F ◦ γ)(t) = q constantly on I, so d p F(v) is the velocity of a constant curve, which is clearly 0
(do you see it?).
■
■ Example 4.11 — Tangent Spaces to the Sphere. Consider the standard Euclidean space Rn+1
with standard coordinates t 1 , . . . ,t n+1 , and the smooth map
∂
V = Vi |P
∂t i
to Rn+1 at some point P ∈ Sn is tangent to Sn if and only if
∂F d d
0 = dP F(V ) = V i i
(P) |F(P) = 2(V 1 P1 +· · ·+V n+1 Pn+1 ) |1 ⇔ V 1 P1 +· · ·+V n+1 Pn+1 = 0,
∂t dt dt
i.e.
ℝ%'(
"# $ %
P
!
$%
Figure 4.3: The tangent space to the sphere at the point P is “orthogonal” to P.
■ Example 4.12 — The Orthogonal Group. The orthogonal group is the following subgroup of
the general linear group:
O(n) := X ∈ GL(n, R) : X t X = In .
where the “(−)t ” denotes transposition and In is the identity matrix. In other words O(n) is the
fiber over In of the map:
i.e. O(n) = F −1 (In ). It is clear that F is a smooth map (do you see it?). If we manage to prove
that the rank of F is constant, it will follow from Corollary 4.2.2 that O(n) is a submanifold in
GL(n, R). Now let A ∈ GL(n, R) be any invertible matrix, and consider the maps:
and
They are smooth for every A (do you see it?). Additionally, they are inverted by the maps ΦA−1 and
ΨA−1 respectively. Hence they are diffeomorphisms. A direct computation shows that the following
diagram
GL(n, R)
ΦA
/ GL(n, R)
F F
M(n, R)
ΨA
/ M(n, R)
dXA F ◦ dX ΦA = dX t X ΨA ◦ dX F.
n(n + 1)
rank F = .
2
We conclude that O(n) ⊆ GL(n, R) is a submanifold of dimension n2 − n(n + 1)/2 = n(n − 1)/2. ■
Exercise 4.3 — The Special Linear and the Special Orthogonal Groups. Prove that
(1) the special linear group
We conclude this section discussing one more corollary of the Rank Theorem.
Proof. The “if part” of the statement is obvious. Let’s discuss the “only if” part. So let F : M → N be
a local diffeomorphism, and let p ∈ M be a point. Choose charts (U, ϕ), (V, ψ) as in the statement
of the Rank Theorem. Then Fb : U b → Vb acts as the identity. In particular, Fb is injective, and
U = F(U) ⊆ V . Put U := ψ (U). It is clear that F(U) ⊆ U ′ , and F : U → U ′ is a diffeomorphism
b b b b ′ −1 b
as claimed (do you see it?). ■
Notice that the Inverse Function Theorem motivates the terminology “local diffeomorphism”.
4.3 Embeddings
If we specialize Proposition 4.2.2 to an immersion i : N → M, we see that, restricting i to a
sufficiently small open subset U ⊆ N we can make i(U) to be a submanifold. In general, we cannot
choose U = N as the following counter-example shows.
82 Chapter 4. Full Rank Smooth Maps
■ Example 4.13 — The Eight Figure. Consider R2 with standard coordinates x, y and the following
curve:
i : R → R2 , t 7→ i(t) := (sin 2t, sint).
The image of i is called the eight figure (for obvious reasons). The eight figure is not a submanifold.
Indeed, around the origin, it is neither of the form y = f (x) nor of the form x = f (y) (for some
smooth function f ). Indeed, every vertical (or horizontal) line through a non-zero point in an open
neighborhood U ⊆ i(N) of 0 intersects U in at least two points (do you see it?). Notice that the
issue is still there if we restrict i to the interval (−π, π) turning it in an injective immersion
i : (−π, π) → R2 .
Indeed i(−π, π) = i(R) which is again the eight figure.
However, for every positive real number ε < π, i(−π + ε, π − ε) is a submanifold and
i : (−π + ε, π − ε) → i(−π + ε, π − ε)
is a diffeomorphism. ■
Proposition 4.3.1 Let i : N → M be an injective immersion. Then the following two conditions
are equivalent:
(1) i(N) ⊆ M is a submanifold of the same dimension as N;
(2) τsub = τi , or, equivalently, i : N → (i(N), τsub ) is a homeomorphism.
Additionally, if (1), or, equivalently, (2), holds true, then i : N → i(N) is a diffeomorphism.
Proof.
(2) ⇒ (1) Assume that i : N → (i(N), τsub ) is a homeomorphism, pick a point q ∈ i(N) and let
p ∈ N be the unique point such that q = i(p). From Proposition 4.2.2, there is an open neighborhood
U ⊆ M of p such that i(U) ⊆ M is a submanifold of the same dimension as N. On the other hand,
we have i(U) ∈ τi = τsub . Hence there is an open subset U ⊆ M such that i(U) = i(N) ∩ U .
Summarizing, we have showed that, for any q ∈ i(N) there exists an open neighborhood U ⊆ M of
q such that i(N) ∩ U is a submanifold. This easily implies that i(N) is itself a submanifold. We
leave the details as Exercise 4.5. ■
Exercise 4.5 Conclude the proof of Proposition 4.3.1, showing that a subset S ⊆ M in a manifold
M is a submanifold if an only if it is locally so, i.e., for every point q ∈ S, there exists an open
neighborhood U ⊆ M of q such that S ∩ U is a submanifold. ■
■ Example 4.15 Let d, n be non-negative integers such that d ≤ n. The smooth map
is an embedding. Indeed it is an injective immersion, and its image is the d-dimensional submanifold
Rd × {0} ⊆ Rd × Rn−d = Rn . ■
Proof. From the section property π ◦ i = idN , it immediately follows that π is surjective and i is
injective. Now, let p ∈ N. The chain rule at p reads
■ Example 4.16 Let M, N be manifolds, and let F : M → N be a smooth map. Consider the map
From Exercise 2.7, ΓF is a smooth map. More precisely, it is a section of the projection
onto the first factor. So, from Proposition 4.3.2, ΓF is an embedding. In particular, ΓF (M) is a
submanifold of the same dimension as M in the product M × N. As ΓF (M) = graph F, the latter
fact is not new (see Example 2.15). The new fact is that ΓF (M) is, additionally, diffeomorphic to
M. ■
■ Example 4.17 — The Segre Embedding. We want to show that the product RP1 × RP1 of two
copies of the projective line RP1 can be embedded into the 3-dimensional projective space RP3 .
More precisely, there is an embedding, called the Segre embedding,
given by
0 1 0 1 0 0 0 1 1 0 1 1 0 1 0 1
σ s :s , t :t := s t : s t : s t : s t , s : s , t : t ∈ RP1 . (4.1)
To see this, first notice that σ is well-defined by (4.1) (do you see it?).
Second, we show that σ is a smooth map. Denote by (U0 , ϕ0 ), (U1 , ϕ1 ) the usual affine charts
on RP1 , and by (Vi , ψi ) the affine charts on RP3 , i = 0, . . . , 3. It is clear that
showing that σ has coordinate representations around every point. Let’s compute the first one:
σb : U b0 = R × R → Vb0 = R3 .
b0 × U (4.2)
which depends smoothly on (P, Q). Similarly, all other coordinate representations are smooth
(check it as an exercise).
b0 × U
Third, we show that σ is an immersion. Denote by x, y the standard coordinates on U b0 =
R × R. Then, the Jacobian matrix of the coordinate representation (4.2) is
∂ σb 1 ∂ σb 1
∂x ∂y 0 1
2
∂ σb ∂ σb 2
∂y =
∂x 1 0 ,
∂ σb 3 ∂σb3 y x
∂x ∂y
4.3 Embeddings 85
whose rank is constant and equal to 2. This shows that rank p σ = 2 for all p ∈ U0 ×U0 . Similarly
the rank of σ is 2 on any other point (check it as an exercise), and σ is an immersion.
Fourth, we show that σ is injective. So let
be two points in RP1 × RP1 such that σ (p) = σ ( p̃). This means that there exists a non-zero scalar
λ ∈ R such that
λ t˜ j i
si = s̃ , for all i ∈ {0, 1}.
tj
Hence, s0 : s1 = s̃0 : s̃1 . Similarly, t 0 : t 1 = t˜0 : t˜1 , so that p = p̃.
We conclude proving that σ (RP1 × RP1 ) ⊆ RP3 is a 2-dimensional submanifold. To do this,
we show that it coincides with the submanifold S from Example 2.17. So let q = P0 : · · · : P3 be a
P0 P3 − P1 P2 = 0. (4.3)
The “only if part” of the claim follows from an easy computation that we leave as an exercise. For
the if part, notice that the condition (4.3) can be re-written in the form
0
P P1
det = 0. (4.4)
P2 P3
Let (P0 , P1 ) ̸= (0, 0). Then, from (4.4), there exists a scalar a ∈ R such that
so that
q = P0 : · · · : P3 = P0 : P1 : aP0 : aP1 = σ [1 : a] , P0 : P1 .
The case (P2 , P3 ) ̸= (0, 0) can be discussed in a similar way and we leave the obvious details to the
reader.
Summarizing, we have proved that σ is an embedding. More generally, for any two non-negative
integers n, m there is an embedding, also called the Segre embedding, σ : RPn × RPm → RPnm+n+m .
We will not discuss the details of the latter construction. ■
II
Vector Fields and Differential
Forms
5 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . 89
5.1 Vector Fields
5.2 Vector Fields and Fields of Vectors
5.3 Vector Fields and Smooth Maps
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5. Vector Fields
In this chapter, we continue developing a differential calculus on manifolds, defining vector fields.
There are two alternative (but equivalent) definitions of a vector field on a manifold. Geometrically,
a vector field on a manifold M is the assignment of a tangent vector Xp to M at the point p, for every
point p ∈ M, what we also call a field of vectors. Additionally, we require that Xp depends smoothly
on p. This can be formalized rigorously via sections of the tangent bundle. There is another possible
algebraic definition of a vector field, which is particularly efficient, and that we adopt as the primary
definition: vector fields on M are derivations of the algebra C∞ (M) of smooth functions on M. This
latter definition can be easily transported to other areas of Geometry like Algebraic Geometry or
Supergeometry. We will provide both definitions and discuss their equivalence. We also discuss
how do vector fields interact with smooth maps. In the next chapter we will discuss two important
(and interrelated) interpretations of vector fields: vector fields as ordinary differential equations,
and vector fields as infinitesimal diffeomorphisms.
X :A→A
90 Chapter 5. Vector Fields
In this section we elaborate on Definition 5.1.2, discussing its algebraic consequences for
the space of vector fields. In the next section, we show that vector fields on M are equivalent to
(smooth) fields of vectors, i.e. (smooth) sections of the tangent bundle T M.
■ Example 5.1 — Coordinate Vector Fields. Let M be an n-dimensional manifold and let
∂ ∂ ∂ fb
: C∞ (U) → C∞ (U), f 7→ f := ◦ ϕ,
∂ xi ∂ xi ∂t i
where, as usual, fb = f ◦ ϕ −1 is the coordinate representation of f . Notice that, for each f ∈ C∞ (U),
∂ ∂ fb
the coordinate representation of ∂ xi
f is exactly ∂t i
:
d∂ ∂ fb
i
f = i. (5.1)
∂x ∂t
The map ∂∂xi is called the i-th coordinate vector field and it is indeed a vector field on U (beware,
not on M!). The proof of this claim is left as Exercise 5.1. ■
Exercise 5.1 Show that the i-th coordinate vector field ∂∂xi is a vector field on U. ■
We now show that the space X(M) of vector fields on M does actually carry some algebraic
structures. We begin defining the sum of two vector fields and the product of a vector field by a
smooth function. So let X,Y ∈ X(M) be two vector fields, and let f ∈ C∞ (M) be a smooth function.
We define new maps:
and
We conclude this section discussing one more algebraic structure on X(M): vector fields form
a Lie algebra. We begin with the definition of Lie algebra.
Definition 5.1.3 — Lie Algebra. A real Lie algebra, or simply a Lie algebra, is a real vector
space g equipped with one more interior operation
[u, [v, w]] + [v, [w, u]] + [w, [u, v]] = 0 for all u, v, w ∈ g.
A Lie subalgebra of a Lie algebra g is a vector subspace h ⊆ g which is preserved by the Lie
bracket, i.e. [v, w] ∈ h, for all v, w ∈ h.
Notice that a Lie subalgebra h of a Lie algebra g is also a Lie algebra with the restricted Lie
bracket. The following example is our main example of a Lie algebra (besides vector fields).
■ Example 5.2 — Lie Algebra of Endomorphisms. Let V be a real vector space. The real vector
space EndR (V ) of (R-linear) endomorphisms of V is a Lie algebra with the following Lie bracket:
Prove this claim as an exercise! The Lie bracket in EndR (V ) is called the commutator, and two
endomorphisms h, l ∈ EndR (V ) are said to commute if their commutator vanishes: [h, l] = 0. ■
Proposition 5.1.1 — Lie Algebra of Vector Fields. Let M be a manifold. Vector fields on M
form a Lie subalgebra of the Lie algebra EndR (C∞ (M)) of R-linear endomorphisms of the real
vector space C∞ (M). In other words, the commutator of two vector fields is a vector field.
In particular, the space X(M) of vector fields, with the commutator, is a Lie algebra. Notice that
the commutator of vector fields is R-bilinear, but it is not C∞ (M)-bilinear (in general). Actually,
the C∞ (M)-module structure and the Lie algebra structure on X(M) interact in a more complicated
(but still nice) way, as explained in the next
92 Chapter 5. Vector Fields
Proposition 5.1.2 Let X,Y ∈ X(M) be vector fields and let f ∈ C∞ (M) be a smooth function.
Then
It can be proved that Xp is a tangent vector to M at p (see Exercise 5.5), called the value at p of the
vector field X.
Exercise 5.5 Prove that, for every p ∈ M, the map X p defined in (5.2) is a tangent vector to M
at the point p. Prove also that, for every X,Y ∈ X(M), and every f ∈ C∞ (M),
(X +Y ) p = Xp +Yp ,
( f X) p = f (p)Xp .
Finally, let (U, ϕ = (x1 , . . . , xn )) be a chart around p. Prove that the value of the coordinate
vector field ∂∂xi at p, is exactly the coordinate tangent vector ∂∂xi | p . ■
The assignment p 7→ Xp is a field of vectors that can be interpreted as a(n a priori non-necessarily
smooth) section sX of the tangent bundle T M:
sX : M → T M, p 7→ sX (p) := Xp .
Proof. We have to check several things. First of all, we have to check that, for every X ∈ X(M),
the section sX is actually smooth. It is enough to check that, for every chart (U, ϕ = (x1 , . . . , xn ))
on M, the components siX of the section sX in the chart (U, ϕ) (as defined at the end of Section
3.3) are smooth functions on U. To do this, we fix an arbitrary point p0 in U and show that siX is
smooth around p0 (the smoothness of siX then follows from the arbitrariness of p0 ). So, let p ∈ U
and compute
siX (p) = ẋi (sX (p)) = ẋi (Xp ) = i-th component of Xp in the coordinate frame = Xp (xi ).
5.2 Vector Fields and Fields of Vectors 93
In order to continue our computation, we need to choose a function xei ∈ C∞ (M) such that xei = xi in
an open neighborhood V ⊆ U of p0 . Such function exists in view of the Local Extension Lemma
(Lemma 2.3.6), and, from the locality of tangent vectors, we have that
siX (p) = Xp (xi ) = Xp (e
xi ) = X(e
xi )(p)
for all p ∈ V . As X maps smooth functions to smooth functions, we conclude that siX |V is smooth,
and (5.3) is well-defined.
We leave it as Exercise 5.6 (Point (1)) to check that the map (5.3) is C∞ (M)-linear. It remains
to check that (5.3) is invertible (the C∞ (M)-linearity of the inverse then follows from general
properties of module isomorphisms). To do this we define its inverse. Namely, given a section
s ∈ Γ(T M) of the tangent bundle, we construct a vector field Xs ∈ X(M) in such a way that
sXs = s and XsX = X, (5.4)
for all s ∈ Γ(T M), and all X ∈ X(M), i.e. in such a way that the map Γ(T M) → X(M), s 7→ Xs
inverts (5.3). For every f ∈ C∞ (M) we define
Xs ( f ) : M → R, p 7→ Xs ( f )(p) := s(p)( f ).
We have to show that
✓ Xs ( f ) ∈ C∞ (M) for all s ∈ Γ(T M) and all f ∈ C∞ (M),
✓ Xs : C∞ (M) → C∞ (M) is a derivation of C∞ (M), hence a vector field, and
✓ the identities (5.4) hold for all s ∈ Γ(T M) and all X ∈ X(M).
We only discuss the first point and leave it to the reader to prove the remaining two points as
Exercise 5.6 (Points (2) and (3)). Let s ∈ Γ(T M) and f ∈ C∞ (M). In order to prove that Xs ( f ) is
smooth, we prove that the restriction Xs ( f )|U is smooth for every chart (U, ϕ). So, let p ∈ U, and
compute
Xs ( f )(p) = s(p)( f ) = s(p)( f |U )
i ∂ i ∂ i ∂
= s (p) i | p f |U = s i f |U = s i ( f |U )(p),
∂x ∂x p ∂x
where, in the first step of the second line, we used Exercise 5.5. As the si are smooth, si ∂∂xi is a
vector field on U, and the last expression depends smoothly on p. This concludes the proof. ■
Proposition 5.2.1 has some important consequences. First of all, a vector field X on M can be
“restricted” to an open submanifold U ⊆ M. Namely, X corresponds to a section s : M → T M of
T M. Restricting s to U in the domain and to T U in the codomain, we get a section s|U : U → T U
of T U . Finally, s|U corresponds to a vector field on U that we denote X|U , and call the restriction
of X to U . Notice that, by construction, for every point p ∈ U , the value of X|U at p agrees with
the value of X at the same point p:
(X|U ) p = Xp , for all p ∈ U .
Hence, we also have X|U ( f |U ) = X( f )|U for all X ∈ X(M), and all f ∈ C∞ (M). Now, let
X ∈ X(M), and f ∈ C∞ (U ). Sometimes, abusing the notation, we will write X( f ) instead of
X|U ( f ). In any case, X( f ) = X|U ( f ) is a smooth function on U (not on M).
