Lecture notes on Analysis and Geometry on

Manifolds

Written by Lecturer
Manuel Hinz Prof. Dr. Laurent Côté
[email protected] or [email protected] lcote[at]math.uni-bonn.de

University of Bonn
Winter semester 2024/2025
Last update: November 17, 2024
Contents

Chapter 0 Manuel’s notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

0.1 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.2 Course overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Chapter 1 Topological manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1 Some point set topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Locally Euclidean spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 Hausdorff spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.3 Basis and covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Topological manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Examples of topological manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.2 Brief interlude: Why do we need Hausdorffness? . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.3 Manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.4 Elementary topological properties of topological manifolds . . . . . . . . . . . . . . . . . . . . . 14
1.3 Classification of topological manifolds (proofs are not examinable) . . . . . . . . . . . . . . . . . . . . 14
1.3.1 Classification of 1-dimensional manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.2 Classification of 2-dimensional manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.3 Classification of high dimensional manifolds (not examinable at all) . . . . . . . . . . . . . . . 16

Chapter 2 Smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1 Basic theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Charts and atlases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 First examples of smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1.3 Smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.4 The category of smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.5 Smooth manifolds with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Partitions of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Preparatory lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Partitions of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Applications of partitions of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

Chapter 3 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Two (equivalent) theories of tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 Definition via equivalence classes of smooth curves . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.2 Definition via derivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.3 Both definitions agree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4 The tangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Chapter 4 Submersions, immersions and embeddings . . . . . . . . . . . . . . . . . . . . . . 37

4.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 The rank theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Chapter 5 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 The “slice lemma” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

Page 1 of 60
 Chapter 0 – CONTENTS

5.3 The (weak) Whitney embedding theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

Chapter 6 Transversality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.1 Basic definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.1.2 Transversality for submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.1.3 Transversality of maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Journal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Page 2 of 60
Chapter 0:
Manuel’s notes

Warning

These are unofficial lecture notes written by a student. They are messy, will almost surely
contain errors, typos and misunderstandings and may not be kept up to date! I do however
try my best and use these notes to prepare for my exams. Feel free to email me any
corrections to [email protected] or [email protected].
Happy learning!

General Information

• Basis: Basis
• Website: https://www.math.uni-bonn.de/∼lcote/V3D3 2024.html
• Time slot(s): Tuesday: 14-16 Nussallee Anatomie B and Friday: 12-14 GHS

• Exams: Tuesday 11.02.2025, 9-11, Großer Hörsaal, Wegelerstraße 10 and Friday

21.03.2025, 9-11, Großer Hörsaal, Wegelerstraße 10
• Deadlines: Friday before noon

0.1 Organization
Start of lecture 01
• Four exercise classes, in the break come to the front and sign up. (08.10.2024)
• First homework is due this Friday

• Exercise sheets are due on Fridays, every week electronically (groups, at most 2)
• No published lecture notes by him!
• 5 Minute break right before the full hour
• Friday after class for questions

0.2 Course overview

He assumes we already know about
• Analysis on Rn

• Basic point set topology

Page 3 of 60
 Chapter 0 – Manuel’s notes

For this class: smooth manifolds based on [2] I would also recommend
[5] and the notes of
• Intersection between analysis and topology Gabriel Ong[3], which are
• Exiting: Connections between those two point of views also based on this course

Main topics:
Topic 00: Topological manifolds
Topic 01: Basic theory of smooth manifolds
Topic 02: Vector fields on smooth manifolds

Topic 03: Tensor calculus and Stokes’ theorem

Topic 04: Lie groups, symplectic and Riemannian geometry

Page 4 of 60
Chapter 1:
Topological manifolds

1.1 Some point set topology

Some (set theoretical) conventions for the whole course:
• A ⊂ B means A subset (not necessarily proper!) of B, i.e. ⊂=⊆
• A neighborhood of some point p ∈ X means an open set U ⊂ X containing p
P2
• Given p = (p1 , . . . , pn ) ∈ Rn , r > 0, Brn (p) := {(x1 , . . . , xn ) | xi −pi < r2 }. Often while
Bs = Bsn (0) ⊂ Rn

1.1.1 Locally Euclidean spaces

Definition. A topological space X is called locally Euclidean of dimension n ≥ 0, if every
point of X is contained in a neighborhood homeomorphic to some open subset of Rn .

Remark. When we speak of a topological space as being locally Euclidean. The dimension is
fixed and implicit.

Definition. Assume that X is locally Euclidean. A chart is a pair U, ϕ, where U ⊂ X, ϕ : U → Rn

is a homeomorphism into its image. Given p ∈ X, we say that U, ϕ is centered at p if p ∈ U and
ϕ(p) = 0 ∈ Rn

Figure 1.1: Sketch 1.01

Lemma 1.1. The following are equivalent (TFAE):

• X is locally Euclidean

• For any p ∈ X, there is a chart U, ϕ centered at p with image ϕ(U ) = B1

Page 5 of 60
 Chapter 1 – Topological manifolds

• For any p ∈ X, there is a chart U, ϕ centered at p with image ϕ(U ) = Rn

x
Proof. 2. and 3. are equivalent, since B1 ≃ Rn are homeomorphic (B1n ∋ x 7→ 1−∥x∥ )
2. =⇒ 1. is tautological
1. =⇒ 2. given p ∈ X, since X is locally Euclidean, there exists some chart U, ψ, p ∈ U .
ψ : U → Rn , homeo onto its image ψ(U ) = O ⊂ Rn . By translativity Rn ∋ x 7→ x − ψ(p), one can
assume ψ(p) = 0 ∈ Rn . By scaling Rn (x 7→ λx, λ > 0), can assume B1 ⊂ ψ(U ). Let
U ′ = ψ −1 (B1 ), then (U, ψ) as claimed.

1.1.2 Hausdorff spaces

Definition. A topological space X is called Hausdorff, if given any p1 ̸= p2 ,p1 , p2 ∈ X, there exist
neighborhoods p1 ∈ U1 , p2 ∈ U2 s.t. U1 ∩ U2 = ∅.

Figure 1.2: Sketch 1.02

Example. • Rn
• CW complexes

• most reasonable spaces

Example (Not Hausdorff). X = {0, 1}, open subsets ∅, {0}, {0, 1}

Remark. X is homeomorphic to R/R∗ (quotient topology), R∗ , (s, x 7→ sx)

Lemma 1.2. Let X be Hausdorff.

(a) point sets {x} are closed

(b) convergent sequences have unique limits. (xn → p, xn → q =⇒ p = q)
(c) compact sets are closed

Proof. (c) =⇒ (a)

For (c): Let K ⊂ X be compact. Want to show K c is open. Pick p ∈ K c . For each q ∈ K, we can
choose Uq ∋ q, Up ∋ p : Uq ∩ Up = ∅ Since K is compact, it can be covered by Uq1 , . . . , Uql . Then
Tl Sl
i=1 Uqi is oen and contains p, disjoint, then i=1 Uqi ⊃ K.

Page 6 of 60
 Chapter 1 – Topological manifolds

Figure 1.3: Sketch 1.03

(b) Suppose for contradiction that xi → p, xi → q and p ̸= q. Since X is Hausdorff,

∃U ∋ p, O ∋ q, U ∩ O = ∅. But for N >> 0xi ∈ U, xi ∈ O∀i > N

1.1.3 Basis and covers

Let X be a topological space.

Definition. A collection B of subsets of X is called a basis(base) for X, if for any p ∈ X and

any neighborhood U ∋ p, there exists an element U ∈ B s.t. p ∈ U ⊂ U .

Page 7 of 60
 Chapter 1 – Topological manifolds

Figure 1.4: Sketch 1.04

Lemma 1.3. B is a basis for X ⇐⇒ every open set of X is a union of elements of B.

Proof. Trivial.
Definition. A topological space X is second-countable if it admits a countable basis.

Example. • Rn , B = {Bsn (p) | s ∈ Q+ , p = (p1 , . . . , pn ) ∈ Qn ⊂ Rn }

Lemma 1.4. The property of being second-countable is closed under

(a) subspaces
(b) countable disjoint unions

(c) countable products

Remark. The property of being second-countable is not closed under arbitrary quotients q : A →
A/B. An obvious sufficient conditions is for q to be an open map. (Since it is a pushforward) When constructing
manifolds via quotients,
Lemma 1.5. If X is second countable, then any open cover of X admits a countable subcover. check that it is still
second-coutable!
Proof. Let B be a countable basis for X. Let C be an open cover. Let B̃ ⊂ B be the collection of
basis elements U , which are contained in some U ∈ C. Observe (key!) B̃ is a cover of X. For each
U ∈ B̃, choose UU ∈ C such that U ⊂ UU . Then {UU } is a countable subcover of C.

Definition. Let X be a topological space. An exhaustion of XSby compact subsets is a se-

∞
quence {Ki }i∈N , where Ki ⊂ X compact and Ki ⊂ int(Ki+1 ) and i=1 Ki = X.

Recall given A ⊂ X. int(A) := {x ∈ A | x in a neighborhood U ⊂ A}.

Page 8 of 60
 Chapter 1 – Topological manifolds

Lemma 1.6. If X is locally Euclidean, Hausdorffa and second countable. Then X admits an
exhaustion by compact subsets.
a not needed

Proof. Since X is locally Euclidean, admits a basis B of open subsets having compact closure. That is take the close of
B 21 ⊂ Rn

Figure 1.5: Sketch 1.05

By Lemma 1.5, one can extract a countable subcover {Ui }∞ i=1 . Set K1 = U1 . Assume that we
. . , Kk such that Uj ⊂ Kj and Kj−1 ⊂ int(Kj ), j ≥ 2. Since
already constructed K1 , .S SmK k is
∞
compact and Kk ⊂ X = i=1 Ui , then there exists some mk such that Kk ⊂ X = i=1 k
Ui by
compactness. Might as well assume that mk ≥ k. Set
m
[k m
[k
Kk+1 = Ui = Ui .
i=1 i=1

By construction Kk+1 is compact, Kk ⊂ int(Kk+1 ). We get {Kj }∞

j=1 , Uj ⊂ Kj (because mj ≥ j)
S∞ S∞
=⇒ i=1 Ui = i=1 Ki
Start of lecture 02
Definition. Let X be a topological space. Let C be a collection of subsets of X. We say that C is (11.10.2024)
locally finite if for every x ∈ X there exists a neighborhood U ∋ x such that the intersection of U
with all but finitely many elements of C is empty.
Example (Example for local finiteness). Take X = R, C = {(i − 1, i + 1)}i∈Z .

Figure 1.6: Sketch 1.06

Example (Non-example for local finiteness). X = R, C = (q − 1, q + 1)q∈Q

Definition. Let X be a topological space. Let C be a cover of X. A cover C ′ of X is called a
refinement of C, if for all elements U ∈ C ′ , there exists such V ∈ C: U ⊂ V .
Example (Example of Refinement). In the proof of lemma 1.5, we showed that any open cover
admits a refinement by basis elements.

