Differential Geometry

Cambridge Part III, Michaelmas 2022

Taught by Jack Smith
Notes taken by Leonard Tomczak

Contents

1 Manifolds and smooth maps 3

1.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Immersions, Submersions and local Diffeomorphisms . . . . . . . . . . . . . 6
1.5 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Vector bundles and tensors 9

2.1 Vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Vector bundles by gluing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Cotangent bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Multilinear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Tensors and forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Index notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7 Pushforward and pullback . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Differential forms 16
3.1 Exterior derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 De Rham cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Orientations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Partitions of unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.6 Stokes’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.7 Applications of Stokes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Connections on vector bundles 23

4.1 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Connections vs End(E) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Curvature algebraically . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.5 Curvature geometrically . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5 Flows and Lie derivatives 29

5.1 Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Lie Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1
 5.3 Homotopy invariance of de Rham cohomology . . . . . . . . . . . . . . . . . 32

6 Foliation and Frobenius integrability 34

6.1 Foliations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.2 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.3 Frobenius integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

7 Connections on vector bundles with extra structure 36

7.1 Connections on T X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.2 Orthogonal vector bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

8 Riemannian geometry 38
8.1 Riemannian metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
8.2 The Levi-Civita connection . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
8.3 The Riemann tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
8.4 Hodge theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

9 Lie groups and principal bundles 42

9.1 Lie groups and Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
9.2 Lie group actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.3 Principal bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2
1 Manifolds and smooth maps

1.1 Manifolds
Definition. A topological n-manifold is a topological space X such that for all p ∈ X
there exists an open neighborhood U of p, an open set V ⊆ Rn and a homeomorphism
φ : U → V . We also require that X is Hausdorff and second countable.
Remark. One can show that for spaces locally homeomorphic to Rn the condition “Haus-
dorff + second countable” is equivalent to “metrisable + has countably many components”
Maps φ as in the definition are called charts for X. A collection of charts whose domains
cover X is called an atlas for X.
If φα : Uα → Vα , φβ : Uβ → Vβ are overlapping charts, then

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

is the transition map. It expresses the φβ coordinates as functions of the φα coordinates.

Definition. Given an atlas A = {φα : Uα → Vα | α ∈ A} on X and a fucntion f : W → R
on an open W ⊆ X, say f is smooth with repsect to A if for all α the map

f ◦ φ−1
α : φα (Uα ∩ W ) → R

is smooth. The atlas is smooth if every transition map φβ ◦ φ−1

α is smooth. Two smooth
atlases are equivalent if their union is smooth.
Here smooth means C ∞ .
Definition. A smooth structure on a topological n-manifold X is an equivalence class of
smooth atlases. A smooth n-manifold is a topological n-manifold together with the choice
of a smooth structure.
Note: Being a topological manifold is a property of a space, but for a differentiable manifold
one needs to choose additional structure, i.e. the smooth structure. There are topological
manifolds with different smooth structures (not even diffeomorphic).
Example. The n-sphere S n has the underlying set

{x ∈ Rn+1 | ∥x∥ = 1} ⊆ Rn+1 ,

3
endowed with the subspace topology. We define an atlas on S n . Let U± = S n \{(0, . . . , 0, ±1)}.
≃
→ Rn by
Define φ± : U± −
1
φ± (x1 , . . . , xn+1 ) = (x1 , . . . , xn ).
1 ∓ xn+1

This defines a smooth structure on S n .

1.2 Tangent spaces

Fix an n-manifold and point p ∈ X.
Definition. A curve based at p is a smooth map of the form γ : I → X sending 0 7→ p,
where I ⊆ R is an open neighborhood of 0. Say two curves γ1 , γ2 (based at p) agree to first
order if there exists a chart φ around p such that

(φ ◦ γ1 )′ (0) = (φ ◦ γ2 )′ (0)

We write πpφ for the map γ 7→ (φ ◦ γ)′ (0).

Lemma 1.1. If γ1 , γ2 satisfy πpφ (γ1 ) = πpφ (γ2 ), then for all charts ψ around p we have

πpψ (γ1 ) = πpψ (γ2 )

Proof. Clear, using chain rule and inserting transition maps.

Corollary 1.2. Agreement to first order is an equivalence relation.

Definition. The tangent space to X at p, denoted Tp X, is

{curves based at p}/agreement to first order.

Proposition 1.3. Tp X is naturally an n-dimensional R-vector space.

Proof. Given a chart φ about p, the map πpφ : {curves at p} → Rn factors through Tp X
and thus induces an injection Tp X ,−→ Rn . It is easily seen to be surjective. This bijection
defines a vector space structure on Tp X. Different charts give rise to different bijections
but they are related by a linear automorphism of Rn , hence they all induce the same vector
space structure on Tp M .

∂
Definition. Given a chart φ about p with coordinates x1 , . . . , xn , define ∂x i
∈ Tp X to be
φ −1 n
(πp ) (ei ) where ei ∈ R is the i-th standard basis vector. We will often abbreviate this
by ∂xi or ∂i .
Warning. ∂xi depends on the choice of the whole coordinate system x1 , . . . , xn , not just
xi .

4
Lemma 1.4. On chart overlaps we have
X ∂xj n
∂ ∂
=
∂yi ∂yi p ∂xj
j=1

∂x
Here ∂yji |p := ∂y∂ i xj := φj (ψ −1 (ψ(p) + tei ))′ (0) where φ, ψ are charts inducing the local
coordinates xj , yi .

Proof. We have
∂
= (πpψ )−1 (ei )
∂yi
= (πpφ )−1 (πpφ ◦ (πpψ )−1 )(ei )
= (πpφ )−1 (t 7→ φ(ψ −1 (p) + tei ))′ (0)
 
Xn
= (πpφ )−1  (t 7→ φj (ψ −1 (p) + tei ))′ (0)ej 
j=0
 
n
X ∂xj
= (πpφ )−1  ej 
∂yi p
j=0
n
X ∂xj ∂
= .
∂yi p ∂yj
j=0

1.3 Derivatives
Fix manifolds X, Y and a smooth map F : X → Y .
Definition. The derivative of F at p ∈ X is the map Dp F : Tp X → TF (p) Y defined by
[γ] 7→ [F ◦γ]. Sometimes we will write it as F∗ ,called pushforward by F on tangent vectors.
Lemma 1.5. The map Dp F is well-defined and linear.

Proof. Let φ, ψ be charts about p, resp. F (p). We have πFψ (p) (F ◦ γ) = (ψ ◦ F ◦ γ)′ (0) =
((ψ ◦ F ◦ φ−1 ) ◦ (φ ◦ γ))′ (0) = T πpφ (γ) where T is the derivative of ψ ◦ F ◦ φ−1 at φ(p).
Dp F
Tp X TF (p) X
πpφ ψ
πF (p)

T
Rdim X Rdim Y

5
Suppose that {xi }, {yj } are coordinates associated to φ resp. ψ. Then ψ◦F ◦φ−1 expresses
∂y
the yj as functions of the xi via F , so T = ∂xji p and
 ∂y 
j
Dp F (∂xi ) = DF ((πpφ )−1 (ei )) = (πFψ (p) )−1 (T ei ) =
X ψ
(πF (p) )−1 ej .
∂xi p
j

Remarks.
(i) For X = Rn , Y = Rm , the new notion of derivative coincides with the standard one
from multivariable calculus.
∂f
(ii) Given f : X → R, we have Dp f (∂xi ) = ∂xi p .

(iii) We can write

[γ] = D0 γ(∂t )
where t is the parameter of γ.
Lemma 1.6 (Chain Rule). Suppose we have smooth maps
F G
X−
→Y −
→Z

Then Dp (G ◦ F ) = DF (p) G ◦ Dp F .

Proof. By definition.

1.4 Immersions, Submersions and local Diffeomorphisms

Definition. A smooth map F : X → Y is a immersion/submersion/local diffeomorphism
(at p ∈ X) if Dq F is injective/surjective/bijective for all q ∈ X (resp. only at q = p).
Say p is a regular point of F if Dp F is surjective (i.e. F is a submersion at p) and a
critical point otherwise. Say that q ∈ Y is a regular value if all p ∈ F −1 (q) are regular
and otherwise critical value.
Lemma 1.7. F is a local diffeomeorphism at p (in the sense of the definition above) iff
there are open neighborhoods U of p, V of F (p) such that F |U : U → V is a diffeomorphism.

Proof. “⇐” is obvious from the chain rule.

“⇒” Pick charts φ : A → B, ψ : C → D about p, F (p). By shrinking φ, WLOG F (A) ⊆ C.
Now consider ψ ◦F ◦φ−1 : B → D. This is a sooth map between open subsets of Euclidean
space with invertible derivative at φ(p). By the inverse function theorem there exist open
neighborhoods B ′ ⊆ B of φ(p) and D′ ⊆ D of ψ(F (p)) such that (ψ ◦F ◦φ−1 )|B ′ : B ′ → D′
′
has a smooth inverse H. Now set U = φ−B , V = ψ −1 (D′ ). Then F |U : U → V has smooth
inverse φ−1 ◦ H ◦ ψ.

6
Notice: We could have chosen ψ ◦ F as φ. W.r.t. the charts ψ ◦ F, ψ the map F looks like
the identity in local coordinates.
Proposition 1.8. Suppose F is an immersion at p. Given local coordinates x1 , . . . , xn
on X about p, there exist local coordinates y1 , . . . , ym on Y about F (p) w.r.t. which F
looks like Rn = Rn × 0 ,−
→ Rn × Rm−n = Rm . Similarly, if F is a submersion of p, then
given coordinates y about F (p) there exist coordinates x about p w.r.t. which F looks like
Rn → Rn /(0 × Rn−m ) = Rm .

Proof. Exercise.

1.5 Submanifolds
Fix an n-manifold X.
Definition. A subset Z ⊆ X is a submanifold of codimension k if for all p ∈ Z there exist
local coordinates x1 , . . . , xn on X about p such that Z is given locally by x1 = · · · = xk = 0.
(Formally: on the domain of the chart Z = {x1 = · · · = xk = 0}).
Z is a properly embedded submanifold if the same holds for all p ∈ X (not merely p ∈ Z).
E.g. 0 × R ⊆ R2 is a properly embedded submanifold. 0 × R∗ ⊆ R2 is a submanifold but
not properly embedded.
Given a codimension k submanifold Z ⊆ X,
• Equip Z with the subspace topology (automatically Hausdorff and 2nd-countable
because X is)
• For p ∈ Z can choose local coordinates x1 , . . . , xn on X as in the definition. Then
xk+1 , . . . , xn define a chart on Z about p.
• Transition functions are smooth since original transition functions on X were smooth.
Proposition 1.9. A codimension k submanifold Z ⊆ X is naturally an (n − k)-manifold.
The inclusion map ι : Z → X is a smooth immersion and a homeomorphism onto its
image. Composition with ι gives a bijection
ι◦−
{smooth maps to Z} −−→ {smooth maps to X
with image ⊆ Z }.