A second consequence of Proposition 5.2.1 (closely related to the previous one) is the following
94 Chapter 5. Vector Fields
Lemma 5.2.2 — Gluing Lemma for Vector Fields. Let M be a manifold, and let C = {U } be
an open cover of M. For every U ∈ C , let XU be a vector field on U . If
XV |U ∩V = XU |U ∩V
for every U , V ∈ C , then there exists a unique vector field X on M such that
X|U = XU
for all U ∈ C .
Exercise 5.7 Prove the Gluing Lemma for Vector Fields (Lemma 5.2.2). ■
A third consequence of Proposition 5.2.1 is that, for every chart (U, ϕ), the C∞ (U)-module of
vector fields on U possesses a finite frame consisting of the coordinate vector fields, according to
the following
Proposition 5.2.3 Let M be an n-dimensional manifold, and let (U, ϕ = (x1 , . . . , xn )) be a chart
on M. Then the coordinate vector fields
∂ ∂
,..., n
∂ x1 ∂x
form a frame in X(U) (called the coordinate frame), i.e. they are independent and generate
X(U). In particular, every vector field X ∈ X(U) can be uniquely written in the form
∂
X = Xi
∂ xi
for some smooth functions X i ∈ C∞ (U).
Proof. We first prove that the coordinate vector fields are independent. So let f 1 , . . . , f n be smooth
functions on U such that
∂
fi = 0.
∂ xi
This means that f i ∂∂xi g = 0 for all g ∈ C∞ (U). In particular, we can chose g = x j , j ∈ {1, . . . , n}.
From (5.1), ∂∂xi x j = δij , hence
∂ j
0 = fi x = f i δij = f j .
∂ xi
It remains to prove that every vector field X ∈ X(U) can be written in the form
∂
X = Xi
∂ xi
for some smooth functions X i ∈ C∞ (U). To do this, consider the section sX ∈ Γ(TU) corresponding
to U, and put X i = siX . This means that, for every p ∈ U
i ∂ i ∂
Xp = X (p) i | p = X ,
∂x ∂ xi p
5.2 Vector Fields and Fields of Vectors 95
where we used Exercise 5.5. As X and X i ∂∂xi share the same values (in points of U), they must
coincide. ■
We can combine the Gluing Lemma 5.2.2 with Proposition 5.2.3, to find that, for every vector
field X on M,
(1) X is completely determined by its restrictions X|U to some coordinate domains U covering
M,
(2) for every chart (U, ϕ = (x1 , . . . , xn )), the restriction X|U can be uniquely written in the form
∂
X|U = X i
∂ xi
for some smooth functions X i ∈ C∞ (U), called the components of X in the chart (U, ϕ).
The proof of Proposition 5.2.3 then shows that the components X i of a vector field X ∈ X(M) in the
chart (U, ϕ = (x1 , . . . , xn )), coincide with the components of the corresponding section, and they
are given by
X i = X(xi ).
Now, let X be a vector field on M. How do the components X i of X in a chart change under a
change of coordinates? To answer this question take two charts (U, ϕ = (x1 , . . . , xn )) and (U, ϕ̃ =
(x̃1 , . . . , x̃n )) with the same coordinate domain U. Then X|U can be written either as
∂ ∂
X|U = X i , or as X|U = X̃ i
∂ xi ∂ x̃i
for some smooth functions X i , X̃ i ∈ C∞ (U), and we want to find the relationship between those. We
have
∂ i
X̃ i = X(x̃i ) = X j x̃ .
∂xj
In its turn, the matrix
∂ i i=1,...,n
x̃
∂xj j=1,...,n
∂ i ∂ i
d ∂ x̃bi ∂ (ϕ̃ ◦ ϕ −1 )i
x̃ = x̃ ◦ ϕ = ◦ ϕ = ◦ ϕ.
∂xj ∂xj ∂t j ∂t j
∂ i ∂ (ϕ̃ ◦ ϕ −1 )i
d
x̃ = .
∂xj ∂t j
96 Chapter 5. Vector Fields
Exercise 5.8 Consider the 2-dimensional standard Euclidean space R2 with standard co-
ordinates x, y, and the polar chart (U, ψ = (r, φ )) on it. Here U = R2 ∖ {y = 0, x ≥ 0},
b = (0, +∞) × (0, 2π) and
U
p
ψ : U → U, b (x, y) 7→ ψ(x, y) = (r, φ ) := x2 + y2 , φ (x, y) ,
where
arccos √ x
if y ≥ 0
x2 +y2
φ (x, y) := .
− arccos √x if y < 0
x2 +y2
ψ −1 : U
b → U, (r, φ ) 7→ ψ −1 (r, φ ) = (r cos φ , r sin φ ) .
∂ ∂ ∂ ∂
X =x +y and Y = x −y .
∂x ∂y ∂y ∂x
Prove that
∂ ∂
X|U = r and Y |U = .
∂r ∂φ
■
Exercise 5.9 Consider the real projective plane RP2 , and denote (Ui , ϕi = (xi , yi )) the standard
affine charts on it, i = 0, 1, 2. Prove that there exists a unique vector field X ∈ X(RP2 ) such that
∂ ∂
X|U0 = 1 + x02 + x0 y0 ,
∂ x0 ∂ y0
∂ ∂
X|U1 = − 1 + x12 − x1 y1 ,
∂ x1 ∂ y1
∂ ∂
X|U2 = x2 − y2 .
∂ y2 ∂ x2
■
X ◦ F ∗ = F ∗ ◦Y.
Before providing examples, we remark that the F-relatedness of two vector fields can be
5.3 Vector Fields and Smooth Maps 97
equivalently defined interpreting vector fields as fields of vectors, according to the following
Proposition 5.3.1 A vector field X on M and a vector field Y on N are F-related if and only if,
for every p ∈ M,
(d p F)(Xp ) = YF(p) .
Proof. Let X ∈ X(M), and Y ∈ X(N). Then X and Y are F-related if and only if X(F ∗ ( f )) =
F ∗ (Y ( f )) for all f ∈ C∞ (N). In turn, this is equivalent to
But
and
■ Example 5.3 In the standard Euclidean plane R2 with standard coordinates x, y, consider the
following curve :
Then X and Y are γ-related. Indeed, for every smooth function f = f (x, y) on R2
d d
X(γ ∗ ( f )) = f ◦ γ = f (cost, sint)
dt dt
∂f d cost ∂ f d sint
= (cost, sint) + (cost, sint)
∂x dt ∂y dt
∂f ∂f
= − sint (cost, sint) + cost (cost, sint)
∂x ∂y
∂ ∂
= x −y f ◦ γ = γ ∗ (Y ( f )).
∂y ∂x
■
■ Example 5.4 Let d, n be non-negative integers, with d ≤ n, and consider the standard Euclidean
ℝ#$" ℝ#
'
ℝ"
pr"
( ℝ"
Figure 5.2: The vector field X and Y from Example 5.4 are prd -related.
on Rn , and
∂
Y = Xi
∂t i
on Rd , are prd -related, for every choice of the X i = X i (t 1 , . . . ,t d ) ∈ C∞ (Rd ), and the Y a =
Y a (s1 , . . . , sd , r1 , . . . , rn−d ) ∈ C∞ (Rn ) (Exercise 5.10).
■
Exercise 5.10 Show that the vector fields X and Y from Example 5.4 are prd -related. ■
Next proposition shows that Definition 5.3.2 agrees with our intuition on a vector field X ∈ X(M)
being tangent to a submanifold S ⊆ M.
Proposition 5.3.2 A vector field X on M is tangent to a submanifold S ⊆ M if and only if, for
every p ∈ S, Xp ∈ Tp S, and, in this case, (X|S ) p = Xp .
Proof. First assume that X is tangent to S, and let X|S be its restriction. Then, from Proposition
5.3.1,
for all p ∈ S. This shows that Xp ∈ Tp S. Conversely, assume that Xp ∈ Tp S for all p ∈ S. Think of X
as a section sX of T M. Then the hypothesis says that sX |S takes values in T S. In particular, sX can
be restricted to S in the domain and to T S in the codomain, to get a section sX |S : S → T S of T S.
As T S ⊆ T M is a submanifold (Proposition 3.3.3), sX |S is a smooth section. Denote by Y ∈ X(S)
the associated vector field. By construction Yp = d p iS (Yp ) = Xp for all p ∈ S, so, from Proposition
5.3.1 again, Y and X are iS -related. ■
5.3 Vector Fields and Smooth Maps 99
M
X S
X|S
It immediately follows from the last part of Proposition 5.3.2 that, if the vector field X ∈ X(M)
is tangent to a submanifold S ⊆ M, then there is a unique restriction of X to S, i.e. a unique vector
field X|S on S such that X|S and X are iS -related.
■ Example 5.5 Consider the standard 3-dimensional Euclidean space R3 with coordinates (x, y, z),
and the vector field
∂ ∂
X = y −z ∈ X(M).
∂z ∂y
Use Proposition 5.3.2 to show that X is tangent to the 2-dimensional sphere S2 ⊆ R3 . So let
P = (x0 , y0 , z0 ) ∈ S2 , i.e. x02 + y20 + z20 = 1, and check that XP ∈ TP S2 . We have
∂ ∂ ∂ ∂
XP = y − z = y0 |P − z0 |P ,
∂z ∂y P ∂z ∂y
which is in TP S2 if and only if its component vector (0, −z0 , y0 ) is orthogonal to P = (x0 , y0 , z0 )
(Example 4.11). This is indeed the case as
0 · x0 − z0 · y0 + y0 · z0 = 0.
■
Exercise 5.11 Consider the vector field X from Example 5.5. As X is tangent to the sphere S2 ,
it can be restricted to S2 , and then further restricted to U+ = S2 ∖ {P+ }. Denote by (A, B) the
stereographic coordinates on U+ . Prove that
∂ B2 − A2 + 1 ∂
X|U+ = AB + .
∂A 2 ∂B
■
Next proposition explores the relationship between F-relatedness and the algebraic structures
on vector fields.
Proposition 5.3.3 Let M, N be manifolds, let F : M → N be a smooth map, let X, X1 , X2 be
vector fields on M, and let Y,Y1 ,Y2 be vector fields on N. If X, X1 , X2 are F-related to Y,Y1 ,Y2
respectively, then
✓ X1 + X2 and Y1 +Y2 are F-related,
100 Chapter 5. Vector Fields
"
2 −A2 +1
Figure 5.4: The vector field AB ∂∂A + B 2
∂
∂B from Exercise 5.11.
■ Example 5.6 Let M be a manifold, let S ⊆ M be a submanifold, and let X1 , X2 , X be vector fields
on M that are tangent to S. According to our notation, denote by X1 |S , X2 |S , X|S the restrictions to S
of X1 , X2 , X respectively. It immediately follows from Proposition 5.3.3 that
✓ X1 + X2 is tangent to S, and (X1 + X2 )|S = X1 |S + X2 |S ,
✓ f X is tangent to S for every f ∈ C∞ (M), and ( f X)|S = f |S X|S ,
✓ [X1 , X2 ] tangent to S, and [X1 , X2 ]|S = [X1 |S , X2 |S ]
(do you see it?). ■
Summarizing
i
∂ j ∂
i i j ∂ i ∂
[X,Y ]|U = X|U (Y ) −Y |U (X ) i
= X j
Y −Y j
X .
∂x ∂x ∂x ∂ xi
In particular
∂ ∂
, = 0, for all i, j ∈ {1, . . . , n}.
∂ xi ∂ x j
■
Φ∗ (Y ) = Φ∗ ◦Y ◦ (Φ−1 )∗
or, equivalently,
Additionally, for any three vector fields Y,Y1 ,Y2 on N and any function f ∈ C∞ (N), we have
✓ Φ∗ (Y1 +Y2 ) = Φ∗ (Y1 ) + Φ∗ (Y2 ),
✓ Φ∗ ( fY ) = Φ∗ ( f )Φ∗ (Y ),
✓ Φ∗ ([Y1 ,Y2 ]) = [Φ∗ (Y1 ), Φ∗ (Y2 )].
Finally, if Ψ : N → Q is another diffeomorphism, and Z ∈ X(Q), then
■ Example 5.8 Let M be an n-dimensional manifold, and let (U, ϕ = (x1 , . . . , xn )) be a chart on M.
Then, every function f ∈ C∞ (U) can be seen as the pull-back ϕ ∗ ( fb) of its coordinate representation
fb along the diffeomorphism ϕ : U → U.
b It follows that Identity (5.1) can be rephrased as follows:
∂ ∂
i
= ϕ∗ (5.6)
∂x ∂t i
(do you see it?). ■
R Let M be a manifold, and let (U, ϕ = (x1 , . . . , xn )) be a chart on it. Identity (5.6) basically
says that, if we identify U and U b via ϕ, then the coordinate vector fields ∂ i identify with
∂x
the standard partial derivatives ∂t∂ i . In particular, the ∂∂xi inherits all standard properties of
standard derivatives, and, to stress this, in the following, for every function g on U, we simply
write
∂g ∂ ∂ 2g ∂ ∂
for g, or for g, etc.
∂ xi ∂ xi ∂ xi ∂ x j ∂ xi ∂ x j
Let’s make one example, take another manifold N, a chart (V, ψ = (y1 , . . . , yn )) on N, and a
smooth map F : M → N such that F(U) ⊆ V . We already know that the restriction F : U → V
102 Chapter 5. Vector Fields
∂ ( f ◦ F) ∂ F a
∂f
= · ◦F (5.7)
∂ xi ∂ xi ∂ ya
(check it as an exercise) which is the chain rule for coordinate vector fields. In the following,
we will apply this and similar formulas without further comments.
6. Flows and Symmetries
In this chapter, we provide two more interpretations of vector fields. First of all, a vector field can
be interpreted as an ordinary differential equation (ODE), more precisely, a system of explicit and
autonomous ordinary differential equestions. The solutions of such system are the so called integral
curves of the vector field, and the Theorem of Existence, Uniqueness, and Smooth Dependence on
Initial Data of Solutions of a System of ODEs from Calculus guarantees that integral curves are
particularly well-behaved. Secondly, a vector field can be seen as an infinitesimal diffeomorphism.
Points of a manifold move along the integral curves of a vector field, and the vector field says
what is the velocity of this displacement. Hence a vector field can be seen as an infinitesimal
displacement of every point of the manifold. This displacement is smooth and (morally) one-to-one,
hence it is a(n infinitesimal) diffeomorphism. This informal discussion will be made more rigorous
along the following sections.
d
◦ γ ∗ = γ ∗ ◦ X,
dt
or, equivalently,
d
γ̇(t0 ) = dt0 γ |t = Xγ(t0 ) , for every t0 ∈ I.
dt 0
dγ i ∂
γ̇(t) = (t) i |γ(t)
dt ∂x
(see (3.13)). On the other hand
∂
X|U = X i ,
∂ xi
∂
Xγ(t) = X i (γ(t)) | .
∂ xi γ(t)
We conclude that γ : J → M is an integral curve of X if and only if
dγ i
(t) = X i (γ(t)) = Xbi (γ 1 (t), . . . , γ n (t)), for all t ∈ J,
dt
which is an ODE for the coordinate representation
γb : J → U,
b t 7→ γb(t) = (γ 1 (t), . . . , γ n (t))
of γ. Most of the properties of integral curves are corollaries of the following standard result in
Calculus, which we state without a proof.
dγ i
(t) = Xbi (γ 1 (t), . . . , γ n (t)), i ∈ {1, . . . , n}, (6.1)
dt
imposed on a curve γb = (γ 1 , . . . , γ n ) : I → U in U. We have the following.
(1) U NIQUENESS . For every t0 ∈ R, every two solutions γ1 : I1 → U, and γ2 : I2 → U of (6.1)
such that t0 ∈ I1 ∩ I2 , and γ1 (t0 ) = γ2 (t0 ), agree on the (non-empty) intersection I1 ∩ I2 of
their domains, i.e. γ1 |I1 ∩I2 = γ2 |I1 ∩I2 .
(2) E XISTENCE . For every t0 ∈ R, and every P0 ∈ U, there is an open interval I containing
t0 , and an open neighborhood V ⊆ U of P0 , such that, for every P ∈ V , there exists a
(necessarily unique) solution γP : I → U of (6.1) such that γP (t0 ) = P.
(3) S MOOTH D EPENDENCE ON I NITIAL DATA . For every t0 ∈ R, and every P0 ∈ U, let I, V ,
P and γP be as in Point (2) E XISTENCE . Then the map Φ : I ×V → U, (t, P) 7→ Φ(t, P) :=
γP (t) depends smoothly on (both t and) P.
Proof. Omitted. ■
6.1 Integral Curves of a Vector Field 105
A first corollary of Theorem 6.1.1 is the existence and uniqueness of maximal integral curves
of a vector field (starting from a prescribed point). Before stating this result, we have to define
maximal integral curves. So let M be a manifold, let X be a vector field on M, and let p ∈ M be
a point. Denote by I p the set of integral curves of X starting from p. It is easy to see that I p is
non-empty, i.e., for every p ∈ M, there exists an integral curve starting from p. Indeed, choose a
chart (U, ϕ) around p and look for an integral curve γ : I → M taking values in U. From Theorem
6.1.1, and the discussion preceding it, such γ exists. Next, any two integral curves γ1 : I1 → M,
γ2 : I2 → M in I p , agree on the intersection I1 ∩ I2 of their domains. This follows from Theorem
6.1.1 again and a little bit of Topology that we omit. Finally, I p is equipped with a partial order ≤
defined as follows: let γ1 : I1 → M and γ2 : I2 → M be in I p , i.e. γ1 , γ2 are integral curves starting
from p, then
A maximal integral curve starting from p is, by definition, a maximal element of I p with respect
to this partial order, i.e. an integral curve γ of X starting from p such that, whenever γ ′ is another
integral curve starting from p such that γ ≤ γ ′ , in fact we have γ = γ ′ . Yet in other words, γ cannot
be extended to an integral curve starting from p and defined on a larger time interval.
Proof (a sketch). Define Ip ⊆ R as the union of the domains of all integral curves starting from p:
[
Ip := I.
