Bachelor’s project for applied mathematics

Faculty of Science and Engineering

The Ricci flow on two-dimensional

almost-Riemannian manifolds
Author: First supervisor:
L. de Ridder Dr. M. Seri

Student ID: Second supervisor:

s3145883 Dr. R.I. van der Veen

July 5, 2021

Abstract
The Ricci flow has shown to be a powerful tool in the study of Riemannian geometry. In this setting, we
provide explicit proofs for the isometric invariance of the Levi-Civita connection, Riemann curvature tensor and,
subsequently, the Ricci curvature. The latter allows one to show that the Ricci flow is invariant under the
infinite dimensional group of diffeomorphisms. This causes the Ricci flow, characterized as a heat-type non-linear
partial differential equation, to be weakly parabolic. In general, existence and uniqueness theorems only apply
to strongly parabolic equations. With DeTurck’s trick, we show that on a closed Riemannian manifold of any
dimension, existence and uniqueness of short-time solutions to the Ricci flow can still be obtained. In addition, we
investigate the evolution of the Ricci flow on two-dimensional almost-Riemannian structures (2-ARS) on compact,
oriented and connected smooth manifolds. These are generalized Riemannian structures on surfaces for which an
orthonormal frame is obtained from a pair of vector fields that satisfy the Hörmander condition. The vector fields
can become collinear at certain points, which as a collection define a singular set. If one removes the singular
set from the connected manifold, one obtains two regular Riemannian structures on both parts of the surface.
However, these are non-complete Riemannian surfaces with boundary, giving rise to difficulties regarding the
evolution of Ricci flow on these surfaces.
Contents
1 Introduction 3

2 Differential geometry 5
2.1 Smooth manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Tangent vectors, covectors and tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Tangent bundle and vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Cotangent bundle and tensor bundle . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Riemannian manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Geodesics and normal coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3 Ricci flow 31
3.1 Time derivatives and deformations of geometric quantities . . . . . . . . . . . . . . . . . . 32
3.2 Evolution of curvature under Ricci flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Short-time existence and uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Almost-Riemannian geometry & Ricci flow 45

4.1 Definitions & properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 Two-dimensional almost-Riemannian structures . . . . . . . . . . . . . . . . . . . . 48
4.2 A discussion on Ricci flow on two-dimensional almost-Riemannian manifolds . . . . . . . . 50
4.2.1 Ricci flow on surfaces with boundary . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.2 Ricci flow on non-complete surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.3 A conjecture on the Ricci flow on two-dimensional almost-Riemannian manifolds . 53

5 Conclusion 55

A A note on Riemannian submanifolds 56

B Covariant derivative of tensor fields in normal coordinates 58

2
1 INTRODUCTION

1 Introduction
The Ricci flow was introduced by Richard Hamilton in his 1982 paper Three-manifolds with positive Ricci
curvature [1]. With the help of essential observations from Shing-Tung Yao, it was soon conjectured as
being the key to proving Thurston’s geometrization conjecture [2], which has the Poincaré conjecture [3]
as a corollary. The former states that every closed three-dimensional manifold can be decomposed in such
a way, that each remaining component admits exactly one of eight types of geometries. The Poincaré
conjecture claims that every simply connected, closed three-dimensional manifold is homeomorphic to the
3-sphere. Both conjectures belonged among the most important open questions in mathematics. Hamilton
proceeded in publishing numerous papers in these directions, but it was the Russian mathematician
Grigori Perelman who was eventually able to prove the geometrization conjecture in his series of papers
[4, 5, 6]. Although he gained more public attention for declining multiple prizes than for the actual proof,
the Ricci flow has since been a thoroughly investigated tool.
Starting with an arbitrary smooth Riemannian manifold, the Ricci flow is a geometric evolution
equation in which one allows the metric to evolve along the vector field of −2 times the Ricci curvature.
By changing the distances between any two points, one also alters the angles between any two points
(except in two-dimensions, when the deformation is conformal) and the volume of the manifold. This
happens in a way that makes the manifold more symmetric, and hence smoothens a manifold’s geometry.
The Ricci curvature can be regarded as a Laplacian of the metric, which causes the Ricci flow to have
a strong resemblance to the usual heat equation. However, it fails to be parabolic due to the isometric
invariance of the Ricci curvature. Nevertheless, without the help of any a priori curvature bounds, we
can still guarantee short-time existence and uniqueness of solutions to the flow.
In [1], Hamilton proves existence and uniqueness for closed manifolds, i.e. compact without boundary,
by using the Nash-Moser inverse function theorem. DeTurck simplified the proof by showing that the
Ricci flow is equivalent to a quasilinear parabolic initial-value problem in [7]. Multiple works provide
similar results for manifolds with distinct topological properties. On non-compact manifolds, existence
was shown by Shi in [8] and uniqueness by Chen and Zhu in [9]. On manifolds with boundary and
an arbitrary initial metric without any curvature constraints or prescriptions, existence and uniqueness
results were established very recently by Chow in [10]. Furthermore, Topping and Giesen [11] obtained
existence and uniqueness results for a non-complete initial manifold. In the generalized context of sub-
Riemannian geometry, little research has been done on the evolution of Ricci flow. The first work in this
direction is from Lovrić, Min-Oo and Ruh [12], who proved existence to Ricci flow on Riemannian foliations
on compact Riemannian manifolds, which can be considered as certain sub-Riemannian manifolds. Due
to recent developments on sub-Riemannian manifolds of e.g. Dong [13], who provided contributions to the
closely related harmonic map heat flow, Baudoin and Garofalo [14], who generalized curvature dimension
inequalities, and Agrachev and Lee [15], who studied sub-Laplacian comparison theorems, more existence
results for Ricci flow on such manifolds are to be expected.
Our aim is to investigate the possibilities of evolving Ricci flow on two-dimensional almost-Riemannian
manifolds. These are smooth connected two-dimensional manifolds endowed with an almost-Riemannian
structure, firstly introduced by Grushin [16] in the context of hypoelliptic operators. They are the proto-
types of rank-varying sub-Riemannian structures. More specifically, they are sub-Riemannian structures
that can be locally defined by a set of smooth vector fields that satisfy the Hörmander condition and
of which the cardinality equals the dimension of the manifold. Almost-Riemannian structures are pri-
marily studied in dimension two, and the first results on its general properties were accomplished by
Agrachev, Boscain and Sigalotti in [17]. With a generic two-dimensional almost-Riemannian structure
there are only three types of points. The most common points are regular Riemannian points, on which
we would like to evolve the Ricci flow. However, the set of Riemannian points is a non-complete surface
with boundary. There have not yet been any results regarding Ricci flow on surfaces with these specific
topological properties.
The thesis is structured as follows: Chapter 2 focusses entirely on differential geometry, since both
the Ricci flow and almost-Riemannian geometry belong to this branch of mathematics. In an attempt to
write a comprehensive piece, we treat everything that is of significant importance for one’s understanding
of the topic with great detail. In particular, we pay close attention to the isometric invariance of the
Levi-Civita connection, Riemann curvature tensor, and, subsequently, the Ricci curvature. Chapter 3

3
is devoted to the Ricci flow. After having discussed time derivatives and deformations of geometric
quantities, we obtain equations for the evolution of the Riemann curvature tensor and Ricci curvature
under Ricci flow. Here, the heat-type nature of the Ricci flow becomes apparent. In the remainder of the
chapter, we devote ourselves to proving short-time existence and uniqueness. We remark on the weakly
parabolicity of Ricci flow, which is due to the isometric invariance of the Ricci curvature. The eventual
proof is in analogy with DeTurck’s proof from [7], but can be considered as a detailed version of the proof
described in [18]. Chapter 4 starts with a brief introduction to sub-Riemannian structures, after which
we quickly focus on (two-dimensional) almost-Riemannian structures. Having established the topology
on the Riemannian points of generic almost-Riemannian manifolds, we investigate whether we can use
existing results on the Ricci flow for our specific case.
We remark that we make use of the Einstein summation convention throughout the thesis. That
is, if an index appears once in a lower, and once in an upper index position of a certain term, then it
is understood to be summed over all possible index values. Furthermore, the reader is assumed to be
acquainted with preliminary notions on topological spaces as from for example [19].

Acknowledgements
First and foremost I would like to thank dr. M. Seri for his excellent guidance throughout the research
and for always being available to answer any of my questions. His precise and detailed knowledge on
a broad variety of mathematical topics has amazed me. I would also like to thank my friend Dave
Verweg, who joined me on weekly study sessions and has inspired me with the beauty of mathematics;
my girlfriend Liselot Visser, who used her talents as a graphic designer to provide me with the illustrations
in this thesis; and my parents, who have always supported me in developing myself as a mathematician.

4
2 DIFFERENTIAL GEOMETRY

2 Differential geometry
The mathematical foundation of Ricci flow and almost-Riemannian geometry is that of differential ge-
ometry. On that account, we devote this first chapter exclusively to the tools needed to define Ricci
flow.
We are particularly interested in the non-Euclidean differential geometry of Riemannian manifolds.
After a brief overview of the most elementary notions regarding differentiable manifolds, we will continue
with providing the tools that allow us to construct Riemannian manifolds: smooth manifolds equipped
with a metric. This process starts with examining tangent vectors, covectors, tensors and some of their
useful properties. We will see that the Riemannian metric is in fact a tensor and enables one to define
multiple (intrinsic) geometric properties of Riemannian manifolds, such as curvature. There are various
kinds of curvature tensors, each with a slightly distinct geometric meaning. To define curvature tensors,
however, one needs a method for taking derivatives of vector fields, which can be achieved by means of
a connection. The unique connection on the tangent bundle of a Riemannian manifold, called the Levi-
Civita connection, is the last necessary tool to define the Riemann curvature tensor. The Ricci curvature
tensor can then easily be obtained, after which the definition of Ricci flow lays at our hands.
The chapter mainly relies upon the educational textbooks on Riemannian Geometry of Lee [20, 21]
and the lecture notes on analysis on manifolds of Seri [22]. To lesser extent, results from [23] and [24] are
used as well.