Definition. A map F : Y → X is an embedding if it is a smooth immersion and a

homeomorphism onto its image.
Lemma 1.10. The image of an embedding F : Y → X is a submanifold of X and F
induces a diffeomorphism from Y to that submanifold.
So: Submanifolds ↔ images of embeddings.

7
Example. The inclusion S n ,−→ Rn+1 is an embedding. So S n is a submanifold of Rn+1
and the induced manifold structure agrees with the one we already defined.
Proposition 1.11. If F : X → Y is a smooth map, and q ∈ Y is a regular value, then
F −1 (q) is a submanifold of X of codimension dim Y .

Proof. Take a point p ∈ F −1 (q). Since q is a regular value, F is a submersion at p, so there

exist local coordinates xi on X about p and yj on Y about q such that y ◦F = (x1 , . . . , xm )
where m = dim Y . By translating the y-coords we may assume that y(q) = 0. Let U be
an open neighborhood of p on which the x-coordinates and y ◦ F are defined. On U we
have U ∩ F −1 (q) = U ∩ (y ◦ F )−1 (0) = {x ∈ U : x1 = · · · = xm = 0}, so the xi give the
required chart about p.

Example. Consider F : Rn+1 → R, y 7→ ∥y∥2 . This is smooth and DF |y = (2y1 . . . 2yn+1 )

which is non-zero (hence surjective) everywhere except at 0. So for all λ ∈ R>0 is a
codimension 1 submanifold of Rn+1 , in particular S n = F −1 (1) is a submanifold.
Theorem 1.12 (Sard’s theorem). The set of critical values of a smooth map F : X → Y
has measure 0 in Y (i.e. for any chart ψ : S → T on Y , ψ({critical values} ∩ S) ⊆ T ⊆
Rdim Y has Lebesgue-measure 0).
Corollary 1.13. The set of regular values is dense in Y .
Warning. This only concerns regular values, e.g. if dim X < dim Y there are no regular
points.
Definition. Submanifolds Y, Z ⊆ X are transverse if for all p ∈ Y ∩ Z we have

Tp Y + Tp Z = Tp X.

Theorem 1.14. If Y, Z are submanifolds of X of codimensions k, l, intersecting trans-

versely, then Y ∩ Z then is a submanifold of codimension k + l.

Proof. Pick p ∈ Y ∩ Z. Since Y, Z are submanifolds, there exist coordinates y1 , . . . , yn ,

z1 , . . . , zn about p such that Y = {y1 = · · · = yk }, Z = {z1 = · · · = zl = 0}. Let U
be an open neighborhood of p on which y and z are defined. Consider F : U → Rk+l
given by (y1 , . . . , yk , z1 , . . . , zl ). Y and Z being transverse at p is equivalent to Dp F
being surjective. So there exist local coordinates x1 , . . . , xn on U about p such that
x1 = y1 , . . . , xk = yk , xk+1 = z1 , . . . , xk+l = zl . Then in these coordinates, Y ∩ Z = {x1 =
· · · = xk+l = 0}.

8
2 Vector bundles and tensors

2.1 Vector bundles

Definition. A vector bundle of rank k over a manifold B consists of the following infor-
mation:
• A manifold E,
• A smooth map π : E → B,
• An open cover (Uα )α∈A of B
• For each α ∈ A a diffeomorphism Φα : π −1 (Uα ) → Uα × Rk such that
– pr1 ◦Φα = π on π −1 (Uα )
– For all α, β ∈ A the map Φβ ◦Φ−1 k k
α : (Uα ∩Uβ )×R → (Uα ∩Uβ )×R has the form
(b, v) 7→ (b, gβα v) for some (necessarily smooth) map gβα : Uα ∩ Uβ → GLk (R)
E is called the total space, π the projection, B the base space, Φα the local trivial-
izations and gαβ the transition functions.The fibres π −1 (b) are denoted by Eb .
Via each Φα the fibres Eb carry the structure of a k-dimensional vector space, independent
of the local trivialization chosen.
Examples.
(i) The trivial bundle (of rank k over B) has E = B × Rk as the total space with the
obvious projection E → B and global trivialization Φ : E → B × Rk .
(ii) The tautological bundle over RPn is the line bundle (i.e. rank 1 vector bundle) over
RPn given by:

E = {(p, v) ∈ RPn × Rn+1 | v lies on the line described by p}

It is a submanifold of RPn × Rn+1 . Define π by π(p, v) = p.

Open cover: Ui = {[x0 : · · · : xn ] | i ̸= 0}. Φi : π −1 (Ui ) → Ui × R is given by ([x0 :
· · · : xn ], (y0 , . . . , yn )) 7→ ([x0 : · · · : xn ], yi ). On Ui ∩ Uj we have Φj ◦ Φ−1
i ([x0 : · · · :
∗
xn ], t) = Φj ([x], t(x0 /xi , . . . , xn /xi )) = ([x], txj /xi ), so gji = xj /xi ∈ R = GL1 (R)
(iii) The tautological complex line bundle over CPn .
(iv) The tangent bundle of an n-manifold X is given by

9
 • Total space: T X = p∈X Tp X. This is a manifold via a pseudo-atlas: Given
F
a
F coordinate patch Un on X with coordinates
P xi we get a pseudo-chart φ :
T
p∈U p X → U × R given by (p, a ∂
i xi ) →
7 (p, (a1 , . . . , an )). These make
T X into a manifold.
• Projection π : (p, v) 7→ p.
• The pseudo-charts give the local trivializations.
Definition. A section of a vector bundle π : E → B is a smooth map s : B → E such
that π ◦ s = idB .
Examples.
(i) Every vector bundle has a zero section given by s(b) = 0 for all b ∈ B.
(ii) A vector field on X is a section of the tangent bundle T X
Definition. Given vector bundles πi : Ei → Bi (i = 1, 2) and a smooth map F : B1 → B2 ,
a morphism of vector bundles E1 → E2 covering F is a smooth map G : E1 → E2 such
that π2 ◦ G = F ◦ π1 and for all p ∈ B1 the restricted map Gb : (E1 )p → (E2 )F (p) is linear.
If B1 = B2 = B, an isomorphism of vector bundles over B is a morphism covering idB
with a two-sided inverse. Equivalently, a diffeomorphism of the total space that induces
linear isomorphisms (E1 )p → (E2 )p .
Example. Consider S 1 = {eiθ | θ ∈ R} ⊆ C. The vector field ∂θ is defined and non-zero
in each fibre. So we get an isomorphism

S1 × R → T S1
(eiθ , a) 7→ (eiθ , a∂θ )

So T S 1 is trivial (i.e. isomorphic to the trivial bundle).

For the trivial bundle of rank k over some (fixed) base we also write Rk .
Remark. A morphism G : B × R → E (covering the identity) is the same thing as a
section s of E. More generally, a morphism G × Rk → E is the same as a k-tuple of
sections. The morphism is an isomorphism iff the sections form a basis in each fibre.
Definition. Given a vector bundle π : E → B of rank k, a subbundle of rank l is a subset
F ⊆ E such that B can be covered by local trivializations Φα : π −1 (Uα ) → Uα × Rk under
which F ∩π −1 (Uα ) = Uα ×(Rl ×{0}). This is naturally a vector bundle of rank l. Similarly
we can define quotient bundle E/F of rank k − l and we have morphisms F → E → E/F .
Example. ORPn (−1) (the tautological bundle) is (by construction) a subbundle of Rn+1
over RPn . We get the Euler sequence ORPn (−1) → Rn+1 → Rn+1 /ORPn (−1) ∼
= T RPn (−1).

10
2.2 Vector bundles by gluing
To define a vector bundle over B (of rank k) it suffices to give an open cover {Uα } of B
and for all α, β a smooth map gβα : Uα ∩ Uβ → GLk (R) satisfying
(i) gαα = constant map with value idRk .
(ii) gγα = gγβ gβα for all α, β, γ (cocycle condition)
−1
Note that from (i) and (ii) it follows that gαβ = gβα .
Given this data, define

× Rk (|{z}
` 
E= α Uα b , v) ∼ (|{z}
b , gβα (b)v)
∈Uα ∈Uα

π : E → B is the obvious map. There are identifications π −1 (Uα ) ∼

= Uα × Rk . These define
pseudo-charts and trivializations.
Example. For r ∈ Z can define a line bundle ORPn (r) on RPn to be trivialized over the
 −r
x
Ui = {[x] | xi ̸= 0} with transition functions gji = xji . Note that ORPn (−1) is the
tautological bundle.
Proposition 2.1. If π : E → B is a vector bundle of rank k, trivialized over {Uα } with
transition functions gβα , then
(a) The gβα satisfy (i) and (ii) above.
(b) E is isomorphic to the bundle constructed above.

Proof. (a) The gβα are defined via Φβ Φ−1

α , so (i) and (ii) are immediate.
∼
× Rk ∼.
` 
(b) The trivializations of E and their inverses define diffeomorphisms E −
→ α Uα
This is linear on fibres, hence a bundle isomorphism.

Corollary 2.2. Two bundles are isomorphic iff they can be trivialized over a common
open cover with the same transition functions.

Proof. If both can be trivialized over {Uα } with transition functions gβα , then they are
both isomorphic to the above construction. Converse is clear.

Example. Define
 the Möbius line bundle M → RP1 to be trivialized over U0 , U1 with
g10 = sign xx01 .

Claim: This is isomorphic to ORP1 (−1). Suffices to show we can modify the trivializations
of M to make the transition function become xx01 instead of sign xx01 . Let’s rescale the
trivialization Φi of M by a smooth map ψi : Ui → R∗ = GL1 (R). Excplicitly, consider the
trivialization
(p, v) 7→ (p, ψi (p) pr2 (Φi (p, v)))

11
This changes g10 to ψ1
g10 . Left to choose ψ0 , ψ1 such that ψ1 /ψ0 = |x1 |/|x0 |. One choice
rψ0 r
x21 x20
that works is ψ1 = x2 +x2 , ψ0 = x2 +x 2.
0 1 0 1

Definition. Given a vector bundle π : E → B and a smooth map F : B ′ → B the pullback

bundle F ∗ E defined as follows: Suppose E is trivialized over {Uα } with transition functions
gβα , then F ∗ E is trivialized over {F −1 (Uα )} with transition functions F ∗ gβα = gβα ◦ F .
The fibre (F ∗ E)p is Ef (p) .
Example. Consider the Hopf map H : S 2n+1 → CPn . Claim: H ∗ OCPn (−1) is trivial.
Proof: It is trivialized by the section S 2n+1 ∋ p 7→ p ∈ line through p.
Definition. Given a vector bundle π : E → B, the dual bundle E ∨ → B has total space
∨ ∨ −1
`
p∈B (Ep ) trivialized over {Uα } with transition functions (gβα )

If E is trivialized over U ⊆ B by a fibrewise basis of sections σ1 , . . . , σk . Then the fibrewise

dual basis σ1∨ , . . . , σk∨ give smooth sections of E ∨ which trivialize it over U .