(γ:I→M)∈I p
Being the union of open intervals containing 0, Ip is itself an open interval containing 0. Finally,
we define a curve
γ p : Ip → M
as follows. For every t ∈ Ip , let γ : I → M be any integral curve starting from p such that t ∈ I, and
put
γ p (t) := γ(t).
From the above discussion, γ p (t) is independent of the choice of γ. Additionally, γ p agrees with γ
in a whole open neighborhood of t in Ip . Hence, from the Gluing Lemma for Smooth Maps, γ p is a
well-defined curve. As γ p agrees locally, around every time in Ip , with an integral curve, it is an
integral curve itself, and, by construction, its domain is the largest possible, so γ p is maximal. ■
■ Example 6.1 On R2 with standard coordinates (x, y) consider the coordinate vector field
∂
X= .
∂x
A curve γ = (x(t), y(t)) : I → R2 is an integral curve of X, if it is a solution of the system of ODEs:
dx
dt (t) = 1 .
dy
dt (t) = 0
106 Chapter 6. Flows and Symmetries
γP : R → R2 , t 7→ γP (t) = (x0 + t, y0 ).
"
∂
Figure 6.1: The coordinate vector field ∂x and its integral curves.
■ Example 6.2 On R2 with standard coordinates (x, y) consider the vector field
∂ ∂
X =x +y .
∂x ∂y
γP : R → R2 , t 7→ γP (t) = (x0 et , y0 et ).
■
6.1 Integral Curves of a Vector Field 107
"
Figure 6.2: The vector field x ∂∂x + y ∂∂y and its integral curves.
■ Example 6.3 On R2 with standard coordinates (x, y) consider the vector field
∂ ∂
X =x −y .
∂y ∂x
A curve γ = (x(t), y(t)) : I → R2 is an integral curve of X, if it is a solution of the system of ODEs:
dx
dt (t) = −y(t) .
dy
dt (t) = x(t)
This means that
x(t) = x(0) cost − y(0) sint
.
y(t) = x(0) sint + y(0) cost
Hence, the maximal integral curve of X starting from P = (x0 , y0 ) is
γP : R → R2 , t 7→ γP (t) = (x0 cost − y0 sint, x0 sint + y0 cost).
■
Exercise 6.1 On R2 with standard coordinates (x, y), find the maximal integral curves of the
vector field
∂ ∂
X =x −y .
∂x ∂y
■
Notice that maximal integral curves need not be defined on the whole R as the following
counter-example shows.
■ Example 6.4 Consider the standard line R with standard coordinate x and the open submanifold
d
(0, +∞) ⊆ R in it. Additionally, consider the coordinate vector field X = dx |(0,+∞) on (0, +∞). Let
x0 ∈ R. Any integral curve γ : I → R of X starting from x0 is of the type
γ : I → (0, +∞), t 7→ γ(t) = x0 + t.
108 Chapter 6. Flows and Symmetries
"
Figure 6.3: The vector field x ∂∂y − y ∂∂x and its integral curves.
In particular, −x0 cannot belong to I, not even when I = Ix0 and γ is the maximal integral curve
starting from x0 . ■
We can change the “initial point” of an integral curve, by composing with a translation. Namely,
let t¯ ∈ R. Then the translation
τt¯ : R → R, t 7→ τt¯(t) := t + t¯
is a diffeomorphism whose inverse is τ−t¯. For any open interval I ⊆ R, we denote by I + t¯ := τt¯(I)
its image under the translation τt¯.
Lemma 6.1.3 — Translation Lemma. Let M be a manifold, let X be a vector field on M, and
let γ : I → M be an integral curve of X. Then, for every t¯ ∈ R, the composition
γ ◦ τt¯ : I − t¯ → M
Proof. For the first part of the statement, we compute the velocity
d(γ ◦ τt¯)
(t)
dt
6.1 Integral Curves of a Vector Field 109
"
of the curve γ ◦ τt¯ at time t ∈ I − t¯. So, let f ∈ C∞ (M) and compute
d(γ ◦ τt¯) d
(t)( f ) = f (γ(τt¯(t)))
dt dt
d
= |s=t f (γ(s + t¯))
ds
d d
= ′ |s′ =t+t¯ f (γ(s′ )) · |s=t (s + t¯)
ds ds
dγ
= (t + t¯)( f )
dt
= Xγ(t+t¯) ( f )
= X(γ◦τt¯)(t) ( f ).
This shows that γ ◦ τt¯ is an integral curve. For the second part of the statement, let p ∈ M be a
point and consider the maximal integral curve γ p : Ip → M that starts from p. For every t¯ ∈ Ip ,
γ ◦ τt¯ : Ip − t¯ → M is an integral curve that starts from γ p (t¯). Indeed, first of all, as t¯ ∈ Ip , then
0 = t¯ − t¯ ∈ Ip − t¯. Now,
It remains to prove that γ ◦ τt¯ is a maximal integral curve (it will then follow that Ip − t¯ = Iγ p (t¯) ). So,
let γ ′ : I ′ → M be another integral curve starting from γ p (t¯) and such that
γ ◦ τt¯ ≤ γ ′ , i.e. Ip − t¯ ⊆ I ′ .
In particular,
I ′ + t¯ ⊃ Ip − t¯ + t¯ = Ip ∋ 0,
110 Chapter 6. Flows and Symmetries
where we used that −t¯ = 0 − t¯ ∈ Ip − t¯. Summarizing, γ ′ ◦ τ−t¯ : I ′ + t¯ → M is an integral curve that
starts from p. As γ p is the maximal integral curve that starts from p, then γ ′ ◦ τ−t¯ ≤ γ p , i.e.
I ′ + t¯ ⊆ Ip ⇒ I ′ ⊆ Ip − t¯.
We conclude that I ′ = Ip − t¯, and γ ′ = γ p ◦ τt¯. This shows that γ p ◦ τt¯ is a maximal integral curve
and concludes the proof. ■
We conclude this section discussing the relationship between integral curves and smooth maps.
Proposition 6.1.4 Let M, N be smooth manifolds, let F : M → N be a smooth map, and let
X,Y be vector fields on M, N respectively. Then X and Y are F-related if and only if F maps
integral curves of X to integral curves of Y , i.e., for every integral curve γ : I → M of X, the
curve F ◦ γ : I → N is an integral curve of Y .
Proof. Assume that X and Y are F-related, and let γ : I → M be an integral curve of X. Then,
d d
◦ (F ◦ γ)∗ = ◦ γ ∗ ◦ F ∗ (the pull-back construction inverts the compositions)
dt dt
= γ∗ ◦ X ◦ F∗ (γ is an integral curve of X)
∗ ∗
= γ ◦ F ◦Y (X is F-related to Y )
∗
= (F ◦ γ) ◦Y (the pull-back construction inverts the compositions),
D := {(t, p) ∈ R × M : t ∈ Ip } ⊆ R × M. (6.2)
6.2 Flow of a Vector Field 111
"# (%)
M
!
We have some remarks on Definition 6.2.1. First of all, (6.2) guarantees that Φ is well-defined. As,
0 ∈ Ip for all p ∈ M, it is clear that D contains {0} × M ⊆ R × M. More precisely, for all p ∈ M,
D ∩ (R × {p}) = Ip × {p}.
In particular, if Ip = R for all p ∈ M, then D = R × M. However, D is usually a proper subset in
R × M (see Example 6.5, and Exercise 6.2).
■ Example 6.5 — A Non-Complete Vector Field. Consider the standard line R, with standard
d
coordinate x, the open submanifold (0, +∞) ⊆ R, and the coordinate vector field dx |(0,+∞) . For any
d
point x0 ∈ (0, +∞), the maximal integral curve γx0 of dx |(0,+∞) starting from x0 is
γx0 : Ix0 = (−x0 , +∞) → (0, +∞), t 7→ γx0 (t) := x0 + t
d
(see also Example 6.4). It follows that the flow domain of dx |(0,+∞) is
D = {(t, x0 ) ∈ R × (0, +∞) : −x0 < t} ,
which is a proper subset in R × (0, +∞). ■
Exercise 6.2 Consider the standard line R, with standard coordinate x, and the open submanifold
d
(−1, +1) ⊆ R. Compute the flow domain of the vector field dx |(−1,+1) . ■
The flow Φ can be encoded in a 1-parameter family of maps Φt , as follows. For any t ∈ R, define
Mt := {p ∈ M : (t, p) ∈ D} = {p ∈ M : t ∈ Ip } ⊆ M.
Notice that Mt might be the empty subset (can you provide an example?). In any case, there is a
well-defined map
Φt : Mt → M, p 7→ Φt (p) := Φ(t, p) = γ p (t),
and Φ is equivalently encoded in the family of the Φt , that we denote {Φt }t and we also call the
flow of X. The main properties of the flow are stated in the next
112 Chapter 6. Flows and Symmetries
(0, +∞)
()
'
−()
d
Figure 6.6: The flow domain of dx |(0,+∞) .
Proof (a sketch). Point (1) is a consequence of Theorem 6.1.1 and we omit the (technical) proof.
Point (2) is a consequence of the Translation Lemma (Lemma 6.1.3), as we now show. First
notice that the left hand side of Equation (6.3) is well-defined when
t + s ∈ Ip . (6.5)
Actually, from the second part of the Translation Lemma, we have that (6.4) implies (6.5). So, let
s,t, p be such that (6.4), and notice that
(Φs ◦ Φt )(p) = Φs (Φt (p)) = γΦt (p) (s) = γγ p (t) (s) = γ p (τt (s)) = γ p (s + t) = Φs+t (p),
where, in the fourth step, we used the Translation Lemma again (the maximal integral curve starting
from γ p (t) is obtained from the maximal integral curve starting from p by composition with the
translation τt ).
For Point (3), first notice that Mt is the preimage of the flow domain D under the map
as claimed. Finally, assume that Mt ̸= ∅. The map Φt : Mt → M can be seen as the composition
of Γt : Mt → D followed by the flow Φ. As both Φ and Γt are smooth, so is Φt : Mt → M, and its
restriction to M−t in the codomain. From the arbitrariness of t, Φ−t : M−t → M is also smooth, and,
from Point (2),
R The flow {Φt }t of a vector field X is sometimes called the local 1-parameter group of local
diffeomorphisms generated by X, and X is called the infinitesimal generator of {Φt }t . This
terminology is motivated by the Fundamental Theorem of Flows.
As a vector field X is completely determined by its maximal integral curves γ p , via Xp = γ̇ p (0),
and the flow of X determines the maximal integral curves via Φt (p) = γ p (t), it follows that a vector
field is completely determined by its flow, in the sense that two vector fields with the same flow
necessarily coincide.
Definition 6.2.2 A vector field X on a manifold M is complete if every maximal integral curve
of X is defined on the whole R, i.e. Ip = R for all p ∈ M. In other words, the flow of X is defined
on the whole R × M.
■ Example 6.6 — The Infinitesimal Generator of Translations. On R2 with standard coordinates
(x, y) consider the coordinate vector field
∂
X= .
∂x
It follows from Example 6.1 that the flow of X is
As, for every t, Φt is a translation, we also say that X generates the translations along the x-axis. ■
∂ ∂
X =x +y .
∂x ∂y
As, for every t, Φt is a dilation, we also say that X generates the dilations. ■
114 Chapter 6. Flows and Symmetries
∂ ∂
X =x −y .
∂y ∂x
It follows from Example 6.3 that the flow of X is
Φ : R × R2 → R2 , (t, (x, y)) 7→ Φ(t, (x, y)) = (x cost − y sint, x sint + y cost).
As, for every t, Φt is a rotation, we also say that X generates the rotations around the origin. ■
d
The vector fields from Examples 6.6, 6.7, and 6.8 are all complete. The vector field dx |(0,+∞)
from Example 6.5 is not complete.
Exercise 6.3 On R2 with standard coordinates (x, y), find the flow of the vector field
∂ ∂
X =x −y
∂x ∂y
(see also Exercise 6.1). Is X complete? ■
We conclude this section describing a new construction involving the flow: the Lie derivative
along a vector field. We begin with some informal remarks on vector fields and their flows. The
flow {Φt }t of a vector field X moves the points of a manifold. At the time 0 this displacement is
trivial: it is just the identity map that does not move any point. In any case, at an arbitrary time, this
displacement is a diffeomorphism. The vector field is the velocity of this displacement at the initial
time: informally
Φt = idM + t · X + O(t 2 ).
d
LX f : M → R, p 7→ (LX f )(p) := |t=0 Φt∗ ( f )(p). (6.6)
dt
This definition requires some explanations. First of all, notice that the expression Φt∗ ( f )(p) under
the t-derivative, does only make sense when p ∈ Mt , or, equivalently, t ∈ Ip . Indeed,
Additionally, (6.7) shows that Φt∗ ( f )(p) depends smoothly on t in Ip which is an open neighborhood
of 0. Hence, we can take the t-derivative of (6.7) at t = 0. This discussion shows that the Lie
derivative LX f is well-defined.
6.2 Flow of a Vector Field 115
LX f = X( f ). (6.8)
Proof. We have
d d
(LX f )(p) = |t=0 Φt∗ ( f )(p) = |t=0 f (γ p (t)) = γ̇ p (0)( f ) = Xp ( f ) = X( f )(p).
dt dt
This concludes the proof. ■
d
X( f ) = LX f = |t=0 Φt∗ ( f ).
dt
d ∗ d
Φt ( f )(p) = |s=t Φ∗s ( f )(p). (6.9)
dt ds
This is slightly more complicated. First of all, (6.9) does only make sense if the expression
Φ∗s ( f )(p) = f (γ p (s)) depends smoothly on s in an open neighborhood of t. This is the case when
t ∈ Ip or, equivalently, p ∈ Mt (do you see it?). In other words, the correspondence
d ∗ d ∗ d
Φ ( f ) : Mt → R, p 7→ Φt ( f )(p) = |s=t Φ∗s ( f )(p)
dt t dt ds
is a well-defined function.
Proposition 6.2.3 Let M be a manifold, let X ∈ X(M) be a vector field with flow {Φt }t , and let
f ∈ C∞ (M) be a smooth function. Then
d ∗
Φ ( f ) = Φt∗ (X( f )), (6.10)
dt t
d ∗
for all t such that Mt ̸= ∅. In particular, dt Φt ( f ) is a smooth function (on Mt ).
d d d d
|s=t Φ∗s ( f )(p) = ′ |s′ =0 f (Φs′ +t (p)) · |s=t (s − t) = ′ |s′ =0 f (Φs′ +t (p)).
ds ds ds ds
Now, we want to use the group property of the flow:
To do this, we have to make sure that t ∈ Ip (exactly what we are already assuming) and s′ ∈ IΦt (p) .
The latter condition can be safely assumed as we are computing the s′ -derivative at 0. So we can go
116 Chapter 6. Flows and Symmetries
R We can also compute dtd Φt∗ ( f ) following a different path. Namely we can use the group
property of the flow in the form:
Φs′ +t (p) = Φt (Φs′ (p)).
To do this, we have to make sure that s′ ∈ I p , which can be safely assumed as we are
computing the s′ -derivative at 0, and t ∈ IΦs′ (p) , i.e. Φs′ (p) = γ p (s′ ) ∈ Mt . The latter condition
can be guaranteed by taking s′ sufficiently close to 0, e.g. in I p ∩ γ p−1 (Mt ) (which is an open
neighborhood of 0). So we can go on with the computation:
d d
|s=t Φ∗s ( f )(p) = ′ |s′ =0 f (Φs′ +t (p))
ds ds
d
= ′ |s′ =0 f (Φt (Φs′ (p)))
ds
d
= ′ |s′ =0 Φ∗s′ (Φt∗ ( f ))(p).
ds
Exactly the same computation as for the Lie derivative (of a function) shows that
d
| ′ Φ∗′ (Φ∗ ( f ))(p) = X|Mt (Φt∗ ( f ))(p),
ds′ s =0 s t
which, adopting our customary notation, we also denote simply by X(Φt∗ ( f ))(p). Summariz-
ing, we proved that, for every t ∈ R such that Mt ̸= ∅,
d ∗
Φ ( f ) = X(Φt∗ ( f )). (6.12)
dt t
Finally, we discuss the Lie derivative of a vector field along a(n other) vector field. So, let M
be a manifold, and let X,Y be vector fields on M. The Lie derivative of Y along X is a new vector
field, denoted LX Y and defined, through its values, as follows:
d
(LX Y ) p := |t=0 Φt∗ (Y ) p . (6.13)
dt
This definition requires some explanations. First of all, what we actually mean by Φt∗ (Y ) is, more
precisely, Φt∗ (Y |M−t ): the pull-back of the vector field Y |M−t ∈ X(M−t ) along the diffeomorphism
Φt : Mt → M−t (if X is complete, then Mt = M and we don’t need this level of precision). In
particular, Φt∗ (Y ) is a vector field on Mt . Hence, the expression Φt∗ (Y ) p under the t-derivative, does
only make sense when p ∈ Mt , or, equivalently, t ∈ Ip , and, in this case, Φt∗ (Y ) p is a t-dependent
tangent vector in Tp M. We can only take the t-derivative if Φt∗ (Y ) p depends smoothly on t (at least
in an open neighborhood of 0). If this is the case, according to our interpretation of the velocity of
a curve in a finite dimensional real vector space (see the end of Section 3.2), the derivative in (6.13)
is again a vector in Tp M.
6.2 Flow of a Vector Field 117
Proposition 6.2.4 — Lie Derivative of a Vector Field. Let M be a manifold, and let X,Y ∈
X(M) be vector fields. Then the Lie derivative of Y along X is a well-defined vector field simply
given by
LX Y = [X,Y ]. (6.14)
Proof. First we need to show that the tangent vector Φt∗ (Y ) p depends smoothly on t in an open
neighborhood of 0 (so that we can take the t-derivative (6.13)). To do this, we work in local
coordinates as follows. Fix a point p0 ∈ M, and let (U, ϕ = (x1 , . . . , xn )) be a chart around p0 . Then
∂ ∂
X|U = X i and Y |U = Y i
∂ xi ∂ xi
for some smooth functions X i ,Y i ∈ C∞ (U). The preimage Φ−1 (U) of U under the flow Φ : D → M,
is an open subset of R × M containing (0, p0 ). Hence, there is an open interval J containing 0,
and a subchart (U0 , ϕ) of (U, ϕ) around p0 , such that J ×U0 ⊆ Φ−1 (U), i.e. Φ(J ×U0 ) ⊆ U. In
particular, we can consider the restriction
Φ : J ×U0 → U (6.15)
(yet in other words, for every t ∈ J, U0 ∈ Mt and Φt (U0 ) ⊆ U). As usual, we denote by Φi = Φ∗ (xi )
the components of (6.15). We will need the following identities:
∂ Φi
(0, p) = X i (p), (6.16)
∂t
and
∂ Φi
(0, p) = δki (6.17)
∂ xk
for all p ∈ U0 . To prove them, notice first that
Φi (t, p) = γ pi (t)
Hence
∂ Φi d d
(0, p) = |t=0 Φi (t, p) = |t=0 γ pi (t) = X i (p),
∂t dt dt
and
∂ Φi ∂ ∂
(0, p) = k | p Φi (0, −) = k | p xi = δki .