2.1 Smooth manifolds

The purpose of this section is to briefly recall or familiarize the reader with smooth manifolds and related
notions in differential geometry. It summarises the most relevant definitions and results for our later
purpose. For more detail on smooth manifolds, we refer to [22, 25, 26]. The main idea is to equip
topological spaces that locally look like Euclidean spaces with a smooth structure and define derivatives
of smooth functions between such spaces. Firstly, we will look at topological spaces of which any point
has an open neighbourhood homeomorphic to an open subset of Rn .

Definition 2.1. A homeomorphism is bijective map f : X → Y between two topological spaces X and
Y such that both f and f −1 are continuous. ♥

With the topological properties of Hausdorffness and second countability, we can already define a
topological manifold.

Definition 2.2. A n-dimensional topological manifold is a topological space M that satisfies the following
properties:

(i) M is a Hausdorff space;

(ii) M is second countable;

(iii) M is locally euclidean of dimension n. That is, for any point p ∈ M there exists an open subset
U ⊆ M with p ∈ U , an open subset V ⊆ Rn and a homeomorphism ϕ : U → V .

Remark 2.3. A n-dimensional topological manifold M with boundary satisfies properties (i) and (ii) of
the above definition, but is instead locally homeomorphic to Hn := {x ∈ Rn | xn ≥ 0}. One could
interpret manifolds without boundary as a special case of manifolds with boundary. Therefore, when we
are discussing a manifold it may either have or may not have a boundary. F
Remark 2.4. A compact manifold is a manifold that is compact as a topological space. A closed manifold
is a compact manifold without boundary. Lastly, a connected manifold is a manifolds that is connected
as a topological space. We will frequently use compact, closed ad connected (Riemannian) manifolds in
the coming chapters. F

5
2.2 Tangent vectors, covectors and tensors

We call the pair (U, ϕ) a coordinate chart, with ϕ : U → V ⊆ Rn the coordinate map and U the
coordinate neighbourhood about a point p ∈ U . In order to define a smooth structure on a topological
S M , we need a way to assemble the charts (U, ϕ) such that they cover M . That is, such that
manifold
M = i∈I Ui , where I is some indexing set. For such a collection to make any sense, two different
coordinate maps of overlapping neighbourhoods must agree on the intersection of these neighbourhoods.
Definition 2.5. Two coordinate charts (Ui , ϕi ) and (Uj , ϕj ) on a manifold M are smoothly compatible
if either Ui ∩ Uj = ∅, or if the transition map

ϕi ◦ ϕ−1
j : ϕj (Ui ∩ Uj ) → ϕi (Ui ∩ Uj )

is C ∞ . ♥
This additional structure enables us to define a smooth structure on a manifold.
Definition 2.6. Let M be a n-dimensional topological manifold, then
(i) the collection
A = {ϕi : Ui → Vi | i ∈ I}
of pairwise smoothly compatible charts that cover M is a smooth atlas;
(ii) we call an equivalence class of smooth atlases a smooth structure on M . Atlases are said to be
equivalent if any two charts of these atlases are smoothly compatible;
(iii) we say that the pair (M, A) is a n-dimensional smooth manifold if A a smooth structure on M .
♥
It is clear that there exists a smooth structure A when we say that M is a n-dimensional smooth
manifold. Of course, we can also define smooth maps between manifolds. These maps can be seen as lifted
maps between Euclidean spaces, and hence differentiability follows from differentiability as a Euclidean
map.
Definition 2.7. A map F : M → N between smooth manifolds of dimension m and n respectively is a
smooth map if for any chart (ϕ, U ) of M and (φ, V ) of N the map

φ ◦ F ◦ ϕ−1 : Rm ⊇ ϕ(U ∩ F −1 (V )) → φ(F (U ) ∩ V ) ⊆ Rn

is C ∞ as a Euclidean function. ♥
Although we will not pay specific attention to integrals on manifolds in this thesis, it is useful for
later to define the orientation of a manifold, which is a required property to apply e.g. Stoke’s theorem
[22, Theorem 8.3.1].
Definition 2.8. An oriented atlas A = {(Ui , ϕi )} is called oriented if all the charts have the same
orientation. That is, if det(Dϕij ) > 0 for all the transition functions ϕij = ϕi ◦ ϕ−1
j . A manifold M with
an oriented atlas is called a oriented manifold 1 . If there exists an orientation on M , then we say that it
is orientable. ♥

2.2 Tangent vectors, covectors and tensors

This section summarizes the notions of tangent vectors, covectors, tensors and the spaces they live
in. We will we see that tangent vectors and covectors are just an example of the latter. Tensors will
eventually allow us to define a Riemannian manifold. In the process, we will also see how to actually define
a derivative of a smooth function between smooth manifolds with which we can define submanifolds.
Moreover, we will briefly examine Lie derivatives of tensor fields and some other useful properties of
tensors.
1 Here, Dϕij denotes the Jacobian matrix of ϕij .

6
2 DIFFERENTIAL GEOMETRY

2.2.1 Tangent bundle and vector fields

Throughout literature, a broad variety of approaches eventually lead to equivalent notions of the actual
derivative of such a smooth map. Without getting to technical, the approach here follows [22]. Recall
that a germ at p ∈ M is an equivalence class in the quotient space of smooth functions on M , i.e. of
Cp∞ (M ) := C ∞ (M )/ ∼p . But as pointed out in [22, Section 2.3], the tangent space at a point p ∈ M
is isomorphic to the space of derivations C ∞ (V ), where V ⊆ M is some open neighbourhood of p. This
allows us to define tangent vectors at p as derivations of C ∞ (V ) for any such neighbourhood V 3 p,
instead of seeing them as linear maps v : Cp∞ (M ) → R that act on equivalence classes. For convenience,
in the next definition we assume V to coincide with whole of a smooth manifold M .
Definition 2.9. A tangent vector at a point p of a smooth manifold M is the linear map

v : C ∞ (M ) → R,

which is also a derivation of C ∞ (M ) at p. That is, it satisfies the Leibniz rule

v(f g) = f (p)v(g) + g(p)v(f ), ∀f, g ∈ C ∞ (M ). (2.1)

We denote the set of all tangent vectors at p by Tp M , the tangent space to M at p. ♥

Alternatively, we can also define a tangent vector by using smooth parametrized curves in a manifold
M.
Definition 2.10. Let M be a smooth manifold, p ∈ M , I = (a, b) ⊂ R with 0 ∈ (a, b) and γ : I → M a
smooth curve with γ(0) = p ∈ M . A tangent vector v at p ∈ M is a map

v : C ∞ (M ) → R,

defined by
d(f ◦ γ(t))
v(f ) := , ∀f ∈ C ∞ (M ). (2.2)
dt t=0
♥
It is important to note that this γ : I → M exists and satisfies γ 0 (0) = v ∈ Tp M [22, Theorem 2.5.5],
meaning that these definitions are in fact equivalent. By using local coordinates of a chart on a manifold
M and with these equivalent definitions at hands, we can show that Tp M is a vector space.
Proposition 2.11. Let M be a n-dimensional smooth manifold, then
(i) the tangent space Tp M at p ∈ M is a vector space of dimension n;
 
∂
(ii) if (U, ϕ) is a chart about p with coordinates2 (xi ) for each i ∈ {1, . . . , n}, the set ∂xi | i ∈ {1, . . . , n}
p
forms a basis for Tp M .
♦
Proof. Let v, w ∈ Tp M and λ ∈ R. Following Definition 2.9, we know that v + λw is a linear mapping
v + λw : C ∞ (M ) → R, since both v and w are linear mappings. Let V ⊆ U ⊆ M be a neighbourhood
about p and let f, g ∈ C ∞ (V ). By using that v, w satisfy Leipniz’s rule we see that

(v + λw)(f g) = v(f g) + λw(f g)

= f (p)v(g) + g(p)v(f ) + λ (f (p)w(f ) + g(p)w(f ))
= f (p)(v + λw)(g) + g(p)(v + λw)(f ),

i.e., that v + λw also satisfies Leipniz’s rule. It follows that Tp M is a real vector space.
2 Here xi = ri ◦ ϕ where ri : Rn → R denote the standard coordinates on Rn .

7
2.2 Tangent vectors, covectors and tensors

To proof (ii), notice that since Definitions 2.9 and 2.10 are equivalent, for v ∈ Tp M we have

d(f ◦ γ(t))
v(f ) =
dt t=0
d(f ◦ ϕ−1 ◦ ϕ ◦ γ(t))
= ,
dt t=0

since ϕ is a bijection. By using the chain rule we can rewrite this to

d(f ◦ ϕ−1 ◦ ϕ ◦ γ(t)) ∂f ◦ ϕ−1 d(xi ◦ γ)

= i
(ϕ(p)) (0).
dt t=0 ∂x dt
d(xi ◦γ) d(xi ◦γ)
Notice that d t (0) ∈ R for each i ∈ {1, . . . , n} . Define v i := dt , then since ϕ−1 ◦ ϕ(p) = p we see
that
∂f
v(f ) = v i ,
∂xi p
and hence that
∂
v = vi .
∂xi p
 
∂
Indeed, the set ∂xi | i ∈ {1, . . . , n} spans the vector space Tp M . Additionally, notice that for each
p
j ∈ {1, . . . n} we have

∂xj
v(xj ) = v i = v i δij ,
∂xi p
 
∂ ∂
showing that v = v i ∂x i
1 n
= 0 if and only if (v , . . . , v ) = 0. Hence ∂xi | i ∈ {1, . . . , n} are linearly
p p
independent and form a basis for the vector space Tp M with dim(Tp M ) = n = dim(M ).