Page 9 of 60
 Chapter 1 – Topological manifolds

Definition. A topological space X is called paracompact if every open cover admits a locally finite
refinement.
Whats up with the word paracompact? It’s like compact, but weaker! It is necessary that it only
admits a locally finite refinement!
Lemma 1.7. Let X be Hausdorff and suppose that X admits an exhaustion by compact subsets.
Then X is paracompact.In fact, we will show that given any basis B of X, any open cover admits
a locally finite refinement by elements of B.
S∞
Proof. By assumption, R {Ki }i∈N , Ki compact, Ki ⊂ int(Ki+1 ), i=1 Ki = X. Let, for Careful! There are many
j ∈ Z : Vj = Kj+1 \ (Kj ) if j ≤ 0 : Kj = ∅1 . definitions of exhaustion
by compact sets . . .

Figure 1.7: Sketch 1.07

Notice:
• Vj is compact, since we take the intersection of a compact set and a closed set. (int(Kj )c is
closed)
S S S
• j∈Z VJ = X, since j≤n = j≤n+1 Kj = Kj+1
• The compact sets Vj are intersecting (along their boundary?)
Vj ∩ Vj−1 = ∂Kj := Kj \ int(Kj )
Evidently {Uα ∩ int(Kj+1 ) ∩ int(Kj−1 )c }α∈A covers Vj = Kj+1 − \Kj−1 c
, where the {Uα }α∈A is Here we use
an open cover. Since B is a basis, we can find a refinement of this cover by basis elements. Since Hausdorffness
Vj are compact, we can extract a finite subcover {Vlj }l=1,...,kj . Let’s consider: {Vlj }j∈Z,l=1,...,kj .
This subcover works, i.e.
• obviously a cover, since the Vj cover X, obviously a refinement of {Uα }
c
• locally finite: given x ∈ X, x ∈ Vj , hence x ∈ int(KKj+2 ) ∩ Kj−1 =: U . If U ∩ Vlk , then we
must have j − 2 ≤ k ≤ j + 2. But {Vlk }j−2≤k≤j+2 is finite.
Corollary 1.8. If X is locally Euclidean, Hausdorff and second countable =⇒ X is paracompact.
Proof. By lemma 1.6 (exhaustion by compact subsets) and lemma 1.7 =⇒ paracompact.
Corollary 1.8’. Let X be Euclidean amd Hausdorff. Then X is second countable iff X has
countably many components and X is paracompact.
Remark. There are different definitions of manifolds. They differ in either forcing second count-
ability or paracompactness. This lemma shows that there only is a difference if there are uncountably
many components.
Proof. Corollary 1.8 and the bonus homework problem from sheet 01.
Remark. Basis elements are open.

1.2 Topological manifolds

1 He writes − for \

Page 10 of 60
 Chapter 1 – Topological manifolds

Definition. A topological n-manifold M is a topological space with the following properties:

(i) M is locally Euclidean (of dimension n)

(ii) M is Hausdorff
(iii) M is second countable

Morally we only really need condition (i). Why do we need the others? For (ii) you will not get a
useful theory without it, while (iii) can be replaced by paracompactness (see corollary 1.8’).

Definition. Let Man0 be the category of topological manifolds with

1. objects: topological manifolds

2. morphisms: continuous functions

Remark. Man0 full subcategory of Top.

Remark. By definition, M, N ∈ Man0 , then M, N are isomorphic iff M, N are homeomorphic.

1.2.1 Examples of topological manifolds

Example (Spaces isomorphic to Rn ). Rn , n ≥ 0 More generally, if V a finite dimensional
R-vector space, then V is a topological n-manifold.
Example. Any open subset of Rn

Example (Graphs). Let U ⊆ Rn open, let f : U → Rn be a continuous function. We set

M := graph(f ) := {(x, y) ∈ U × Rn | y = f (x)}.

Then M is a manifold. The map M → U by (x, y) 7→ U gives a global chart.

Example (Spheres). Let S n := {x20 + · · · + x2n = 1} ⊂ Rn+1 . Then S n is a manifold. We define Here we no longer have a
charts global chart (for
ϕ± ± n n
i : Ui = {(x0 , . . . , xn ) ∈ S | ±xi > 0} → B1 (0) topological reasons)
by (x0 , . . . , xn ) 7→ (x0 , . . . , x̂i , . . . , xn ) := (x0 , . . . , xi−1 , xi+1 , . . . , xn )

Page 11 of 60
 Chapter 1 – Topological manifolds

Figure 1.8: Sketch 1.08

Example (spheres’). Let C n := ∂([−1, 1]n+1 ) = [−1, 1]n+1 \ int([−1, 1]n+1 ). Homework:
C n ≃ S n (homeomorphic)
Example (n-torus). Let Πn := Rn /Zn with the quotient topology. Then this is a manifold
(exercise).

Figure 1.9: Sketch 1.09

Example (RPn := S n /{x ∼ −x}). RPn are also manifolds (called the real projective spaces).

Page 12 of 60
 Chapter 1 – Topological manifolds

Figure 1.10: Sketch 1.10

Example (Klein bottle).

Remark. RP2 or generally RP2n and the Klein bottle are not orientable.

1.2.2 Brief interlude: Why do we need Hausdorffness?

• Back in the day (Riemann)
• There is no hope to classify even 1d locally Euclidean, second-countable NOT Hausdorff
spaces (See the line with two origins)
• With Hausdorff: Only 1d manifolds are R, S 1 (see website)

Why do we need second countability?

• Subspaces of Rn are second countable
• We want partitions of unity (paracompactness suffices for that)

1.2.3 Manifolds with boundary

Let Hn := {(x1 , . . . , xn ) ∈ Rn | xn ≥ 0}.

Definition. A manifold with boundary is a topological space with the following properties:
(i) Every point has a neighborhood homeomorphic to an open subset of Hn

(ii) Hausdorff
(iii) second countable

Clearly every manifold is also a manifold with boundary.

Example. Hn is a manifold with boundary, but not a manifold. Since for points on the
boundary, there are no neighborhoods homeomorphic to Euclidean space.

Example. S n ∩ Hn+1 , S n ⊂ Rn+1 , [a, b], [0, ∞)

Page 13 of 60
 Chapter 1 – Topological manifolds

Figure 1.11: Sketch 1.12

Definition. If M manifold with boundary, we say x is a boundary point, if x ∈ M \ int(M ) (i.e.

it has no neighborhood homeomorphic to Euclidean space?), otherwise x is an iterior point. We let
∂M := {boundary points}.
Start of lecture 03
Remark. Most of what he says in the course can be generalized to manifolds with boundary (unless (15.10.2024)
it makes no sense). Those results are only stated (and proofed) for manifolds. it might be a good
exercise to go through the notes and generalize the statements to manifolds with boundary.

1.2.4 Elementary topological properties of topological manifolds

Not proved here, but we
• A manifold is connected iff it is path connected are welcome to use
• For manifolds, all forms of compactness (ordinary compactness (every open cover has a
finite subcover), limit point compactness, sequential compactness) are equivalent
• All manifolds are metrizable (Urysohn metrization theorem + second countable =⇒
metrizable) The first two point were
proven on the first sheet.
• Any manifold is homotopy equivalent to a countable CW complex (Milner?) πk (M ) are
The last two use
countable
countability

1.3 Classification of topological manifolds (proofs are not

examinable)
1.3.1 Classification of 1-dimensional manifolds
Theorem 1.9. Any connected one dimensional manifold is homeomorphic to
• R1 or
• H1
Proof. See Course website: [1] in the form of a take-home exam

Page 14 of 60
 Chapter 1 – Topological manifolds

Remark. If you allow a boundary, then you also have [0, 1], [0, 1).

1.3.2 Classification of 2-dimensional manifolds

P2 2-dimensional manifolds
• S 2 = {(x0 , x1 , x2 ) ∈ R3 | 2
i=0 xi = 1} are often called surfaces
• Π2 := R2 /Z2
• RP2 = S 2 /{x ∼ −x}

Construction(Connected sum of surfaces):

Figure 1.12: Sketch 1.14

ϕi
Let M1 , M2 be surfaces (i.e. 2-dimensional manifolds). Choose charts Mi ⊃ Ui → B1 ⊂ R2 . Let
◦ ◦ ◦ ◦ ◦
Mi = Mi \ ϕ−1 −1
i (B 21 ). Let M1 #M2 := M1 ⊥ M2 / ∼, where X ∈ M1 ∼ y ∈ M2 if x ∈ ϕ1 (∂B 12 )
and y = (ϕ−1
2 ◦ ϕ1 )(x)
Facts:
• If M1 , M2 are connected, then M1 #M2 is well defined up to homeomorphism.
• The operation of connected sum is also well defined for connected n-manifolds

• (for the future) The operation of connected sum also works in the smooth category.

Theorem 1.10 (Classification of surfaces). Every compact, connected surface is homeomorphic to

one of the following manifolds:

• S2
• Π2 # . . . #Π2
| {z }
k times

• RP # . . . #RP2 (non-orientable)
2
| {z }
l times

Page 15 of 60
 Chapter 1 – Topological manifolds

Figure 1.13: Sketch 1.15

Remark. Surfaces are classified by the following invariants:

(a) orientability
(b) Euler characteristic
For later: This classification also works in the smooth category.

1.3.3 Classification of high dimensional manifolds (not examinable at

all)
Poincaré conjecture (now theorem of G. Perelman (2003), W. Thurston (1980s)): Any
compact connected 3 dimensional manifold which is simply connected is homeomorphic to S 3 .
This paper is all about PDEs and Ricci flows.
Generalized Poincaré conjecture: Any n-manifold, which is homotopy equivalent to S n is
homeomorphic to S n . This is true in all dimensions. for n ≥ 5 Smale in the 1960s, for n = 4
Freedsen in the 1980s.
Unlike in dimension 1,2,3 the classification of n ≥ 3-dimensional manifolds is complicated and not
complete.
Example. Any finitely presented group arises as the fundamental group of a compact connected
4-manifold(Which is provably too hard).

Page 16 of 60
Chapter 2:
Smooth manifolds

2.1 Basic theory

2.1.1 Charts and atlases
Definition. Given U ⊂ Rn open, a function f : U → Rm , f = (f1 , . . . , fm ) is called smooth (or
C∞ or infinitely differentiable), if the component functions fi admit all partial derivatives
of all orders and all these partial derivatives are continuous.

In other words f smooth: ⇐⇒ ∀1 ≤ i ≤ m, α = (α1 , . . . , αn ) ∈ Nn , ∂α f := ∂xα11 . . . ∂xαnn f exists.

Remark. Given k ≥ 0, we can similarly say hat f is k-times continuously differentiable and
write (f ∈) and write f ∈ C k (U, Rm ), if for all α = (α1 , . . . , αn ) ∈ Nn , αi ≤ k ∂xα fi is continuous
P
for all i.

Definition. Let M be a topological manifold. We say that two charts (U1 , ϕ1 ), (U2 , ϕ2 ) are
smoothly compatible if the map ϕ2 ◦ ϕ−1
1 : ϕ1 (U1 ∩ U2 ) → ϕ2 (U1 ∩ U2 ) is smooth. We call
−1
ϕ2 ◦ ϕ1 a transition function.

Figure 2.1: Sketch 2.01

Definition. Let M be a topological manifold. An (smooth) atlas A of M is a collection of charts

{Uα , ϕα }α∈A such that
• the {Uα } cover M

• the charts are pairwise smoothly compatible (i.e. for all α, β ∈ A(Uα , ϕα ), (Uβ , ϕβ ) are
smoothly compatible).