∂ xk ∂x ∂x
Now let (t, p) ∈ J ×U0 , and, using (5.5), compute
∂ Φi ∂
Φt∗ (Y ) p = dΦt (p) Φt−1 (YΦt (p) ) = dΦt (p) Φ−t (YΦt (p) ) = j
(−t, Φ(t, p))Y j (Φ(t, p)) i | p .
∂x ∂x
The coefficients in the last expression depend smoothly on t. This shows that we can take the
derivative in (6.13). It remains to compute
i
d ∗ d ∂Φ k ∂
(LX Y ) p = |t=0 Φt (Y ) p = |t=0 k
(−t, Φ(t, p))Y (Φ(t, p)) |p.
dt dt ∂t ∂ xi
118 Chapter 6. Flows and Symmetries
d
[X,Y ] = LX Y = |t=0 Φt∗ (Y ).
dt
It is also useful to compute the following derivative at an arbitrary time t:
d ∗ d
Φt (Y ) p = |s=t Φ∗s (Y ) p . (6.18)
dt ds
Similarly as for functions, (6.18) does only make sense when t ∈ Ip or, equivalently, p ∈ Mt , and
the correspondence
d ∗ d ∗ d
Φ (Y ) : Mt → T Mt , p 7→ Φt (Y ) p = |s=t Φ∗s (Y ) p
dt t dt ds
is a well-defined (a priori non-necessarily smooth) section of T Mt . One can show that it is actually
a smooth section, hence it corresponds to a vector field on Mt . To see this, let p ∈ Mt , and compute
d d d
|s=t Φ∗s (Y ) p = ′ |s′ =0 Φ∗s′ +t (Y ) p · |s=t (s − t)
ds ds ds
d
= ′ |s′ =0 Φt∗ (Φ∗s′ (Y )) p
ds
d
= ′ |s′ =0 dΦt (p) Φt−1 (Φ∗s′ (Y )Φt (p) ). (6.19)
ds
At this point we need a
Lemma 6.2.5 Let V,W be finite dimensional real vector spaces, let γ : I → V be a curve in V ,
and let A : V → W be a linear map. Then A ◦ γ : I → W is a smooth curve and
d d
|s=s0 (A ◦ γ)(s) = A |s=s0 γ(s)
ds ds
for all s0 ∈ I.
Proof. Let R = (e1 , . . . , en ) be a frame of V , n = dimV . Then γ(s) = γ i (s)ei for all s ∈ I, where
the γ i : I → R, s 7→ γ i (s) are smooth functions. If
j=1,...,n
Aij
i=1,...,n
6.3 Symmetries and Infinitesimal Symmetries 119
d d
|s=t Φ∗s (Y ) p = ′ |s′ =0 dΦt (p) Φt−1 (Φ∗s′ (Y )Φt (p) )
ds ds
−1 d ∗
= dΦt (p) Φt |s =0 Φs′ (Y )Φt (p)
′
ds′
= dΦt (p) Φt−1 (LX Y )Φt (p)
= Φt∗ ([X,Y ]) p .
d ∗
R As for functions, we can also compute dt Φt (Y ) following a different path. Namely,
d d d
|s=t Φ∗s (Y ) p = ′ |s′ =0 Φ∗s′ +t (Y ) p = ′ |s′ =0 Φ∗s′ (Φt∗ (Y )) = (LX Φt∗ (Y )) p = [X, Φt∗ (Y )] p ,
ds ds ds
for all p ∈ Mt . This shows that
d ∗
Φ (Y ) = [X, Φt∗ (Y )].
dt t
We leave to the reader to check that every single step in the above computation makes sense.
■ Example 6.9 In R2 with standard coordinates (x, y), consider a 1-dimensional subspace ℓ ⊆ R2 .
Let λ ∈ R be a non-zero real number, and consider the dilation
■ Example 6.10 In R2
with standard coordinates (x, y), consider the circle S1 ⊆ R2 . Let θ ∈ R,
and consider the rotation around the origin by an angle θ :
■ Example 6.11 In R2 with standard coordinates (x, y), consider a 1-dimensional subspace ℓ ⊆ R2
and the circle S1 ⊆ R2 . Consider also the vector fields
∂ ∂ ∂ ∂
X1 = x +y and X2 = x −y .
∂x ∂y ∂y ∂x
It follows from Examples 6.7 and 6.9 that X1 is an infinitesimal symmetry of ℓ. Similarly, it follows
from Examples 6.8 and 6.10 that X2 is an infinitesimal symmetry of S1 . ■
Proposition 6.3.2 Let M be a manifold, let S ⊆ M be a submanifold, and let X be a vector field
on M.
✓ If X is an infinitesimal symmetry of S, then X is tangent to S.
✓ Conversely, if X is tangent to S and, additionally, the restriction X|S is a complete vector
field, then X is an infinitesimal symmetry of S.
We now pass to symmetries of smooth functions, noticing that the group Diffeo(M) of dif-
feomorphisms of a manifold M acts on the algebra of smooth functions C∞ (M) (by algebra
isomorphisms) via the pull-back, i.e., for any diffeomorphism Φ : M → M, we have an algebra
isomorphism Φ∗ : C∞ (M) → C∞ (M). This suggests the following
Definition 6.3.3 A symmetry of a smooth function f ∈ C∞ (M) on a manifold M is a diffeo-
morphism Φ : M → M preserving f in the sense that Φ∗ ( f ) = f . A local symmetry of f
is a diffeomorphism Φ : U → Φ(U ) between open submanifolds U , Φ(U ) ⊆ M such that
Φ∗ ( f |Φ(U ) ) = f |U .
■ Example 6.12 On R2 with standard coordinates (x, y), consider a smooth function f ∈ C∞ (R2 )
depending on the sole y, i.e. there is a functions g ∈ C∞ (R) such that f (x, y) = g(y) for all (x, y) ∈ R2 .
Let x0 ∈ R, and consider the translation along the x-axis
Exercise 6.6 On R2 with standard coordinates (x, y), consider a smooth function f ∈ C∞ (R2 )
depending on the sole x2 + y2 , i.e. there is a functions g ∈ C∞ (R) such that f (x, y) = g(x2 + y2 )
for all (x, y) ∈ R2 . Let θ ∈ R, and consider the rotation around the origin by an angle θ :
Proposition 6.3.3 Let M be a manifold, and let f ∈ C∞ (M) be a smooth function on M. The
subset
■ Example 6.13 On R2 with standard coordinates (x, y), consider a smooth function f1 depending
on the sole y, and a smooth function f2 depending on the sole x2 + y2 . Consider also the vector
fields
∂ ∂ ∂
X1 = and X2 = x −y .
∂x ∂y ∂x
It follows from Examples 6.6 and 6.12 that X1 is an infinitesimal symmetry of f1 . Similarly, it
follows from Example 6.8 and Exercise 6.6 that X2 is an infinitesimal symmetry of f2 . ■
We postpone a little bit the proof of Proposition 6.3.5. First we propose an exercise and discuss
a consequence. Both should convince the reader of the relevance of characterizing infinitesimal
symmetries of smooth functions in a “flow-independent” way.
Exercise 6.9 On R2 with standard coordinates (x, y) consider a smooth function f ∈ C∞ (R2 )
and the vector field
∂f ∂ ∂f ∂
X= − .
∂x ∂y ∂y ∂x
Show that X is an infinitesimal symmetry of f . ■
Corollary 6.3.6 Let M be a manifold, and let f ∈ C∞ (M) be a smooth function of M. The
subset
Proof of Proposition 6.3.5. Let M, f , X be as in the statement, and assume first that X is an in-
finitesimal symmetry of f . This means that
Φt∗ ( f ) = f |Mt
d d
X( f )(p) = LX f (p) = |t=0 Φt∗ ( f )(p) = |t=0 f (p) = 0.
dt dt
6.3 Symmetries and Infinitesimal Symmetries 123
Conversely, let X( f ) = 0, and let t ∈ R be such that Mt ̸= ∅. Then, for every p ∈ Mt , we have
d ∗
Φ ( f )(p) = Φt∗ (X( f ))(p) = 0, (6.21)
dt t
where we used (6.10). Identity (6.21) now shows that
Φt∗ ( f )(p) = const = Φ∗0 ( f )(p) = f (p),
i.e. Φt∗ ( f ) = f |Mt . This concludes the proof. ■
Corollary 6.3.7 Let M be a manifold, let X ∈ X(M), and let f , g ∈ C∞ (M). If X is an in-
finitesimal symmetry of both f and g, then it is also an infinitesimal symmetry of f + g, and
f g.
■ Example 6.14 On R2 with standard coordinates (x, y), consider the vector field Y = ∂∂y . Let
x0 ∈ R, and consider the translation along the x-axis
Φ : R2 → R2 , (x, y) 7→ Φ(x, y) := (x + x0 , y).
Then Φ is a symmetry of Y . To see this, let’s compute Φ∗ (Y ). We have
∂ ∂
Φ∗ (Y ) = A +B ,
∂x ∂y
for some functions A, B ∈ C∞ (R2 ) given by
A = Φ∗ (Y ) (x) = Φ∗ (Y ((Φ−1 )∗ (x))), and B = Φ∗ (Y ) (y) = Φ∗ (Y ((Φ−1 )∗ (y)))
Now Φ−1 : R2 → R2 is given by
Φ−1 : R2 → R2 , (x, y) 7→ Φ−1 (x, y) := (x − x0 , y),
and (Φ−1 )∗ (x) is the first component of Φ−1 , i.e. x − x0 . Hence
∗ −1 ∗ ∗ ∗ ∂ (x − x0 )
A = Φ (Y ((Φ ) (x))) = Φ (Y (x − x0 )) = Φ = Φ∗ (0) = 0.
∂y
Similarly, (Φ−1 )∗ (y) is the second component of Φ−1 (y), and
∗ ∗ −1 ∗ ∗ ∗ ∂y
B = Φ (Y ) (y) = Φ (Y ((Φ ) (y))) = Φ (Y (y)) = Φ = Φ∗ (1) = 1.
∂y
We conclude that
∂ ∂ ∂
Φ∗ (Y ) = 0 · +1· = = Y,
∂x ∂y ∂y
i.e. Φ is a symmetry of Y as claimed. ■
124 Chapter 6. Flows and Symmetries
Exercise 6.12 On R2 with standard coordinates (x, y), consider the vector field
∂ ∂
Y =x +y .
∂x ∂y
Additionally, let θ ∈ R, and consider the rotation around the origin of an angle θ :
Proposition 6.3.8 Let M be a manifold, and let Y ∈ X(M) be a vector field on M. The subset
Corollary 6.3.9 Let M be a manifold, let Y, Z ∈ X(M), let f ∈ C∞ (M) and let Φ : M → M be a
diffeomorphism. If Φ is a symmetry of Y , Z and f , then it is also a symmetry of Y + Z, Y ( f ),
fY and [Y, Z].
Beware that the two vector fields X,Y in Definition 6.3.6 play completely different roles.
■ Example 6.15 On R2 with standard coordinates (x, y), consider the vector fields
∂ ∂ ∂
Y1 = and Y2 = x +y ,
∂y ∂x ∂y
and the vector fields
∂ ∂ ∂
X1 = and X2 = x −y .
∂x ∂y ∂x
It follows from Examples 6.6 and 6.14 that X1 is an infinitesimal symmetry of Y1 . Similarly, it
follows from Example 6.8 and Exercise 6.12 that X2 is an infinitesimal symmetry of Y2 . ■
Similarly as for functions, checking whether or not a vector field X is an infinitesimal symmetry
of an other vector field Y using the definition might be problematic, but as for functions, infinitesimal
symmetries of a vector field can be characterized in a rather efficient way.
6.3 Symmetries and Infinitesimal Symmetries 125
Proposition 6.3.10 Let M be a smooth manifold, and let Y, X ∈ X(M) be vector fields. Then X
is an infinitesimal symmetry of Y if and only if [Y, X] = 0.
Proof. The proof is very similar to that of Proposition 6.3.5 and we leave it as Exercise 6.15. ■
Corollary 6.3.11 Let M be a smooth manifold, let Y, X ∈ X(M) be vector fields. Then X is an
infinitesimal symmetry of Y if and only if Y is an infinitesimal symmetry of X.
Proof. Obvious. ■
Corollary 6.3.12 Let M be a manifold, and let Y ∈ X(M) be a vector field on M. The subset
The following infinitesimal version of Corollary 6.3.9 is a corollary of either Corollary 6.3.9
itself, or (in the part about the commutator) Proposition 6.3.10 and the Jacobi identity for the
commutator of vector fields.
Corollary 6.3.13 Let M be a manifold, let X,Y, Z ∈ X(M), and let f ∈ C∞ (M). If X is an
infinitesimal symmetry of both Y , Z and f , then it is also an infinitesimal symmetry of Y + Z,
Y ( f ), fY , and [Y, Z].
We conclude this chapter discussing the flows of two commuting vector fields, i.e. two vector
fields X,Y such that their commutator vanishes: [X,Y ] = 0. Let M be a manifold, and let X,Y ∈
X(M) be two vector fields on M. Denote by {Φt }t and {Ψs }s the flows of X and Y , respectively.
We want to show that X and Y commute if and only if their flows commute. First we have to explain
what does it mean that {Φt }t and {Ψs }s commute. We will only discuss in details the case when
both Φ and Ψ are defined on the whole R × M, however, we report a definition in the general case
for completeness.
Definition 6.3.7 The flows {Φt }t and {Ψs }s commute if, for every p ∈ M, we have that,
whenever I, J ⊆ R are open intervals containing 0 such that one of the expressions
makes sense for all (t, s) ∈ I × J, then the other one also makes sense and they are equal.
126 Chapter 6. Flows and Symmetries
Proposition 6.3.14 — Commuting Vector Fields. The vector fields X and Y commute if and
only if so do their flows {Φt }t and {Ψs }s .
Proof. We provide a proof only in the case when both X and Y are complete. We begin remarking
that, in this case, Definition 6.3.7 boils down to the condition
Assume first that X and Y commute. Then, from Proposition 6.3.10, X is an infinitesimal
symmetry of Y , i.e. Φt is a symmetry of Y for all t ∈ R. It now follows from Proposition 6.1.4
that Φt maps integral curves of Y to integral curves. In particular, for all p ∈ M, the maximal
integral curve of Y starting from p, say γ p : R → M, is mapped to an integral curve Φt ◦ γ p : R → M.
Notice that, being defined on the whole R, not only Φt ◦ γ p is an integral curve, but it is also
maximal. As it starts from Φt (γ p (0)) = Φt (p), it must coincide with the maximal integral curve
γΦt (p) , i.e. Φt ◦ γ p = γΦt (p) : R → M. In other words
This conclude the proof of the “only if” part of the statement.
For the if part, suppose that
Then, similarly as in the proof of Proposition 6.1.4, for all p ∈ M, and all t ∈ R,
d
d p Φt (Yp ) = d p Φt |s=0 Ψs (p) (s 7→ Ψs (p) is an integral curve of Y )
ds
d
= |s=0 Φt (Ψs (p)) (tangent map applied to the velocity of a curve)
ds
d
= |s=0 Ψs (Φt (p)) ({Φt }t and {Ψs }s commute)
ds
= YΦt (p) (s 7→ Ψs (Φt (p)) is an integral curve of Y ).
This shows that Φt is a symmetry of Y for all t, hence X is an infinitesimal symmetry of Y and,
from Proposition 6.3.10, X and Y commute. ■
7. Covectors and Differential 1-Forms
In this chapter, we introduce (tangent) covectors and differential 1-forms. Covectors and tangent
vectors are dual objects. Similarly, differential 1-forms and vector fields are dual objects. Covectors
are the points of a new manifold, the cotangent bundle, and differential 1-forms are equivalent to
sections of the cotangent bundle, in a very similar way as for tangent vectors and vector fields.
However, there are important differences between vector fields and differential forms. Among
the most relevant, unlike vector fields, differential forms on a manifold can be transformed into
differential forms on another manifold along any smooth map, via the pull-back construction.
of the tangent space. Vectors in Tp∗ M are called tangent covectors, or cotangent vectors, or,
simply, covectors. In other words, covectors are R-linear maps θ : Tp M → R. The frame of
Tp∗ M dual to the coordinate frame
∂ ∂
1
|p, . . . , n |p
∂x ∂x
is denoted
d p x1 , . . . , d p xn
Similarly as for the coordinate frame, if we change coordinates around p, the coordinate
coframe changes. So, let (Ũ, ϕ̃ = (x̃1 , . . . , x̃n )) be another chart on M around p. Denote by t 1 , . . . ,t n
the standard coordinates on ϕ(U), and by t˜1 , . . . , t˜n the standard coordinates on ϕ̃(Ũ). Finally let
P = ϕ(p). The transition matrix
MR ∗ ,R̃ ∗ ∈ GL(n, R)
is the transpose of the transition matrix MR̃,R between the coordinate frames
∂ ∂ ∂ ∂
R= 1
,..., n and R̃ = 1
,..., n .
∂x ∂x ∂ x̃ ∂ x̃
In other words
∂ x̃i ∂ (ϕ̃ ◦ ϕ −1 )i
d p x̃i = (p)d p x j
= (P)d p x j ,
∂xj ∂t j
for all i = 1, . . . , n.
The following notion of differential of a smooth function f at a point p encodes partial
derivatives of f at p in a coordinate-free manner.