2.3 Cotangent bundle

Fix a n-manifold X.
Definition. The cotangent bundle of X, denoted T ∗ X is the dual of the tangent bundle.
The fibre of p is Tp∗ X, the cotangent space at p.
Dual to the picture of Tp X via curves R → X we can describe Tp∗ X via functions X → R:
Say that functions f1 , f2 about p agree to first order if Dp f1 = Dp f2
Proposition 2.3. There is a natural isomorphism
∼
{functions about p}/agreement to
first order −→ Tp∗ X

Proof. Define the map e : {functions about p} → Tp∗ X, f 7→ ([γ] 7→ (f ◦ γ)′ (0)) = Dp f . In
P P ∂f
local coordinates: f 7→ ( ai ∂xi 7→ ai ∂xi ). We see that e(xj ) are the dual basis to ∂xj ,
so e is surjective and by definition e(f1 ) = e(f2 ) iff f1 , f2 agree to first order about p.

So for any smooth function f : U → R we get an element of Tp∗ X at each p ∈ U . This is

denoted dp f .
Lemma 2.4. The dp f define a smooth section of T ∗ X, denoted df , the differential of f .
P ∂f
Proof. We have df = ∂xi dxi . We saw above that the dxi are fibrewise dual to the ∂xi ,
so the dxi are smooth. So df is a smooth linear combination of smooth sections, hence
smooth itself.

Note, by construction df (v) = directional derivative of f in direction v.

12
Definition. A section of T ∗ X is a 1-form.
Unwarning! dxi only depends on xi , not the other xj (unlike ∂xi ).
Definition. Given a smooth map F : X → Y . The map (Dp F )∨ : TF∗ (p) Y → Tp∗ X is
called pullback by F , denoted F ∗ .
Lemma 2.5. Given F : X → Y and a smooth function g on Y , we have F ∗ dg =
d(F ∗ g) := d(g ◦ F ).

Proof. Given [γ] ∈ Tp X, we have (F ∗ dg)[γ] = dg(Dp F [γ]) = dg([F ◦ γ]) = (g ◦ F ◦ γ)′ (0) =
d(g ◦ F )[γ].

2.4 Multilinear algebra

See handout.

2.5 Tensors and forms

We can apply any functorial operations to transition functions of existing bundles to build
new ones.
Example. Dual bundle above.
Example. Given vector bundles E, F → B, trivialized over {Uα } with transition functions
gβα , hβα , can define E ⊕ F → B with fibres Ep ⊕ Fp , trivialized over {Uα } with transition
functions gβα ⊕ hβα .
Similarly can define E ⊗ F .
Given a smooth map F : X → Y , DF defines a section of T ∗ X ⊗ F ∗ T Y . For each p ∈ X,
Dp F ∈ HomR (Tp X, TF (p) Y ) = Tp∗ X ⊗ TF (p) Y .
Similarly can take tensor or exterior powers of a given vector bundle.
Definition. A tensor V(field) of type (p, q) on X is a section of (T X)⊗p ⊗ (T ∗ X)⊗q . An
r-form is a section of r T ∗ X. The space of r-forms on an open set U ⊆ X is denoted
Ωr (U ).
Examples. A tensor of type
• (0, 0) is a section of R, i.e. a smooth function (or scalar field).
• (1, 0) is a vector field.
• (0, q) is something which “eats q vectors multilinearly and spits out a number”.

13
2.6 Index notation
From now on, indices on local coordinates will be superscripts: x1 , . . . , xn .
A section T of T X ⊗ T ∗ X ⊗ T X (a specific kind of tensor of type (2, 1)) can be written
in local coordinates xi uniquely as
X ik
T = T j ∂i ⊗ dxj ⊗ ∂k
i,j,k

for locally defined smooth functions T ij k .

Horizontal position of indices refer to the ordering of tensor factors. Vertical position
denotes T X vs T ∗ X. We will often use summation convention where repeated indices
once up and once down are summed over, e.g.

T = T ij k ∂i ⊗ dxj ⊗ ∂k

Often we just write T ij k for T .

Tensor product corresponds to juxtaposition, e.g.

(T ij k ∂i ⊗ dxj ⊗ ∂k ) ⊗ (Slm dxl ⊗ dxm ) = T ij k Slm ∂i ⊗ dxj ⊗ ∂k ⊗ dxl ⊗ dxm

or (T ⊗ S)ij ksm = T ij k Slm .

Contraction corresponds to summation. E.g. contraction of third factor of T with second
factor of S is

T ij k Slm dxm (∂k ) ∂i ⊗ dxj ⊗ dxl = T ij k Slk ∂i ⊗ dxj ⊗ dxl ,

| {z }
∂ mk

i.e. the result is T ij k Slk = T ij k Slk .

P
k

Similarly, in local coordinates xi an r-form α can be written uniquely as

X X
αI dxI = αI dxi1 ∧ · · · ∧ dxir
I

where the sum is over multi-indices I = (i1 < i2 < · · · < ir ).

Given r vectors v(1) , . . . , v(r) , we can feed them to α to give the number
X iσ(1) iσ(r)
ε(σ)αI v(1) . . . v(r)
I
σ∈Sr

This is equivalent to viewing α as the tensor

X
ε(σ)αI dxiσ(1) ⊗ · · · ⊗ dxiσ(r)
I
σ∈Sr

14
of type (0, r) and contracting with v(1) , . . . , v(r) .
1
Warning. Some people include the factor r! .

We can refer to the components of this tensor as αi1 ...ir (this is the coefficient of dxi1 ⊗
· · · ⊗ dxir ). When the ij form a multi-index I, i.e. i1 < i2 · · · < ir , this agrees with αI .
Example. On R2 we view dx1 ∧ dx2 as dx1 ⊗ dx2 − dx2 ⊗ dx1 . A general 2-form looks
like α12 dx1 ∧ dx2 = αij dxi ⊗ dxj where α21 = −α12 and α11 = α22 = 0.
In summation convention, it is correct to say α = αi1 ...ir dxi1 ⊗ · · · ⊗ dxir but NOT
α = αi1 ...ir dxi1 ∧ · · · ∧ dxir (the last sum would be r!α).
If α = α1 ∧ · · · ∧ αr , then α(v) = det(αi (v(j) )).

2.7 Pushforward and pullback

Fix manifolds X, Y and a smooth map F : X → Y .
• Given p ∈ X and a tensor of type (r, 0) at p (i.e. an element of (Tp X)⊗r ). We can
push this forward to (TF (p) Y )⊗r by applying Dp F on each tensor factor. Denoted
F∗ .
• Given p ∈ X and a tensor of type (0, r) at F (p), we can pullback to (Tp∗ X)⊗r using
((Dp F )∨ )⊗r . We can do the same for r-forms at F (p) using ∧r (Dp F )∨ . Denoted F ∗
• Given a tensor T of type (0, r) on Y can pull back to a tensor F ∗ T on X by (F ∗ T )p =
F ∗ (TF (p) ). Similarly for r-forms.
Summary: Can pushforward “up” tensors at a point and can pullback “down” tensors or
forms at a point or across the whole manifold.
If F is a diffeomorphism, then can pushforward or pullback any tensor over the whole
manifold, e.g. let T be of type (1, 1) on X. Then

(F∗ T )q = F∗ (TF −1 (q) )

where we apply DF −1 (q) F on the T X factor and (Dq (F −1 ))∨ on the T ∗ X factor.
In this setting F∗ = (F −1 )∗ and vice versa.

15
3 Differential forms

3.1 Exterior derivative

Take a 1-form α = αi dxi on X. Let us try to differentiate naively. We get:
∂αi j
dx ⊗ dxi .
∂xj
∂y i i
Suppose we change to different local coords y i . Then α = αi′ dy i = αi′ ∂x j ′ ∂y
j dx , so αj = αi ∂xj .
Hence
k ∂αk′ y k 2 k
 
∂αi j i ∂ ′ ∂y j i j i ′ ∂ y
dx ⊗ dx = α i dx ⊗ dx = dx ⊗ dx + αk dxj ⊗ dxi
∂xj ∂xj ∂xi ∂xj ∂xi ∂xj ∂xi
∂αk′ j k ′ ∂ y
2 k
= dy ⊗ dy + α k dxj ⊗ dxi
∂y j ∂xj ∂xi

∂αi j i i
Definition. The exterior derivative dα is ∂x j dx ∧ dx = dαi ∧ dx . By the above calcu-
lation this is well-defined (independent of local coordinates).
∂αI
More general, given a p-form α = αI dxI we define dα := dαI ∧dxI = ∂xj
dxj ∧dxI . Again,
this is well-defined.
Proposition 3.1. d satisfies the following:
(i) It is R-linear.
(ii) It agrees with the differential on 0-forms.
(iii) d2 = 0.
(iv) It commutes with pullback, i.e. F ∗ (dα) = d(F ∗ α).
(v) Graded Leibniz rule: Given a p-form and a q-form β:

d(α ∧ β) = (dα) ∧ β + (−1)p α ∧ (dβ)

Proof. (i) and (ii) are immediate from the definition.