Page 17 of 60
 Chapter 2 – Smooth manifolds

Definition. We say that two atlases A, A′ (on a fixed topological manifold) are equivalent, if
their union A ∪ A′ is still an atlas.
Fact(Sheet 03): This defines an equivalence relation.

Definition. A smooth manifold M = (M, [A]) consists of the following data:

(i) a topological manifolds M
(ii) an equivalence class of smooth atlases

Remark. • typically, we will designate smooth manifolds by a capital letter, e.g. M . But we
always mean (M, [A]). Note being a smooth manifolds is extra structure on a topological
space, while being a topological manifold is a property

• Using Zorn’s lemma, it can be shown that any atlas is contained in a unique maximal atlas.
Uniqueness here does not use Zorn’s lemma, only existence needs that! Equally well define a
smooth manifold to be a topological manifold and a maximal atlas. Typically we are given an
atlas, since the maximal
• ∀0 ≤ k ≤ ∞, we can define the notion of a C k -atlas, simply by requiring that the transition atlases have uncountably
functions are C k functions. This yields the definition of C k -Manifolds. Two extreme cases: mani charts, which is why
C 0 -manifold (topological manifolds) and C ∞ -manifolds. Any k ≥ 1 is not more interesting we work with equivalence
than C ∞ ! classes, rather than
maximal
Start atlases 04
of lecture
(18.10.2024)
2.1.2 First examples of smooth manifolds
Example (Example 1: The cannoical smooth manifold). Rn , n ≥ 0 is canonically a smooth
id
manifold. The canonical atlas is induced by the topological chart U = Rn , ϕ : U → Rn .
Example (Example 2: Another canonical smooth manifold). Let V be a finite dimensional real
vector space . Then V is canonically a smooth manifold. Pick a vector space basis B. This basis
induces a homeomorphism ϕB : V → Rn . If we had picked another basis B ′ , then then the
−1
transition map ϕB′ ◦ ϕB ∈ GL(n, R). Hence ϕB′ ◦ ϕ−1 B is smooth.
Pn
Example (Example 3: Spheres). We have Scn := {(x0 , . . . , xn ) ∈ Rn+1 | i=0 x2i = c2 } for c > 0.
Let ϕ±
i : Ui± → Bcn . Then
|{z}
:={(x0 ,...,xn )∈Scn |±xi >0}
 
pm −1
ϕj ◦ (ϕi ) (y1 , . . . , yn ) = ϕ±
±
p
, where (y1 , . . . , yn ) ∈ Bcn .
2−
P
j y 1 , . . . , ± c y i , . . . , y n


(y1 , . . . , yp
 n)
P
i=j
= (y1 , . . . , c2 − yk , . . . , yˆj , . . . , yn ) j>i (1)
 p P
ˆ , . . . , c2 − y k , . . . , y n )
(y1 , . . . , yj+1 j<i


We conclude {Ui± , ϕ±
i } is a smooth atlas.

Example (Example 4: Level sets). Let Φ : Rn+1 → R be a smooth function. Fix c ∈ R. Recall
that the set Φ−1 (c) = {x ∈ Rn+1 | Φ(x) = c} is called a level set of value c. Suppose that,
∀p ∈ Φ−1 (c) : D Φ(p) ̸= 0. This means that ∃0 ≤ i ≤ n s.t. ∂xi Φ(c) ̸= 0. By the
| {z }
=(∂x0 Φ(p),...,∂xn Φ(p))
implicit function theorem (Lee, Theorem C.40, Course website), there exists a neighborhood
U of p such that U ∩ Φ−1 (p) = {(x0 , . . . , f (x0 , . . . , x̂i , . . . , xn ), xn )}.

Page 18 of 60
 Chapter 2 – Smooth manifolds

Figure 2.2: Sketch 2.02

Let M = ϕ−1 (c). We define π̂i : Rn+1 → Rn , (x0 , . . . , xn ) 7→ (x0 , . . . , x̂i , . . . , xn ).

{(U, π̂i ) | U ⊂ M, π̂i |U homeomorphism, ∂xi Φ ̸= 0 on U }

Remains to check the formula:


(y1 , . . . , f, . . . , yˆj , . . . , yn )
 j>i
πˆj ◦ π̂i −1 (y1 , . . . , yn ) = (y1 , . . . , yj+1
ˆ , . . . , f, . . . , yn ) i<j

(y1 , . . . , yn ) i=j


Remark. The condition DΦ ̸= 0 is very explicit! It is very easy to generate lots of manifolds. For
example: Φ(x) = λi x2i
P

Example (Example 5: Subset of smooth manifold). Let M be a smooth manifold. Then U ⊂ M

open, is also a smooth manifold. (Take charts of M and intersect / restrict each chart)
Example (Example 6: Product of manifolds). Let M, N be smooth manifolds. Then M × N is
also a smooth manifolds. Take as charts This takes care of the
torus!
{(U × V, (ϕ, ψ)) | (U, ϕ), (V, ψ) charts of M,N respectively}

Example (Example 7: ). Let’s consider R. We define a chart R → R, x 7→ x3 . Observe that This is one to pay
attention to!
id
M = (U = R, U → R)

and
x7→x3
N = (U = R, U → R)
are smooth manifolds, which are different! Since the transition functions between them are not
smooth:
1
Indeed id ◦ (x 7→ x3 )−1 = (x 7→ x 3 ), which is not smooth!

2.1.3 Smooth maps

Definition. Let M be a smooth manifold. A map f : M → Rm is said to be smooth, if for all
p ∈ M , there exists a chart (U, ϕ) containing p, such that

f ◦ ϕ−1 : ϕ(U ) → Rm
| {z }
⊂Rn

is smooth.

Page 19 of 60
 Chapter 2 – Smooth manifolds

Definition. Let M, N be manifolds. We say f : M → N is smooth if, for all p ∈ M there exists manifolds = smooth
charts (U, ϕ) with p ∈ U ⊂ M and (V, ψ) with V ⊂ N such that: manifolds as always
(unless otherwise stated)
• V ⊃ f (U )
• ψ ◦ f ◦ ϕ−1 : ϕ(U ) → Rm is smooth
| {z }
⊂Rn

Reality check.
Lemma 2.1. Smooth maps are continuous.
Proof. Enough to show that ∀p ∈ M , there exists a neighborhood of p on which f : M → N is
continuous, for f smooth. By definition ∃(U, ϕ), p ∈ U, (V, ψ), V ⊂ N s.t.
ψ ◦ f ◦ ϕ−1 : ϕ(U ) → Rm smooth.
Observe f = ψ −1 ◦ (ψ ◦ f ◦ ϕ−1 ) ◦ ϕ on U .

Lemma 2.2. f : M → N is smooth if and only if each p ∈ M has a neighborhood U such that f |U
is smooth.
Proof. Sheet 03.

Lemma 2.3 (Properties of smooth maps). (i) Any constant map c : M → N is smootha
(ii) The identity map id : M → M is smooth

(iii) If U ◦ M open, then the inclusion i : U ,→ M is smooth

(iv) Compositions of smooth functions are smooth
a Since it sends M to a point in N

Proof. Sheet 03.

Definition. Let M, N be manifolds. A diffeomorphism f : M → N is a smooth map, which is In particular,

bijective and admits a smooth inverse. diffeomorphisms are
homeomorphism!
Example. f : R → R, x 7→ x + 3 is a diffeomorphism with inverse x 7→ x − 3.
Example. Let A ∈ GL(n, R). Define a map
fA : Rn → Rn , x 7→ Ax.
This is a diffeomorphism (smooth, since linear) with inverse fA−1 = fA−1 .
Pn
Example. Let Scn := {(x0 , . . . , xn ) | i=0 x2i = c2 } ⊂ Rn+1 . Given d > c > 0, we define a
diffeomorphism.
d
Scn → Sdn , (x0 , . . . , xn ) 7→ (x0 , . . . , xn ).
c
1
Example. M = (R, id), N = (R, x 7→ x3 ). The map M → N, x 7→ x 3 is a diffeomorphism.
Indeed,
1
(x 7→ x3 ) ◦ (x 7→ x 3 ) ◦ id−1 = id

2.1.4 The category of smooth manifolds

Definition. Let Man∞ be the category of smooth manifolds. The objects are the smooth manifolds.
The morphisms are the smooth maps.
Exercise: M, N objects in Man∞ are isomorphic if and only if they are diffeomorphic.
Observe that there is a forgetful functor: Man∞ → Man0 by (M, [A]) → M and f : M → N 7→ f .
In general:
• not full
• not essentially surjective

Page 20 of 60
 Chapter 2 – Smooth manifolds

Remark (Hierarchy of categories). • for k = 0, . . . , ∞, we can consider the category Mank

with objects C -Manifolds, and morphisms C k -maps. for k ≤ l there is a forgetful functor
k

Manl → Mank
• if k ≥ 1,then the forgetful functor Man∞ → Mank is essentially surjective. This is different
from the C 0 case. For this reason, we mainly focus on Man0 , Man∞ . This is a theorem by
Whitney
• there are other interesting categories: ManReal-analytic , ManCplx-analytic , . . ., which both come
with a forgetful functor to Man∞

Remark (Classification of manifolds (not examinable)). • all topological manifolds of dimen-

sion ≤ 3 admit a unique smooth structure

• S 7 , as a topological manifold, admits 15 pairwise non-diffeomorphic smooth structures. These

are called exotic spheres. They also exist in higher dimensions (Milan-Kervaire?)

• R4 admits uncountably many pairwise non-diffeomorphic smooth structures (Taubes 1980s)

• Open problem(Smooth 4 dimensional Poincaré conjecture): Prove or disprove: any

smooth 4-manifold, which is homeomorphic to S 4 is diffeomorphic to S 4 . Most experts believe
this is false!
Start of lecture 05
(22.10.2024)
2.1.5 Smooth manifolds with boundary
Definition. A function f : Hn ⊃ U → Rk is smooth if every p ∈ U admits an open neighborhood
p ∈ Up ⊂ Rn on which f extends to a smooth function. (i.e. there exists f˜p : Up → Rk , f˜p smooth
and f˜p |Hn ∩U = f )

Figure 2.3: Sketch 2.03

Example. n = 1, H1 = [0, ∞), f (x) = x2

√
Example (Non-Example). n = 1, H1 = [0, ∞), f (x) = x has no smooth extension to 0, since
the derivative goes to ∞.
Give a topological manifold with boundary, we can define unproblematically the notions of
• smoothly compatible charts: (U, ϕ) : M → Hn , ϕα ◦ ϕ−1
β : ϕβ (Uα ∩ uβ ) → H
n

• smooth atlases
Definition. A smooth manifold with boundary M = (M, [A]) is the data of
• a topological manifold with boundary

Page 21 of 60
 Chapter 2 – Smooth manifolds

• an equivalence class of atlases

Remark. Every smooth manifold is a smooth manifold with boundary. This is an enlargement of Similarly we cna
Man∞ . generalise even more to
manifolds with corners ...
2.2 Partitions of unity
2.2.1 Preparatory lemmas This section is technical,
but also very important!
Lemma 2.4. The function f : R → R,
( 1
e− t t>0
f (t) =
0 t≤0

is smooth.