Definition 7.1.2 — Differential at a Point. The differential at the point p is the map:
d p : C∞ (M) → Tp∗ M, f 7→ d p f
defined by
The differential d p is well-defined in the sense that d p f is a covector for every f ∈ C∞ (M).
Indeed, let v, w ∈ Tp M, and let a, b ∈ R. Then
dp f = dpg
Let f ∈ C∞ (M), let p ∈ M, and let (U, ϕ = (x1 , . . . , xn )) be a chart on M around p. We now
compute the components of d p f in the coordinate coframe. If d p f = θi d p xi , then
∂ ∂ ∂f
θi = d p f i
|p = | p f = i (p),
∂x ∂ xi ∂x
hence
∂f
dp f = (p)d p xi , (7.1)
∂ xi
showing that d p f encodes, in a coordinate-free way, partial derivatives of f , as announced. It
follows from Equation (7.1) that the elements of the coordinate coframe are exactly the differentials
at the point p of the coordinate functions, motivating the notation d p xi .
Proposition 7.1.2 Let M be a manifold and let p ∈ M be a point. For any covector θ ∈ Tp∗ M,
there exists a (non-unique) smooth function f such that θ = d p f .
Proof. Choose a chart (U, ϕ = (x1 , . . . , xn )) around p, and let θ ∈ Tp∗ M. There exist real numbers θi
such that θ = θi d p xi . Consider the smooth function g = θi xi ∈ C∞ (U). In view of the Local Exten-
sion Lemma, there exists a smooth function f ∈ C∞ (M) agreeing with g in an open neighborhood
of p, and we have
∂g
dp f = dpg = (p)d p xi = θi d p xi = θ .
∂ xi
■
Similarly as tangent vectors, covectors on a manifold M can be organized into a manifold called
the cotangent bundle to M, and denoted T ∗ M. Let’s construct T ∗ M. First of all, we put
T ∗ M :=
∏ T ∗ M = (p, θ ) : p ∈ M and θ ∈ T ∗ M .
p p
p∈M
A point (p, θ ) in T ∗ M will be sometimes simply denoted by θ . The set T ∗ M comes with a
natural surjection π : T ∗ M → M, (p, θ ) 7→ p, and the pair (T ∗ M, π) is the cotangent bundle to M
130 Chapter 7. Covectors and Differential 1-Forms
(sometimes T ∗ M itself is referred to as the cotangent bundle). The smooth structure on M induces
an atlas on T ∗ M as follows. Begin with a chart (U, ϕ = (x1 , . . . , xn )) on M, and define a chart
(T ∗U, T ∗ ϕ) on T ∗ M as follows. First of all, put
T ∗U := π −1 (U) =
∏ T ∗ M = {(p, θ ) ∈ T ∗ M : p ∈ U}.
p
p∈U
Notice that the last n entries of T ∗ ϕ(p, θ ) are the components of θ in the coordinate coframe
d p x1 , . . . , d p xn .
Clearly, (T ∗U, T ∗ ϕ) is a 2n-dimensional chart on T ∗ M. Every such chart is called a standard chart.
The first n components of the coordinate map T ∗ ϕ will be again denoted by (x1 , . . . , xn ), and the
last n components will be denoted by (p1 , . . . , pn ).
We now show that any two standard charts on T ∗ M are compatible. So let (U, ϕ = (x1 , . . . , xn ))
and (Ũ, ϕ̃ = (x̃1 , . . . , x̃n )) be charts on M, and let (T ∗U, T ∗ ϕ), (T ∗Ũ, T ∗ ϕ̃) be the associated
standard charts on T ∗ M. If U ∩ Ũ = ∅, then T ∗U ∩ T ∗Ũ = ∅ and there is nothing else to prove. If
U ∩ Ũ ̸= ∅, then T ∗U ∩ T ∗Ũ = T ∗ (U ∩ Ũ), and we can consider the transition map
Now,
is an open subset and similarly for T ∗ ϕ(T ∗U ∩ T ∗Ũ). Additionally, take a point (P̃; a1 , . . . , an ) ∈
(U ∩ Ũ) × Rn , and compute
(T ∗ ϕ ◦ (T ∗ ϕ̃)−1 )(P̃; a1 , . . . , an )
= T ∗ ϕ ϕ̃ −1 (P̃), ai dϕ −1 (P̃) x̃i
−1 i ∂ i ∂
= (ϕ ◦ ϕ̃ )(P̃); ai dϕ −1 (P̃) x̃ | −1 , . . . , ai dϕ −1 (P̃) x̃ | −1
∂ x1 ϕ̃ (P̃) ∂ xn ϕ̃ (P̃)
∂ x̃i −1 ∂ x̃i −1
−1
= (ϕ ◦ ϕ̃ )(P̃); ai 1 (ϕ (P̃)), . . . , ai n (ϕ (P̃)) .
∂x ∂x
As (U, ϕ), (Ũ, ϕ̃) are compatible, the first n entries depend smoothly on P̃. The last n entries
depend smoothly on P̃, and linearly, hence smoothly, on a1 , . . . , an . This shows that T ∗ ϕ ◦ (T ∗ ϕ̃)−1
is smooth. By the same argument, its inverse is also smooth, so they are both diffeomorphisms
and the charts (T ∗U, T ∗ ϕ), (T ∗Ũ, T ∗ ϕ̃) are compatible. Finally, standard charts cover T ∗ M, so
they form an atlas called the standard atlas. The standard atlas induces a Hausdorff and second
countable topology on T ∗ M, and this can be proved exactly as in the case of the tangent bundle. So
T ∗ M is a manifold, and one can also show that π : T ∗ M → M is a smooth map. We leave details to
the readers. Summarizing, we have proved the following
7.2 Differential 1-Forms and Fields of Covectors 131
θ : X(M) → C∞ (M).
formed by coordinate vector fields. Accordingly, the dual module Ω1 (U) does also possess a finite
frame, the dual frame, denoted by
dx1 , . . . , dxn
(7.2)
∂
X = Xi
∂ xi
is a vector field on U, we have
θ = θi dxi ,
If (U, ϕ̃ = (x̃1 , . . . , x̃n )) is another chart with the same coordinate domain, then θ can be also written
as
θ = θ̃i d x̃i
and
∂ x̃ j ∂ ∂ x̃ j ∂ x̃ j
∂ ∂
θi = θ =θ = θ = θ̃ j .
∂ xi ∂ xi ∂ x̃ j ∂ xi ∂ x̃i ∂ xi
■
d : C∞ (M) → Ω1 (M), f 7→ d f
defined by
The differential d is well-defined in the sense that d f is a 1-form for every f ∈ C∞ (M). Check it as
an exercise!
Proposition 7.2.1 — Properties of the Differential. The differential d : C∞ (M) → Ω1 (M)
enjoys the following properties:
✓ d is R-linear;
✓ d satisfies the following Leibniz rule:
Let (U, ϕ = (x1 , . . . , xn )) be a chart on M, and let f ∈ C∞ (U). We want to compute the
components of d f ∈ Ω1 (U) in the coordinate coframe. If d f = θi dxi , then
∂ ∂f
θi = d f i
= i,
∂x ∂x
hence
∂f i
df = dx .
∂ xi
7.2 Differential 1-Forms and Fields of Covectors 133
So d f encodes the gradient of f in a coordinate free manner. In particular the differential of the i-th
coordinate function is exactly the i-th coordinate 1-form, motivating the notation dxi .
In the second part of this section we show that differential 1-forms are equivalent to sections of
the cotangent bundle. More precisely, there is a canonical C∞ (M)-module isomorphism Ω1 (M) ∼ =
Γ(T ∗ M). We begin proving that a differential 1-form θ ∈ Ω1 (M) determines a covector θ p ∈ Tp∗ M
for every point p ∈ M. In order to define θ p we need two lemmas.
Lemma 7.2.2 Let θ ∈ Ω1 (M), let p0 ∈ M, and let X ∈ X(M). Then the value
θ (X)(p0 )
of the function θ (X) ∈ C∞ (M) at p0 , does only depend on (θ and) the value Xp0 of the vector
field X at p0 .
Step I: Differential 1-Forms are Local Operators For every open subset U ⊆ M, the func-
tion θ (X)|U does only depend on (θ and) the restriction X|U . To see this, let X ′ be another vector
field on M such that X ′ |U = X|U , let p ∈ U , and let β ∈ C∞ (M) be a bump function relative to the
data (U , p). In other words, β = 1 in an open neighborhood of p, and supp β ⊆ U , in particular,
β = 0 outside U . Put η = 1 − β . Then
X − X ′ = η(X − X ′ ),
hence
so that
θ (X)|U = θ (X ′ )|U .
Step II: Local Extension Lemma for Vector Fields Let S ⊆ M be a submanifold, and let X
be a vector field on S. For every point p0 in S, there exists an open neighborhood U ⊆ M of p0
e U is tangent to S ∩U, and X|
and a vector field Xe on M such that X| e U |S∩U = X|S∩U . We prove this
claim in the case when S = U is an open submanifold, and leave the general case as an exercise for
the reader (the proof is not very different from that of the Local Extension Lemma for functions).
So, let U ⊆ M be an open submanifold, let p0 ∈ U , and let X ∈ X(U ). Choose a bump function
β ∈ C∞ (M) relative to the data (U , p0 ), and consider the vector field Xe defined (thorugh its values)
as follows:
β (p)Xp if p ∈ U
Xp := .
/U
e
0 if p ∈
In view of the Gluing Lemma for Vector Fields, Xe is a well-defined vector field on M. Additionally,
Xe agrees with X on an open neighborhood of p0 (specifically, the open neighborhood where β = 1).
This concludes the proof of the Local Extension Lemma (for S = U an open submanifold).
134 Chapter 7. Covectors and Differential 1-Forms
Step III: Vector Fields Vanishing at a Point Let X be a vector field vanishing at a point p,
i.e. Xp = 0, then there exist vector fields X1 , . . . , Xn , and smooth functions f1 , . . . , fn such that: 1)
the fi all vanish at p, i.e. fi (p) = 0, and 2)
X = f1 X1 + · · · + fn Xn
in an open neighborhood of p. To see this, consider (U, ϕ = (x1 , . . . , xn )) a chart around p. Then
∂
X|U = gi
∂ xi
∂
0 = Xp = gi (p) |p.
∂ xi
Hence, gi (p) = 0 for all i = 1, . . . , n. Now, let f1 , . . . , fn ∈ C∞ (M) be smooth functions agreeing with
g1 , . . . , gn in an open neighborhood of p. In particular, fi (p) = 0, for all i = 1, . . . , n. Finally, use
Step II to find vector fields X1 , . . . , Xn ∈ X(M) agreeing with ∂∂x1 , . . . , ∂∂xn in an open neighborhood
of p, and notice that
X = f1 X1 + · · · + fn Xn
Step IV: Conclusion Let p0 ∈ M be a point, and let X, X ′ ∈ X(M) be vector fields such that
Xp0 = Xp′ 0 . We want to show that θ (X)(p0 ) = θ (X ′ )(p0 ). To do this notice that
0 = Xp0 − Xp′ 0 = (X − X ′ ) p0 .
Hence, it follows from Step III, that there are functions f1 , . . . , fn ∈ C∞ (M) and vector fields
X1 , . . . , Xn ∈ X(M) such that fi (p0 ) = 0, and
X − X ′ = f1 X1 + · · · + fn Xn
θ (X − X ′ ) = θ ( f1 X1 + · · · + fn Xn ) = f1 θ (X1 ) + · · · + fn θ (Xn )
θ (X −X ′ )(p0 ) = ( f1 θ (X1 ) + · · · + fn θ (Xn )) (p0 ) = f1 (p0 )θ (X1 )(p0 )+· · ·+ fn (p0 )θ (Xn )(p0 ) = 0.
Lemma 7.2.3 Let p ∈ M, and let v ∈ Tp M. Then there exists a vector field X ∈ X(M) such that
Xp = v.
7.2 Differential 1-Forms and Fields of Covectors 135
Xp = Yp = v.
We are now ready to present the main result of this section. Begin with a differential 1-form θ
on a manifold M. For any point p ∈ M we want to define a covector θ p ∈ Tp∗ M. So, let v ∈ Tp M,
and let X ∈ X(M) be any vector field such that Xp = v. The vector field X exists in view of Lemma
7.2.3. We put
θ p (v) := θ (X)(p).
From Lemma 7.2.2, θ p (v) does only depend on (θ and) v. Hence, θ p is a well-defined map
θ p : Tp M → R.
It is easy to see that θ p is a covector. Indeed, let v, w ∈ Tp M, let a, b ∈ R, and let X,Y ∈ X(M) be
vector fields such that Xp = v and Yp = w. Then, from Exercise 5.5,
and
θ p (av + bw) = θ (aX + bY )(p) = (aθ (X) + bθ (Y )) (p)
= aθ (X)(p) + bθ (Y )(p) = aθ p (v) + bθ p (w).
The covector θ p ∈ Tp∗ M is called the value at p of the 1-form θ .
Exercise 7.3 Prove that, for every θ , κ ∈ Ω1 (M), every f ∈ C∞ (M), and every p ∈ M,
(θ + κ) p = θ p + κ p ,
( f θ ) p = f (p)θ p ,
(d f ) p = d p f . (7.3)
Additionally, let (U, ϕ = (x1 , . . . , xn )) be a chart around p. Prove that the value of the coordinate
1-form dxi at p is exactly the coordinate covector d p xi . ■
sθ : M → T ∗ M, p 7→ sθ (p) := θ p .
136 Chapter 7. Covectors and Differential 1-Forms
Proof. First we prove that for every θ ∈ Ω1 (M), the section sθ is actually smooth. As for vector
fields, it is enough to check that, for every chart (U, ϕ = (x1 , . . . , xn )) on M, the components (sθ )i
of sθ in the chart (U, ϕ) (see the end of Section 7.1) are smooth functions on U. To do this, we fix
p0 ∈ U and show that (sθ )i is smooth around p0 . So, let p ∈ U and compute
∂
(sθ )i (p) = pi (sθ (p)) = pi (θ p ) = i-th component of θ p in the coordinate coframe = θ p |p .
∂ xi
To continue the computation, we need to choose a vector field Xi ∈ X(M) such that (Xi ) p = ∂∂xi | p .
It is convenient to choose Xi in a more precise way. Namely, we choose Xi such that it agrees with
∂
∂ xi
in an open neighborhood V ⊆ U of p0 . Such vector field exists in view of the Local Extension
Lemma for Vector Fields, and we have
∂
(sθ )i (p) = θ p | p = θ (Xi )(p)
∂ xi
for all s ∈ Γ(T ∗ M), and all θ ∈ Ω1 (M). For every X ∈ X(M) we define
∂
X|U = X i
∂ xi
for some smooth functions X i on U, and, for p ∈ U, we have
∂
θs (X)(p) = s(p)(Xp ) = s(p) X (p) i | p = X i (p)si (p) = (X i si )(p).
i
∂x
As the si are smooth, the last expression depends smoothly on p. This concludes the proof. ■
7.2 Differential 1-Forms and Fields of Covectors 137
As for vector fields, Proposition 7.2.4 has some important consequences. First of all, a 1-form
θ on M can be “restricted” to an open submanifold U ⊆ M by restricting the associated section,
exactly as for vector fields (we leave details to the reader). We denote by θ |U ∈ Ω1 (U ) the
restricted 1-form. Clearly we have θ |U (XU ) = θ (X)|U for all θ ∈ Ω1 (M) and all X ∈ X(M).
A second consequence of Proposition 7.2.4 is a Gluing Lemma for Differential 1-Forms. The
statement and the proof are formally identical to that of the Gluing Lemma for Vector Fields.
Finally, we can combine the Gluing Lemma for Differential 1-Forms with Example 7.1 to find that,
for every 1-form θ on M,
(1) θ is completely determined by its restrictions θ |U to some coordinate domains U covering
M,
(2) for every chart (U, ϕ = (x1 , . . . , xn )), the restriction θ |U can be uniquely written in the form
θ |U = θi dxi
for some smooth functions θi ∈ C∞ (U), called the components of θ in the chart (U, ϕ).
(3) given a vector field X ∈ X(M) locally given by
∂
X|U = X i ,
∂ xi
we have
θ (X)|U = θ |U (X|U ) = θi X i .
The proof of Proposition 7.2.4 shows that the components θi of a 1-form in the chart (U, ϕ =
(x1 , . . . , xn ))
coincide with the components of the corresponding section. When the chart (U, ϕ)
changes, the components of θ change. The transformation rule has been already discussed in
Example 7.1.
Exercise 7.5 Consider R2 with standard coordinates x, y, and the polar chart (U, ψ = (r, φ )) on
it. Recall that U = R2 ∖ {y = 0, x ≥ 0}, U
b = (0, +∞) × (0, 2π) and
p
ψ : U → U,b (x, y) 7→ ψ(x, y) = (r, φ ) := x2 + y2 , φ (x, y) ,
with
arccos √ x
if y ≥ 0
x2 +y2
φ (x, y) := .
− arccos √x if y < 0
x2 +y2
ψ −1 : U
b → U, (r, φ ) 7→ ψ −1 (r, φ ) = (r cos φ , r sin φ ).
Prove that
Let (U, ϕ = (x1 , . . . , xn )) be a chart on M. Example 7.1 shows not only that the C∞ (U)-module
Ω1 (U) possesses a finite frame, but also that it is generated by the differentials dx1 , . . . , dxn of the
coordinate functions x1 , . . . , xn . In particular, it is generated by 1-forms of the type d f for some
f ∈ C∞ (U). The last fact is true even globally according to the following
θ = f1 dg1 + · · · + fk dgk
Proof. The proof is non-trivial, in the sense that it is based on more involved results on smooth
manifolds, and we omit it. ■
d ∗p iS : Tp∗ M → Tp∗ S
to the tangent map to the inclusion iS : S ,→ M is a surjection consisting in restricting a covector
θ : Tp M → R on M at p ∈ S to tangent vectors to S. ■
Proposition 7.3.1 — Naturality of the Differential at a Point. For every smooth function
f ∈ C∞ (N) we have
d p (F ∗ ( f )) = d ∗p F dF(p) f .