Figure 1: The tangent space Tp S2 at a point p ∈ S2 .

We now have the right tools and knowledge to actually define the derivative of a smooth map between
manifolds, rather than stating results on differentiability. It turns out that the derivative of a smooth
map is nothing more than a linear map between tangent spaces.

8
2 DIFFERENTIAL GEOMETRY

Definition 2.12. The differential or total derivative of a smooth map F : M → N between smooth
manifolds M and N at p ∈ M is the linear map

dFp : Tp M → TF (p) N, dFp (v)(f ) := v(f ◦ F ), ∀f ∈ C ∞ (N ). (2.3)

♥
With the total derivative of Definition 2.12, we can define two particular kinds of smooth maps
between manifolds. These maps will in turn provide us with a definition for submanifolds.
Definition 2.13. Let F : M → N be a smooth map between smooth manifolds of respectively dimension
m and n. Then
(i) F is an immersion if dFp is injective for all p ∈ M (i.e. then, m ≤ n);
(ii) F is an embedding if F is an injective immersion that is also a homeomorphism onto its range, i.e.
F (M ) = N .
♥
Due to Definition 2.13, we can make a distinction between two kinds of submanifolds.
Definition 2.14. Let ι : M ,→ N be the inclusion map between two smooth manifolds M and N such
that M ⊂ N . Then we say that
(i) M is an immersed submanifold if the inclusion map is an immersion;
(ii) M is an embedded submanifold if the inclusion map is an embedding.
♥
In the above definition, we see that M is always a manifold ‘sitting inside’ a larger manifold N since
the dimensions satisfy m ≤ n. With the Implicit Function Theorem for Manifolds, we can also detect
submanifolds when m ≥ n (see, for example, [22, Theorem 2.8.14]).
Let us move back back to tangent spaces again. Rather than talking about a tangent space at one
point of a manifold, it is often convenient to talk about the set of tangent spaces as a whole.
Definition 2.15. The tangent bundle T M of a smooth manifold M is the disjoint union of tangent
spaces G
T M := ({p} × Tp M ) .
p∈M

Elements of T M are pairs (p, v) where p ∈ M is a base point and v ∈ Tp M is a tangent vector. ♥
An important fact is that the tangent bundle itself is also a smooth manifold. More precisely, if M is
a n-dimensional smooth manifold, then T M is a 2n-dimensional smooth manifold of which the smooth
structure is naturally obtained from the smooth structure of M [22, Theorem 2.6.3]. But, actually this
is just the prototype of the more general notion of a vector bundle.
Definition 2.16. A smooth vector bundle of rank r is a triple (E, M, π) where E and M are manifolds
and π : E → M a smooth surjective map such that, for all p ∈ M , the following properties hold:
(i) the fibre over p, Ep := π −1 (p), has the structure of a vector space of dimension r;
(ii) there exists a neighbourhood U ⊆ M of p and a diffeomorphism ϕ : π −1 (U ) → U × Rr such that
(a) π1 ◦ ϕ = π, where π1 : U × Rr → U is the projection on the first factor,
(b) for all q ∈ U , the map ϕ Eq
: Eq → {q} × Rr is an isomorphism of vector spaces.

The space E is called the total space, M the base space, π its projection and each ϕ is a local trivialisation.
♥

9
2.2 Tangent vectors, covectors and tensors

Remark 2.17. Sometimes we just say the E is a vector bundle over M . F

Remark 2.18. One could also define an orientation on a vector bundle3 . An orientation on E means
that for each fibre Ep , there exists an orientation such that each trivialization ϕ : π −1 (U ) → U × Rr is
fibrewise orientation-preserving. We will use this to define orientation on almost-Riemannian manifolds
in Chapter 4. F
To emphasize the earlier statement that the tangent bundle is an example of a vector bundle, notice
that the projection π : T M → M is a surjective map such that the tangent spaces are fibres: π −1 (p) =
Tp M . Not surprisingly, we can also define sub-bundles.
Definition 2.19. Let (E, M, π) be a smooth vector bundle of rank n and F ⊂ E a submanifold. If
Fp := F ∩ Ep is k-dimensional subspace of the vector space Ep for all p ∈ M and π F : F → M defines a


smooth vector bundle of rank k, then F, M, π F is called a sub-bundle of E. ♥
For our quest on defining a Riemannian manifold, we still need more tools. The first thing that we
will look at are vector fields. A vector field, however, is itself an example of another notion.
Definition 2.20. A section of a vector bundle π : E → M is a smooth map S : M → E such that
π ◦ S = idM . We denote the set of all smooth sections on E by Γ(E) ♥
For alocal chart (U, ϕ) about p ∈ M with local coordinates (xi ), we say that the family of n local
∂ ∂
sections ∂x 1 , . . . , ∂xn is a smooth local frame of the tangent bundle T M , since these sections yield a
basis for Tp M for each p ∈ U . If U = M , we call this set a global frame. We can now define vector fields
as sections of the tangent bundle: a smooth map from a manifold to the tangent bundle that assigns a
tangent vector to each point of a manifold.
Definition 2.21. A smooth vector field is a smooth map X : M → T M with π ◦ X = idM . We denote
the set of smooth vector fields by X(M ). ♥
By using the basis and chart about a point p ∈ U ⊆ M of a smooth manifold as in Proposition 2.11,
we can express the value of a smooth vector field X : M → T M as
∂
Xp = X i (p) , (2.4)
∂xi p
with X i : U → R the component functions of X. A tool that will be frequently used throughout this
thesis is the Lie bracket of two smooth vector fields.
Definition 2.22. Let X, Y ∈ X(M ). The Lie bracket [·, ·] : X(M ) × X(M ) → X(M ) of X and Y is the
derivation given by their commutator:
[X, Y ] := XY − Y X. (2.5)
♥
Proposition 2.23. Let (U, ϕ) be a chart on M with local coordinates (xi ) and let X, Y ∈ X(U ). If
∂ j ∂
X = X i ∂x i and Y = Y ∂xj are the coordinate expressions for X and Y , then

∂Y j ∂X j
 
∂
[X, Y ] = X i i − Y i i
. (2.6)
∂x ∂x ∂xj
♦
∞
Proof. Let f ∈ C (U ), then we can compute the lie bracket of X, Y ∈ X(U ) by using (2.5) as
[X, Y ]f = X(Y (f )) − Y (X(f ))
   
∂ ∂f ∂ ∂f
= Xi i Y j j − Y j j Xi i
∂x ∂x ∂x ∂x
j 2 i 2
∂Y ∂f i j ∂ f j ∂X ∂f j i ∂ f
= Xi i + X Y − Y − Y X ,
∂x ∂xj ∂xi ∂xj ∂xj ∂xi ∂xj ∂xi
3 Remember that we already defined an oriented manifold in Definition 2.8. An orientation can also be defined on a

vector space [22, Definition 8.1.3]. Moreover, one could also define orientation preserving maps [22, Page 128].

10
2 DIFFERENTIAL GEOMETRY

where we used the product rule in the third step. Since the order of taking partial derivatives can be
interchanged, notice that the second and fourth term cancel out. Therefore, we get

∂Y j ∂f i
j ∂X ∂f
[X, Y ]f = X i i j
− Y
 ∂x ∂x ∂xj ∂xi
j j
∂Y ∂X ∂f
= Xi i − Y i i
,
∂x ∂x ∂xj

where we interchanged the dummy indices i and j in the last step.

The Lie bracket is actually another way of writing the Lie derivative of a vector field Y with respect
to X, i.e. LX Y = [X, Y ]. Later, we will also define the Lie derivative of tensor fields. For that purpose,
let us briefly introduce integral curves of vector fields [22, Definition 3.31].
Definition 2.24. Let M be a smooth manifold and X ∈ X(M ). A smooth curve γ : R ⊃ (a, b) → M is
an integral curve of X if
γ 0 (t) = Xγ(t) , ∀t ∈ (a, b). (2.7)
It is often assumed that 0 ∈ (a, b) and that y(0) = p ∈ M , such that γ is an integral curve through p. ♥
Definition 2.25. Let M be a smooth manifold and X ∈ X(M ). For a given p ∈ M we denote Ip =
(t− (p), t+ (p)) ⊂ R, with 0 ∈ Ip , as the maximal interval on which the unique integral curve γp : Ip → M
of X through p is defined. We call γp the maximal integral curve of X through p. ♥
Existence and uniqueness of (maximal) integral curves follow from existence and uniqueness of Eu-
clidean theorems [22, Theorem 3.3.5, Theorem 3.3.11] on ODE’s. Theorem 3.3.11 of [22] guarantees
the existence of a unique map ϕ : D → M with D ⊂ R × M̊ such that for all p ∈ M one has
D ∩ (R × {p}) = Ip × {p} and ϕ(t, p) = γp (t) for all (t, p) ∈ D. The map ϕX is called the flow of
X.