Figure 2.4: Sketch 2.04

Proof. It is enough to proof, that f has well defined derivatives of all orders, since f is a function
on R.
f 0 = f , for k ≥ 1, assume
1. f (k−1) exists
2. f (k−1) |(−∞,0] = 0
1
3. f (k−1) |(−∞,0] (t) = Pk−1 ( 1t )e− t for some polynomial P(k−1) .
Clearly this holds for k = 1.
We have
f (k−1) (t) − f (k−1) (0) f (k−1) (t)
lim = lim
t→0+ t t→0+ t
1 1 1
= lim+ P(k−1) ( ) e− t
t→0 t t
= lim P(k−1) (x) · x · e−x = 0
x→∞

′
Therefore f (k−1) is differentiable at the origin, the derivative f (k−1) (0) = 0. and
f (k−1) |(−∞,0] = 0. Therefore f (k−1) is differentiable. Therefore we only have to check 3., which
only takes place on R+ !
Finally  
1
− 1t 1 1
f (k−1) |(0,∞) (t) = P(k−1) ( 1t )e− t =⇒ P(k−1)
′
( 1t ) −1
t2 e + P(k−1) ( 1t )e− t =: P(k) ( 1t )e− t .

Lemma 2.5. Fix real numbers r1 < r2 . Then there exists a smooth function h : R → R such that

1. h ≡ 1 on(−∞, r1 ]

Page 22 of 60
 Chapter 2 – Smooth manifolds

2. 0 < h < 1 on r1 , r2

3. h ≡ 0 on [r2 , ∞)
f (s2 −t)
Proof. h(t) := f (s2 −t)+f (t−s1 ) , since the denominator never goes to 0.

Lemma 2.6 (Existence of cutoff functions). Given 0 < r1 < r2 , there exists a smooth function
H : Rn → R such that
1. H ≡ 1 on Br1

2. 0 < H < 1 on Br2 \ Br1

3. H ≡ 0 on Rn \ Br2
p
Proof. Set H(x) := h(|x|), where h is defined as in lemma 2.5. (Recall: |x| := x21 + · · · + x2n ).
Then H is smooth, since it is a composition of smooth functions on Rn \ Br1 and constant on
B r1 .

2.2.2 Partitions of unity

Definition. Given a topological space X and a function f : X → R, the support of f is the set

supp(f ) := {x ∈ X | f (x) ̸= 0} ⊂ X

Example. If f : R → R has the form f (x) = a0 + a1 x + . . . , an xn =⇒ supp(f ) = R. In fact, by

Taylor’s theorem, if f analytic, then supp(f ) either R or ∅. In contrast, the function h : R → R
defined in lemma 2.5 has support (−∞, r2 ] ⫋ R.

Definition. Let M be a smooth manifold. Let {Uα }α∈A be an open cover. A

partition of unity subordinate to the cover is the data of a collection of smooth functions
{ψα }α∈A , ψα : M → R such that
(1) 0 < ψα < 1

(2) supp(ψα ) ⊂ Uα
(3) {supp(ψα )}α∈A is locally finite
P
(4) α∈A ψα ≡ 1

Remark. There is an analogous notion in the category Top,Man0 , Mank , etc.,. . .

Example. M = R, U1 = (−∞, r2 + 1), U2 = (r1 − 1, ∞), where r1 < r2 as in lemma 2.5.
Similarly let h as in lemma 2.5. and set ψ1 = h, ψ2 = 1 − h

Figure 2.5: Sketch 2.05

Theorem 2.7 (Existence of partitions of unity). Let M be a smooth manifold. Let {Uα }α∈A be
an open cover. Then there exists a partition of unity subordinate to this cover.

Page 23 of 60
 Chapter 2 – Smooth manifolds

Remark. The same theorem works in Top, Man0 , Mank . It will not work in
ManAnalytic , ManCplx-Analytic , Varieties /C.

Proof. Step 1: Construction of the Vi An open supset U ⊂ M is called a

regular coordinate ball if there exists U ⊂ Ũ, (Ũ, ϕ̃) a chart such that ϕ̃(U ) = Br1 , ϕ̃(Ũ ) = Br2 .

Figure 2.6: Sketch 2.06

By lemma 1.6 M admits an exhaustion by compact sets. By lemma 1.7, given any basis, any open
cover, one can find a locally finite, countable basis refinement of this cover by basis elements.
Claim: {regular coordinate balls whose closure is contained in some Uα } basis of M The claim is easy to
These tree points imply that {Uα }α∈A admits a countable, locally finite refinement by regular verify
coordinate balls {Vi }i∈I .
By sheet 2, exercise 1 (a) {Vi } is still locally finite.
Step 2: Construction of the fi For each Vi ∃Vi ⊃ Ṽi , ϕ̃i : Ṽi → Rn such that
ψ̃i (Vi ) = Br1i , ψ̃i (Ṽi ) = Br2i with 0 < r1i < r2i , Ṽi ⊂ Uα for some α. Using lemma 2.6, let
Hi : Rn → R be a cutoff function, i.e. Hi |Br1 > 0, H = 0 on R \ Br1i . Let us set Finging Ṽi s.t. Ṽi ⊂ Uα is
(
Hi ◦ ϕ̃i on Ṽi the reason we considered
fi : M → R, fi = regular coordinate balls
0 M \ Vi
P whose
Step 3: Construction of the gi Let us set f = i∈I fi . This is well defined by local finiteness closure is contained in
P
of the Vi Note also that f > 0.. We set gi = fi /f . Then clearly we have 0 ≤ gi ≤ 1, i∈I gi ≡ 1 some Uα
Step 4: Reindexing and conformation Since Ṽi ⊂ Uα , for some α, we can choose for each
i ∈ I, α(i) ∈ A s.t. Vi ∈ Uα(i) . Let us set Here the empty sum is 0
X
ψα := gi
i|α=α(i)

Observe for (2):

Exercise 2.1
[ [
supp(ψα ) = Vi = Vi ⊂ Uα
α(i)=α α(i)α

We still have 0 ≤ ψα ≤ 1, which is (1)

and supp(ψα ) are locally finite: for each op ∈ M , since {Vi } locally finite, there exists a
neighborhood Up of p which only intersects finitely many of the {Vi }, call them V1 , . . . , Vk . Then
the only ψα which have a chance of being non-zero must satisfy α ∈ {α(1), . . . , α(k)} (this is (3)).
Lastly  
X X X X
ψα =  gi  = gi ≡ 1,
α∈A α i:α=α(i) i∈I

which confirms (4).

Start of lecture 06
(25.10.2024)
Page 24 of 60
 Chapter 2 – Smooth manifolds

2.2.3 Applications of partitions of unity

Definition. Let X be a topological space. Let A ⊂ X be closed, U ⊂ X, A ⊂ U be open. A
bump function for A supported in U is a function

ϕ:X→R

such that ψ |A ≡ 1, supp(ϕ) ⊂ U .

Figure 2.7: Sketch 2.07

Proposition 2.8. Let M be a smooth manifold. Fix A ⊂ M closed, U ⊂ M, A ⊂ U ⊂ M open.

Then there exists a smooth bump function for A supported in U
Proof. Let V = M \ A. Then {U, V } is a covering and by theorem 2.7, there exist {ΨU , ΨV }
partitions of unity subordinate to this cover. Now ΨU does the job.
Definition. Let M, N be smooth manifolds. Let A ⊂ M be closed. We say that f : A → N is
smooth if it admits a smooth extension in a neighborhood of each point p ∈ A. I.e. for any p ∈ A there
exists Up ∋ p, a smooth
Proposition 2.9. Let M be a smooth manifold. Let A ⊂ M be closed and f : A → Rk , k ≥ 0 function f˜p : Up → N s.t.
be smooth. Then for any open U ⊂ M, A ⊂ U , there exists f˜ : M → Rk , such that f˜ |A = f and f˜p |Up ∩A = f |Up ∩A
supp(f˜) ⊂ U

Remark. This would be false if we replaced Rk by an arbitrary smooth manifold N . E.g. take
f =id
R2 ←- A = S 1 → S 1
Proof. For each p ∈ A, choose a neighborhood p ∈ Wp ⊂ U , f˜p : Wp → Rk smooth extension of We maybe need Wp ⊂ U ?
f|A∩Wp . Then observe that {Wp }p∈A ∪ (M − A) forms an open cover of M . {ψp }p∈A ∪ ψ0 be a Prob. not?
partition of unity subordinate to the cover. Now we set f˜ = p∈A ψp f˜p . By local finiteness f˜ is
P
smooth. Also
X
f˜|A = ψp|A f˜p |A
p∈A |{z}
=f
X
=f ψp|A = f|A · 1 = f|A .
p∈A

Definition. Let X be a topological space. An exhaustion function f : X → R is a continuous

function such that ∀C ∈ R, f −1 (−∞, c] is compact. If X is compact every f
is an exhaustion function
f
Example. X = R, f : R → R, x 7→ x2 ...
Example (NON-EXAMPLE). X = R, f (x) = x

Page 25 of 60
 Chapter 2 – Smooth manifolds

Proposition 2.10. Every smooth manifold admits a smooth exhaustion function.

1
Proof. Pick a countable partition of unity {Ui }i∈N+ by open
Psubsets having compact closure .
Let {Ψi }i∈N+ be a subordinate partition of unity. Let f := i∈N+ iψi .
Observe that for any c ∈ R, c < N ∈ N that
N
[
−1 −1
f (−∞, c] ⊂ f (−∞, c] ⊂ Ui
i=1
SN SN
Why f −1 (−∞, c] ⊂ i=1 Ui ? Let q ̸∈ i=1 Ui . Then
N
X ∞
X
f (q) = iψi (q) + iψi (q)
i=1 i=N +1
| {z }
=0
∞
X ∞
X
≥ (N + 1) ψi (q) = (N + 1) ψi (q)
i=N +1 i=1
| {z }
=1
=N +1

Proposition 2.11. Let M be a smooth manifold. Let A ⊂ M be a closed subset. Then there exists
a smooth function
f : M → R, f −1 (0) = A
In fact, the prove shows
one can assume f ≥ 0
E.g. take M = R, A = Cantor set, shows that this is non-trivial.
Proof. Assume M = Rn (general case: Sheet 04).
Choose a countable cover of Rn \ A by balls {Bri (xi )}∞
i=1 with ri < 1. By Lemma 2.6 there exists
a cutoff function
H : Rn → R
s.t. H ≡ 1 on B 12 (0) and 0 < H < 1 on B1 (0) \ B 12 (0) and H ≡ 0 on Rn \ B1 (0). For each i ≥ 1
let Ci ≫ 1 be large enough so that
∈Nn
z }| {
Ci > sup{∂xα H | α = (α1 , . . . , αn ), |α| ≤ i}

Let
∞
rii
 
X x − xi
f := H .
i=1
2i ci ri
 
We need to argue that f is smooth. Observe that, since ri < 1 2ri ci i H x−x
ri
i
≤ 21i It follows from
Analysis 2 that f is continuous. To prove that f is smooth assume for k ≥ 1 that all partial of
order k < 1 exist and are continuous. If |α| = k, then
i
rii−k α
   
α ri x − xi x − xi
∂ i H = i ∂ H
2 Ci ri 2 Ci ri
If i > k, then

rii−k α
 
x − xi 1
∂ H <
2i Ci ri 2i

rii
  
x−xi
Again follows by Analysis 2 that ∂ α f exists and equals ∂α
P
2i Ci H ri .