■
7.3 Differential 1-Forms and Smooth Maps 139
Now let (U, ϕ = (x1 , . . . , xm )) be a chart on M around p, and let (V, ψ = (y1 , . . . , yn )) be a chart
on N around F(p) such that F(U) ⊆ V . Put P = ϕ(p). The representative matrix of the dual map
d ∗p F : TF(p)
∗ N → T ∗ M in the coordinate coframes
p
dF(p) y1 , . . . , dF(p) yn d p x1 , . . . , d p xm
and
is the transpose of the representative matrix of the tangent map d p F : Tp M → TF(p) N in the
coordinate frames
∂ ∂ ∂ ∂
|p, . . . , m |p and | , . . . n |F(p) .
∂ x1 ∂x ∂ y1 F(p) ∂y
Hence it is
i=1,...,m !i=1,...,m
∂ Fa ∂ Fba
(p) = (P)
∂ xi a=1,...,n ∂t i
a=1,...,n
∂ Fa
d ∗p F(dF(p) ya ) = (p)d p xi = d p F a = d p F ∗ (ya ),
∂ xi
for all a ∈ {1, . . . , n}, which is duly consistent with Proposition 7.3.1.
We now come to the main result of this section.
Proposition 7.3.2 — Pull-Back of 1-Forms. Let M, N be smooth manifolds and let F : M → N
be a smooth map. Then there exists a unique map
F ∗ : Ω1 (N) → Ω1 (M),
called the pull-back of differential 1-forms along F such that, for all θ , κ ∈ Ω1 (N), and all
f ∈ C∞ (N),
(1) F ∗ (θ + κ) = F ∗ (θ ) + F ∗ (κ),
(2) F ∗ ( f θ ) = F ∗ ( f )F ∗ (θ ), and
(3) F ∗ (d f ) = dF ∗ ( f ).
Proof. Let θ ∈ Ω1 (N). We define F ∗ (θ ) ∈ Ω1 (M) through its values. So, for every p ∈ M, put
F ∗ (θ ) p = d ∗p F(θF(p) ).
θ |V = θa dya ,
140 Chapter 7. Covectors and Differential 1-Forms
F ∗ (d f ) p = d ∗p F((d f )F(p) )
= d ∗p F(dF(p) f ) (Equation (7.3) in Exercise 7.3)
∗
= d p (F ( f )) (Proposition 7.3.1)
∗
= (dF ( f )) p . (Equation (7.3) in Exercise 7.3)
θ = f1 dg1 + · · · + fk gk .
Hence
′
F ∗ (θ ) = ′ F ∗ ( f1 dg1 + · · · + fk gk )
= F ∗ ( f1 )′ F ∗ (dg1 ) + · · · + F ∗ ( fk )′ F ∗ (dgk )
= F ∗ ( f1 )dF ∗ (g1 ) + · · · + F ∗ ( fk )dF ∗ (gk )
= F ∗ (θ ).
Exercise 7.6 Complete the proof of Proposition 7.3.2 showing that the map F ∗ : Ω1 (N) →
Ω1 (M) enjoys Properties (1) and (2) in the statement. ■
Let F : M → N be a smooth map, and let (U, ϕ) and (V, ψ) be charts as in the proof of
Proposition 7.3.2. It immediately follows from that proof that
∂ Fa i
F ∗ (θ )|U = F ∗ (θ |V ) = F ∗ (θa ) dx .
∂ xi
7.3 Differential 1-Forms and Smooth Maps 141
■ Example 7.3 Consider R2 with standard coordinates (u, v), and R3 with standard coordinates
(x, y, z). Consider also the smooth map
We want to compute the pull-back along F of the 1-form θ = ydx − xdy ∈ Ω1 (R3 ):
an open submanifold, and θ ∈ Ω1 (M) is a 1-form, then i∗U (θ ) = θ |U is the (usual) restriction of θ
to U .
More generally, we can “restrict” a 1-form θ ∈ Ω1 (M) to any submanifold S ⊆ M by pulling it
back along the inclusion iS : S ,→ M. The pull-back i∗S (θ ) is sometimes denoted by θ |S and called
the restriction of θ to S. In practice, it is obtained by restricting the values of θ at points of S to
tangent vectors to S.
For instance, let M be R3 with standard coordinates (x, y, z), let S = S2 ⊆ R3 be the 2-
dimensional sphere, and let
We want to show that θ |S2 = 0. We do this in two different ways. First of all let P = (a, b, c) ∈ S2
and let
∂ ∂ ∂
v=A |P + B |P +C |P ∈ TP S2 .
∂x ∂y ∂z
This means that
aA + bB + cC = 0.
Hence
∂ ∂ ∂
θP (v) = (adP x + bdP y + cdP z) A |P + B |P +C |P = aA + bB + cC = 0.
∂x ∂y ∂z
Exercise 7.7 Consider R2 with standard coordinates (x, y), and let (U+ , ϕ+ ) be the stereographic
chart (from the north) on the circle S1 ⊆ R2 . Consider the 1-form
θ = ydx − xdy.
142 Chapter 7. Covectors and Differential 1-Forms
In this chapter we define higher degree differential forms. Differential forms are of great importance
in modern Differential Geometry. Numerous interesting geometric structures (and physical systems)
can be formalized via differential forms, and they play a key role in Algebraic Topology: the branch
of Mathematics that uses algebraic constructions to study topological properties of spaces (in our
case, smooth manifolds and related geometric objects). Additionally, differential forms serve as
integrands in a coordinate free integration theory (not discussed in these notes). The basic rules of
the calculus with vector fields and differential forms are collectively called Cartan calculus, after
the French mathematician Èlie Cartan (in the head picture of the chapter), who first defined higher
degree differential forms as we know them in modern Differential Geometry.
In this chapter we will skip, or just sketch, most of the proofs.
ω : |M × ·{z
· · × M} → A
k times
such that
✓ ω is A-multilinear,
✓ ω is alternating,
i.e. ω(µ1 , . . . , µk ) = 0 whenever two arguments among µ1 , . . . , µk ∈ M coincide. When k = 0
we interpret this definition requiring ω to be simply an element in A.
where we denoted by (−)σ the sign of the permutation σ . As we are working over a field of
0 characteristic, the converse is also true, i.e. skewsymmetric forms are alternating.
n
det : R · · × R}n → R,
| × ·{z (A1 , . . . , An ) 7→ det(A1 · · · An )
n times
is an n-form on Rn . ■
(aω)(µ1 , . . . , µk ) = aω(µ1 , . . . , µk ),
for all µ1 , . . . , µk ∈ M . When k is a negative integer, we put Altk (M , A) = 0: the zero module. In
other words, by definition, negative degree alternating forms vanish.
There are more operations on alternating forms. In order to define them, we need one
more
Definition 8.1.2 For any two non-negative integers k, l, a (k, l)-unshuffle, or simply an unshuffle,
is a permutation σ ∈ Sk+l such that
Now, for every k, l there is a wedge product (also called exterior product):
(ω ∧ ρ)(µ1 , . . . , µk+l ) = ∑ (−)σ ω(µσ (1) , . . . , µσ (k) )ρ(µσ (k+1) , . . . , µσ (k+l) ), (8.2)
σ ∈Sk,l
where, as indicated, the sum runs over the (k, l)-unshuffles. When k = 0, ω = a is just an element
in A, and we interpret Formula (8.2) by putting ω ∧ ρ = aρ. Similarly, when l = 0, ρ = b ∈ A, and
we put ω ∧ ρ = bω. Finally, when either k or l is negative, then either ω or ρ vanish and we put
ω ∧ ρ = 0.
It is not too hard to see that ω ∧ ρ is indeed a well-defined (k + l)-form: the A-multilinearity of
ω ∧ ρ immediately follows from the A-multilinearity of ω and ρ, and the skew-symmetry follows
from both the skew-symmetry of ω and ρ, and the sign (−)σ in the defining formula (8.2).
8.1 Algebraic Preliminaries: Alternating Forms 145
1 2 3 4 5 6 7 8 9 1 3 4 8 9 2 5 6 7
We stress that, in (8.3), the sum runs over all permutations, not just the unshuffles. However,
as ω and ρ are alternating, hence skew-symmetric, several terms in (8.3) are counted several
times, and we have to normalize by the factor 1/(k!l!). Using the unshuffles, instead, is a
way to avoid such repetitions. Additionally, it makes it easier to prove several properties of
the wedge product.
(ω ∧ ρ) ∧ τ = ω ∧ (ρ ∧ τ);
ρ ∧ ω = (−)|ω||ρ| ω ∧ ρ.
Proof. The bilinearity should be clear. The associativity and the graded commutativity can be
easily proved by playing with unshuffles. We omit the technical details. ■
It immediately follows from the graded commutativity of the wedge product that, if ω is any
alternating form of odd degree, then
ω ∧ω = 0
The wedge products (8.1) can be combined into a single interior operation in the direct sum
M
Alt• (M , A) := Altk (M , A).
k∈Z
Before discussing this, we briefly recall what is the direct sum of a sequence of modules. So let
(Nk )k∈Z be a sequence of A-modules. By definition, the direct sum
N• = Nk
M
(8.4)
k∈Z
consists of sequences
with the additional property that only finitely many of the νk are non-zero. The direct sum N• is an
A-module with the following two term-wise operations:
and
for all (νk )k∈Z , (νk′ )k∈Z ∈ N• , and all a ∈ A. Notice that the zero vector in N• is the zero sequence
(0)k∈Z , and the opposite of an element (νk )k∈Z is the opposite sequence (−νk )k∈Z . A module of the
form (8.4) is also called a graded module. If A = R we rather speak about a graded vector space.
■ Example 8.2 The real vector space R[t 1 , . . . ,t n ] of real polynomials in the real indeterminates
(t 1 , . . . ,t n ) is a graded vector space. Indeed, denote by R[t 1 , . . . ,t n ]k the space of homogeneous
polynomials of degree k. We also put R[t 1 , . . . ,t n ]k = 0 for k < 0. Then R[t 1 , . . . ,t n ] is (canonically
isomorphic to) the direct sum of the R[t 1 , . . . ,t n ]k :
M
R[t 1 , . . . ,t n ] = R[t 1 , . . . ,t n ]k
k∈Z
where
ν if k = k0
νk := .
0 if k ̸= k0
The image of ιk0 consists of sequences whose terms all vanish except, possibly, for the k0 -th one. If
(νk )k∈Z ∈ N• is a sequence whose only non-zero entries are νk1 , . . . , νkℓ , then, clearly,
In the following, we will always interpret the Nk as subspaces of N , identifying them with
their images under the ιk (if we do so, then N• is exactly the direct sum of its subspaces Nk , in the
usual sense of the direct sum of subspaces). For instance, rather than (8.5), we will simply write
will be also referred to as alternating forms. Homogeneous forms are then those with a “definite
number of entries”.
Proposition 8.1.2 — Wedge Product. There exists a unique R-bilinear map
extending the wedge products (8.1), and also called the wedge product. The wedge product (8.6)
enjoys the following properties:
✓ (bilinearity) the wedge product is A-bilinear;
✓ (associativity) for any three alternating forms ω, ρ, τ,
(ω ∧ ρ) ∧ τ = ω ∧ (ρ ∧ τ);
ρ ∧ ω = (−)|ω||ρ| ω ∧ ρ;
1 ∧ ω = ω ∧ 1 = ω.
Proof. Let ω = (ωk )k∈Z and ρ = (ρl )l∈Z be two (non-necessarily homogeneous) alternating forms.
By definition, their wedge product is
ω ∧ ρ = (τm )m∈Z ,
where
τm := ∑ ωk ∧ ρl ,
k+l=m
where, in the last formula, we used the wedge products (8.1). It is easy to see that the operation
in Alt• (M , A) so defined enjoys all the required properties. We leave the simple details to the
reader. ■
The vector space Alt• (M , A) of alternating forms, together with the wedge product (8.6), is an
important instance of a graded algebra.
Definition 8.1.3 — Graded Algebra. A real, associative, graded commutative algebra with
unit, or simply a graded R-algebra, or just a graded algebra, is a graded vector space
k∈Z Ak , +, ·)
L
(A• =
|α ∧ β | = |α| + |β |,
β ∧ α = (−)|α||β | α ∧ β ,
|H(α)| = |α|;
H(1) = 1.
As usual, the identity is a graded algebra homomorphism, and the composition of graded
algebra homomorphisms is a graded algebra homomorphism.
(ε1 , . . . , εn ) ,
and let
ε 1, . . . , ε n
be the dual frame of Hom(M , A) = Alt1 (M , A). Then, for every 0 ≤ k ≤ n, the A-module
Altk (M , A) of alternating k-forms on M is generated by
ε i1 ∧ · · · ∧ ε ik , i1 , . . . , ik ∈ {1, . . . , n}.
ε i1 ∧ · · · ∧ ε ik i <···<i
1 k
In the hypothesis of Theorem 8.1.3, let i1 , . . . , ik ∈ {1, . . . , n} be any k-tuple of indexes, and
consider the multiple wedge product
ε i1 ∧ · · · ∧ ε ik .
We notice preliminarily that, from the graded commutativity of the wedge product, when at least two
indexes among i1 , . . . , ik coincide, then ε i1 ∧ · · · ∧ ε ik = 0. So ε i1 ∧ · · · ∧ ε ik can only be non-trivial
when the i j are all different, j = 1, . . . , k, for instance when i1 < · · · < ik .
Proof of Theorem 8.1.3. The proof generalizes the proof of the existence of the dual frame and we
only sketch it. It is enough to prove the “more precise version” of the statement. Let k ≥ 0. First
we prove that the system
ε i1 ∧ · · · ∧ ε ik
i1 <···<ik
(8.7)
generates Altk (M , A). So let ω be a k-form. For every 1 ≤ i1 < · · · < ik ≤ n, put
Denote by ω ′ the right hand side of (8.8). By multilinearity it is enough to show that ω ′ acts as ω
on elements in the frame (ε1 , . . . , εn ). By skew-symmetry it is enough to check that
ω ′ (ε j1 , . . . , ε jk ) = ω(ε j1 , . . . , ε jk ) = b j1 ··· jk
ω ′ (ε j1 , . . . , ε jk )
!
= ∑ bi1 ···ik ε i1 ∧ · · · ∧ ε ik (ε j1 , . . . , ε jk )
i1 <···<ik
bi1 ···ik ε i1 ∧ · · · ∧ ε ik (ε j1 , . . . , ε jk )
= ∑
i1 <···<ik
ε i1 ∧ · · · ∧ ε ik (ε j1 , . . . , ε jk )
come from
ε i1 (ε j1 ) · · · ε ik (ε jk ) = δ ji11 · · · δ jikk .
Hence
ω ′ (ε j1 , . . . , ε jk )
bi1 ···ik ε i1 ∧ · · · ∧ ε ik (ε j1 , . . . , ε jk )
= ∑
i1 <···<ik
It remains to show that the system (8.7) consists of linearly independent k-forms. So, let bi1 ···ik ∈ A
be scalars such that
ω′ = ∑ bi1 ···ik ε i1 ∧ · · · ∧ ε ik = 0.
i1 <···<ik
In particular, ω ′ vanishes on elements in the frame (ε1 , . . . , εn ), and, for every 1 ≤ j1 < · · · < jk ≤ n,
the same computation as above, reveals that
0 = ω ′ (ε j1 , . . . , ε jk ) = b j1 ··· jk .
Theorem 8.1.3 says that, when M possesses a finite frame (ε 1 , . . . , ε n ), then every k-form
ω ∈ Altk (M , A) can be written in the form
for some ai1 ···ik ∈ A, where the sum (omitted according to the Einstein convention) runs over all
indexes i1 , . . . , ik . On the other hand ω can be uniquely written in the form
Notice that the ai1 ···ik can be uniquely chosen in such a way that they enjoy the following skew-
symmetry property
for every permutation σ ∈ Sk (do you see it?). In this case, the relationship between the ai1 ···ik and
the bi1 ···ik is easily obtained by appropriately reordering the factors in (8.9), and it is
It is sometimes convenient to expand ω as in (8.9) (with skew-symmetric coefficients ai1 ···ik ) rather
than as in (8.10).
Corollary 8.1.4 — Alternating Forms on a Module with a Finite Frame. Let A be an algebra,
let M be an A-module possessing a finite frame
(ε1 , . . . , εn ) ,
and let
ε 1, . . . , ε n ,
be the dual frame of Hom(M , A) = Alt1 (M , A). Then the graded R-algebra Alt• (M , A) of
alternating forms is generated by
(1) 0-forms, i.e. elements in Alt0 (M , A) = A, and
(2) the 1-forms ε 1 , . . . , ε n .
This means that every form ω ∈ Alt• (M , A) can be written in the form
ω= ∑ ai ···i ε i
1 k
1
∧ · · · ∧ ε ik , (8.12)
k∈Z
for some ai1 ···ik ∈ A, where only finitely many of the ai1 ···ik are non-zero.
8.2 Higher Degree Differential Forms 151
Proof. Obvious. ■
Notice that, from the above discussion, the ai1 ···ik can be uniquely chosen so that they enjoy the
skew-symmetry property (8.11). Additionally, any form ω ∈ Alt• (M , A) can be uniquely written
in the form
ω= ∑ ∑ bi1 ···ik ε i1 ∧ · · · ∧ ε ik ,
k∈Z i1 <···<ik
A = A1 ε 1 + A2 ε 2 + A3 ε 3 , and B = B1 ε 1 + B2 ε 2 + B3 ε 3 .
ε 2 ∧ ε 3, ε 3 ∧ ε 1, ε 1 ∧ ε 2.
whose coefficients are the components of the vector product of the coordinate vectors (A1 , A2 , A3 )
and (B1 , B2 , B3 ). This shows that the wedge product generalizes the standard vector product in R3 .
■
of alternating k-forms on the tangent space Tp M. Forms in ∧k Tp∗ M are k-covectors. In other
words, a k-covector at the point p is an R-multilinear, alternating map:
ω : Tp M × · · · × Tp M → R.
| {z }
k times
The frame
d p x i1 ∧ · · · ∧ d p x ik
i1 <···<ik
Notice that 0-covectors are just real numbers, while 1-covectors are just covectors.