2.2.2 Cotangent bundle and tensor bundle

By using vector fields we can eventually endow a smooth manifold with an inner product to define lengths
of and angles between vectors of a tangent space. As will later become clear, we therefore first need to
define the dual of a tangent space.
Definition 2.26. The cotangent space Tp∗ M := (Tp M )∗ of a smooth manifold M at p ∈ M is the dual
of the tangent space Tp M . Elements of Tp∗ M are called covectors or differential 1-forms at p, linear
functionals from Tp M to R. ♥
Not surprisingly, if M is n-dimensional, then the dual space Tp∗ M at p ∈ M is also n-dimensional.
Also similar to Proposition
 2.11, if (U, ϕ)is a chart about p with local coordinates (xi ), then the set of
covectors given by dxi | i ∈ {1, . . . , n} forms a basis for Tp∗ M . Moreover, we can define the cotangent
p
bundle, which is then also a 2n-dimensional smooth manifold.
Definition 2.27. The cotangent bundle T ∗ M of a smooth manifold M is the disjoint union of the
cotangent spaces G
T ∗ M := {p} × Tp∗ M .

p∈M

Elements of T M are pairs (p, ω) where p ∈ M is a base point and ω ∈ Tp∗ M is a covector.
∗
♥
Remark 2.28. Note that just as with the tangent bundle and vector bundle, the cotangent bundle is a
specific example of the dual bundle E ∗ of a vector bundle E over M . Its fibres are the dual spaces of the
fibres of E [18, Section 2.3.4]. F
The smooth section from a manifold to the cotangent bundle that assigns a covector to each point of
a manifold is what we call a covector field.

11
2.2 Tangent vectors, covectors and tensors

Definition 2.29. A smooth covector field is a smooth map ω : M → T ∗ M with π ◦ ω = idM . We denote
the set of smooth covector fields by X∗ (M ). ♥
Similarly to (2.4), we can express the value of a smooth vector field ω : M → T ∗ M at p ∈ M as

ωp = ωi (p)dxi , (2.8)
p

with ωi : U → R. Both tangent vectors as covectors are, however, examples of more general objects,
called tensors. These multilinear maps will be one of our primary tools throughout this thesis.
Definition 2.30. The tensor space
r times
z }| {
:= Mult(Tp M, . . . , Tp∗ M , Tp M, . . . , Tp M )
Tsr (Tp M ) ∗
| {z }
s times

of a smooth manifold M at p ∈ M is the space of multilinear maps τ . That is, the space of tensors of
type (r, s):
r times
z }| {
τ : Tp M × · · · × Tp∗ M × Tp M × · · · × Tp M → R.
∗
| {z }
s times
♥
If r = 0 and s = k ≥ 1, then we say that τ ∈ Tk0 (Tp M ) is a covariant k-tensor on Tp M . In contrast,
if r = k ≥ 1 and s = 0, we call τ ∈ T0k (Tp M ) a contravariant k-tensor on Tp M . A tensor of type (r, s),
i.e. τ ∈ Tsr (Tp M ), is sometimes defined as the pairing
τ ω 1 , . . . , ω r ; v1 , . . . , vs =: τ | ω 1 , . . . , ω r ; v1 , . . . , vs , ω 1 , . . . , ω r ∈ Tp∗ M,



v1 , . . . , vs ∈ Tp M.
(2.9)
Of course, we can also define the tensor bundle over a manifold M .
Definition 2.31. The (r, s)-tensor bundle Tsr M of a smooth manifold M is the disjoint union of the
tensor spaces G
Tsr M := ({p} × Tsr (Tp M )) .
p∈M
♥
Remark 2.32. Note again that the tensor bundle is an example of a tensor product of vector bundles and
dual bundles. More specifically, if E1∗ , . . . , Er∗ are dual bundles and E1 , . . . , Es are vector bundles, then
the tensor product E1∗ ⊗ · · · ⊗ Er∗ ⊗ E1 ⊗ · · · ⊗ Ek is a vector bundles with fibres (E1∗ )p ⊗ · · · ⊗ (Er∗ )p ⊗
(E1 )p ⊗ · · · ⊗ (Ek )p [18, Section 2.3.5]. F
Note that therefore T M = T01 M and T ∗ M = T10 M . For completeness, we also define tensor fields.
The vector and covector field that we saw before are just an example of this more general object, as one
would expect.
Definition 2.33. A smooth tensor field of type (r, s) is a smooth map τ : M → Tsr M such that
π ◦ τ = idM . We denote the space of smooth tensor fields by Tsr (M ). ♥
Then clearly X(M ) = T01 (M ) and X∗ (M ) = T10 (M ). If (U, ϕ) is a chart about p ∈ M with local
coordinates (xi ), then we can express the value of a smooth tensor field τ : M → Tsr (M ) as the tensor
product
···jr ∂ ∂
τp = τij11···i (p) j1 ⊗ · · · ⊗ jr ⊗ dxi1 ⊗ · · · ⊗ dxis , (2.10)
s
∂x p ∂x p p p
···jr
with τij11···i s
: U → R the component defined by
 
···jr ∂ ∂
τij11···i := τ dxj1
, . . . , dx jr
, , . . . , (2.11)
s
∂xi1 ∂xis
The following lemma allows us to conveniently characterize tensor fields as C ∞ (M ) multilinear maps.

12
2 DIFFERENTIAL GEOMETRY

Lemma 2.34. A map

r times
z }| {
τ : X (M ) × · · · × X∗ (M ) × X(M ) × · · · × X(M ) → C ∞ (M )
∗
| {z }
s times

is induced by a (r, s)-tensor field τ ∈ Tsr (M ) if and only if it is multilinear over C ∞ (M ). Similarly, a map

r times
z }| {
τ : X∗ (M ) × · · · × X∗ (M ) × X(M ) × · · · × X(M ) → X(M )
| {z }
s times

is induced by a (r + 1, s)-tensor field τ ∈ Tsr+1 (M ) if and only if it is multilinear over C ∞ (M ). ♠

Proof. The proof of this lemma is not of significant importance for our purpose. Therefore, it is omitted
and we refer to the tensor characterization lemma in [21, Lemma B.6] or [25, Lemma 12.24].

As promised, we define the Lie derivative of tensor fields [22, Remark 7.6.6], which we will use in our
proof for existence and uniqueness of the Ricci flow.

Definition 2.35. Let M be a smooth manifold and τ ∈ Trs (M ). The Lie derivative of τ along X ∈ X(M )
for p ∈ M is
d 
∗ 
(LX τ )p := ϕX
t τ . (2.12)
dt t=0 p

Notice that here, R 3 t 7→ ϕXt (p) := ϕ(t, p) denotes the maximal integral curve for X starting from p.
Particular interesting classes of tensors are symmetric and alternating tensors. If V is a real n-
dimensional vector space and ω ∈ Tk0 (V ), then we say that ω is a symmetric covariant k-tensor on V if
its value is unchanged by interchanging any pair of arguments. That is, if v1 , . . . , vk ∈ V , we have

ω(v1 , . . . , vi , vj , . . . , vk ) = ω(v1 , . . . , vj , vi , . . . , vk ), (2.13)

for all 1 ≤ i < j ≤ k. Let us denote the set of all symmetric k-tensors on V by Σk (V ). Via the projection
Sym : Tk0 (V ) → Σk (V ) defined by

1 X
Sym ω(v1 , . . . , vk ) := ω(vσ(1) , . . . , vσ(k) ), (2.14)
k!
σ∈Sk

one is able to symmetrize any ω ∈ Tk0 (V ). Here, Sk denotes the symmetric group on k elements and
σ ∈ Sk is a permutation. The symmetric product of ω ∈ Σk (V ) and η ∈ Σl (V ) is the (k + l)-tensor given
by
ωη := Sym (ω ⊗ η). (2.15)

Notice that if ω and η are covectors on V , they are always symmetric since they act on only one vector
v ∈V.

Lemma 2.36. If α and β are covectors on a n-dimensional real vector space V , i.e. α, β ∈ T10 (V ), then

1
αβ = (α ⊗ β + β ⊗ α) . (2.16)
2
♠

13
2.3 Riemannian manifolds

Proof. Let v1 , v2 ∈ V , then by (2.15) we have

αβ(v1 , v2 ) = Sym (α ⊗ β)(v1 , v2 )
1 X
= α(vσ(1) )β(vσ(2) )
2
σ∈Sk
1 1
= α(v1 )β(v2 ) + β(v1 )α(v2 )
2 2
1
= (α ⊗ β + β ⊗ α)(v1 , v2 ),
2
which proves the proposition.
Contrastingly, we call ω an alternating covariant k-tensor if it changes sign whenever two arguments
are interchanged:
ω(v1 , . . . , vi , vj , . . . , vk ) = −ω(v1 , . . . , vj , vi , . . . , vk ), (2.17)
for all 1 ≤ i < j ≤ k. These are also called exterior forms, or k-covectors.