(iii) Let α = αI dxI . We have

∂ 2 αI
 
2 ∂αI j I
d α=d dx ∧ dx = dxk ∧ dxj ∧ dxI = 0
∂xj ∂xk ∂xj

16
 ∂ 2 αI
since ∂xk ∂xj
is symmetric in j, k, but dxk ∧ dxj is antisymmetric.
(iv) Write α locally as αI dy I . Then

F ∗ (dα) = F ∗ (dαI ∧ dy i1 ∧ · · · ∧ dy ip )
= F ∗ (dαI ) ∧ F ∗ (dy i1 ) ∧ · · · ∧ F ∗ (dy ip )
= d((F ∗ αI )d(F ∗ y i1 ) ∧ · · · ∧ d(F ∗ y ip )
= d(F ∗ α)

using (iii) and (v) (we sum over multi-indices i1 < · · · < ip ).
(v) Let α = αI dxI , β = βJ dxJ . Then

d(α ∧ β) = d(αI βJ dxI ∧ dxJ )

= d(αI βJ ) ∧ dxI ∧ dxJ
= (dαI )βJ ∧ dxI ∧ dxJ + αI dβJ ∧ dxI ∧ dxJ
= (dαI )βJ ∧ dxI ∧ dxJ + (−1)p (αI ∧ dxI )(dβJ ∧ dxJ )
= (dα) ∧ β + (−1)p α ∧ dβ

Definition. A form α is closed if dα = 0, exact if there exists β such that α = dβ. We

write Z r (X), B r (X) ⊆ Ωr (X) for the spaces of closed resp. exact r-forms.
Aside: The exteriors derivative is the unique map Ω∗ (X) → Ω∗+1 (X) satisfying the prop-
erties in the proposition.

3.2 De Rham cohomology

Since d2 = 0, we have B r (X) ⊆ Z r (X).
r (X), is Z r (X)/B r (X).
Definition. The r-th de Rham cohomology group of X, denoted HdR
r (X) = 0 for r > dim X or r < 0.
Note that HdR
Example. (trivial cases)
0 (X) = Z 0 (X)/B 0 (X) = {functions f : df = 0}/0 = {locally constant functions},
(i) HdR
0 (X) = R{components of X} .
so HdR
(
0 R r = 0,
(ii) HdR (point) =
0 r ̸= 0.

17
Example. Let X = S 1 . We know that

R
 r = 0,
r 1
HdR (S ) = ? r = 1,

0 r ̸= 0, 1.


A 1-form on S 1 can be written uniquely as f (θ)dθ. All 1-forms are closed. Define a map

I : Ω1 (S 1 ) −→ R,
Z 2π
f (θ)dθ 7−→ f (θ)dθ
0

This is linear and non-zero, hence surjective. Claim: ker I = B 1 (S 1 ). Proof: If f dθ = dg,
then f = ∂g ∂θ , so I(f dθ) = g(2π) − g(0) = 0. Conversely, if I(f dθ) = 0, define g(θ) =
Rθ 1 1
0 f (t)dt. Then we have f dθ = dg. Thus we get an isomorphism I : HdR (S ) ≃ R.

Proposition 3.2. If F : X → Y is smooth, then F ∗ induces a linear map F ∗ : HdR

∗ (Y ) →
∗
HdR (X).

Proof. Immediate from the fact that d commutes with pullback.

E.g. consider F : S 1 → S 1 given by eiθ 7→ einθ . The map F ∗ : HdR

1 (S 1 ) → H 1 (S 1 ) is
dR
multiplication by n.
Proposition 3.3. Wedge product of forms descends to HdR∗ (X). This makes H ∗ (X)
dR
into a unital, graded-commutative associative algebra.

∗ (X), we need to show that d(α ∧ β) = 0 and [α ∧ β] depends

Proof. Given [α], [β] ∈ HdR
only on [α] and [β]. d(α ∧ β) = 0 follows from the Leibniz rule. If α′ = α + dγ, β ′ = β + dδ,
then α′ ∧β ′ = α∧β +α∧dδ +(dγ)∧β +(dγ)∧(dδ) = α∧β +d((−1)|α| α∧δ +γ ∧β +γ ∧(dδ)),
so [α ∧ β] only depends on the classes.

Since F ∗ commutes with ∧ and F ∗ 1 = 1, the map F ∗ : HdR

∗ (Y ) → H ∗ (X) is a unital
dR
algebra homomorphism.
Proposition 3.4 (Homotopy invariance). If F0 , F1 : X → Y are smoothly homotopic.,
∗ (Y ) → H ∗ (X) are equal.
then the maps F0∗ , F1∗ : HdR dR

Proof. See Section 5.3.

Corollary 3.5. If F : X → Y is a homotopy equivalence, then F ∗ : HdR

∗ (Y ) → H ∗ (X)
dR
is an isomorphism.
∗ (Rn ) = H ∗ (point).
Example. HdR dR

18
3.3 Orientations
Vn
Definition. An orientation of a n-dimensional vector space is a non-zero element of V
modulo positive rescalings.
An orientation of a vector bundle E → X of rank k is a nowhere-zero section of k E,
V
modulo rescaling by positive smooth functions. E is orientable if there exists an orientation
for it, and oriented if it is equipped with a choice of orientation.
Note: E is orientable iff k E is trivial. E.g. any trivial bundle is orientable. The tauto-
V
logical bundle over RPn is not orientable.
Definition. A manifold X is orientable/oriented if T X → X is.
E.g. S n is orientable for all n. RPn is orientable iff n is odd.
Definition. A volume form on n-manifold X is a nowhere-zero n-form.
A volume form ω defines an orientation (basis e1 , . . . , en for Tp X is positively oriented iff
ω(e1 , . . . , en ) > 0) and conversely an orientation defines a volume form modulo rescaling
by positive smooth functions.

3.4 Partitions of unity

Definition. Given an open over {Uα } of X, a partition of unity subordinate to the cover
is a collection of smooth functions {ρα : X → R≥ } such
• supp ρα ⊆ Uα .
• locally finite: For all p ∈ X there exists an open neighborhood V of p such that on
V all but finitely many ρα are ≡ 0.
•
P
α ρα = 1.

Lemma 3.6. For any open cover {Uα } of X, there exists a partition of unity subordinate
to it.

3.5 Integration
Fix an oriented n-manifold X and a compactly supported n-form ω on X.
R
Definition. The integral of ω over X, denoted X ω, is defined as follows:
• Cover X by coordinate patches Uα with coordinates xiα . WLOG these are positively
oriented, i.e. ∂x1α ∧ · · · ∧ ∂xnα represents the orientation.
• Pick a Rsubordinate
P Rpartition1 of unity {ρα }. Write ρα ω = f/ alphadx1α ∧ · · · ∧ dxnα .
Define X ω = α Rn fα dxα . . . dxnα .

19
 R
Lemma 3.7. The integral X ω is well-defined.

Proof. Suppose we cover X by patches Vβ with coords yβ1 , . . . , yβn . Take a partition of
unity σβ subordinate to this cover. Locally write σβ ω = gβ dyβ1 ∧ · · · ∧ dyβn . We want to
show α Rn fα dx1α . . . dxnα = β Rn gβ dyβ1 . . . dyβn . On overlaps Uα ∩ Vβ we have
P R P R

σβ fα dx1α ∧ · · · ∧ dxnα = σβ ρα ω = ρα gβ dyβ1 ∧ · · · ∧ dyβn .

∂y j
So σβ fα = ρα gβ det( ∂xβi ). Since both y and x are oriented in the same way, we have
α
∂y j ∂y j
det( ∂xβi ) = | det( ∂xβi )|. So
α α

XZ XZ
fα dx1α . . . dxnα = σβ fα dx1α . . . dxnα
α Rn α,β Rn

XZ ∂yβj
= ρα gβ det( )dx1α . . . dxnα
Rn ∂xiα
α,β
XZ
= ρα gβ dyβ1 . . . dyβn
α,β Rn
XZ
= ρα gβ dyβ1 . . . dyβn
β Rn

Note: Since ω is compactly supported and partitions of unity are locally finite, all sums
appearing are actually finite.

3.6 Stokes’s Theorem

Definition. A (smooth) n-manifold-with-boundary X is defined in the same way as an
ordinary n-manifold, except the codomain of each chart φ : U → V may be an open set in
Rn or in R≥0 × Rn−1 . Given p ∈ X and a chart φ : U → V at p, say p is in the boundary,
∂X, if V ⊆ R≥0 × Rn−1 and φ(p) ∈ {0} × Rn−1 . Otherwise p is in the interior X ◦ .
The notion of boundary/interior is independent of the chart.
Examples.
(i) An ordinary n-manifold X is a manifold-with-boundary with ∂X = ∅.
(ii) The closed ball X = {p ∈ Rn | ∥p∥ ≤ 1} is a manifold-with-boundary with ∂X =
S n−1 and X ◦ = {p | ∥p∥ ≤ 1}.

20
(iii) If X is a m-w-b and Y is an ordinary manifold, then X × Y is a manifold with
boundary. Then X × Y is a manifold-with-boundary. ∂(X × Y ) = (∂X) × Y and
(X × Y )◦ = X ◦ × Y .
If both X and Y are manifolds-with-boundary, then in general X × Y is a manifold-
with-corners.
Definition. If X is an oriented
V n-manifold with boundary, then ∂X is oriented as follows.
Given p ∈ ∂X, pick oX ∈ n Tp X representing the orientation of X. Pick a vector
n ∈ Tp X transverse
Vn−1 to ∂X and pointing outwards. Orient ∂X at p by the unique o∂X ∈
Tp ∂X ≤ n−1 Tp X satisfying oX = n ∧ o∂X .
V

Theorem 3.8 (Stokes’s Theorem). Given an oriented n-manifold-with-boundary X, and

a compactly supported (n − 1)-form ω on X, we have
Z Z Z
dω = ω := i∗ ω.
X ∂X ∂X

Proof. Step 1: Cover X by coordinate patches and pick a subordinate partition of unity
ρα . Then Z Z X XZ XZ
ω= ρα ω = ρα ω = ρα ω
∂X ∂X α α ∂X α ∂Uα

and Z Z X XZ
dω = d( ρα ω) = d(ρα ω).
X X α α Uα

So it suffices to prove the result when X is a coordinate patch. WLOG X = R≥0 × Rn−1 .

Step 2: Take ω = i ωi dx1 ∧ · · · ∧ dx

ci ∧ · · · ∧ dxn . Want to show that
P
Z Z
X ∂ωi
(−1)i−1 i dx1 . . . dxn = ω
R≥0 ×Rn−1 ∂x ∂(R≥0 ×Rn−1 )
i

The left side is

Z Z ∞  XZ Z ∞ 
∂ω1 1 2 n ∂ωi i
dx dx . . . dx + dx dx1 . . . dx
ci . . . dxn
Rn−1 0 ∂x1 R≥0 ×Rn−2 −∞ ∂xi
i≥2
Z
= −ω1 dx2 . . . dxn
Rn−1

Orientation of ∂(R≥0 ×Rn−1 ) is −∂x2 ∧· · ·∧∂xn . So the right side above is exactly this.