Every k-covector ω ∈ ∧k Tp∗ M can be uniquely written as
ω = ai1 ···ik d p xi1 ∧ · · · ∧ d p xik ,
for some ai1 ···ik ∈ R with the skew-symmetry property (8.11), where the (omitted) sum runs over all
multi-indexes. The k-covector ω can also be uniquely written as
ω= ∑ bi1 ···ik d p xi1 ∧ · · · ∧ d p xik .
i1 <···<ik
for some bi1 ···ik ∈ R related to the ai1 ···ik by the formula
∂ ∂
bi1 ···ik = k!ai1 ···ik = ω | p , . . . , i | p , i1 < · · · < ik .
∂ x i1 ∂x k
As already for tangent vectors and covectors, k-covectors can be organized into a manifold
called the bundle of k-covectors of M, and denoted ∧k T ∗ M, defined as
∧k T ∗ M :=
∏ ∧k T ∗ M = n(p, ω) : p ∈ M and ω ∈ ∧k T ∗ M o .
p p
p∈M
As usual, a point (p, ω) ∈ ∧k T ∗ M will be often simply denoted by ω. The set ∧k T ∗ M comes with
an obvious surjection π : ∧k T ∗ M → M, (p, ω) 7→ p, and the pair (∧k T ∗ M, π) is the bundle of k-
covectors of M. The smooth structure on M induces an atlas on ∧k T ∗ M as follows. Begin with a
chart (U, ϕ = (x1 , . . . , xn )) on M and define a chart (∧k T ∗U, ∧k T ∗ ϕ) on ∧k T ∗ M as follows. First
of all, put
∧k T ∗U := π −1 (U) =
∏ ∧k T ∗ M = n(p, ω) ∈ ∧k T ∗ M : p ∈ U o .
p
p∈U
Next put
n
N(k, n) = ,
k
and define a map:
∧k T ∗ ϕ : ∧k T ∗U → U
b × RN(k,n) , (p, ω) 7→ ∧k T ∗ ϕ(p, ω) := ϕ(p); (bi1 ···ik )i1 <···<ik ,
where
∂ ∂
bi1 ···ik := ω |p, . . . , i |p , i1 < · · · < ik .
∂ x i1 ∂x k
Recall that the bi1 ···ik are the components of ω in the coordinate frame
d p xi1 ∧ · · · ∧ d p xik i <···<i .
1 k
R Notice that a point in the bundle ∧0 T ∗ M of 0-covectors is simply a pair (p, a) where p ∈ M,
and a ∈ R. Accordingly, as a set, ∧0 T ∗ M is simply the cross product M × R, and it is easy
to see that the manifold structure on ∧0 T ∗ M agrees with the product manifold structure on
M × R. Additionally, the discussion preceding Proposition 8.2.1, shows that the bundle of
1-covectors, with its manifold structure, is exactly the cotangent bundle.
The C∞ (M)-module of k-forms is denoted Ωk (M). For k < 0, we also put Ωk (M) = 0. The
graded algebra of alternating forms on X(M) is denoted
Ω• (M) =
M
Ωk (M).
k∈Z
for some bi1 ···ik ∈ C∞ (U) related to the ai1 ···ik by the formula
∂ ∂
bi1 ···ik = k!ai1 ···ik = ω , . . . , i , i1 < · · · < ik .
∂ x i1 ∂x k
■
Differential k-forms are equivalent to sections of the bundle of k-covectors. More precisely,
there is a canonical C∞ (M)-module isomorphism Ωk (M) ∼ = Γ(∧k T ∗ M). To see this, first notice that
a k-form ω ∈ Ω (M) determines a k-covector ω p ∈ ∧ Tp∗ M, for every point p ∈ M. In order to
k k
Lemma 8.2.2 Let ω ∈ Ωk (M), let p0 ∈ M, and let X1 , . . . , Xk ∈ X(M). Then the value
ω(X1 , . . . , Xk )(p0 )
of the function ω(X1 , . . . , Xk ) ∈ C∞ (M) at p0 does only depend on (ω and) the values
(X1 ) p0 , . . . , (Xk ) p0
Proof. The proof is essentially the same of that of Lemma 7.2.2 and we omit it. ■
ω p : Tp M × · · · × Tp M → R.
| {z }
k times
Exercise 8.1 Let k, l be non-negative integers. Prove that, for every ω, ω ′ ∈ Ωk (M), every
ρ ∈ Ωl (M), and every p ∈ M,
(ω + ω ′ ) p = ω p + ω p′
(ω ∧ ρ) p = ω p ∧ ρ p . (8.13)
R Consider Exercise 8.1. Notice that when l = 0, then ρ = f is a function. Then Equation
(8.13) says that
( f ω) p = f (p)ω p .
Additionally, if (U, ϕ = (x1 , . . . , xn )) is a chart around p, then the values at p of the k-forms
in the coordinate frame are
sω : M → ∧k T ∗ M, p 7→ sω (p) := ω p .
Proof. The proof is essentially the same as that of Proposition 7.2.4 and we omit it. ■
As for vector fields and 1-forms, Proposition 8.2.3 has important consequences. Differential
forms can be restricted to open submanifolds, by restricting the associated sections, exactly as
vector fields and 1-forms, and we use the same notation ω|U for the restriction of a differential form
ω to an open submanifold U . Second, there is a(n obvious) Gluing Lemma for Differential Forms,
and we leave it to the reader to state and prove it. Finally, given a differential k-form ω ∈ Ωk (M),
we have that
(1) ω is completely determined by its restrictions ω|U to some coordinate domains U covering
M;
(2) for every chart (U, ϕ = (x1 , . . . , xn )), the restriction ω|U (as already remarked) can be uniquely
written as
for some ai1 ···ik ∈ C∞ (U) with the skew-symmetry property (8.11), where the (omitted) sum
runs over all multi-indexes. The k-form ω|U can also be uniquely written as
∂
Xα |U = Xαi , α = 1, . . . , k,
∂ xi
we have
ω(X1 , . . . , Xk )|U = ω|U (X1 |U , . . . Xk |U ) = k!ai1 ···ik X1i1 · · · Xkik = ∑ bi1 ···ik X1i1 · · · Xkik ;
i1 <···<ik
(4) if k > n = dim M, then ω|U = 0 for all carts (U, ϕ), hence ω = 0, i.e. there are no non-trivial
differential forms on M of degree greater than dim M.
Exercise 8.2 Investigate how does the local coordinate expression of a differential form change
under a change of coordinates. ■
Now, let (U, ϕ = (x1 , . . . , xn )) be a chart on M. From Example 8.4, the graded R-algebra
Ω• (U) is generated by functions C∞ (U) and 1-forms dx1 , . . . , dxn . In particular, it is generated by
functions, and 1-forms of the type d f for some f ∈ C∞ (U). This is true even globally according to
the following analogue of Theorem 7.2.5.
156 Chapter 8. Differential Forms and Cartan Calculus
Theorem 8.2.4 Let M be an n-dimensional manifold. The graded R-algebra Ω• (M) is generated
by functions C∞ (M) and differentials d f of functions f ∈ C∞ (M), i.e. every differential form
ω ∈ Ωk (M) can be written (in a possibly non-unique way) as a finite sum of forms of the type
f dg1 ∧ · · · ∧ dgk , 0 ≤ k ≤ n,
where f , g1 , . . . , gk ∈ C∞ (M).
Proof. Omitted. ■
We conclude this section showing that the pull-back of functions and differential 1-forms along
a smooth map F : M → N extends uniquely to a homomorphism of graded algebras
F ∗ : Ω• (N) → Ω• (M),
Proposition 8.2.5 — Pull-Back of Differential Forms. Let M, N be smooth manifolds and let
F : M → N be a smooth map. Then there exists a unique homomorphism of graded algebras
F ∗ : Ω• (N) → Ω• (M)
such that, on functions and 1-forms, F ∗ agrees with the pull-backs already defined.
Proof. First of all, let ω ∈ Ωk (N) be a homogeneous differential form (of degree k). When k = 0,
then ω = f is a functions and we put F ∗ ( f ) = f ◦ F. When k > 0, we define F ∗ (ω) ∈ Ωk (M)
through its values as follows. For every p ∈ M, and every v1 , . . . , vk ∈ Tp M, we put
In particular, when k = 1, we recover the pull-back of 1-forms already defined. For generic k,
Formula (8.14) defines a section s : p 7→ F ∗ (ω) p of ∧k T ∗ M. One can prove that s is a smooth section
in a very similar way as for 1-forms and we leave the details to the reader. In this way we defined
the pull-back of homogeneous forms. We extend the definition to possibly non-homogeneous forms
by additivity in the obvious way, i.e. for (ωk )k∈Z ∈ Ω• (M) a generic form, we put
F ∗ : Ω• (N) → Ω• (M)
is R-linear and preserves the wedge product of differential forms. We leave this easy check as
Exercise 8.3. Finally, we have to prove uniqueness. This immediately follows from Theorem 8.2.4
(do you see it?). ■
Exercise 8.3 Complete the proof of Proposition 8.2.5 showing that the pull-back
F ∗ : Ω• (N) → Ω• (M)
defined via (8.14) and (8.15) is R-linear and preserves the wedge product. (Hint: to show
that F ∗ is R-linear, discuss first linear combinations of homogeneous differential forms of the
same degree. To show that F ∗ preserves the wedge product, discuss first the wedge product of
8.2 Higher Degree Differential Forms 157
homogeneous forms (of possibly different degrees). To do this use Exercise 8.1. Then discuss the
general case.) ■
Exercise 8.4 Let F : M → N and G : N → Q be smooth maps between manifolds. Show that,
for every ω ∈ Ω• (Q),
id∗Q (ω) = ω,
and
■ Example 8.5 Consider R2 with standard coordinates (u, v), and R3 with standard coordinates
(x, y, z). Consider also the smooth map
ω = dx ∧ dy + dy ∧ dz + dz ∧ dx ∈ Ω2 R3 .
Exercise 8.5 Consider R3 with standard coordinates (x, y, z) and the differential 2-form
Let (U+ , ϕ+ = (X+ ,Y+ )) be the stereographic chart on the 2-dimensional sphere S2 ⊆ R3 .
Compute
l ∈ Z be an integer.
Definition 8.3.1 — Graded Derivation. A degree l, graded derivation of A• , or, simply, a
graded derivation, is an R-linear map ∆ : A• → A• such that
(1) ∆(Ak ) ⊆ Ak+l , for all k ∈ Z,
(2) ∆ satisfies the following graded Leibniz rule: for all homogeneous elements α, β ∈ A•
The (possibly negative) integer l is called the degree of ∆ and is also denoted by |∆|.
Proposition 8.3.1 Degree l graded derivations of A• form a real vector space under the sum
and the product by a scalar defined in the following (obvious) way:
Now, let ∆, ∇ be graded derivations (of possibly different degrees), and let β ∈ A• be a
homogeneous element. We define new operators:
β ∧ ∆ : A• → A• and [∆, ∇] : A• → A• ,
by putting
(β ∧ ∆)(α) := β ∧ ∆(α),
[∆, ∇](α) := ∆(∇(α)) − (−)|∆||∇| ∇(∆(α)).
Proposition 8.3.2 Both β ∧ ∆ and [∆, ∇] are graded derivations. Their degrees are
[∆, [∇, □]] + (−)|∆|(|∇|+|□|) [∇, [□, ∆]] + (−)|□|(|∆|+|∇|) [□, [∆, ∇]] = 0.
(β ∧ γ) ∧ ∆ = β ∧ (γ ∧ ∆);
(6)
The derivation [∆, ∇] is called the graded commutator of ∆ and ∇. Notice that, while the graded
commutator of graded derivations is a graded derivation, the usual commutator ∆ ◦ ∇ − ∇ ◦ ∆ is not
a graded derivation in general.
R The reader has probably already understood that the wedge product in a graded algebra and
the graded commutator of graded derivations are graded analogues of the product in an algebra
and the usual commutator of usual derivations. These operations enjoy similar properties as
their standard cousins. However the “graded formulas” differ from the usual ones by signs
depending on the degrees of the objects involved. There is a (mnemonic) meta-mathematical
rule called the Koszul sign rule that might help the reader remembering these signs:
Koszul Sign Rule: A graded formula differs from its standard (non-graded) analogue by
signs appearing everytime two (homogeneous) graded objects swap. Namely, everytime that
′
two graded objects O, O′ swap, there appears a sign (−)|O||O | .
The reader is invited to check that the Koszul Sign Rule applies to all cases discussed so far:
commutativity rule of the product, Leibniz rule, definition of the commutator, skew-symmetry
of the commutator, Jacobi identity, etc..
then ∆ = ∆′ .
160 Chapter 8. Differential Forms and Cartan Calculus
Proof. First of all, we clarify/recall what we mean by a family of homogeneous generators {αi }i∈I
of A• . This means that the αi are homogeneous elements of A• , and every other element α ∈ A•
can be written as a finite sum of products of the form
αi1 ∧ · · · ∧ αik , i1 , . . . , ik ∈ I, k ∈ Z.
α = ∑ αi1 ∧ · · · ∧ αik i1 , . . . , ik ∈ I, k ∈ Z,
and
∆′ (α) = ∆′
∑ αi 1 ∧ · · · ∧ αik
k
|∆′ |(|αi1 |+···|αi j−1 |)
= ∑ ∑ (−) αi1 ∧ · · · ∧ αi j−1 ∧ ∆′ (αi j ) ∧ αi j+1 ∧ · · · ∧ αik
j=1
k
|∆|(|αi1 |+···|αi j−1 |)
= ∑ ∑ (−) αi1 ∧ · · · ∧ αi j−1 ∧ ∆(αi j ) ∧ αi j+1 ∧ · · · ∧ αik
j=1
= ∆(α),
where we used the graded Leibniz rule. This concludes the proof. ■
Theorem 8.3.4 — Interior Product. Let M be a manifold, and let X ∈ X(M) be a vector field
on M. There exists a unique degree −1 graded derivation iX of the graded algebra Ω• (M) of
differential forms on M such that
✓ iX f = 0,
✓ iX d f = X( f ),
for all smooth functions f ∈ C∞ (M) = Ω0 (M).
Proof. First, we define iX on homogeneous forms ω. If |ω| = 0, then ω = f ∈ C∞ (M), and we put
iX f := 0.
iX ω := ω(X, −, . . . , −),
for all X1 , . . . , Xk−1 ∈ X(M). It is clear that iX ω is a (k − 1)-form, indeed the expression in the right
hand side of (8.16) is obviously C∞ (M)-multilinear and alternating in the arguments X1 , . . . , Xk−1 ,
and it is R-linear (actually even C∞ (M) linear) in the argument ω. Now we extend iX to all forms
by additivity in the following obvious way: for all (ωk )k∈Z ∈ Ω• (M) we put
The map
iX : Ω• (M) → Ω• (M)
8.3 Cartan Calculus 161
defined in this way is R-linear and maps k-forms to (k − 1)-form. We now prove that it is a graded
derivation (of degree −1). So, let ω ∈ Ωk (M) and ρ ∈ Ωl (M). We assume k, l > 0 and we leave
the simpler cases k = 0 or l = 0 to the reader. Let X1 , . . . , Xk+l−1 ∈ X(M) and compute
iX (ω ∧ ρ)(X1 , . . . , Xk+l−1 )
= ω ∧ ρ(X, X1 , . . . , Xk+l−1 )
= ∑ (−)σ ω(X, Xσ (1) , . . . , Xσ (k−1) )ρ(Xσ (k) , . . . , Xσ (k+l−1) )
σ ∈Sk−1,l
iX d f = d f (X) = X( f ).
This shows that iX satisfies all the required properties. Uniqueness immediately follows from
Proposition 8.3.3 and Theorem 8.2.4. ■
The graded derivation iX is called the insertion of X, or contraction with X, or also interior
product with X. We want to describe it in local coordinates. So, let (U, ϕ = (x1 , . . . , xn )) be a chart
on M, let ω ∈ Ωk (M) be locally given by
for some ai1 ···ik ∈ C∞ (U) enjoying the usual skew-symmetry property (8.11), and let X ∈ X(M) be
locally given by
∂
X|U = X i .
∂ xi
First of all, it is easy to see using the definition of iX , that (iX ω)|U = iX|U ω|U (do you see it?). Now
162 Chapter 8. Differential Forms and Cartan Calculus
compute
iX|U ω|U
= iX|U ai1 ···ik dxi1 ∧ · · · ∧ dxik
k
= ∑ (−) j−1 ai ···i dxi
1 k
1
∧ dxi j−1 ∧ iX dxi j ∧ dxi j+1 ∧ · · · ∧ dxik
j=1
k
= ∑ (−) j−1 ai ···i dxi
1 k
1
∧ dxi j−1 ∧ X(xi j ) ∧ dxi j+1 ∧ · · · ∧ dxik
j=1
k
= ∑ (−) j−1 ai ···i X i dxi
1 k
j 1 di j ∧ · · · ∧ dxik
∧ · · · ∧ dx
j=1
k
= ∑ ai i ···ib ···i X i dxi
j 1 j k
j 1 di j ∧ · · · ∧ dxik
∧ · · · ∧ dx (reordering the indexes)
j=1
k
= ∑ ai i ···i X i dxi
1 2 k
1 2
∧ · · · ∧ dxik (renaming the indexes)
j=1
■ Example 8.7 On R3 with standard coordinates (x, y, z), consider the 2-form
iX+Y = iX + iY
(8.17)
i f X = f iX .
Proof. Both the left hand sides and the right hand sides of (8.17) are degree −1 graded derivations
of Ω• (M). In order to check that they agree, it is enough to check that they agree on generators,
and this is an easy consequence of Theorem 8.3.4. We leave details to the reader. ■
8.3 Cartan Calculus 163
Theorem 8.3.6 — Lie Derivative. Let M be a manifold, and let X ∈ X(M) be a vector field
on M. There exists a unique degree 0 graded derivation LX of the graded algebra Ω• (M) of
differential forms on M such that
✓ LX f = X( f ),
✓ LX d f = dX( f ),
for all smooth functions f ∈ C∞ (M) = Ω0 (M).
Proof. First, we define LX on homogeneous forms ω. So let |ω| = k. Denote by {Φt }t the flow of
X. We define LX ω through its values, as follows:
d
(LX ω) p := |t=0 Φt∗ (ω) p , (8.18)
dt
for all p ∈ M. As already for the Lie derivative of a vector field, we have to show that Formula
(8.18) defines correctly a smooth section of the bundle ∧k T ∗ M of k-covectors. We work in local
coordinates. So, fix a point p0 ∈ M, and let (U, ϕ = (x1 , . . . , xn )) be a chart around p0 . Then
∂
X|U = X i and ω|U = ai1 ···ik dxi1 ∧ · · · ∧ dxik ,
∂ xi
for some smooth functions X i , and ai1 ···ik ∈ C∞ (U) (enjoying the usual skew-symmetry property).