1 Like in the proof of 2.7

Page 26 of 60
Chapter 3:
Tangent Vectors

3.1 Motivation
Consider the following pictures

Figure 3.1: Sketch 3.01

Page 27 of 60
 Chapter 3 – Tangent Vectors

Figure 3.2: Sketch 3.02

Ap the affine hyperplane tangent to S 1 (Π2 ) at the point p. Let Tp M := Ap − p ⊂ Rn+1 . This is a
vector subspace of Rn+1 . It is called the tangent space of M at p. Consider
a
TM = Tp M,
p∈M

called the tangent bundle. Observe that there is a map Think of π as a map of
p, Tp M
π : TM → M

by
x ∈ Tp M 7→ p
π
the data T M → forms a vector bundle.
Problems with this approach:

• not very intrinsic (depends on Rn+1 . . . )

• need to prove that manifolds can always embedded into RN

This is really the picture / intuition we should have, but we will construct it in a different way.

3.2 Two (equivalent) theories of tangent vectors

3.2.1 Definition via equivalence classes of smooth curves
Let M be a smooth manifold. Fix p ∈ M . I could not quite make
out what he called this
Definition. The tangent space of M at p denoted by Tp M is the set of equivalence classes of chapter, so I named it
smooth curves gamma : [−ϵ, ϵ] → M, γ(0) = p with γ1 ∼ γ2 ⇐⇒ for any smooth function f according to [4]
defined neara p, we have (f ◦ γ1 )′ (0) = (f ◦ γ2 )′ (0). Here the ϵ > 0 is any positive real number,
which depends on γ.
a in a neighborhood of

Page 28 of 60
 Chapter 3 – Tangent Vectors

Figure 3.3: Sketch 3.03

Definition. Given a smooth map F : M → N , let

dFp : Tp M → TF (p) N

be given by This is clearly well

[γ] 7→ [F ◦ γ]. defined
This map dFp is called the differential of F at p.

Remark. The map is also called the tangent map of M at p and the total derivative. It is
also denoted by
DFp , T Fp , ∇Fp , Fp′ , DF (p), T F (p), . . .

Lemma 3.1 (Fundamentality of the differential). Let F 1 : M1 → M2 , F 2 : M2 → M3 smooth.

Then:
(i) dFF2 1 (p) ◦ dFp1 = d(F 2 ◦ F 1 )p

(ii) If F : M → M is the identity, then dFp = id

Proof. Exercise.
Start of lecture 07
(29.10.2024)
Lemma 3.2. Let γ : [−ϵ, ϵ] → Rn and σ : (−δ, δ) → Rn with γ(0) = σ(0) = p ∈ Rn . Then
γ ∼ σ ⇐⇒ γ ′ (0) = σ ′ (0)
| {z }
(γ1′ (0),...γn
′ (0))∈Rn

Proof. By abusive notation, we denote by xi the map Rn → R, (x1 , . . . , xn ) 7→ xi .. If γ ∼ σ, then xi might be better (in the
′ Def. ′
γ i (0) = (xi ◦ γ)′ (0) = (xi ◦ σ)′ (0) = σ i (0) =⇒ γ ′ (0) = σ ′ (0). sense of the dual space),
but xi is used in practice

Page 29 of 60
 Chapter 3 – Tangent Vectors

Conversely, suppose σ ′ (0) = γ ′ (0). Given any f smooth defined near p, we have
′ ′
(f ◦ γ)′ (0) = (∂x1 f (p), . . . , ∂xn f (p)) · (γ 1 (0), . . . , γ n (0))
′ ′
= (∂x1 f (p), . . . , ∂xn f (p)) · (σ 1 (0), . . . , σ n (0))
= (f ◦ σ ′ )(0).

Corollary 3.3. Let V be a finite dimensional R vector space. Then, for any p ∈ V , the canonical
map

V → Tp V
w 7→ [t 7→ p + tw]

is a bijection.
Proof. If V = Rn , then this is immediate from lemma 3.2. In general pick a basis to define an
isomorphism1 F : V → Rn . Then the following diagram commutes:

Figure 3.4: Sketch 3.04

using lemma 3.1.

3.2.2 Definition via derivations

Definition. Let M be a smooth manifold. A derivation at p ∈ M is a linear map

ν : C ∞ (M ) → R

satisfying the property

ν(f g) = f (p)ν(g) + ν(f )g(p), (1)
which is also called the Leibniz rule.

Remark. Here C ∞ (M ) is the set of smooth functions from f : M → R. It is naturally an R-vector

space. Similarly we have C 0 (M ) the space of continuous functions f : M → R and C k (M ) the
space of k-times differentiable function f : M → R.

Definition. The set of derivations at p shall be also called the tangent space of M at p, denoted
by Tp M .

Lemma 3.4. Tp M is naturally a vector subspace of C ∞ (M )∨ C ∞ (M )∨ denotes the

dual space of C ∞ (M )
1 In particular a diffeomorphism

Page 30 of 60
 Chapter 3 – Tangent Vectors

Proof. Given derivations ν1 , ν2 ∈ Tp M we must show that aν1 + ν2 is still an element of

Tp M ∀a ∈ R. We compute we compute
(aν1 + ν2 )(f g) = aν1 (f g) + ν2 (f g) = a[ν1 (f )g(p) + f (p)ν1 (g)] + [ν2 (f )g(p) + f (p)ν2 (g)]
= f (p)[aν1 + ν2 ] + [aν1 + ν2 ](f )g(p)

Definition. Given a smooth map F : M → N , we let dFp : Tp M → TF (p) N be the map

ν 7→ dFp (ν) := C ∞ (N ) ∋ f 7→ ν(f ◦ F )

Lemma 3.5. (i) the previous definition gives a derivation

(ii) dFF2 1 (p) ◦ dFp1 = d(F 2 ◦ F 1 )p By (ii) and (iii) d is a
Functor
(iii) If F : M → M is the identity, then dFp = id

Lemma 3.6. Let ν be a derivation at p ∈ M . Then

(a) f ≡ C, then ν(f ) = 0. That is ν annihilates constant functions.
(b) if f (p) = g(p) = 0, then ν(f g) = 0
Proof. (a): Since ν is linear, it is enough to prove ν(f ) = 0 for f ≡ 1. But then

ν(f ) = ν(f 2 ) = f (p)ν(f ) + ν(f )f (p) = 2ν(f ).

(b) is obvious by the Leibniz rule (1).

Lemma 3.7. Let V be a finite dimensional vector space over R. A derivation ν ∈ Tp V is entirely
determined by its action on any dual basis {ξ 1 , . . . , ξ n }.. This should remind us of
lemma 3.2
Proof. Fix a basis ({e1 , . . . , en }) to identify V ≡ Rn . It is enoughPn to show that ν(f ) = 0 if
{∂x1 f (p), . . . , ∂xn f (p)} all vanish(Indeed, consider f → f − k=1 ∂xi f (p)ξi ). By Taylor’s formula with ξi : Rn → R as
(Appendix C.15, Lee), we have before
X n Xn Z 1
f (x) = f (p) + ∂xi f (p)(xi − pi ) + (xi − pi ) (xj − pj ) (1 − t)∂xi xj f (p + t(x − p))dt .
|{z}
i=1
| {z }
i,j=1 constant at p | 0
constant | {z } {z }
constant at p
=0

Then by lemma 3.6 ν(f ) = 0.

Corollary 3.8. The canonical map V → Tp V , p ∈ V defined by

d
w 7→ (C ∞ (V ) ∋ f 7→ f (p + tw))
dt |t=0

is an isomorphism of vector spaces. This should remind us of

corollary 3.3
Proof. We define These are really
canonically equal! No
Tp V → V choice needed
n
X
ν 7→ ν(ξ i )ei , ξ i : V → R
i=1

By lemma 3.7 this map is injective and hence dim Tp V ≤ dim V . So it is enough to show that
V 7→ Tp V is also injective. Suppose for contradiction that V ∋ w ̸= 0, that maps to the zero
derivation.
d
0= f (p + tw)∀f
dt |t=0
d d
=⇒ 0 = w∨ (p + tw) = t=1
dt dt |t=0
Page 31 of 60
 Chapter 3 – Tangent Vectors

3.2.3 Both definitions agree

(1)
Temporary notation: Let Tp M (1) , dFp ,. . . , be those objects defined in section 3.2.1 and
(2)
Tp M (2) , dFp ,. . . , the analogous objects defined in 3.2.2
Key observation: There is a canonical map, for any p ∈ M ,

KP : Tp M (1) → Tp M (2)
γ 7→ (C ∞ (M ) ∋ f 7→ (f ◦ γ)′ (0)).
(1)
Note that this commutes with dF (i) , i.e. dF (2) ◦ Kp = KF (1) (p) ◦ dFp (exercise).

Proposition 3.9. Kp is a bijection.

Proof. Choose a chat (U, φ), p ∈ U . Then we have a map
U → φ(U ) ⊂ Rn

Figure 3.5: Sketch 3.05

Finally, we have

Figure 3.6: Sketch 3.06

Page 32 of 60
 Chapter 3 – Tangent Vectors

3.3 Coordinates
Definition. (1) Given a point p ∈ Rn let (∂xi )p ∈ Tp Rn be the vector represented by the curve
t 7→ p + t (0, . . . , 1, . . . , 0).
| {z }
ei
(2) Given p ∈ M , we shall abusenotation by writing (∂xi )p := dφ−1
φp (∂xi )p for some chart ((U, ϕ))

Remark. 1. Various authors also write ∂xi (p)

2. {(∂x1 )p , . . . , (∂xn )p } form a basis for Tp M , by construction
3. {∂x1 , . . . , ∂xn } very much depend on the chart (U, φ)

Suppose now that F : M → N smooth map. Let (U, φ), (V, ψ) be charts, F (U ) ⊂ V . Let
p̂ := ϕ(p) ∈ Rm . Then we have

Figure 3.7: Sketch 3.07

where F̂ = ψ ◦ F ◦ φ−1 .
Note that dF̂p̂ : Tp̂ Rm → TF̂ (p̂) Rn is a linear map. We want to find an expression ofr the matrix
dF̂p̂ w.r.t the basis {∂x1 , . . . , ∂xm } and {∂y1 , . . . , ∂yk }.
Well, by definition

dF̂p̂ ((∂xi )p̂ ) := [F̂ (p̂ + (0, . . . , 1, 0, . . . 0))]

X n
= ∂xi F j (p̂)(∂yj )F̂ (p̂)
j=1

and therefore
∂x1 F̂ 1 (p̂) ∂xm F̂ 1 (p̂)
 
···
dF̂p̂ =  .. ..
.
 