Historical fact: Stokes put this theorem as a problem in the Cambridge exam (basically
Part III).
Example. X = {x ∈ R2 : ∥x∥ ≤ a}. Area of X =
Z Z
1
dx ∧ dy = d(xdy − ydx)
X X 2

21
 Z
1
= xdy − ydx
2 ∂X
Z
1
= r2 dθ
2 ∂X
a2
Z
= dθ
2 ∂X
= πa2

3.7 Applications of Stokes

Proposition 3.9 (Integration by parts). Given an oriented n-manifold-with-boundary X,
a (p − 1)-form α on X and an (n − p)-form β such that at least one is compactly supported,
we have Z Z Z
(dα) ∧ β = α ∧ β + (−1)p α ∧ (dβ)
X ∂X X

Proof. By the Leibniz rule d(α ∧ β) = (dα) ∧ β + (−1)p−1 α ∧ dβ. Integrating and applying
Stokes gives Z Z Z
α∧β = (dα) ∧ β + (−1)p−1 α ∧ dβ
∂X X X

Proposition 3.10. If X is a compact oriented n-manifold (without boundary), then in-

tegration over X defines a linear map
Z
n
: HdR (X) → R.
X

n (X) ̸= 0.
Corollary 3.11. If X is a compact, orientable n-manifold, then HdR

Proof. Fix an orientation on X and choose a volume form ω representing

R this orientation.
Then ω integrates to a positive number in every chart, hence X ω ̸= 0 and thus 0 ̸= [ω] ∈
n (X).
HdR

22
4 Connections on vector bundles

Notation and terminology:

Vr
• Given a vector bundle E → B, an E-valued r-form is a section of E ⊗ T ∗ B.
• Given a vector space V , a V -valued r-form is V -valued r-form.
• Ωr (E) is the space of E-valued r-forms. We write Γ(E) for the space of sections of
E (i.e. Ω0 (E))
• Write gl(k, R) for the space of k × k-real matrices.

4.1 Connections
Fix a rank k vector bundle π : E → B. Given a section s, we can view it locally under each
trivialization Φα as an Rk -valued function which we will denote by vα (= pr2 ◦Φα ◦ s|Uα ).
The naive derivative is dvα , which we can view as a local E-valued 1-form via Φ−1
α . Under
a different trivialization Φβ , s becomes vβ = gβα vα . Taking the naive derivative and
transferring the answer to the α-trivialization gives
−1 −1
gβα d(gβα vα ) = gβα (dgβα )vα + dvα

So the answer is trivialization-dependent via the action of the gl(k, R)-valued 1-form
−1
gβα dgβα on vα .
Definition. A connection A on E comprises a gl(k, R)-valued 1-form Aα on Uα for each
∼
trivialization Φα : π −1 (Uα ) −
→ Uα × Rk such that on overlaps we have
−1 −1
Aα = gβα Aβ gβα + gβα dgβα (∗)

Given a connection A on E, the covariant derivative of a section s is the E-valued 1-form

dA s given under Φα by dvα + Aα vα .
By the calculations with the naive derivative, this is well-defined (i.e. consistent on over-
laps). The section s is horizontal or covariantly constant if dA s = 0
The Aα are the local connection 1-forms. Note that the zero section is always horizontal.
But non-zero horizontal sections may not exist, even locally.

23
Example (Trivial connection). Suppose E → B admits a global trivialization Φα . We
can define a connection A by Aα = 0, then defining Aη for all other trivializations by (∗).
A section is horizontal iff it is locally constant under Φα .
Lemma 4.1. Given a connection A on E → B, the covariant derivative

dA : Γ(E) → Ω1 (E)

is R-linear and satisfies the Leibniz-rule dA (f s) = f dA s + s ⊗ df .

Conversely, any R-linear map Γ(E) → Ω1 (E) satisfying this, arises from a unique con-
nection in this way.

Proof. R-linearity os obvious. We can check Leibniz under trivializations:

LHS = d(f vα ) + aα f vα = vα ⊗ df + f dvα + f Aα vα = f (dvα + Aα vα ) + vα ⊗ df = RHS

The converse is on sheet 3.

Example. Given a submanifold i : X ,− → RN , ι∗ T RN has a standard trivialization and

hence a trivial connection A0 . Now consider
dA0 orthogonal projection
→ Γ(ι∗ T RN ) −−→ Ω1 (ι∗ T RN ) −−−−−−−−−−−−−→ Ω1 (T X)
Γ(T X) ,−

It is clearly R-linear and inherits the Leibniz rule from dA0 . So it corresponds to a unique
connection on T X.
Lemma 4.2. Any vector bundle admits a connection.

Proof. Trivialize E over Uα with transition functions gβα as usual. Pick a partition of
unity ρα subordinate to this cover. Now define
X
−1
Aα = ργ gγα dgγα
γ

It remains to prove that this satisfies the transformation law (∗). We have
X
−1 −1
gβα Aβ gβα = ργ gβα (gγβ dgγβ )gβα
γ
X
−1
= ργ gγα (d(gγβ gβα ) − gγβ dgβα )
γ
X X
−1 −1
= ργ gγα dgγα − ργ gβα dgβα
γ γ
−1
= Aα − gβα dgβα

24
4.2 Connections vs End(E)
Fix rank k vector bundle E → B. Let ρ : GL(k, R) → GL(gl(k, R)) be the representation
ρ(A)(M ) = AM A−1 for A ∈ GL(k, R) and M ∈ gl(k, R).
Definition. End(E) is the vector bundle over B of rank k 2 with total space
a
End(Eb )
b∈B

If E is trivialized over Uα with transition functions gβα , then End(E) is trivialized over
the same sets with transition functions ρ(gβα ).
−1
A section M of End(E) is locally a gl(k, R)-valued function Mα such that Mβ = gβα Mα gβα .
Equivalently End(E) = E ⊗ E . ∨

Lemma 4.3. Given a connection A on E, and a section ∆ of Ω1 (End(E)), there exists

a connection A + ∆, defined locally by Aα + ∆α . Conversely, every connection A′ on E
can be written uniquely as A + ∆ for some ∆. Hence the set of connections on E is an
affine space for Ω1 (End(E)).

Proof. Just prove that everything is compatible with the transition functions.
−1 −1 −1 −1 −1
Aα + ∆α = gβα Aβ gβα + gβα dgβα + gαβ ∆β gαβ = gβα (Aβ + ∆β )gβα + gβα dgβα

For the other direction, verify that A′ − A transforms correctly, i.e. like a section of
Ω1 (End(E)).

4.3 Curvature algebraically

Definition. The exterior covariant derivative is the unique R-linear map dA : Ω• (E) →
Ω•+1 (E) satisfying the Leibniz rule

dA (s ⊗ ω) = (dA s) ∧ ω + s ⊗ dω

for sections s of E and forms ω. Locally in trivializations, an E-valued p-form σ becomes

an Rk -valued p-form σα , then dA σ is given by dσα + Aα ∧ σα .
Warning. (dA )2 ̸= 0 in general.
Proposition 4.4. There exists a unique End(E)-valued 2-form F on B such that for all
E-valued forms σ:
(dA )2 σ = F ∧ σ.

Proof. Locally in a trivialization (dA )2 σ is given by

d(dσα + Aα ∧ σα ) + Aα ∧ (dσα + Aα ∧ σα ) = (dAα ) ∧ σα − Aα ∧ dσα + Aα ∧ dσα + Aα ∧ Aα ∧ σα

25
 = Fα ∧ σα

where Fα = dAα +Aα ∧Aα . Then one can check that this transforms like a End(E)-valued
2-form.

Definition. F is the curvature of A. A is flat if F = 0.

Examples.
(i) Trivial connections are flat. Conversely, if A is flat, then it is locally trivial.
 
1 0
(ii) Consider R2 → R × S 1 with a connection A given by Aα = f dx +
0 −1
 
0 −1
g dθ under the standard trivialization. Then
1 0
     
1 0 0 −1 0 −1
Fα = dAα + Aα ∧ Aα = df ∧ dx + dg ∧ dθ + 2f g dx ∧ dθ
0 −1 1 0 −1 0
      
∂f 1 0 ∂g 0 −1 0 1
= − + − 2f g dx ∧ dθ
∂θ 0 −1 ∂x 1 0 1 0

4.4 Parallel transport

Fix E → [0, 1] with connection A.
Lemma 4.5. Given v0 ∈ E0 , there exists a unique horizontal section s with s(0) = v0 .
This s depends linearly on v0 .

Proof. Locally in trivializations the condition that s is horizontal says dvα + Aα vα = 0

(∗). We have Aα = Mα dt for some gl(k, R)-valued function Mα where t is the coordinate
on [0, 1]. Then (∗) ⇔ vdtα + Mα vα = 0. This is a linear ODE. By standard ODE theory
solutions exist locally and are unique (locally, hence globally). The solution depends
linearly on the initial condition. Left to prove global existence: Local existence says that
for all p ∈ [0, 1] there exists a fibrewise basis of horizontal sections locally about p. By
compactness of [0, 1] there exist 0 = a0 < a1 < · · · < aN P = 1 such that on [ai , ai+1 ]
we have such a local fibrewise basis s1i , . . . , ski . Write v0 = kj=1 λ0j sj0 (0). Then define
s on [a0 , a1 ] by j λ0j sj0 . Now write s(a1 ) = kj=1 λ1j sj0 (0) and extend s to [a1 , a2 ] as
P P

s(a1 ) = kj=1 λ1j sj0 . Then keep going.

P

Definition. The linear map E0 → E1 , v0 7→ s(1) is the parallel transport of v0 along

[0, 1] (w.r.t. A).
Now go back to general vector bundles E → B with connection A. Suppose γ : [0, 1] → B
is a path. Then γ ∗ Aα defines a connection on γ E , denoted γ ∗ A.