Additionally, let J ⊆ R be an open interval containing 0, and let (U0 , ϕ) be a subchart of (U, ϕ)
around p0 exactly as in the proof of Proposition 6.2.4, i.e. such that Φ(J × U0 ) ⊆ U. Now, let
(t, p) ∈ J × U0 and, using the same notation and formulas as in the proof of Proposition 6.2.4,
compute
∂ Φi1 ∂ Φik
d ∗ d
(LX ω) p = |t=0 Φt (ω) p = |t=0 ai1 ···ik (Φ(t, p)) j (t, p) · · · j (t, p) d p x j1 ∧· · ·∧d p x jk .
dt dt ∂x 1 ∂x k
The coefficients are all smooth in the argument p, and this shows that the section
p 7→ (LX ω) p
of ∧k Tp∗ M is well-defined and smooth. For future use, we write explicitly the coefficient labelled
by ( j1 , . . . , jk ). A direct computation exploiting (6.16) and (6.17) shows that
∂ Φi1 ∂ Φik
d
|t=0 ai1 ···ik (Φ(t, p)) j (t, p) · · · j (t, p)
dt ∂x 1 ∂x k
(8.20)
∂ Xi ∂ Xi
i ∂ a j1 ··· jk
= X + j ai j2 ··· jk + · · · + j a j1 ··· jk−1 i (p).
∂ xi ∂x 1 ∂x k
We leave the computational details to the reader.
Next we extend LX to all forms by additivity as follows: for all (ωk )k∈Z ∈ Ω• (M) we put
The map
LX : Ω• (M) → Ω• (M)
defined in this way is R-linear and maps k-forms to k-forms. We now prove that it is a graded
derivation (of degree 0). So, let ω, ρ be homogeneous forms, let p ∈ M, and compute
d d d
(LX (ω ∧ ρ)) p = |t=0 Φt∗ (ω ∧ ρ) p = |t=0 (Φt∗ (ω) ∧ Φt∗ (ρ)) p = |t=0 Φt∗ (ω) p ∧ Φt∗ (ρ) p .
dt dt dt
Now we apply Lemma 8.3.7 below to the case B = ∧ to find
d
(LX (ω ∧ ρ)) p = |t=0 Φt∗ (ω) p ∧ Φt∗ (ρ) p
dt
d d
= |t=0 Φt (ω) p ∧ ρ p + ω p ∧ |t=0 Φt∗ (ρ) p
∗
dt dt
= (LX ω) p ∧ ρ p + ω p ∧ (LX ρ) p
= ((LX ω) ∧ ρ + ω ∧ LX ρ) p .
The graded Leibniz rule now follows from the arbitrariness of p.
Next, let f ∈ C∞ (M). Property LX f = X( f ) is provided by Proposition 6.2.2. In order to
compute LX d f , fix p0 ∈ M, and let (U, ϕ = (x1 , . . . , xn )) and U0 be as in the first part of the proof.
Then, for all p ∈ U0 ,
d
(LX d f ) p = |t=0 Φt∗ (d f ) p
dt
d
= |t=0 d p Φt∗ ( f )
dt
d ∂ Φt∗ ( f )
= |t=0 (p)d p xi
dt ∂ xi
2
∂
= |t=0,q=p f (Φ(t, q)) d p xi
∂t∂ xi
∂ d
= i | p |t=0 Φt∗ ( f )d p xi
∂ x dt
= (dX( f )) p ,
(can you reproduce all the above steps?) and, from the arbitrariness of p, we have LX d f = dX( f ).
Finally, uniqueness immediately follows from Proposition 8.3.3 and Theorem 8.2.4. ■
Lemma 8.3.7 Let V1 ,V2 ,V be finite dimensional real vector spaces, and let
B : V1 ×V2 → V
be a bilinear map. Let I ⊆ R be an open interval. Then, for any two smooth curves γ1 : I → V1 ,
γ2 : I → V2 ,
(1) the map I → V , t 7→ B(γ1 (t), γ2 (t)) is a smooth curve, and
(2) for all t0 ∈ I, the following Leibniz rule holds:
d d d
|t=t B(γ1 (t), γ2 (t)) = B |t=t γ1 (t), γ2 (t0 ) + B γ1 (t0 ), |t=t0 γ2 (t) .
dt 0 dt 0 dt
The graded derivation LX is called the Lie derivative along X. We want to describe it in local
coordinates. So, let (U, ϕ = (x1 , . . . , xn )) be a chart on M, let ω ∈ Ωk (M) be locally given by
with the ai1 ···ik skew-symmetric, and let X ∈ X(M) be locally given by
∂
X|U = X i .
∂ xi
It immediately follows from (8.20) that
■ Example 8.8 On the 3-dimensional standard Euclidean space with standard coordinates (x, y, z)
consider the 2-form
LX+Y = LX + LY
(8.22)
L f X = f LX + d f ∧ iX .
Proof. Both the left hand sides and the right hand sides of (8.22) are degree 0 graded derivations of
Ω• (M). In order to check that they agree it is enough to check that they agree on generators, and
this is an easy consequence of Theorem 8.3.6. ■
Theorem 8.3.9 — Exterior Differential. Let M be a manifold. There exists a unique degree 1
graded derivation d of the graded algebra Ω• (M) of differential forms on M such that
✓ d f is exactly the differential of f ,
✓ dd f = 0,
for all smooth functions f ∈ C∞ (M) = Ω0 (M).
166 Chapter 8. Differential Forms and Cartan Calculus
for all X1 , . . . , Xk+1 ∈ X(M). It is clear that dω is R-multilinear. The fact that dω is alternating
easily follows from (the fact that ω is R-multilinear and) the signs (−)i and (−)i+ j in (8.23) and
we leave the details to the reader. Next we have to show that dω is C∞ (M)-multilinear. It is enough
to show that dω is C∞ (M)-linear in the first argument. The C∞ (M)-linearity in the other arguments
then follows from skew-symmetry. So we have to show that
for all X1 , X2 , . . . , Xk+1 ∈ X(M), and all f ∈ C∞ (M). To see this we compute
dω( f X1 , X2 , . . . , Xk+1 )
= f X1 (ω(X2 , . . . , Xk+1 )) + ∑ (−)i+1 Xi ω( f X1 , X2 , . . . , Xbi , . . . , Xk+1 )
1<i
j+1
+ ∑ (−) ω([ f X1 , X j ], X1 , . . . , Xbj , . . . , Xk+1 )
1< j
= f dω(X1 , X2 , . . . , Xk+1 ).
d : Ω• (M) → Ω• (M)
defined in this way is clearly R-linear and maps k-forms to (k + 1)-forms. We now prove that it is a
graded derivation (of degree 1). So, let ω ∈ Ωk (M) and ρ ∈ Ωl (M). If k = 0, then ω = f ∈ C∞ (M),
8.3 Cartan Calculus 167
This shows that the graded Leibniz rule for d(ω ∧ ρ) works when |ω| = 0. We leave it to the
reader to show that it also works when |ω| = 1 as Exercise 8.8 (it’s a long computation but, if well
organized, it is not conceptually complicated). This allows us to prove the graded Leibniz rule for
d(ω ∧ ρ) by induction on |ω| (notice that the graded Leibniz rule could also be proved by a direct,
but cumbersome, computation). So assume that the graded Leibniz rule for d(ω ∧ ρ) works when
|ω| < r − 1, and let |ω| = r > 1. By Theorem 7.2.5, ω can be written as a sum of terms of the form
θ1 ∧ · · · ∧ θr ,
d(ω ∧ ρ) = ∑ d(θ1 ∧ θ2 ∧ · · · ∧ θr ∧ ρ)
= ∑(dθ1 ) ∧ θ2 ∧ · · · ∧ θr ∧ ρ − ∑ θ1 ∧ d(θ2 ∧ · · · ∧ θr ∧ ρ) (base of induction)
= ∑(dθ1 ) ∧ θ2 ∧ · · · ∧ θr ∧ ρ − ∑ θ1 ∧ d(θ2 ∧ · · · ∧ θr ) ∧ ρ
− ∑(−)r−1 θ1 ∧ θ2 ∧ · · · ∧ θr ∧ dρ (induction hypothesis)
r
= (dω) ∧ ρ + (−) ω ∧ dρ (base of induction).
Finally, specializing the Chevalley-Eilenberg Formula (8.23) to k = 1, we get
for all X,Y ∈ X(M). This concludes the proof of the existence. The uniqueness again follows from
Theorem 7.2.5. ■
The graded derivation d is called the exterior differential, or De Rham differential, or just
differential. We want to describe it in local coordinates. So, let (U, ϕ = (x1 , . . . , xn )) be a chart on
M, and let ω ∈ Ωk (M) be locally given by
1 k+1 j+1 ∂
di1 ···ik+1 = ∑ (−) a .
k + 1 j=1 ∂ xi j i1 ···i j ···ik+1
b
We leave the details to the reader. We now prove (8.24). First of all, it is easy to see, using the
definition of the exterior differential, that (dω)|U = d(ω|U ). Now,
∂
d(ω|U ) = d ai1 ···ik dxi1 ∧ · · · ∧ dxik = dai1 ···ik ∧ dxi1 ∧ · · · ∧ dxik = i ai1 ···ik dxi ∧ dxi1 ∧ · · · ∧ dxik ,
∂x
where we used that
∂
dai1 ···ik = i ai1 ···ik dxi
∂x
and that ddxi = 0. After renaming the indexes this proves (8.24).
■ Example 8.9 On R3 with standard coordinates (x, y, z) consider the 2-form
Now, we want to compute the graded commutators of interior products, Lie derivatives, and the
exterior differential. These formulas are the main formulas in Cartan calculus. We collect them in
the following table of graded commutators:
[−, −] iY LY d
iX 0 i[X,Y ] LX
LX i[X,Y ] L[X,Y ] 0
d LY 0 0
Theorem 8.3.10 — Cartan Calculus. Let M be a manifold and let X,Y ∈ X(M) be vector fields
on M. The interior products iX , iY , the Lie derivatives LX , LY and the exterior differential d fit
in the Table 8.1 of graded commutators.
Proof. All the graded commutators in Table 8.1 can be computed in the same way. We discuss one
example and leave all the rest to the reader as Exercise 8.9. The formula
[d, iX ] = LX (8.25)
is known as Cartan Magic Formula and can be proved as follows. Both sides of (8.25) are graded
derivations of degree 0. To show that they coincide, it is enough to show that they agree on
generators. So let f ∈ C∞ (M), and compute
[d, iX ] f = diX f + iX d f = iX d f = X( f ) = LX f ,
and
[d, iX ]d f = diX d f + iX dd f = dX( f ) = LX d f .
This concludes the proof of the Cartan Magic Formula. ■
Exercise 8.9 Complete the proof of Theorem 8.3.10 computing the remaining graded commu-
tators in Table 8.1. ■
Notice that the formula [d, d] = 0, in the bottom-right corner of Table 8.1, is not at all trivial.
Indeed, as d is a degree 1 derivation, we have
[d, d] = d ◦ d + d ◦ d = 2d ◦ d.
Hence [d, d] = 0 is the same as
d ◦ d = 0,
which is in turn equivalent to
im d ⊆ ker d.
A form in the kernel of d is called a closed form or a cocycle. A form in the image of d is called an
exact form or a coboundary. So [d, d] = 0 means that exact forms are also closed. The converse
might not be true. The quotient vector space
ker d
HdR (M) :=
im d
is a graded vector space called the De Rham cohomology of M, and contains important topological
information about M.
170 Chapter 8. Differential Forms and Cartan Calculus
of the Lie algebra of vector fields is a Lie subalgebra. To prove this last claim it is convenient
to characterize infinitesimal symmetries of ω in a suitable way. Namely, a vector field X on
M is an infinitesimal symmetry of ω if, and only if, LX ω = 0. This statement can be proved,
exactly as Proposition 6.3.5, after proving the following formulas:
d ∗
Φ (ω) = Φt∗ (LX ω) = LX (Φt∗ (ω)) , (8.26)
dt t
where {Φt }t is the flow of X. By definition, given t such that Mt ̸= ∅, the left hand side of
(8.26) is the differential form on Mt defined through its values as follows: for all p ∈ Mt , the
value of dtd Φt∗ (ω) at p is
d ∗ d
Φ (ω) p = |s=t Φt∗ (ω) p .
dt t ds
One can show that this a well-defined form and that Formulas (8.26) indeed hold. We leave
details to the reader. As already mentioned, from those formulas, one can easily prove that X is
an infinitesimal symmetry of ω if and only if LX ω = 0. Finally, to see that X(M, ω) ⊆ X(M)
is a Lie subalgebra, take X,Y ∈ X(M, ω), and a ∈ R. Then
LX+Y ω = LX ω + LY ω = 0.
Additionally,
LaX ω = aLX ω = 0.
The last two formulas show that X(M, ω) is a vector subspace of X(M). We conclude
computing
L[X,Y ] ω = [LX , LY ]ω = LX LY ω − LY LX ω = 0,
where we used the commutator in the center of Table 8.1. Hence X(M, ω) is also a Lie
subalgebra of X(M), as claimed.
Besides the definition provided in the proof of Theorem 8.3.6, and the one provided by the
Cartan Magic Formula, there is yet another equivalent definition of the Lie derivative.
Proposition 8.3.11 Let M be a manifold, and let X ∈ X(M). The Lie derivative along X satisfies
the following Leibniz rule with respect to insertions of vector fields: for any degree k differential
form ω, and any X1 , . . . , Xk ∈ X(M),
k
(LX ω)(X1 , . . . , Xk ) = X (ω(X1 , . . . , Xk )) − ∑ ω(X1 , . . . , [X, Xi ] , . . . , Xk ). (8.27)
i=1
| {z }
i-th place
Proof. The proof is by induction on k. The case k = 1 is easily obtained applying the formula
to a 1-form (do you see it?). Now, rewrite the left hand side of (8.27) as follows:
(LX ω)(X1 , . . . , Xk ) = (iX1 LX ω)(X2 , . . . , Xk )
= ([iX1 , LX ]ω)(X2 , . . . , Xk ) + (LX iX1 ω)(X2 , . . . , Xk )
= (i[X1 ,X] ω)(X2 , . . . , Xk ) + (LX iX1 ω)(X2 , . . . , Xk )
= ω([X1 , X], X2 , . . . , Xk ) + (LX iX1 ω)(X2 , . . . , Xk ).
and use induction on the last summand. ■
We conclude this section, this chapter, and these notes discussing how do the interior product,
the Lie derivative, and the exterior differential interact with smooth maps. So let M, N be smooth
manifolds and let F : M → N be a smooth map. We begin with the most important case: that of the
exterior differential.
Proposition 8.3.12 — Naturality of the Exterior Differential. The exterior differential “com-
mutes” with pull-backs, i.e.
d ◦ F ∗ = F ∗ ◦ d. (8.28)
Before proposing a proof we stress that the reader should not be confused by the terminology
“the exterior differential commutes with pull-backs”. Namely, the exterior differential in the left
hand side of (8.28) is that in Ω• (M) while the one in the right hand side is that in Ω• (N).
Proof of Proposition 8.3.12 (a sketch). The statement could be proved by a direct computation. We
propose an alternative proof exploiting Theorem 8.2.4. Consider the R-linear map
We have to prove that D = 0. To do this, we first list some properties of D. Besides the R-linearity,
D enjoys the following properties:
(1) D is a degree 1 map, i.e. D(Ωk (N)) ⊆ Ωk+1 (M), for all k ∈ Z,
(2) D satisfies the following (F-relative) version of the graded Leibniz rule: for all homogeneous
forms ω, ρ ∈ Ω• (N),
Property (1) is obvious. Property (2) can be easily checked using that d is a graded derivation and
that F ∗ is a graded algebra homomorphism, and we leave the details to the reader. In a very similar
way as in the proof of Proposition 8.3.3, it follows from Properties (1) and (2) that, if D vanishes
on generators, then D = 0. So it remains to check that
D( f ) = 0,
D(d f ) = 0,
i.e.
dF ∗ ( f ) = F ∗ (d f ), (8.29)
∗ ∗
dF (d f ) = F (dd f ), (8.30)
for all f ∈ C∞ (M). Now, the first identity has already been proved as Proposition 7.3.2 Point (3),
while the second identity follows from Proposition 7.3.2 Point (3), and d ◦ d = 0. This concludes
the proof. ■
Proposition 8.3.13 Let X ∈ X(M) and Y ∈ X(N) be F-related vector fields. Then
iX ◦ F ∗ = F ∗ ◦ iY ,
LX ◦ F ∗ = F ∗ ◦ LY .
iΦ∗ (Y ) ◦ Φ∗ = Φ∗ ◦ iY ,
LΦ∗ (Y ) ◦ Φ∗ = Φ∗ ◦ LY .
Exercise 8.10 Prove Proposition 8.3.13. (Hint: there is a proof very similar to the proof of
Proposition 8.3.12.) ■
Corollary 8.3.14 Let M be a manifold, let Y ∈ X(M), let ω, ω1 , ω2 ∈ Ω• (M), and let Φ :
M → M be a diffeomorphism. If Φ is a symmetry of ω, ω1 , ω2 , then it is also a symmetry of
dω, ω1 + ω2 , ω1 ∧ ω2 . If, additionally, Φ is a symmetry of Y , then it is also a symmetry of iY ω
and LY ω.
The following infinitesimal version of Corollary 8.3.14 is a corollary of either Corollary 8.3.14
itself, or Theorem 8.3.10.
Corollary 8.3.15 Let M be a manifold, let X,Y ∈ X(M), and let ω, ω1 , ω2 ∈ Ω• (M). If X is
an infinitesimal symmetry of ω, ω1 , ω2 , then it is also an infinitesimal symmetry of dω, ω1 +
ω2 , ω1 ∧ ω2 . If, additionally, X is an infinitesimal symmetry of Y , then it is also an infinitesimal
symmetry of iY ω and LY ω.
Zero locus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43