. .
∂x1 F̂ n (p̂) · · · ∂xm F̂ n (p̂)
Start of lecture 08
(05.11.2024)
Remark. By abuse of notation we often write F ≡ F̂, p ≡ p̂, ∂xi f ≡ ∂xi F̂, dFp ≡ dF̂p̂

Page 33 of 60
 Chapter 3 – Tangent Vectors

Figure 3.8: Sketch 3.08

Remark. dF̂· : x
|{z} 7→ dF̂x ∈ Mat(n × m) ≡ Rn×m . This it clearly a smooth map.
∈ϕ(U )⊂Rm

3.4 The tangent bundle

`
Definition. Given a smooth manifold M , let TM := p∈M Tp M . We write elements of TM as
pairs (p, v), where v ∈ Tp M . Note that we have a map

π : TM → M, (p, v) 7→ p.

Remark (Added by Manuel, was an answer to my question). For p ∈ M the preimage of p under
π is called a fiber. He also highlighted, the condition that π −1 (p) is a vector space (namely Tp M ),
which seems to be important in our context, but not generally required for fibers.
A priori, TM is just a set. We will exhibit natural smooth manifold structure.
Special case: M ⊂ U Rn . Then
a
T U := Tp U ≡ U × Rn
p∈U

(t 7→ p + tv) 7→ (p, v)
Remember that this is a
General construction Given a smooth chart (U, ϕ) for a smooth manifold M , we have a map dϕ canonical identification!

Figure 3.9: Sketch 3.09

Page 34 of 60
 Chapter 3 – Tangent Vectors

where
dϕ(p, v) := (ϕ(p), dϕp (v)).
Define a subset S ⊂ TM to be open, if, for any chart U, ϕ, dϕ(S ∩ T U ) open in
T ϕ(U ) ≡ ϕ(U ) × Rn . This is a pullback

Lemma 3.10. This prescription defines a topological space on TM. Moreover, TM is a topological
manifold.
Proof. Omitted. Check transition maps

dψ ◦ dϕ−1 : ϕ(U ∩ V ) × Rn → ψ(U ∩ V ) × Rn

is it an elementary, but tedious proof.

Remark. Alternatively define the same topology on TM by taking the basis the union over all
charts (U, ϕ) in your atlas of {dϕ−1 (V ) | V ⊂ T (ψ(U )) open}.

To make TM into a smooth manifold, we take as our atlas the set {(T U, dϕ)}(U,ϕ) , where (U, ϕ)
runs over the smooth charts of M .
Lemma 3.11. this is a smooth atlas.
Proof. Fix charts (U, ϕ), (V, ψ). Then the transition functions take the form

Figure 3.10: Sketch 3.11

Check if both components are smooth:

• The first component x 7→ ψ ◦ ϕ−1 (x) is smooth, since M is a smooth manifold an
(U, ϕ), (V, ψ) are smooth Since we can write the
second component as a
• For the second component can be fractured as follows: concatination of maps, it
is smooth
(x, v) 7→ (d( ψ ◦ ϕ−1 , v)) 7→ d(ψ ◦ ϕ−1 )x v
| {z x }
∈Man(n×n)≡R2n

Exercise: the map Mat(m × n) × Mat(n × p) → Mat(m × p) by A, B 7→ AB is smooth.

Remark. We will see later that (ϕ : TM → M ) forms a vector bundle. It can be shown that given
F : M → N the map dF : TM → TN, (p, v) 7→ (F (p), dFp (v)) is smooth.. This is the exact same
In fact, we have computation as in the
proof of lemma 3.11

Page 35 of 60
 Chapter 3 – Tangent Vectors

Figure 3.11: Sketch 3.12

commutes. This can be restated as follows: There is a functor Man∞ → Smooth vector bundles by

M 7→ (π : TM → M )

F : M → N 7→ dF : TM → TN

Page 36 of 60
Chapter 4:
Submersions, immersions and
embeddings

4.1 Basic definitions

Definition. Let F : M → N be smooth. The rank of F at p ∈ M is the rank of the linear map
dFp : Tp M → TF (p) M .

Smooth maps, which have full rank (highest possible rank, i.e. rankF = max(m, n)) are
particularly important:
Definition. Let F : M m → N n be smooth. We say M m , N n means M, N are
m, n dimensional
• F is a submersion if dFp is surjective, for all p ∈ M (m ≥ n) manifolds
• F is an immersion if dFp is injective, for all p ∈ M (m ≤ n)

Lemma 4.1. Given (m, n) ∈ N+ × N+ , let Mat(m × n) ≡ Rm×n . The subset Mat(m × n)full rank :=
{A ∈ Mat(m × n) | A has full rank} is open in Mat(m × n).

Proof. Fix M ∈ Mat(m × n)full rank . Without loss of generality m ≤ n, otherwise apply
Mat(m × n) → Mat(n × m), A 7→ AT . By definition there exists a submatrix M ′ , obtained by
deleting n − m columns, which is invertible. Now the map M is fixed and F depends
on M , but it does not
F :M 7→M ′ det(·)
Mat(m × n) → Mat(m × m) → R matter here!

is continuous, since booth the forgetful F and det is smooth.

M ∈ (det ◦F )−1 (R \ {0}) ⊂ Mat(m × n)full rank
| {z }
open

since M was arbitrary this completes the proof.

Figure 4.1: Sketch 4.00

Page 37 of 60
 Chapter 4 – Submersions, immersions and embeddings

Lemma 4.2. Fix F : M m → N n , p ∈ M . The property of full rank

is stable under small
1. If dFp is injective, then there exists a neighborhood of p on which dF· is injective. pertubation!
2. If dFp is surjective, then there exists a neighborhood of p on which dF· is surjective.

Proof. This is a local statement. We can therefore assume that M, N are open subsets of Rm , Rn
respectively. Then
dF(·) : M → Mat(n × m)
is smooth, hence continuous. By assumption dFp ∈ Mat(n × m)full rank . But Mat(n × m)full rank ,
so the preimage is open (by lemma 4.1) and contains p.

Remark. important: contains both

a definition and a
1. If F : M → N is both an immersion and a submersion, then we say that F is a counterexample!
local diffeomorphism. We will see (by the rank theorem 4.3) that F is a local diffeo-
morphism ⇐⇒ ∀p ∈ M ∃p ∈ U : F|U is a diffeomorphism.
2. be warned. local diffeomorphism need not be global:

R2 ≡ C ⊃ S 1 ={|z| = 1} →S 1

(x, y) 7→ x + iy z 7−−−−−−→ z 2

Definition. An immersion is an embedding if it is a homeomorphism onto its image with the

subspace topology.

Figure 4.2: Sketch 4.01

Example. Another example:

X
S n = {(x0 , . . . , xn ) ∈ R1+n | x2i = 1} ⊂ R1+n

with
i : S n ,→ R1+n
Non-examples

Page 38 of 60
 Chapter 4 – Submersions, immersions and embeddings

Figure 4.3: Sketch 4.02

parametrized by
t 7→ (sin t, sin 2t)
and

R 7→ R2 /Z2 = S 1 × S 1
t 7→ (t, αt), α ∈ R \ Q

Can show1 that the image is dense. It is an immersion, but no a homeomorphism!

Start of lecture 09
(08.11.2024)

4.2 The rank theorem

Theorem 4.3 (rank theorem). Let F : M m → N n be a smooth map of constant rank r. For each This is arguably the most
p ∈ M , there exist charts (U, φ) : p ∈ U and (V, ψ) : F (U ) ⊂ V , such that important result of the
first half of the course.
F̂ := ψ ◦ F ◦ φ−1 : There is a lot of results in
[2], what is actually
useful? Implied answer:
Figure 4.4: Sketch 4.03 Rank theorem
F̂ takes the form

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

Remark. By lemma 4.2, if F has full rank at p ∈ M , then Up to diffeomorphism

there is only one map of
• if m = r ≥ n, then F is an submersion near p and constant, full, rank
F̂ (x1 , . . . , xn , xn+1 , . . . , xm ) = (x1 , . . . , xn )

• m = r ≤ n, then F is an immersion near p, and

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

• m = n =⇒ up to the diffeomorphism, F̂ is just the identity

Remark. This theorem is a non-linear generalization of the following linear algebra fact: L :
≃ ≃
V m → W n , then there are linear maps φ : V m → Rm , ψ : W n → Rn , such that L̂ := ψ ◦ L ◦ ϕ−1
takes the form
L̂(x1 , . . . , xr , xr+1 , . . . , xm ) = (x1 , . . . , xr , 0, . . . , 0),
where r = rank(L).
1 not obvious, non-examable

Page 39 of 60
 Chapter 4 – Submersions, immersions and embeddings

Proof of theorem 4.3. Step 0: We might as well assume that M = U ⊂ Rm , N = V ⊂ Rn , since see [2]
we only make a local statement up to diffeomorphism. We may also assume, up to reordering the
coordinates, that the matrix (∂xi F j (p))1≤i,j≤r is invertible for p ∈ U . We label our coordinates:
source coordinates in U
(x1 , . . . , xr , y1 , . . . , ym−r )
Tarkget coordiantes
(v1 , . . . , vr , . . . , w1 , . . . , wn−r )
and wlog F (0, 0) = (0, 0).
We write F (x, y) = Q(x, y) , R(x, y) . Notice that (∂xi Qj ) is non-singular.
| {z } | {z }
v-coordinates w-coordinates
Step 1: Define φ : U → Rm , φ(x, y) = (Q(x, y), y). Then
∈Mat(r×r) 
z }| {
 ∂ Qj ∂yi Qj 
dφ(0,0) =  xi
 
0 1
|{z}

∈Mat((n−r)×(n−r))

=⇒ by the inverse function theorem, there exist connected neighborhoods

U0 ⊂ U, U˜0 ⊂∈ Mat((n − r) × (n − r))φ(U ), such that φ|U0 : U0 → U˜0 . We may as well assume
that U˜0 is a cube, i.e. (−ϵ, ϵ)n .

Figure 4.5: Sketch 4.04

While φ−1 (x, y) = (A(x, y), B(x, y)), for some A : U˜0 → Rr , B : U˜0 → Rm−r . We compute
x=Q(A(x,y),B(x,y))
(x, y) = φ ◦ φ−1 (x, y) = φ (A(x, y), B(x, y)) = (Q(A(x, y), B(x, y)), B(x, y)) =⇒ y=B(x,y)

Hence φ−1 (x, y) = (A(x, y), y).

Step 2: Observe that

F ◦ φ−1 (x, y) = (Q(φ−1 (x, y)), R(φ−1 (x, y))) = (x, R̃(x, y)),

where
R̃(x, y) = R(φ−1 (x, y)).
Then
 
∈Mat(r×r)
z}|{
−1
 1 0 
d(F ◦ φ ) = 
 
∂xi R̃(x, y)
j
∂yi R̃j 

| {z }
∈Mat((m−r)×(m−r))

But the rank of d(F ◦ φ−1 ) is r, because φ−1 is a diffeomorphism and F has rank r
Page 40 of 60
 Chapter 4 – Submersions, immersions and embeddings

• Since 1r×r has rank r, we must have ∂yi R̃ ≡ 0

We write S(x) := R̃(x, y), we now have

F ◦ φ−1 (x, y) = (x, S(x)) (1)

. to make clear R̃ does not

Step 3: Recall really depend on y

F : U → V ⊂ Rn
F (0, 0) = (0, 0)

Let V0 ⊂ V be defined as follows:

V0 := {(v, w) ∈ V | (v, 0) ∈ Ũ0 }

By (1), F ◦ φ−1 (Ũ0 ) ⊂ V0 . Hence F (U0 ) ⊂ V0 . Set ψ : V0 → Rn , ψ(v, w) = (v, w − S(v)). Clearly S(v) makes perfect sense,
ψ is a diffeomorphism, since since both x, v have r
(v, w) 7→ (v, w + S(w)) entries
is am inverse. =⇒ (V0 , ψ) is a smooth chart.