26
Definition. Given a vector v0 ∈ Eγ(0) , the unique horizontal section s of γ ∗ E starting at
v0 is the horizontal lift of γ to E (starting at v0 ). The vector s(1) ∈ Eγ(1) is the parallel
transport of v0 along γ. Doing this for all v0 gives a linear map Pγ : Eγ(0) → Eγ(1) . If γ
is a loop, i.e. γ(0) = γ(1), then Pγ : Eγ(0) → Eγ(0) is the monodromy or holonomy of A
around γ.
Examples.
(i) Consider T S 2 with the “orthogonal projection” connection. Given path γ on S 2 and
v0 ∈ Tγ(0) S 2 , the horizontal lift is the map v : [0, 1] → T S 2 such that
• v(t) ∈ Tγ(t) S 2 for all t.
• v̇(t) in R3 is orthogonal to Tγ(t) S 2 , so that the orthogonal projection to Tγ(t) S 2
is 0
   
2 1 1 0 0 −1
(ii) Returning to R → R × S with connection Aα = f dx + g dθ.
0 −1 1 0
 
1 0
Consider γ(t) = (t, 0). Horizontal lift v of γ starting at v0 satisfies v̇+f v=
0 −1
 −λ 
e 0 Rt
0. So v(t) = λ v0 where λ = 0 f (x, 0)dx. Similarly, the monodromy
0 e
 
cos φ sin φ R 2π
around γ(t) = (0, 2πt) is where φ = 0 g(0, θ)dθ.
− sin φ cos φ

4.5 Curvature geometrically

Fix E → B, with connection A. Fix also a point p ∈ B, a trivialization Φα around p, and
local coordinates xi about p.
Let Aα = Ai dxi . where the Ai are gl(k, R)-valued functions. Similarly let Fα = Fij dxi ⊗
dxj . WLOG p = (0, . . . , 0). For a, b ∈ R small, let γ1 (t) = atei , γ2 (t) = aei + btej in x
coordinates. Then let γ3 , γ4 be the other two sides of the rectangle.
Proposition 4.6. Letting Pa,b = Pγ4 Pγ3 Pγ2 Pγ1 ∈ End(Ep ) we have

∂ 2 Pa,b
|a=b=0 = −Fij (p)
∂a∂b
Proof. For formal proof see Sheet 3. We will give an intuitive sketch proof ignoring analysis
details, but these details can be filled in (e.g. see Nicolaescu Proposition 3.3.14).
Parallel transport in the xi direction satisfies v̇ = −Ai v, so Pγ1 = I −aAi (p)+. . . where . . .
means higher order terms which will wash out. Similarly Pγ2 = I − bAj (γ1 (1)) + · · · = I −
∂A
b(Aj (p)+a ∂xij (p))+. . . . So Pγ2 ◦Pγ1 = I −aAi (p)−bAj (p)+abAj (p)Ai (p)−ab ∂A ∂xi
i
(p)+. . . .

27
 ∂bi
Similarly Pγ4 ◦ Pγ3 = I + aAi (p) + bAj (p) + ab ∂x j
(p) + abAj Ai (p) + . . . . So
 
∂Ai ∂Aj
Pa,b = I + ab (p) − (p) + Aj (p)Ai (p) − Ai (p)Aj (p) + a(. . . ) + b(. . . ) + . . .
∂xj ∂xi

So
∂ 2 Pa,b ∂Ai ∂Aj
|a=b=0 = j
(p) − (p) + Aj (p)Ai (p) − Ai (p)Aj (p) = −Fij (p)
∂a∂b ∂x ∂xi

V2
Corollary 4.7. If v ∈ Ep is such that F (p)v ̸= 0 in Ep ⊗ Tp∗ B, then there does no
exist a local horizontal section s about p with s(p) = v.

Proof. If such an s exists, then Pγ1 (v) = s(γ1 (1)), similarly for the other paths, so Pa,b (v) =
s(p) = v for all a, b. So by the Proposition for all i, j we have −Fij (p)v = 0, hence
F (p)v = 0.

Example. Consider R → R2 with Aα = Cx1 dx2 . Let γ1 , . . . , γ4 be as before. Then

∂ 2 Pa,b
Pγ1 = id = Pγ3 = Pγ4 and Pγ2 = e−Cab . Hence Pa,b = e−Cab . Then ∂a∂b |a = b = 0 = −C.
1 2
So F12 = C which is of course also clear from F = Cdx ∧ dx = dAα + Aα ∧ Aα .
Explicitly, if s were a local horizontal section about p, given by vα in our trivialization,
then we would have dvα +Aα vα = 0, i.e. dvα +Cx1 vα dx2 = 0, i.e. ∂v α
∂x1
= 0, ∂v
∂x2
α
= −Cx1 vα .
2
Hence 0 = ∂x∂ 2v∂x
α
1 = −Cvα . If C ̸= 0, then the only horizontal local section is 0.

Example. Consider R → S 1 with Aα = Cdθ. Local horizontal sections exist and have
the form vα = Ke−cθ . This extends to a global section if C = 0. If C ̸= 0, then this does
not extend, due to the presence of non-trivial monodromy e−2πC .
Summary. Curvature is the local obstruction from the existence of horizontal sections,
monodromy is the global obstruction.

28
5 Flows and Lie derivatives

5.1 Flows
Fix a manifold X and a vector field v on X. Given a point p ∈ X, can try to flow along
v from p, i.e. solve the ODE γ̇(t) = v(γ(t)) and γ(0) = p. By standard ODE theory,
solutions exist locally and are unique. Solutions are called integral curves of v.
Definition (Non-standard). A flow domain is an open neighborhood U of X × 0 in X × R
such that for all p ∈ X the set U ∩ (p × R) is connected.
Definition. A local flow of v comprises a flow domain U and a smooth map Φ : U → X
such that
• Φ(−, 0) = idX .
• d
dt Φ(p, t) = v(Φ(p, t)) for all (p, t) ∈ U .
We write Φt (p) for Φ(p, t).
Previous ODE discussion plus smooth dependence on initial conditions, tells us that ocal
flows exist and are unique in the sense that if (U1 , Φ1 ) and (U2 , Φ2 ) are local flows, then
Φ1 = Φ2 on U1 ∩ U2 .
A vector field is complete if it has a global flow, i.e. one with U = X × R. Not all vector
fields are complete, e.g. x2 ∂x on R. But if v is compactly supported, then v is complete.
(Idea: for each p ∈ X, there exists Up neighborhood of p and εp > 0 such that a flow exists
on Up × (−εp , εp ). By compactness, get local flow on U = X × (−ε, ε) for some ε > 0. Can
then define a global flow by Φt = (Φt/N )N for n ≫ 0.)
Lemma 5.1. If Φ is a local flow of v, then Φs+t = Φs ◦ Φt whenever this makes sense.
So in particular (Φt/N )N = Φt when this makes sense, and Φ−t = (Φt )−1 .

Proof. Fix p ∈ X, fix t. Let q ∈ Φt (p). Then γ1 (s) := Φs+t (p), γ2 (s) = Φs ◦ Φt (p). These
two curves both satisfy γ̇i = v ◦ γi and γi (0) = q. So by uniqueness of solutions to ODEs
get γ1 = γ2 .

5.2 Lie Derivatives

Fix X and v, and let Φ be a local flow of v.

29
Definition. For a tensor or form T on X, its Lie derivative is
d
Lv T = |t=0 (Φt )∗ T
dt

Lemma 5.2. We have

d t ∗
(Φ ) T = (Φt )∗ Lv T
dt
Proof. We have
d t ∗ d d
(Φ ) T = |h=0 (Φt+h )∗ T = |h=0 (Φt )∗ (Φh )∗ T = (Φt )∗ Lv T
dt dh dh

Lemma 5.3. For a function f we have

Lv f = df (v).

For a 1-form α we have

∂v i
 
∂αj
Lv α = vi i + αi j dxj .
∂x ∂x

d t d
Proof. At each point p we have Lv f = dt |t=0 f (Φ (p)) = df ( dt |t=0 Φt (p)) = df (v).
We have
d
Lv α = |t=0 (Φt )∗ α
dt
d
= |t=0 (αi ◦ Φt )d(xi ◦ Φt )
dt
= (Lv αi )dxi + αi d(Lv xi )
∂αi
= v j j dxi + αi dv i
∂x
∂αi ∂v j
= (v j j + αj i )dxi
∂x ∂x

Lemma 5.4. For a 1-form α and a vector field w we have

d(αi wi )(v) = Lv (αi wi ) = (Lv α)i wi + αi (Lv w)i

For any tensors S, T we have

Lv (S ⊗ T ) = (Lv S) ⊗ T + S ⊗ (Lv T )

30
Proof. Pullback by Φt commutes with contraction and with ⊗. Then proceed as in the
proof of the ordinary Leibniz rule.

Corollary 5.5. For a vector field w we have

i i
 
j ∂w j ∂v
Lv w = v −w ∂ xi
∂xj ∂xj

Proof. By first part of the previous lemma, for any 1-form α we have Lv (αi wi ) = (Lv α)i wi +
αi (Lv w)i , so we get by the lemma before
i j
 
j ∂(αi w ) j ∂αi j ∂v
v = v +α wi + αi (Lv w)i
∂xj ∂xj ∂xi

Hence
∂αi ∂wi j i ∂αi i ∂v
j
v j wi + v j
α i = v w + α j w + αi (Lv w)i
∂xj ∂xj ∂xj ∂xi
and thus
∂wi i ∂v
j
αi (Lv w)i = v j αi − α j w .
∂xj ∂xi
This holds for all α, so
∂wi j ∂v
i
(Lv w)i = v j − w .
∂xj ∂xj

Definition. The Lie bracket of v and w is

[v, w] := Lv w = −Lw v.

This operation makes the Γ(T X) into a Lie algebra, i.e. a vector space equipped with a
bilinear operation [·, ·] satisfying
• [v, v] = 0 for all v (alternating)
• [u, [v, w]] + [v, [w, u]] + [w, [u, v]] = 0 for all u, v, w (Jacobi identity)
Lemma 5.6. If F : X → Y is a diffeomorphism, then for any vector field v on Y , and
any tensor T on Y , we have
F ∗ (Lv T ) = LF ∗ v (F ∗ T )

Proof. We have
d
F ∗ (Lv T ) = F ∗ |t=0 (Φt )∗ T
dt
d
= |t=0 F ∗ (Φt )∗ T
dt

31
 d
= |t=0 F ∗ (Φt )∗ (F ∗ )−1 F ∗ T
dt
d
= |t=0 (F −1 ◦ Φt ◦ F )∗ F ∗ T
dt
But F −1 ◦ Φt ◦ F is a flow of F ∗ v.

5.3 Homotopy invariance of de Rham cohomology

Definition. Given an r-form α and a vector field v, the (r −1) form ιv α or v⌟α is defined
to by
(ιv α)i1 ···ir−1 = v j αji1 ···ir−1

The Lie derivative and exterior derivative are related as follows:

Proposition 5.7 (Cartan’s magic formula). For a vector field v, an r-form α, we have

Lv α = (dιv α) + ιv dα

Proof. Example Sheet 3.