2.3 Riemannian manifolds

Since we have acquired the right tools, we can start off with the definition [27, Definition 3.11].
Definition 2.37. A Riemannian manifold is a pair (M, g) where M is a smooth manifold and g is a
Riemannian metric that provides each p ∈ M with an inner product gp : Tp M × Tp M → R such that for
all X, Y ∈ X(M ), the map p 7→ gp (Xp , Yp ) is smooth. ♥
Remark 2.38. When it is more convenient, we say that M is an Riemannian manifold. F
Remark 2.39. More generally, one could also define metrics on a vector bundle, dual bundle and tensor
bundle. In the latter case for two tensor fields α, β ∈ Tsr (M ), their inner product is given by
j1 ···js
hα, βi = g a1 b1 · · · g ar br gi1 j1 · · · gis js αai11···i
···ar βb1 ···br .
s
(2.18)
F
Remark 2.40. In the more general pseudo-Riemannian manifold, we require the metric g to be non-
degenerate, smooth, symmetric and bilinear. Almost all of the upcoming theory is also applicable to
pseudo-Riemannian manifolds, but some is not or requires extra information. For simplicity we therefore
only consider Riemannian manifolds. F
We see from Lemma 2.34 that the metric g : X(M ) × X(M ) → C ∞ (M ) is a (0, 2)-tensor field, and
hence lives in T20 (M ). Sometimes, it is more convenient to use the usual inner product notation, i.e.
hXp , Yp i := gp (Xp , Yp ). In our usual notation for a chart (U, ϕ) with local coordinates (xi ) about a point
p ∈ M , we can write
gp = gij (p)dxi ⊗ dxj , (2.19)
p p
with each  
∂ ∂
gij (p) = gp , . (2.20)
∂xi p ∂xj p
The matrix [gij ] can therefore be seen as the n × n matrix with inner products as its entries. We denote
the inverse of this inner product matrix by [g ij ]. We can alternatively write (2.19) as
g = gij dxi ⊗ dxj
1
gij dxi ⊗ dxj + gji dxi ⊗ dxj


=
2
1
gij dxi ⊗ dxj + gij dxi ⊗ dxj


=
2
= gij dxi dxj , (2.21)
where the final step follows from Lemma 2.36. As showed in for example [25, Proposition 13.3], every
smooth manifold can be endowed with a Riemannian metric.

14
2 DIFFERENTIAL GEOMETRY

Example 2.41. The most straightforward example of a Riemannian manifold is Euclidean space Rn . The
metric is then the usual inner product on the tangent space Tp Rn = Rn [20, Section 3]. In standard
coordinates we usually write
g = δij dxi dxj .
♣
∂ ∂
 n
In the example above, we also see that ∂xi , . . . , ∂xn
is an orthonormal frame for T R . More
generally, {E1 , . . . , En } is a smooth orthonormal frame for a tangent bundle T M on a open set U ⊆ M
if and only if
hEi , Ej i = δij . (2.22)
Since we can apply the Gram-Schmidt algorithm on any smooth local frame for the tangent bundle over
a subset U ⊆ M , we can always find such a smooth orthonormal frame defined on some neighbourhood
of any p ∈ M [21, Proposition 2.8].
A useful property of a Riemannian metric g on M are the musical isomorphisms
[
: T M → T ∗M and ]
: T ∗M → T M
 ∂ ∂
used to convert vectors to covectors and covectors to vectors respectively. If ∂x 1 , . . . , ∂xn is smooth
local frame for T M on U and {dx , . . . , dx } its dual coframe, then by writing g = gij dx dxj according
1 n i
∂
to equation (2.21) and X(U ) 3 X = X i ∂x i , we define its flat by

X [ := gij X i dxj = Xj dxj . (2.23)

Similarly, if X∗ (U ) 3 ω = ωi dxi , we define its sharp by
∂ ∂
ω ] := g ij ωi = ωj j . (2.24)
∂xj ∂x
The latter raising operator will be of use when applying the trace operator on the metric.
∂
Definition 2.42. The trace or (a, b)-contraction of a tensor4 ∂x1 ⊗ · · · ⊗ ∂x∂ r ⊗ dx1 ⊗ · · · ⊗ dxs ∈ Tsr (M ),
with a ≤ r and b ≤ s, is the linear map
r−1
trba : Tsr (M ) → Ts−1 (M ),
defined by5
   
∂ ∂ ∂ ∂ ∂ ∂ ∂
trba 1
⊗ · · · ⊗ r
⊗ dx1
⊗ · · · ⊗ dx s
:=dxb
a 1
⊗ · · · ⊗ a−1 ⊗ a+1
⊗ ··· ⊗ r
∂x ∂x ∂x ∂x ∂x ∂x ∂x
⊗ dx1 ⊗ · · · ⊗ dxb−1 ⊗ dxb+1 ⊗ · · · ⊗ dxs .
(2.25)
♥
Example 2.43. A specifically relevant example is the trace of a symmetric (0, 2)-tensor field ωij dxi dxj
through the metric tensor g of some Riemannian manifold M :
trg (ω) := tr11 (ω ] )
 
∂
= tr11 g jk ωij dxi ⊗ by equation (2.24)
∂xk
= g ij ωij . by equation (2.25)
Equivalently, we may write
 
∂ ∂
tr1,2 −1
tr1,2 ij
⊗ ωij dxi ⊗ dxj = g ij ωij .


trg (ω) = 1,2 g ⊗ω = 1,2 g i
⊗ (2.26)
∂x ∂xj
♣
4 We use a slightly more
 convenient
  notationfor this definition.
5 Here ∂ ∂
we use R 3 dxk ∂x h =: dxk | ∂x h as the dual pairing.

15
2.3 Riemannian manifolds

A characteristic of Riemannian manifolds that will be used a lot is that its properties are preserved
by isometries.
Definition 2.44. Let (M, g) and (M f, g̃) be Riemannian manifolds. An isometry ϕ : (M, g) → (M
f, g̃) is
a diffeomorphism such that6 ϕ∗ g̃ = g. ♥
A map ϕ : (M, g) → (M
f, g̃) between Riemannian manifolds (M, g) and (M
f, g̃) is called a local isometry
if for each point p ∈ M there exists a neighbourhood U ⊆ M such that ϕ U is an isometry onto an open
subset of Ue ⊆M f.
When considering submanifolds of Riemannian manifolds, one naturally equips this submanifold with
an induced Riemannian metric. Mathematically, this phenomena is depicted in the following lemma.
Lemma 2.45. Let (M
f, g̃) be a Riemannian manifold, M a smooth manifold, and F : M ,→ Mf a smooth
map. The smooth (0, 2)-tensor field g = F ∗ g̃ is a Riemannian metric on M if and only if F is an
immersion. ♠
Proof. We start by assuming that g = F ∗ g̃ is a Riemannian metric on M . Recall that F is an immersion
if its derivative is injective at every p ∈ M . Since g is assumed to be a Riemannian metric, it is positive
definite. Let v ∈ ker dFp ⊂ Tp M be nonzero, then by definition of the pullback of tensor fields we have

(F ∗ g̃)p (v, v) = dFp∗ (g̃F (p) (v, v)) = g̃F (p) (dFp (v), dFp (v)) ≥ 0.

This means that dFp (v) = 0 if and only if v = 0, i.e. that dFp is injective for every p ∈ M . Therefore, F
is an immersion.
Next, we assume F to be an immersion. We must check whether the induced metric g = F ∗ g̃ satisfies
the conditions of an inner product. Therefore, let v1 , v2 ∈ Tp M and observe that

gp (v1 , v2 ) = (F ∗ g̃)p (v1 , v2 )

= dFp∗ (g̃F (p) (v1 , v2 ))
= g̃F (p) (dFp (v1 ), dFp (v2 ))
= g̃F( p) (dFp (v2 , ), dFp (v1 )) by linearity of g̃
∗
= (F g̃)p (v2 , v1 )
= gp (v2 , v1 ).

Indeed, g is symmetric. Secondly, let a, b ∈ R, and v1 , v2 , v3 ∈ Tp M . Then

gp (av1 + bv2 , v3 ) = (F ∗ g̃)p (av1 + bv2 , v3 )

= dFp∗ (g̃F (p) (av1 + bv2 , v3 ))
= g̃F (p) (dFp (av1 + bv2 ), dFp (v3 ))
= g̃F (p) (adFp (v1 ) + bdFp (v2 ), dFp (v3 )) by linearity of dFp
= ag̃F (p)(dFp (v1 ), dFp (v3 ) + bg̃F (p) (dFp (v2 ), dFp (v3 )) by linearity of g̃
= agp (v1 , v3 ) + bgp (v2 , v3 ),

which shows that g is a linear. Since F is an immersion, i.e. its derivative dFp is injective for all p ∈ M ,
notice that for v ∈ Tp M we have

gp (v, v) = (F ∗ g̃)p (v, v) = dFp∗ (g̃F (p) (v, v)) = g̃F (p) (dFp (v), dFp (v)) ≥ 0,

and equality follows if and only if dFp (v) = 0, i.e. if and only if v = 0. Therefore g = F ∗ g̃ is an
Riemannian metric.
Definition 2.46. A Riemannian submanifold M is an immersed or embedded submanifold of a Rieman-
f with the metric g = ι∗ g̃ induced by the inclusion map ι : M ,→ M
nian manifold M f. We call M
f the
ambient manifold. ♥
6 Here ϕ∗ g̃ denotes the pullback of g̃ by ϕ. For more detail, see [22, Section 6.2].