F̂ := ψ ◦ F ◦ ϕ−1 = Ψ(x, S(x)) = (x, S(x) − S(x)) = (x, 0)

Remark. This is one theorem you should really not forget! If you continue to think about Manifolds
in your life, this is really useful! Do not remember the proof, remember the statement!

Page 41 of 60
Chapter 5:
Submanifolds

5.1 Basic definitions

Definition. Let M be a topological manifolds. A subset S ⊂ M is a topological submanifold,
if S is a topological manifold with the subspace topology.

Example. S n = {(x0 , . . . , xn ) | x2i = 1} ⊂ R1+n

P

Example (Non-example). {(x, y) | x = 0 ∨ y = 0} ⊂ R2 , since this is not a manifold (see sheet

01).

Definition. Let M be a smooth manifold. A topological submanifold S ⊂ M is a

smooth submanifold, if it is equipped with a smooth structure, s.t. the embedding i : S ,→ M is
smooth.
Example. If M is a smooth manifold and U ⊂ M open, then U ⊂ M is a smooth manifold. With the restricted
smooth structure of M
Remark. Some authors (including Lee’s textbook) use the term embedded submanifold to dis-
tinguish from immersed submanifolds. For use “submanifolds”≡“embedded submanifold”.

Lemma 5.1. Suppose that f : M → N smooth embedding. Let S := F (M ) ⊂ N . Then S

admits a unique smooth structure making it a smooth submanifold, with the property that f is a
diffeomorphism onto its image.
Proof. By definition of f being an embedding, f is a homeomorphism onto it’s image, with the
subspace topology. =⇒ S is a topological manifold.
We define a smooth atlas on S by taking {(f (U ), φ ◦ f −1 )}, as (U, φ) ranges over the set of charts
for M .
Clearly f is a diffeomorphism, since φ ◦ f ◦ f −1 ◦ ψ −1 , for (U, φ), (V, ψ) smooth charts, this
follows from the fact that (U, φ), (V, ψ) are smoothly compatible on M .
This is the only smooth at with the property that f is a diffeomorphism, if B is an other such
atlas, then the fact that f is a diffeomorphism for (S, B) ⇐⇒ (S, A) compatible.
Finally

Page 42 of 60
 Chapter 5 – Submanifolds

Figure 5.1: Sketch 5.01

so i is a smooth embedding.
Definition. A embedded submanifold S is called properly embedded, if the inclusion map i ,→ N
is proper (i.e. the preimage of a compact set is compact).

Example. S n ,→ Rn+1 properly embedded.

Example (Non-example). S n \ {pt} ⊂ Rn+1

Figure 5.2: Sketch 5.02

Lemma 5.2. A topological submanifold S ⊂ N is properly embedded iff S is closed.

Proof. Exercise. Elementary exercise in

point set topology
Page 43 of 60
 Chapter 5 – Submanifolds

5.2 The “slice lemma”

Theorem 5.3 (Slice lemmaa ). (a) Suppose S k ⊂ M n is a submanifold of codimension n − k.
Then, for all p ∈ S, there exists a chart (V, ψ), p ∈ U ⊂ N , such that this is also a definition of
codimension:
ψ(V ∩ S) = {(x1 , . . . , xk , xk+1 , . . . , xn ) ∈ ψ(V ) | xk+1 = ck+1 , . . . , xn = cn }. dim M − dim S

Figure 5.3: Sketch 5.03

Start of lecture 10
(12.11.2024)
(b) Suppose that S ⊂ N is a subset with the property that, for all p ∈ S, there exists a slice chart The converse of (a)
(V, ψ), p ∈ V ⊂ N , such that

ψ(V ∩ S) = {(x1 , . . . , xk , xk+1 , . . . , xn ) ∈ ψ(V ) | xk+1 = ck+1 , . . . , xn = cn },

then S admits a smooth manifold structure making it a smooth submanifold of N .

a Lee [2] calls it a theorem

Remark. • We get an equivalent theorem by requiring ck+1 = · · · = cn = 0

• Part (b) of theorem 5.3 tells us, that being a smooth submanifold S ⊂ N of ambient smooth
manifold N is a property property of the subset. It suffices to check, pointwise, the local
property described above!

Proof. (a): By assumption S ,→ N is an immersion. By theorem 4.3 (rank theorem), there exists Locally, all immersions
charts (U, φ), (V, ψ) such that i(U ) ⊂ V and look the same

î = ψ ◦ i ◦ φ−1 : φ(U ) → ψ(V )

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

Up to shrinking ψ (restricting the image of φ), we find that

ψ(V ∩ S) = {(x1 , . . . , xk , xk+1 , . . . , xn ) | xk+1 = · · · = xn = 0}

Warning: What can go wrong here? Consider

Page 44 of 60
 Chapter 5 – Submanifolds

Image
Figure 5.4: Sketch 5.04

Show that there is no more stuff in the set!

(b): We have to check that the local charts given form an atlas. Which is almost a tautology and
quite tedious, as we can use {S ∩ V, ψ|S } as the atlas.

Remark (+Exercise). In section 2.1.2, example 4, we considered Φ : R1+n → R. We assumed dΦ

is nonzero on the set Φ−1 (0) ⊂ R1+n . Under this assumption, we proved that Φ−1 (0) is a naturally
smooth manifold. Using theorem 5.3 (or by hand) Φ−1 (0) is a smooth submanifold.

A priori, S ⊂ N could admit multiple smooth structures making it a submanifold. We know seek
to show that this is not the case.
Lemma 5.4. Let S ⊂ N be a submanifold. If F : M → N is a smooth map which factors through
S ,→ N as a continuous map, then F is smooth as a map M → S.

Image
Figure 5.5: Sketch 5.05
Proof. By theorem 5.3, there exists U ⊂ S ,→ N ⊃ V
Page 45 of 60
 Chapter 5 – Submanifolds

More by the proof of the

theorem ...

Image
Figure 5.6: Sketch 5.06

∨ ∨ ∨ ∨
Let us call F : M → S, F (x) = F (x). Since F is continuous, F −1 (U ) ⊂ M open. So, we can
∨
write, for (W, u), W ⊂F −1 (U )

Image
Figure 5.7: Sketch 5.07
∨
were, a priori, F i are continuous.
Concatenating the two diagrams, we find that
∨ ∨ ∨
F (x1 , . . . , xm ) = i◦ F (x1 , . . . , xm ) = (F 1 (x1 , . . . , xm ), . . . , F k (x1 , . . . , xm ), 0, . . . , 0). But then
∨ ∨
each F i has to be smooth and therefore F is smooth.
Lemma 5.5. Let S ⊂ M be a subset satisfying the conditions of theorem 5.3 (b), then the smooth
structure produced by the theorem is the unique smooth structure, such that S ,→ M is a smooth
submanifold.

Page 46 of 60
 Chapter 5 – Submanifolds

Proof. Let S̃ be a copy of S, but endowed with some possibly different smooth structure s.t.
S̃ ,→ M is an embedding. Ergo it is a smooth
id
S̃ ,→ M factors through S, so S̃ → S smooth. Similarly S → S̃ smooth.
id submanifold
This of M5.4
uses lemma

5.3 The (weak) Whitney embedding theorem

Theorem 5.6 (Whitney). Every compact n-dimensional smooth manifold admits an embedding
into RN for N ≫ 1 large enough.

Remark. Later (probably this month), we will remove the compactness assumption and also argue
that one can take N = 2n + 1.
Whitney proofed that one can take N = 2n. Don’t sue him, if he is off
by one :)
Added remark. This is a very philosophically pleasing statement, since we recover our intuition
of embedded manifold from the abstract theory. It is also true, that there is only one embedding (up
to isotopy).

Proof of theorem 5.6. Fix a finite cover of M {B1 , . . . , Bk }, Bi ⊂ M open. We may as well
assume that there exist charts (Bi′ , ϕi ), Bi ⊂ Bi′ , φi (Bi′ ) = B1 (0) ⊂ Rm .
Let ρi : M → R be a cutoff function for (B i ⊂ Bi′ ), i.e. ρi|Bi =≡ 1, supp(ρi ) ⊂ Bi′ , 0 ≤ ρi ≤ 1. The
existence of the ρi follows from proposition 2.8.
We now define Notice the k comes from
compactness, i.e. we have
F : M → Rmk+k no control over it, as it its
p 7→ (ρ1 (p) φ1 (p), . . . , ρk (p)φk (p), ρ1 (p), . . . , ρk (p)) non-constructive
| {z }
∈Rm

We will now see that F is an embedding. First, we will argue F is an injective immersion.
If F (p) = F (q) =⇒ ρi (p) = ρi (q)∀i = 1, . . . , k. Let i0 be such that p ∈ Bi0 . Then
ρi0 (p) = 1 = ρi0 (q) =⇒ q ∈ supp(ρi0 ) ⊂ Bi′0 . But now
φi0 (p) = ρi0 (p)φi0 (p) = ρi0 (q)φi0 (q) = φi0 (q). Hence p, q ∈ Bi′0 =⇒ p = q.
| {z }
∈Rm
F is an immersion: Choose p ∈ M . Then p p ∈ Bi0 , for some i0 . Hence ρi0 ≡ 1 for some
neighborhood of p.

Image
Figure 5.8: Sketch 5.08

Page 47 of 60
 Chapter 5 – Submanifolds

Hence d(ρi0 φi0 ) = dρi near p =⇒ dF is injective near p, but p was arbitrary.
|{z}0
invertible m×m
Finally, since M is compact, the theorem follows from the following lemma 5.7. Kind of cheating ...
I.e. it is enough to show that F −1 : F (M ) → M is continuous, i.e. F : M → F (M ) is a closed
lemma 5.7
map. But since M is compact, F is proper =⇒ F closed.

Lemma 5.7 (Lee Appendix A: 57). Let X be a topological space. Let Y be locally compact (e.g. a
topological manifold), then any proper continuous map is closed.

Page 48 of 60
Chapter 6:
Transversality

6.1 Basic definition

6.1.1 Motivation
Let l1 , l2 ⊂ R2 be (linear) lines. We will say that l1 , l2 are transverse, if T0 l1 ⊕ T1 l2 = T0 R2 ≡ R2
|{z} |{z}
l1 l2

Figure 6.1: Sketch 6.01

Page 49 of 60
 Chapter 6 – Transversality

Figure 6.2: Sketch 6.02

Page 50 of 60
 Chapter 6 – Transversality

Figure 6.3: Sketch 6.03

Observations:

1. transversality is stable (slight changes to the lines don’t change transversality) Similarly to being full
rank
2. transversality is generic(for pretty much any lines l1 , l2 they are transverse)
One goal: Develop non-linear theory of transversality. I.e. replace l1 , l2 ⊂ R2 by manifolds.
Both of the above observations will still be true.
Announcement On Tuesday, November 26, there will be a course evaluation. Start of lecture 11
(15.11.2024)
• Please show up that day!