Proof of Proposition 3.4 (Homotopy Invariance of de Rham cohomology). Let F : [0, 1] ×

X → Y be a homotopy between F0 , F1 . Write Ft for F (t, −). Let it : X → [0, 1] × X
be the inclusion x 7→ (t, x). Note that it = Φt ◦ i0 where Φt is the flow of ∂t . Note that
Ft = F ◦ it . For any form α on Y we have
Z 1
∗ ∗ d ∗
F1 α − F0 α = Ft αdt
0 dt
Z 1
d
= (F ◦ Φt ◦ i0 )∗ αdt
0 dt
Z 1
d
= i∗0 (Φt )∗ F ∗ αdt
0 dt
Z 1
= i∗0 (Φt )∗ L∂t (F ∗ α)dt
0

Now suppose α is closed. By Cartan’s magic formula we have

Lv (F ∗ α) = d(ι∂t F ∗ α) + 0

So
Z 1
F1∗ α − F0∗ α = i∗0 (Φt )∗ d(ι∂t F ∗ α)dt
0
Z 1
= i∗t d(ι∂t F ∗ α)dt
0

32
 Z 1
=d i∗t ι∂t F ∗ αdt
0

So F1∗ α − F0∗ α is exact and thus F0 , F1 induce the same map on de Rham cohomology.

33
6 Foliation and Frobenius integrability

6.1 Foliations
If F : X → Y is a submersion, then X decomposes into slices F −1 (q) which are submani-
folds of dimension dim X − dim Y .
A k-foliation on X is a local decomposition of X into k-dimensional slices, but the slices
need not globally form submanifolds.
Example. Consider X = T 2 = R2 /Z2 . For any α ∈ R, we can locally slice X into lines
of slope α. If α is irrational, then the slices do not globally form submanifolds.
Definition. An atlas on X is k-foliated if the transition functions φβ ◦ φ−1 α are locally
of the form Rk × Rn−k ∋ (x, y) 7→ (ζ(x, y), η(y)) ∈ Rk × Rn−k . Two k-foliated atlases are
equivalent if their union is k-foliated, and a k-foliation is an equivalence class of k-foliated
atlases. We will usually write associated local coordinates as x1 , . . . , xk , y 1 , . . . , y n−k . Slices
are given locally by y = const.
Example. If F : X → Y is a submersion, then foliated charts correspond to local
coordinates on X in which F corresponds to projection onto the last n − k components.

6.2 Distributions
Fix an n-manifold X.
Definition. A k-plane distribution on X is a rank k subbundle D of T X.
Example. ⟨∂x , ∂y ⟩ and ⟨∂x + y∂z , ∂y ⟩ each define a 2-plane distribution on R3 . Note that
⟨∂x , ∂y ⟩ = ker dz and ⟨∂x + y∂z, ∂y ⟩ = ker(dz − ydx).
In general a k-plane distribution can be written locally as the kernel as the kernel of n − k
1-forms.
Example. If X is equipped with a k-foliation, with foliated coordinates x, y as usual, then
Tn−k
⟨∂x1 , . . . , ∂xk ⟩ = i=1 ker dy i is a k-plane distribution. It describes the tangent spaces to
the slices.
Definition. A k-plane distribution D is integrable if it arises from a k-foliation in this
way

34
If k = 1, then every distribution is integrable: Locally D = ⟨v⟩ for some vector field v,
then X is foliated by integral curves of v.

6.3 Frobenius integrability

Theorem 6.1 (Frobenius integrability). A distribution D on X is integrable iff D is
closed under [·, ·], i.e. for all vector fields v, w tangent to D, [v, w] is also tangent to D.

Proof. Both conditions are local, so it suffices to work near a point p. If D is integrable
with local foliated coordinates x, y, then D = ⟨∂x1 , . . . , ∂xk ⟩. Can easily check by hand
that for any smooth coefficients f i , g i , [f i ∂xi , g j ∂xj ] ∈ D.
Conversely, suppose D is closed under [·, ·]. We want to show that there exist local coordi-
nates x, y such that D = ⟨∂x1 , . . . , ∂xk ⟩. First choose local coordinates s1 , . . . , sk , t1 , . . . , tn−k
about p such that D = ⟨∂s1 , . . . , ∂sk ⟩ at p. WLOG p corresponds to s = 0, t = P0. Locally
there exist uniquely determined smooth functions aij such that vi := ∂si + aij ∂tj lies
in D. Let Φti be the flow of vi . Now define F : open neighborhood of 0 in Rn → X by
1 k
F (x, y) = Φx1 ◦ · · · ◦ Φxk (s = 0, t = y). This has F (0) = p and D0 F (∂xi ) = vi = ∂si and
D0 F (∂yi ) = ∂tj . So D0 F is an isomorphism, hence F defines a parametrization near p. It
suffices to show that ∂xi = vi . Suppose the Φi all commute, then for each i we would have

d 1 i k
∂xi = |h=0 Φx1 · · · Φxi +h · · · Φxk (0, y)
dh
d i 1
ci . . . Φxk (0, y)
= |h=0 Φxi +h Φx1 · · · Φ k
dh
i 1 k
= vi (Φxi Φx1 · · · Φ
ci . . . Φx (0, y))
k
= vi

so we would be done.
i j j i
Left to show: Φxi ◦Φxj = Φxj ◦Φxi for all i, j, i.e. that [vi , vj ] =
P0 for all i, j. We know that
D is closed under [·, ·], so there exist bijl such that [vi , vj ] = l bijl vl . Equate coefficients
of ∂sl to get all bijl = 0.

Example. Consider D = ⟨∂x + y∂z , ∂y ⟩ on R3 . This is not closed under [, ] since [∂x +
y∂z , ∂y ] = −∂z ∈
/ D. So D is not integrable.
∂f ∂f ∂f
By hand: If D were tangent to a surface f = const, then we would have ∂x +y ∂z = ∂y = 0.
∂2f ∂2f
So 0 = ∂x∂y + ∂f∂z + y ∂y∂z =
∂f
∂z . Then also ∂f
∂x = −y ∂f
∂z = 0 and so df = 0. So {f = const}
is not a surface!

35
7 Connections on vector bundles with extra
structure

7.1 Connections on T X
Suppose A is a connection on E = T X → X. Given local coordinates x1 , . . . , xn on X, we
get a trivialization of E by ∂x1 , . . . , ∂xn . Call this a coordinate trivialization. We typically
write the induced local connection 1-form as Γijk dxk where i, j are the matrix indices on
gl(n, R). So for a vector field v we have (dA v)i = dv i + Γijk v j dxk .

Warning. The Γijk do not transform like a tensor of type (1, 2). But the space of
connections on E is an affine space for Ω1 (End E) = Γ(E ⊗ E ∨ ⊗ T ∗ X) = Γ(T X ⊗ T ∗ X ⊗
T ∗ X), i.e. the space of tensors of type (1, 2).
Definition. The solder form θ is the E-valued 1-form that corresponds to the fibrewise
identity map under E ⊗ T ∗ X = T X ⊗ T ∗ X = End(T X).
The torsion T of A is the E-valued 2-form dA θ. A is torsion-free if T = 0.
In a coordinate trivialization θ = ei ⊗ dxi , so T = d(ei ⊗ dxi ) + Aα ∧ (ei ⊗ dxi ) =
Γj ik ej ⊗ dxk ∧ dxi .

So A is torsion-free iff Γijk = Γikj .

Proposition 7.1 ((First) Bianchi identity). dA T = F ∧ θ.

Proof. We have dA T = (dA )2 θ = F ∧ θ.

Definition. A curve γ in X is a geodesic (w.r.t. A) if γ̇ is covariantly constant as a

section of γ ∗ T X. This is equivalent to the geodesic equation

γ̈ i + Γijk γ̇ j γ̇ k = 0.

Note that
• A connection on T X induces connections on T ∗ X and all bundles of tensors and
forms. If we had taken the covariant derivative of θ as a tensor of type (1, 1), we
would have got 0 automatically.

36
 • The curvature of A is an End(E)-valued 2-form, which we can view as a tensor of
type (1, 3) F ijkl that is antisymmetric in k, l.

• Often dA or A itself is called ∇ and the contraction of dA with a vector or vector

field v is written ∇v .

7.2 Orthogonal vector bundles

Fix a vector bundle E → B.
Definition. An inner product on E is a section of (E ∨ )⊗2 which is fibrewise symmetric
and positive definite.
Lemma 7.2. E admits an inner product.

Proof. Define locally and glue using a partition of unity.

Definition. An orthogonal vector bundle is a vector bundle equipped with an inner prod-
uct g. A trivialization Φα is orthogonal if under Φα , g becomes the standard inner product
on Rk .
Note: Transition functions between orthogonal trivializations take values in O(k).
Fix an orthogonal vector bundle (E, g) → B.
Lemma 7.3. E can be covered by orthogonal trivializations.

Proof. We can locally trivialize E by sections s1 , . . . , sk . Apply Gram-Schmidt fibrewise

to make the si orthonormal. The corresponding trivialization is then orthogonal.

Definition. A connection A on E is orthogonal if g is covariantly constant w.r.t. to the

induced connection on (E ∨ )⊗2 .
Lemma 7.4. E admits an orthogonal connection, and the space of orthogonal connections
on E is an affine space for Ω1 (o(E)) ⊆ Ω1 (End(E)) where o(E) ≤ End(E) is the bundle
of skew-adjoint endomorphisms of E
Lemma 7.5. If A is an orthogonal connection on (E, g), then its curvature is an o(E)-
valued 2-form.

37
8 Riemannian geometry

8.1 Riemannian metrics

Fix an n-manifold X.
Definition. A (Riemannian) metric on X is an inner product on T X. A Riemannian
manifold is a pair (X, g) where X is a manifold and g is a Riemannian metric on X.
Since every vector bundle admits an inner product, every manifold admits a Riemannian
metric.
Given a Riemannian metric gij , we write g ij for the dual metric on T ∗ X. This satisfies
(and is defined by) g ij = g ji and g ij gjk = δ ik . We denote contraction with gij or g ij by
raising or lowering indices, e.g. gil T ijk = Tl jk or g ik Sij = S kj .

A section T ij of End(T X) lies in o(T X) iff T ij gik = −T ik gji , i.e. Tkj = −Tjk . When
dxi ⊗dxj +dxj ⊗dxi
writing coordinate expressions, we use dxi dxj to mean 2 , e.g. the standard
Riemannian metric on Rn is gEucl = i (dxi )2
P

8.2 The Levi-Civita connection

Fix a Riemannian manifold (X, g).
Theorem 8.1 (Fundamental theorem of Riemannian geometry). There exists a unique
torsion-free orthogonal connection on T X.