16
2 DIFFERENTIAL GEOMETRY

Example 2.47. Consider Sn := {x ∈ Rn+1 | kxk = 1} ⊆ Rn+1 . We can equip Sn with the Euclidean
metric of Rn+1 induced on Sn making it into a Riemannian submanifold.
For S2 , we can find a local expression of the metric by using spherical coordinates as in Figure 2.
Notice that S2 (except for the semi-circle from the point (0, 0, 1) to (0, 0, −1) in the xz-plane) can be
parametrized by
x = sin θ cos ϕ, y = sin θ sin ϕ, z = cos θ, {(θ, ϕ) | 0 < θ < π, 0 < ϕ < 2π}.
2
From this we can derive the basis vectors for Tp S at a point p = (sin θ cos ϕ, sin θ sin ϕ, cos θ):
∂
∂

= cos θ cos ϕ cos θ sin ϕ − sin θ , = − sin θ sin ϕ sin θ cos ϕ 0 .
∂θ p ∂ϕ p
 
∂ ∂
This yields the matrix representation of the metric on Tp S2 with respect to the basis ∂θ , ∂ϕ given
p p
by
 
∂ ∂ ∂ ∂
h ∂θ , ∂θ i h ∂θ , ∂ϕ i 
1 0

p p p p
gp =  = . (2.27)
∂
h ∂ϕ ∂
, ∂θ i ∂
h ∂ϕ ∂
, ∂ϕ i 0 sin2 θ
p p p p
Alternatively, we may write
g = dθ2 + sin2 θdϕ2 . (2.28)
♣

Figure 2: Parametrization of S2 .

Given a a Riemannian manifold (M f, g̃) and a smooth submanifold M ⊂ M f, we call a tangent vector
v ∈ Tp M a normal to M if g̃(v, w) = 0 for every w ∈ Tp M . The set of all vectors normal to M at a point
f
p is the normal space at p denoted by Np M = (Tp M )⊥ ⊂ Tp M f. In particular, the orthogonal direct sum
Tp M ⊕ Np M = Tp M f at each p ∈ M . To conclude the section, we define the normal bundle of M as
G
N M := ({p} × Np M ). (2.29)
p∈M

The theory of curvature that is dealt with in Section 2.6 can also be applied to Riemannian subman-
ifolds. We briefly discuss the extend to which a submanifold curves within its ambient manifold, i.e. the
second fundamental form of Riemannian submanifolds, in Section A.

17
2.4 Connections

2.4 Connections
In our quest on defining the Riemannian and Ricci curvature tensors, we firstly need to study connections.
Connections allow one to define the equivalent of Euclidean straight lines on Riemannian manifolds, called
geodesics, as we will see in Section 2.5. As illustratively described in [21, Section 4], a connection can be
seen as a set of rules for taking directional derivatives of vector fields. We are particularly interested in
the Levi-Civita connection, which enables us to properly express the properties of a Riemannian metric.
This unique connection on a Riemannian manifold must satisfy certain properties that will be discussed
throughout the remainder of this section.
Definition 2.48. Let π : E → M be a smooth vector bundle over a smooth manifold M . A connection
in E is a map
∇ : X(M ) × Γ(E) → Γ(E), (X, Y ) 7→ ∇X Y,
satisfying the following properties:
(i) ∇X Y is linear over C ∞ (M ) in X, i.e. for f, g ∈ C ∞ (M ) and X, Z ∈ X(M ),
∇f X+gZ Y = f ∇X Y + g∇Z Y. (2.30)

(ii) ∇X Y is linear over R in Y , i.e. for a, b ∈ R and Y, Z ∈ Γ(E),

∇X (aY + bZ) = a∇X Y + b∇X Z. (2.31)

(iii) ∇ satisfies the product rule, i.e. for f ∈ C ∞ (M ),

∇X (f Y ) = f ∇X Y + (Xf )Y. (2.32)

We call ∇X Y the covariant derivative of Y in the direction of X. ♥

Remark 2.49. Following [18, Remark 2.26], we could equivalently interpret a connection as the linear
map ∇ : Γ(E) → Γ(T ∗ M ⊗ E) which also obeys the product rule. F
This definition of a connection is rather broad, and applicable to more geometric structures. In the
Riemannian setting, we define a linear connection in the tangent bundle as the map ∇ : X(M ) × X(M ) →
X(M ) that satisfies the conditions of Definition 2.48. Although at first glance one would assume through
Lemma 2.34 that ∇ is then  ∂a (1, 2)-tensor field, this is not the case since it is not linear over C ∞ (M ) in
∂
the second argument. If ∂x1 , . . . , ∂xn is some smooth local frame for T M on U ⊆ M of dimension n,
we have
∂ ∂
∇ ∂ i j = Γkij k , (2.33)
∂x ∂x ∂x
or equivalently
∂
Γkij = dxk ∇ ∂ i j , (2.34)
∂x ∂x

with Γkij : U → R the n3 connection coefficients or Christoffel symobls of ∇ for each i, j, k ∈ {1, . . . , n}.
∂
For some smooth vector fields X, Y ∈ X(U ) written in terms of the frame, i.e. as X = X i ∂x i and
j ∂
Y = Y ∂xj , we then have
∂
∇X Y = ∇X (Y j )
∂xj
∂ ∂
= Y j ∇X j + (XY j ) j by equation (2.32)
∂x ∂x
∂ ∂
= Y j ∇X i ∂ i j + (XY j ) j
∂x ∂x ∂x
∂ ∂
= X i Y j ∇ ∂ i j + (XY j ) j by equation (2.30)
∂x ∂x ∂x
∂ ∂
= X i Y j Γkij k + (XY j ) j by equation (2.33)
∂x ∂x

∂
= X i Y j Γkij + XY k . by interchanging dummy index of the second term (2.35)
∂xk

18
2 DIFFERENTIAL GEOMETRY

Remark 2.50. From now on, we may write ∇i = ∇ ∂ . F

∂xi

As showed in [21, Proposition 4.2], the tangent bundle of every smooth manifolds admits a linear
connection. Moreover, a connection in the tangent bundle naturally induces a connection in all tensor
bundles over a smooth manifold. Therefore, we can also compute covariant derivatives of tensor fields.

Definition 2.51. A connection in the tensor bundle Tsr M is a map ∇ : X(M ) × Tsr (M ) → Tsr (M ), such
that the following conditions are satisfied:

(i) In T01 M = T M , ∇ agrees with the definition of a linear connection in T M .

(ii) In T00 M = M × R with X ∈ X(M ) and f ∈ C ∞ (M ), ∇ is equivalent to regular differentiation of

functions. That is,
∇X f = Xf. (2.36)

(iii) The connection ∇ obeys the product rule with respect to tensor products. That is, for two smooth
tensor fields F, G on M , we have

∇X (F ⊗ G) = (∇X F ) ⊗ G + F ⊗ (∇X G). (2.37)

(iv) The connection ∇ commutes with all contractions, i.e. if tr denotes the trace on any pair of indices
(h, k), then
∇X (trba Y ) = trba (∇X Y ) . (2.38)

The following lemma shows that a unique connection in each tensor bundle exists for a given linear
connection in the tangent bundle.

Lemma 2.52. Let ∇ be a linear connection in T M of a smooth manifold M . Then there exist a unique
connection in each tensor bundle Tsr M , also denoted by ∇, that satisfies the conditions of Definition 2.51.
Moreover, it satisfies the following properties.

(i) The product rule with respect to the dual pairing between ω ∈ X∗ (M ) and Y ∈ X(M ) is obeyed.
That is,
∇X (ω | Y ) = (∇X ω | Y ) + (ω | ∇X Y ) . (2.39)

(ii) For all τ ∈ Tsr (M ), ω 1 , . . . , ω s ∈ X∗ (M ) and Y1 , . . . , Yr ∈ X(M ), we have

(∇X τ ) ω 1 , . . . , ω r , Y1 , . . . , Ys =X τ ω 1 , . . . , ω r , Y1 , . . . , Ys




r
X
τ ω 1 , . . . , ∇X ω i , . . . , ω r , Y1 , . . . , Ys


−
i=1 (2.40)
s
X
τ ω 1 , . . . , ω r , Y1 , . . . , ∇X Yj , . . . , Ys .


−
j=1

Proof. Using the properties of Definition 2.51, the proof for (i) follows from a direct computation and
∂


by noticing that the dual pairing (in terms of the coordinate basis) (ω | Y ) = ωi Y j dxi | ∂x j = ωi Y j
∂
equals tr11 (ω ⊗ Y ) = tr11 ωi dxi ⊗ Y j ∂x


j = ωi Y j . Hence,

∇X (ω | Y ) = ∇(tr11 (ω ⊗ Y )) = tr11 (∇X ω ⊗ Y + ω ⊗ ∇X Y ) = (∇X ω | Y ) + (ω | ∇X Y ) .

19
2.4 Connections

For (ii), since τ ∈ Tsr (M ), we have τ ⊗ ω 1 ⊗ · · · ⊗ ω r ⊗ Y1 ⊗ · · · ⊗ Ys ∈ T2s 2r

(M ). Therefore, we can
7
apply the trace operator r + s times to obtain


∇X τ ω 1 , . . . , ω r , Y1 , . . . , Ys =∇X tr1,...,r,r+1,...,s 1 r



1,...,r,r+1,...,s τ ⊗ ω ⊗ · · · ⊗ ω ⊗ Y1 ⊗ · · · ⊗ Ys
h
1,...,r,r+1,...,s
=tr1,...,r,r+1,...,s (∇X τ ) ⊗ ω 1 ⊗ · · · ⊗ ω r ⊗ Y1 ⊗ · · · ⊗ Ys
r
X
+ τ ⊗ ω 1 ⊗ · · · ⊗ ∇X ω i ⊗ · · · ⊗ ω r ⊗ Y1 ⊗ · · · ⊗ Ys
i=1
s
X i
+ τ ⊗ ω 1 ⊗ · · · ⊗ ω r ⊗ Y1 ⊗ · · · ⊗ ∇X Yj ⊗ · · · ⊗ Ys
j=1
r

X
= (∇X τ ) ω 1 , . . . , ω r , Y1 , . . . , Ys + τ ω 1 , . . . , ∇X ω i , . . . , ω r , Y1 , . . . , Ys


i=1
s
X
τ ω 1 , . . . , ω r , Y1 , . . . , ∇X Yj , . . . , Ys ,


+
j=1

where we exploited (iv) of Definition 2.51 in the second step and reversed the trace operator in the final
step.
For uniqueness, assume that ∇ indeed satisfies (i) and (ii) and the conditions of Definition 2.51.
Rewriting equation (2.39) shows that the covariant derivative of any smooth covector field ω ∈ X∗ (M )
can be computed as
(∇X ω | Y ) = ∇X (ω | Y ) − (ω | ∇X Y ) .