• Bring a phone / computer

Please add to your notes:

Page 51 of 60
 Chapter 6 – Transversality

Figure 6.4: Sketch 6.04

Proof of theorem 5.3. Since S ,→ N is an embedding, i(U ) is open in the subspace topology, so
there exists W ⊂ N such that i(U ) = S ∩ W .

6.1.2 Transversality for submanifolds

Let M be a smooth manifold.
Definition. We say that a pair of submanifolds K, L ⊂ M are transverse at p ∈ K ∩ L if Here the sum is a gain
the span of both of them
Tp K + Tp L = Tp M.

We say that K, L are transverse and write K ⋔ L.

Remark. In the literature, we also see “transversal”, “transversally intersecting”.

Page 52 of 60
 Chapter 6 – Transversality

Figure 6.5: Sketch 6.05

Example. K, L are transverse.

Figure 6.6: Sketch 6.06

Tp K = TP L, transversality fails.

Lemma 6.1. Let K k , Ll be submanifolds of M . If K, L are transversal, then K ∩ L ⊂ M is a Key lemma for
submanifold. transversality

Remark. In general, if S, T are submanifolds of N , then S∩T need not be a topological submanifold.
For example: f : R2 → R
f (x, y) = x2 − y 2 .
Let g : R2 → R
g(x, y) = 0.
Let S = {(x, y, z) | z = f (x, y)} ⊂ R2+1 and T = {(x, y, z) | z = g(x, y)} ⊂ R2+1 . But

S ∩ T = {(x, y, z) | z = 0, x2 − y 2 = 0}

Page 53 of 60
 Chapter 6 – Transversality

Figure 6.7: Sketch 6.07

Look at the derivative at 0 ....

Proof. This is a local question, e.g. by theorem 5.3. So we may as well assume that
M = U ⊂ Rn . We can also assume that 0 ∈ U .
It is enough to check that K ∩ L smooth submanifold in a neighborhood of p = 0. By rank
theorem (4.3), we may assume (after possibly further shrinking U ∋ 0) that
K = f −1 (0), f : U → Rn−k , L = g −1 (0), g : U → Rn−l where f, g have full rank.

Figure 6.8: Sketch 6.08

Now we consider H = (f, g) : U → Rn−k ⊕ Rn−l . It is enough to prove that dH0 is surjective (by
the rank theorem). Note that H −1 (0) = f −1 (0) ∩ g −1 (0) = K ∩ L.
To see surjectivity of dH0 , we consider the exact sequences:

Page 54 of 60
 Chapter 6 – Transversality

Figure 6.9: Sketch 6.09

The horizontal map T0 L + T0 K → TL /(T0 L ∩ T0 K) ⊕ TK /(T0 L ∩ T0 K) sends v + w to (v, w).

This is well defined, because if v + w = v ′ + w′ =⇒ v − v ′ = w − w′ ∈ T0 L ∩ T0 K. (Equivalently,
this map is just quotient by T0 L ∩ T0 K)
Clearly the R.H vertical arrow is injective: the kernel of df0 = T0 K, so (df0 )|T0 L/(T0 L∩T0 K) and
similarly for dg0 . To prove the R.H. vertical arrow is an isomorphism, do a dimension count: Exact sequence

v7→(v,v) (u,w)7→u−w
0 → T0 K ∩ T0 L → T0 K + T0 L → T0 U ≡ Rn → 0

=⇒ dim(T0 K ∩ T0 L) + n = k + l =⇒ dim(T0 L/(T0 K∩T0 L) ) = l − (k + l − n) = n − k and

dim(T0 K/(T0 K∩T0 L) ) = k − (k + l − n) = n − l. We conclude that the R.H. vertical arrow is an
isomorphism.
Remark. We have

Figure 6.10: Sketch 6.10

where the left vertical arrow is an isomorphism, due to the 5’lemma or diagram chasing.
Hence ker(dH0 ) = T0 L ∩ T0 K = T0 (L ∩ K).

6.1.3 Transversality of maps

Definition. Let

Y
g
f
X Z

Figure 6.11: Sketch 6.11

be a diagram in Top (the category of topological spaces). We let X × z Y := {(x, y) | f (x) = g(y)} ⊂
X × Y , endowed woth the subspace topology. We call X × z Y the fiber product (of the diagram).

Remark (for enthusiasts only). It can be shown that given any topological space W ∈ Top and
maps

Page 55 of 60
 Chapter 6 – Transversality

W Y

X Z

W
∃!

X×
ZY Y

X Z

Figure 6.12: Sketch 6.12

there exists a unique map W → X ×

z Y commutes.(Universal property)

Lots of categories admit fiber products! This is a good property for categories to have.
Bad news: The (not-full) subcategory Man∞ ⊂ Top does not admit fiber products (nor does
Man0 ⊂ Top).
Example (Non-example). Z = R2+1 , X = graph(x2 − y 2 ), Y = graph(0).

Definition. Let

Y
g
f
X Z

Figure 6.13: Sketch 6.13

be a diagram in Man∞ . We say that f, g are transverse at z = f (x) = g(y) if

im dfx + im dgy = Tz Z.

We say that f, g are transverse and say f ⋔ g if this holds for all such z.

Remark. Transversality for maps generalizes transversality for submanifolds. Take the diagram

X Z

Figure 6.14: Sketch 6.14

i
Proposition 6.2. If f ⋔ g, then X ×
z Y → X × Y is a smooth embedding.

The diagonal arrows are

Y the obvious projections

X×
ZY X ×Y

Figure 6.15: Sketch 6.15

Page 56 of 60
 Chapter 6 – Transversality

Proof. Some observations:

• exercise sheet 07: X ×
z X × Y is proper.

• similarly to the proof of theorem 5.6, it is enough to prove that i is an injective immersion.
By definition i is injective. Therefore we need to check that i is smooth and the differential
is injective.
Consider

∆ := (X, Y, Z, Z)

π
X ×Y ×Z ×Z X ×Y

W = graph(f, g)

Figure 6.16: Sketch 6.16

where
graph(f, g) =:= {(x, y, x2 , z1 , z2 ) | z1 = f (x), z2 = g(y)}.
Then
W ∩ ∆ = {(x, y, z1 , z2 ) | z1 = z2 = f (x) = g(y)} = X ×
z Y.

We have:
j
W ∩∆ X ×X ×Z ×Z
α i
∼

X×
Z X ×Y

Figure 6.17: Sketch 6.17

α is clearly bijective and continuous. It is elementary that α is a closed map. That means we
have to check the limit points. W ∩ ∆ is closed, i.e. contains the same limit points. . . Therefore α
is a homeomorphism.
j
By lemma 6.1, if we can show that W ⋔ ∆, then W ∩ ∆ X × Y × Z × Z is smooth embedding.
Hence i := π ◦ j smooth. Let us now check that W ⋔ ∆ at some arbitrary point
p = (x, y, z, z) ∈ W ∩ ∆ ⊂ X × Y × Z × Z. Note that z = f (x) = g(y). We have

Tp W = {(v, w, dfx (v), dgy (w))}

and
Tp ∆ = {v ′ , w′ , u, u},
where v, v ′ ∈ Tx X, w, w′ ∈ Ty Y, u ∈ Tz Z. We need to check: Tp W + Tp ∆ = Tp (X × Y × Z × Z).
We must show that for an arbitrary (a, b, c, d) ∈ Tp (X × Y × Z × Z) = TX X ⊕ Ty Y ⊕ Tz Z ⊕ Tz Z,

Page 57 of 60
Journal

• Lecture 01: Covering: Introduction, locally Euclidean, Hausdorff, second countable spaces,
their covers and exhaustions by compact sets .
Starting in ‘Organization’ on page 3 and ending in ‘Basis and covers’ on page 9. Spanning 6
pages
• Lecture 02: Covering: Local finiteness, refinements, paracompactness, introduction to
topological manifolds and examples .
Starting in ‘Basis and covers’ on page 9 and ending in ‘Manifolds with boundary’ on
page 14. Spanning 5 pages
• Lecture 03: Covering: Topological properties of topological manifolds, classification of
topological manifolds, introduction to smooth manifolds .
Starting in ‘Manifolds with boundary’ on page 14 and ending in ‘Charts and atlases’ on
page 18. Spanning 4 pages

• Lecture 04: Covering: Examples of smooth manifolds, smooth maps, the category of
smooth manifolds, hierarchy of categories of manifolds .
Starting in ‘Charts and atlases’ on page 18 and ending in ‘The category of smooth
manifolds’ on page 21. Spanning 3 pages
• Lecture 05: Covering: Smooth manifolds with boundary, partitions of unity .
Starting in ‘The category of smooth manifolds’ on page 21 and ending in ‘Partitions of
unity’ on page 24. Spanning 3 pages
• Lecture 06: Covering: Applications of partitions of unity, motivation of tangent vectors,
definition of tangent vectors via equivalence classes of smooth curves, definition of
differentials, fundamentality of the differential .
Starting in ‘Partitions of unity’ on page 24 and ending in ‘Definition via equivalence classes
of smooth curves’ on page 29. Spanning 5 pages
• Lecture 07: Covering: Definition of tangent vectors via derivations, equivalence of both
definitions, coordinates .
Starting in ‘Definition via equivalence classes of smooth curves’ on page 29 and ending in
‘Coordinates’ on page 33. Spanning 4 pages
• Lecture 08: Covering: Coordinates (continued), tangent bundles, submersions, immersions
and embeddings .
Starting in ‘Coordinates’ on page 33 and ending in ‘Basic definitions’ on page 39. Spanning
6 pages

• Lecture 09: Covering: The rank theorem as a generalization of a linear algebra fact and it’s
proof, basic definitions of submanifolds .
Starting in ‘Basic definitions’ on page 39 and ending in ‘Slice lemma1 ’ on page 44.
Spanning 5 pages
1 Lee [2] calls it a theorem

Page 58 of 60
 Chapter 6 – Transversality

• Lecture 10: Covering:

.
Starting in ‘Slice lemma2 ’ on page 44 and ending in ‘Motivation’ on page 51. Spanning 7
pages

• Lecture 11: Covering:

.
Starting in ‘Motivation’ on page 51 and ending in ‘Transversality of maps’ on page 57.
Spanning 6 pages

2 Lee [2] calls it a theorem

Page 59 of 60
Bibliography

[1] David Gale. “The Classification of 1-Manifolds: A Take-Home Exam”. In: (). url:
https://www.math.uni-bonn.de/~lcote/1_man_classification.pdf.
[2] John M. Lee. Smooth Manifolds. New York, NY: Springer New York, 2012. isbn:
978-1-4419-9982-5. doi: 10.1007/978-1-4419-9982-5. url:
https://doi.org/10.1007/978-1-4419-9982-5.
[3] Wern Juin Gabriel Ong. Notes for F4D1: Analysis and Geometry on Manifolds. 2024. url:
https://wgabrielong.github.io/notes/.
[4] Alex Taylor. “Equivalent definitions of the tangent space”. en. In: (). url:
https://art-math.github.io/files/tangentspace.pdf.
[5] Marco Zambon and Gilles Castel. Differential Geometry. 2020. url:
https://castel.dev/notes.

Page 60 of 60