Proof. We will prove the more generally statement that the map {orthogonal connections} →
Ω2 (T X) sending a connection to its torsion, is a bijection.
Fix an arbitrary orthogonal connection A0 . Any other orthogonal connection A can be
written uniquely as A0 + ∆ for an o(E)-valued 1-form ∆. We will show that the map
Ω1 (o(E)) → Ω2 (T X), ∆ 7→ TA0 +∆ − TA0 is a bijection. This map sends ∆ to ∆ ∧ θ, i.e
∆ikj − ∆ijk . (If A0 is locally Γijk , then A0 + ∆ is Γijk + ∆ijk , so (TA0 +∆ − TA0 )ijk =
(Γ + ∆)ikj − (Γ + ∆)ijk − (Γikj − Γijk )).
It is induced by the bundle morphism F : o(T X)⊗T ∗ X → T X ⊗∧2 T ∗ X given by wedging
with θ. So it suffices to show that
F is an isomorphism which we can do fibrewise.
n ∗
Note both bundles have rank n 2 since o(T X) ⊗ T X = {∆ijk | ∆ijk = −∆jik } and

38
T X ⊗ ∧2 T ∗ X = {T ijk : T kjk = −T ikj }. So it is enough to show that ∆ 7→ ∆ ∧ θ is injective,
i.e. that if ∆ijk satisfies ∆ijk = −∆jik and ∆ijk = ∆ikj (∆ ∈ ker), then ∆ = 0. But if
∆ satisfies these two conditions, then ∆ijk = −∆jik = −∆jki = ∆kji = ∆kij = −∆ikj =
−∆ijk .

Definition. This is the Levi-Civita connection on (X, g). Its components Γijk are called
Christoffel symbols.
The explicit coordinate expressions are
1
Γijk = (∂j gik + ∂k gji − gi gjk ) .
2

Proposition 8.2. If ι : X ,→ RN is an embedding, then

• X inherits a metric ι∗ gEucl , hence has an induced Levi-Civita connection.
• T X carries the ”orthogonally project from ι∗ T RN ” connection.
The connections coincide.

Proof. Example Sheet 4.

8.3 The Riemann tensor

Fix (X, g).
Definition. The curvature of the Levi-Civita connection ∇ is the Riemann tensor Rijkl .
This is an o(T X)-valued 2-form, viewed as a tensor of type (1, 3).
The Riemann tensor has the following properties:
• Rijkl = −Rijkl since it is a 2-form.
• Rijkl = −Rjikl since it takes values in o(T X).
• First Bianchi identity R ∧ θ = d∇ T = 0, i.e. Rijkl + Riklj + Riljk = 0.

• Second Bianchi dEnd ∇ R = 0.

8.4 Hodge theory

Let (X, g) be an oriented Riemannian manifold. The dual metric g ij gives an inner product
on T ∗ X and induces inner products on ∧p T ∗ X for all p. Explicitly, if α1 , . . . , αn is a local
fibrewise orthonormal basis of 1-forms, then the αI = αi1 ∧ · · · ∧ αip are a fibrewise
orthonormal basis of p-forms.

39
In particular, there is a distinguished unit volume form ω.
Given a p-form β there exists a unique (n − p)-form ∗β such that for all p-forms α
α ∧ ∗β = ⟨α, β⟩ω
E.g. ∗αI = ±αJ where J = {1, . . . , n} \ I.
Definition. The map
∗ : Ωp (X) → Ωn−p (X)
is the Hodge star operator.
By considering its action on the αI , can see that it is a fibrewise isometry and ∗2 =
(−1)p(n−p) idΩp (X) .
Example. Take R3 with the standard metric and orientation. Then ω = dx1 ∧ dx2 ∧ dx3 ,
so ∗dx1 = dx2 ∧ dx3 , ∗(dx2 ∧ dx3 ) = dx1 and cyclically.
assume X is compact. Define an inner product on Ωp (X) by ⟨α, β⟩X = X ⟨α, β⟩ω =
R
Now
R
X α ∧ ∗β. For (p − 1)-form α, p-form β we have
Z
⟨dα, β⟩X = (dα) ∧ ∗β
Z X

= d(α ∧ ∗β) − (−1)p−1 α ∧ d ∗ β

X Z
p
= (−1) α∧d∗β
X
= ⟨α, (−1)p ∗−1 d ∗ β⟩X
So the operator δ : Ωp → Ωp−1 (X) given by (−1)p ∗−1 d∗ is adjoint to d.
Definition 8.3. This δ is the codifferential. A form α is coclosed if δα = 0, coexact if
∃β such that α = δβ.
NB: δ = (−1)np+n+1 ∗ d∗ and the definition of δ also makes sense for non-compact X.
Notice δ 2 = − ∗−1 d ∗ ∗−1 d∗ = − ∗ d2 ∗ = 0.
Definition. The Laplace-Beltrami operator ∆ : Ωp (X) → Ωp (X) is defined by dδ + δd =
(d + δ)2 .
A form α is harmonic if ∆α = 0. This is equivalent to α being closed and coclosed (Sheet
4). We denote the space of harmonic forms by Hp (X).
Theorem 8.4. The map
p
Hp (X) −→ HdR (X)
α 7−→ [α]
is an isomorphism, i.e. every cohomology class has a unique harmonic representative.

40
Idea: Hp (X) = ker ∆ = ker d ∩ ker δ = ker d ∩ (im d)⊥ ∼ p
= ker d/ im d = HdR (X)
Theorem 8.5 (Hodge decomposition). The space Hp (X) is finite-dimensional and we
have orthogonal decompositions

Ωp (X) = Hp (X) ⊕ dδΩp (X) ⊕ δdΩp (X)

= Hp (X) ⊕ δΩp−1 (X) ⊕ δΩp+1 (X)

Proof. See Section 10.4.3 in Nicolaescu.

Proof of Theorem 8.4. It suffices to show that

ker d = Hp (X) ⊕ dΩp−1 (X)

LHS⊇RHS: harmonic and exact forms are both closed.

LHS⊆RHS: by Hodge decomposition RHS = (im δ)⊥ , so it suffices to prove ⟨ker d, im δ⟩ =
0. Given α ∈ ker d, we have for all β, ⟨α, δβ⟩ = ⟨dα, β⟩ = 0.

41
9 Lie groups and principal bundles

9.1 Lie groups and Lie algebras

Definition. A Lie group is a manifold G equipped with a group structure such that mul-
tiplication and inversion m : G × G → G, i : G → G are smooth.
An embedded Lie subgroup of G is a submanifold H that is also a subgroup. The restric-
tions of the operations from G to h make H into a Lie group.
Examples. GL(n, R) is a Lie group. SL(n, R), O(n), SO(n) are embedded Lie subgroups.
Similarly SL(n, C), U (n), SU (n) are embedded Lie subgroups of GL(n, C).
Definition. For each g ∈ G we get diffeomorphisms Lg , Rg , Cg : G → G defined by
Lg (h) = gh, Rg (h) = hg, Cg (h) = ghg −1 for all h.
A tensor T is left/right/conjugation invariant iff (Lg )∗ T = T for all g etc. It is bi-
invariant if it is both left and right invariant.
Lemma 9.1. For any h ∈ G, the map

{left-invariant tensor field of type (p, q)} −→ {tensors of type (p, q) at h}

T 7−→ Th

is an isomorphism. Similarly for right-invariant.

Proof. The inverse map is define by Tg = (Lgh−1 )∗ Th .

Definition. The Lie algebra g of a Lie group G is Te G.

Examples.
• gl(n, R) = {n × n matrices}.
• sl(n, R) = {A ∈ gl(n, R) | tr A = DI det A = 0}
• o(n) = {A ∈ gl(n, R) | AT + A = 0}.
For ξ ∈ g let ℓξ denote the corresponding left-invariant vector field, i.e. ℓξ (g) = (Lg )∗ ξ.
Lemma 9.2. The Lie bracket of left-invariant vector fields is left-invariant.

42
Proof. Given left-invariant vector fields v, w, we have for all g ∈ G that

(Lg )∗ [v, w] = [(Lg )∗ v, (Lg )∗ w] = [v, w]

where we used the diffeomorphism-invariance of the Lie derivative.

Definition. The Lie bracket on g is defined by [ξ, η] = ζ where ζ is the unique element
of g such that [ℓξ , ℓη ] = ℓζ . It inherits alternating, bilinear, Jacobi from the Lie bracket of
vector fields.

9.2 Lie group actions

Definition. An action of G on a manifold X is smooth if the action map σ : G × X → X
is smooth. Similarly for right actions.
E.g. GL(n, R) acting on Rn , G acting on itself by conjugation, O(n) acting on S n−1 ⊆ Rn .
Example. The adjoint action/representation of G on g is

Adg (ξ) := (Cg )∗ ξ.

Definition. Given a smooth left action of G on X, the infinitesimal action of ξ ∈ g on

x ∈ X is
ξ · X := D(e,g) σ(ξ, 0) = [γ(t)x]
where γ is any curve representing ξ. Similarly for right actions but with [xγ(t)].

9.3 Principal bundles

Fix a Lie group G.
Definition. A (principal) G-bundle P over B is defined the same way as a vector bun-
≃
dle except trivializations are Φα : π −1 (Uα ) −
→ Uα × G and on overlaps Φβ Φ−1
α (b, g) =
(b, gβα (b)g) for (necessarily smooth) maps gβα : Uα ∩ Uβ → G.
Example. Given a rank k vector bundle E → B, its frame bundle F (E) → B is the
principal GL(k, R)-bundle with F (E)b := {ordered bases in Eb }. Similarly, if E has an
inner product, can consider the orthonormal frame bundle F0 (E), which is a principal
O(k)-bundle.
Note that
• Many definitions transfer from vector bundles, e.g. sections, constructions by gluing
etc.
• P admits a right G-action, defined in trivializations, i.e. Φ−1 −1
α (b, x)g := Φα (b, xg).

43
 • Sections s over U ⊆ B correspond to trivializations Φ over U :
– Given Φ, define s by s(b) = Φ−1 (b, e)
– Given s, define Φ by Φ(s(b)g) = (b, g).

9.4 Connections
Fix a principal G-bundle P → B. Write Rg : P → P for the right action of g.
Definition. A connection on P is a g-valued 1-form A on P , satisfying:
• A(p · ξ) = ξ for p ∈ P, ξ ∈ g.
• Rg∗ A = Adg−1 A (A is equivariant).
Given a local section sα (or equivalently a trivialization Φα ), the local connection 1-form
Aα is s∗α A.
Lemma 9.3. On overlaps we have Aα = Adg−1 Aβ + (Lg−1 )∗ dgβα .
βα βα

Conversely, given Aα transforming this way, they arise from a unique connection A on P .

Proof. Sheet 4.

N.B. If P = F (E), then a connection on P is equivalent to a connection on E.

Definition.
P The curvature P of A is theP g-valued 2-form F on P given by F = dA+ 12 [A∧A]
where [( i ξi ⊗ αi ) ∧ ( j ηj ⊗ βj )] = i,j [ξi , ηj ] ⊗ (αi ∧ βj ).
A is flat if F = 0.

44