We see that the connection on ω is uniquely determined by the linear connection. Moreover, (ii) shows
that the covariant derivative of any tensor field is determined by the covariant derivatives of every smooth
covector and vector field. Since these are determined by the linear connection on T M as well, it follows
that the connection in everytensor bundle is uniquely determined.
Existence follows by checking that ∇X τ is a smooth tensor field, i.e. multilinear over C ∞ (M ), and
that ∇ satisfies the properties of a connection as in Definition 2.48 [21, Proposition 4.15].

It is sometimes convenient to consider the covariant derivative of a tensor field in any direction as a
tensor field itself by using Lemma 2.34.

Definition 2.53. The total covariant derivative of a smooth tensor field τ ∈ Tsr (M ) is the smooth
(r, s + 1)-tensor field on M

r times
z }| {
∇τ : X (M ) × · · · × X∗ (M ) × X(M ) × · · · × X(M ) → C ∞ (M ),
∗
| {z }
s+1 times

given by
(∇τ ) ω 1 , . . . , ω r , Y1 , . . . , Ys , X = (∇X τ ) ω 1 , . . . , ω r , Y1 , . . . , Ys .



(2.41)

One could repetitively take total covariant derivatives of tensor fields. That is, given τ ∈ Tsr (M ) and
X, Y ∈ X(M ), we can obtain the smooth (r, s + 2)-tensor field as

∇2X,Y τ ω 1 , . . . , ω r , Z1 , . . . , Zs = ∇2 τ ω 1 , . . . , ω r , Z1 , . . . , Zs , Y, X .




(2.42)
7 Here we use that tr1,...,r,r+1,...,s i
1,...,r,r+1,...,s acts r times on the ω ’s and
∂
∂xi
components of τ , and s times on the Yj ’s and the
dxj components of τ .

20
2 DIFFERENTIAL GEOMETRY

Remark 2.54. It is important to notice that ∇X ∇Y τ 6= ∇2X,Y τ . In fact, we have (see [21, Proposition
4.21])
∇2X,Y τ = ∇X ∇Y τ − ∇∇X Y τ. (2.43)
F
Connections in the tensor bundles can be generalized even more to any arbitrary (tensor product of)
vector bundle(s) (recall Remark 2.32). For example, if ∇(1) is a connection on the dual bundle E1∗ , ∇(2)
is a connection on the vector bundle E2 , then there is a unique connection ∇ on E1∗ ⊗ E2 such that8
 
(2) (1)
(∇X τ ) (Y ) = ∇X (τ (Y )) − τ ∇X Y , X ∈ X(M ), Y ∈ Γ(E1 ). (2.44)

Moreover, we can also define the pullback connection. If f : M → N is a smooth map and ∇ a
connection on the vector bundle E over N , then there is a unique connection f ∇ on f ∗ E such that9
f
∇X (Y ) := ∇f∗ X Y, X ∈ X(M ), Y ∈ Γ(E). (2.45)
We are one step closer to the Levi-Civita connection since we can now define whether a linear con-
nection in the tangent bundle is compatible with the Riemannian metric. There are multiple equivalent
ways to do this, but the approach here only requires the tools that we already acquired. As opposed to
[24, Proposition 3.2], where compatibility with the metric is defined with parallel vector fields along a
smooth curve, we provide the following definition10 .
Definition 2.55. Let (M, g) be a Riemannian manifold. A Riemannian connection ∇ is a connection
on the tangent bundle such that ∇g = 0. We say that ∇ is compatible with g. ♥
Proposition 2.56. A connection ∇ is Riemannian if and only if
∇X g(Y, Z) = g (∇X Y, Z) + g (Y, ∇X Z) , ∀X, Y, Z ∈ X(M ). (2.46)
♦
Proof. By equations (2.40) and (2.41), for some smooth vector fields X, Y, Z ∈ X(M ) we have
(∇g)(Y, Z, X) = (∇X g)(Y, Z) = X(g(Y, Z)) − g(∇X Y, Z) − g(Y, ∇X Y ),
which equals zero if and only if (2.46) holds.
Compatibility with the metric is not enough to define a unique connection on a Riemannian manifold.
As in e.g. [23, Theorem 13.9], the additional required condition can be expressed by means of the torsion
tensor T : X(M ) × X(M ) → X(M ) of the connection defined by
T (X, Y ) := ∇X Y − ∇Y X − [X, Y ], ∀X, Y ∈ X(M ). (2.47)
Alternatively, we can require the connection to be symmetric.
Definition 2.57. A connection ∇ on the tangent bundle T M of a smooth manifold M is symmetric if
∇X Y − ∇Y X = [X, Y ], ∀X, Y ∈ X(M ), (2.48)
♥
Clearly, the connection is symmetric if and only if the torsion tensor vanishes identically. Compatibility
with the metric and symmetry are enough to proof existence and uniqueness of the connection on a
Riemannian manifold. Hence, we have arrived at the Fundamental Theorem of Riemannian Geometry.
The proof of this theorem can be found in e.g. [21, Theorem 5.10]. The result of this theorem is translated
in the following definition.
8 Thisresult can be derived from [18, Proposition 2.32-2.33] by using the properties of Lemma 2.52
9 Heref∗ X denotes the pushforward of X by f . For more detail, see [22, Section 6.2]. For more detail on the pullback
bundle structure, see [18, Section 2.8]
10 In the next chapter we will define the derivative of a vector field along a smooth curve. Parallel vector fields along

smooth curves are defined by using this map as elements of its kernel.

21
2.4 Connections

Definition 2.58. Let (M, g) be a Riemannian manifold. The Levi-Civita connection of g is the unique
connection ∇ on T M that is compatible with g and symmetric. ♥

Proposition 2.59. If (U, (xi )) is a smooth local coordinate chart of a Riemannian manifold M , then
the coefficients of its Levi-Civita connection ∇ are given by
 
k 1 kl ∂ ∂ ∂
Γij = g gjl + gil − gij . (2.49)
2 ∂xi ∂xj ∂xl

Proof. Let X, Y, Z ∈ X(M ), then by compatibility of ∇ with g we have

∇X g(Y, Z) = g (∇X Y, Z) + g (Y, ∇X Z) .

Since ∇ is symmetric, notice that

g(Y, [X, Z]) = g(Y, ∇X Y ) − g(Y, ∇Z X),

and therefore that

∇X (g(Y, Z)) = g(Y, ∇X Y ) + g(Y, ∇Z X) + g(Y, [X, Z]).
We can derive similar expression for ∇Y g(Z, X) and ∇Z g(X, Y ). By adding ∇X g(Y, Z) and ∇Y g(Z, X)
and subtracting ∇Z g(X, Y ), we can solve the expression for g(∇X Y, Z) to obtain
1
g(∇X Y, Z) = (∇X g(Y, Z) + ∇Y g(Z, X) − ∇Z g(X, Y ) − g(Y, [X, Z]) − g(Z, [Y, X]) + g(X, [Z, Y ])) .
2
Written in terms of the coordinate vector fields, the terms with Lie brackets cancel out by Proposition 2.23.
Therefore, the equation above reduces to
        
∂ ∂ 1 ∂ ∂ ∂ ∂ ∂ ∂
g ∇i j , l = ∇i g , + ∇j g , − ∇l g , .
∂x ∂x 2 ∂xj ∂xl ∂xl ∂xi ∂xi ∂xj

By equations (2.20) and (2.33), the above equation can be rewritten to

 
k 1 ∂ ∂ ∂
Γij gkl = gjl + gil − gij .
2 ∂xi ∂xj ∂xl

Multiplying both sides by the inverse matrix g kl gives the desired expression.

A nice property of the Levi-Civita connection that will be used later on, is that it is invariant under
(local) isometries.

Proposition 2.60. Let ϕ : (M, g) → (M f, g̃) be an isometry between two Riemannian manifolds equipped
with Levi-Civita connections ∇ for g and ∇e for g̃. Then ϕ∗ ∇
e = ∇, or equivalently,

ϕ∗ (∇X Y ) = ∇
e ϕ X (ϕ∗ Y ) ,
∗ ∀X, Y ∈ X(M ). (2.50)

Proof. Notice that equation (2.50) is equivalent to

 
∇X Y = ϕ−1 ∗
∇
e ϕ X (ϕ∗ Y ) .
∗

Now let ϕ∗ ∇
e : X(M ) × X(M ) → X(M ) be the map defined by
 
 
ϕ∗ ∇
e Y := ϕ−1 ∗ ∇ e ϕ X (ϕ∗ Y ) .
∗
X

22
2 DIFFERENTIAL GEOMETRY

Firstly, we check whether ϕ∗ ∇e satisfies the properties of a connection from Definition 2.48. On that
∞
account, note that for f, g ∈ C (M ) and X, Z ∈ X(M ) the pushforward of vector fields by ϕ yields

ϕ∗ (f X + gZ) = (f ◦ ϕ−1 )ϕ∗ X + (g ◦ ϕ−1 )ϕ∗ Z.

Therefore, if also Y ∈ X(M ), then

 
(ϕ∗ ∇)
e f X+gZ Y = ϕ−1
∗
∇e ϕ (f X+gZ) (ϕ∗ Y )
∗

−1

 
= ϕ ∗
∇e (f ◦ϕ−1 )ϕ X+(g◦ϕ−1 )ϕ Z (ϕ∗ Y )
∗ ∗
 
= ϕ−1 ∗ (f ◦ ϕ−1 )∇ e ϕ X (ϕ∗ Y ) + (g ◦ ϕ−1 )∇


e ϕ Z (ϕ∗ Y ) ,
∗ ∗

e is a connection. By now applying the pushforward ϕ−1

where the last step follows from the fact that ∇ ∗
on the vectors (f ◦ ϕ−1 )∇e ϕ X (ϕ∗ Y ) , (g ◦ ϕ−1 )∇
∗
e ϕ Z (ϕ∗ Y ) ∈ X(M
∗
f), we see that

 
 
 
ϕ−1 ∗
(f ◦ ϕ−1 )∇
e ϕ X (ϕ∗ Y ) + (g ◦ ϕ−1 )∇
∗
e ϕ Z (ϕ∗ Y ) = f ϕ−1
∗ ∗
e ϕ X (ϕ∗ Y ) + g ϕ−1
∇ ∗ ∗
∇
e ϕ Z (ϕ∗ Y )
∗

= f (ϕ∗ ∇)
e X Y + g(ϕ∗ ∇)
e Z Y,

since clearly f ◦ ϕ−1 ◦ ϕ = f and g ◦ ϕ−1 ◦ ϕ = g. Indeed, ϕ∗ ∇ e satisfies (i) of Definition 2.48.
∗e
Next, we shows that ϕ ∇ is linear over R in Y . Let a, b ∈ R and X, Y, Z ∈ X(M ), then the results is
again a direct consequence of ∇e being a connection on T M f:

 
(ϕ∗ ∇)
e X (aY + bZ) = ϕ−1
∗
∇
e ϕ X (ϕ∗ (aY + bZ))
∗

−1

 
= ϕ ∗
∇
e ϕ X (aϕ∗ Y + bϕ∗ Z))
∗

−1

 
 
=a ϕ ∇e ϕ X (ϕ∗ Y ) + b ϕ−1 ∇
e ϕ X (ϕ∗ Z)
∗ ∗ ∗ ∗

= a(ϕ∗ ∇)
e X Y + b(ϕ∗ ∇)
e X Z,

as desired.
For (iii) of Definition 2.48, let f ∈ C ∞ (M ) and X, Y ∈ X(M ). By following similar reasoning as
above, we have

 
(ϕ∗ ∇)
e X f Y = ϕ−1
∗
∇
e ϕ X (ϕ∗ f Y )
∗

−1



= ϕ e ϕ X (f ◦ ϕ−1 )ϕ∗ Y
∇
∗ ∗
 
= ϕ−1 ∗ (f ◦ ϕ−1 )∇


e ϕ X (ϕ∗ Y ) + (ϕ∗ Xf )Y
∗

 
= f ϕ−1 ∗ ∇ e ϕ X (ϕ∗ Y ) + ϕ−1 (ϕ∗ Xf ) Y


∗ ∗

= f (ϕ∗ ∇)
e X Y + (Xf )Y,

where the last step follows from observing that ϕ−1 ∗ (ϕ∗ Xf ) = dϕ−1 ◦ (ϕ∗ Xf ) ◦ ϕ = dϕ−1 ◦ dϕ ◦ X ◦

ϕ−1 ◦ ϕ = Xf .
The remainder of this proof aims to show that ϕ∗ ∇ e is compatible with g and symmetric. For the
former, recall Proposition 2.56 and Definition 2.44. For X, Y, Z ∈ X(M ), we have

 
(ϕ∗ ∇)
e X g(Y, Z) = ϕ−1
∗
∇
e ϕ X (ϕ∗ g(Y, Z))
∗
 
= ϕ−1 ∗ ∇


e ϕ X (g̃(ϕ∗ Y, ϕ∗ Z)) ,
∗

23
2.5 Geodesics and normal coordinates

since ϕ is an isometry. Consequently, since ∇ e is compatible with the metric g̃, we get

 
    
ϕ−1 ∗ ∇ e ϕ X (g̃(ϕ∗ Y, ϕ∗ Z)) = ϕ−1
∗ ∗
g̃ ∇
e ϕ X (ϕ∗ Y ), ϕ∗ Z + g̃ ϕ∗ Y, ∇
∗
e ϕ X (ϕ∗ Z)
∗

−1

 
 
= ϕ g̃ e ϕ X (ϕ∗ Y ), ϕ∗ Z + ϕ−1 g̃ ϕ∗ Y, ∇
∇ e ϕ X (ϕ∗ Z)
∗ ∗ ∗ ∗
   
= g ϕ−1 ∗ ∇ e ϕ X (ϕ∗ Y ), Z + g Y, ϕ−1 ∇



e ϕ X (ϕ∗ Z)
∗ ∗ ∗
   
∗e ∗e
= g (ϕ ∇)X Y, Z + g Y, (ϕ ∇)X Z ,

as desired. Symmetry then follows from the following computation:

 
 
(ϕ∗ ∇)
e X Y − (ϕ∗ ∇)
e Y X = ϕ−1
∗
e ϕ X (ϕ∗ Y ) − ϕ−1
∇ ∗ ∗
∇
e ϕ Y (ϕ∗ X)
∗

= ϕ−1 ∗ [ϕ∗ X, ϕ∗ Y ]

= [X, Y ].

Indeed, ϕ∗ ∇
e is a Levi-Civita connection of g, and by uniqueness, we must have that ϕ∗ ∇
e = ∇.

Example 2.61. In continuation of Example 2.47, the nonzero coefficients of the Levi-Civita connection ∇
for S2 are given by

g θθ
 
θ ∂ ∂ ∂
Γϕϕ = gϕθ + gϕθ − gϕϕ
2 ∂ϕ ∂ϕ ∂θ
1
= (0 + 0 − 2 sin θ cos θ)
2
= − sin θ cos θ,

and (after similar computations)

cos θ
Γϕ ϕ
θϕ = Γϕθ = .
sin θ
♣

2.5 Geodesics and normal coordinates

Geodesics are the equivalent of Euclidean straight lines on Riemannian manifolds. As pointed out in
[21, Chapter 4], it turned out to be the most useful way to define them as curves with zero acceleration.
To define them, we need another interpretation of a connection. Instead of interpreting the covariant
derivative as a rule for taking directional derivatives of vector fields, we instead need a way to differentiate
vector fields along curves. For our purpose, we do not necessarily need geodesics. However, we do need
them to define the exponential map, which in turn will be used to define normal coordinates. Normal
coordinates will come out handy for proving various properties in the next chapter. By keeping this in
mind, we will keep this section relatively short and straight to the point. Therefore, we will leave out
any proofs and refer to [21] when desired.
For the next couple of definitions, we are not necessarily working on a Riemannian manifold with its
Levi-Civita connection, but we move slightly back to our ordinary smooth manifold M with its linear
connection on the tangent bundle ∇. To begin with, let us define smooth vector fields along a curve.
Definition 2.62. Let γ : R ⊃ I → M a smooth curve. A smooth vector field along γ is a smooth map
V : I → T M such that V (t) ∈ Tγ(t) M for all t ∈ I. The space of all smooth vector fields along γ is
denoted by X(γ). ♥

More generally, we can also define smooth tensor fields along a curve γ in a similar way. If Ve : M ⊆
U → T U ⊆ T M is a smooth vector field such that γ(I) ⊆ U , then we can define V : I → M by setting
V (t) = Veγ(t) for all t ∈ I. Since then V = Ve ◦ γ, it is clearly smooth. If such a smooth vector field Ve
exists on a neighbourhood of γ(I), we say that V is extendible.

24
2 DIFFERENTIAL GEOMETRY

As shown in [21, Theorem 4.24], a connection ∇ in T M determines a unique operator Dt : X(γ) →

X(γ), which we call the covariant derivative along γ. This map is linear over R, satisfies the product rule,
and has the property that Dt V (t) = ∇γ 0 (t) Ve if V is an extendible vector field along γ. With this map
we can define the acceleration of γ and consequently geodesics.
Definition 2.63. The acceleration of a smooth curve γ : I → M is the smooth vector field X(γ) 3 Dt γ 0
along γ. ♥
Definition 2.64. A smooth curve γ : I → M is a geodesic (with respect to ∇) if its acceleration is zero,
i.e. if Dt γ 0 ≡ 0. ♥
Alternatively, one could characterize γ as a geodesic if its components satisfy the geodesic equation.
That is, if (xi ) are smooth local coordinates such that γ i = xi ◦ γ(t), then
γ̈ k + Γkij γ̇ i γ̇ j = 0. (2.51)
By using theory of ordinary differential equations, one could proof that for every p ∈ M , v ∈ Tp M
and t0 ∈ R, there exists an open interval I ⊆ R and a unique geodesic γ : I → M such that γ(t0 ) = p
and γ 0 (t0 ) = v (see [21, Theorem 4.27]). We say that a geodesic γ : I → M is maximal, if there does not
exists a geodesic γ e : Ie → M such that γe I = γ. Moreover, for each p ∈ M and v ∈ Tp M there exists a
unique maximal such that γ(0) = p and γ 0 (0) = v (see [21, Corollary 4.28]). Alternatively, we call this
geodesic the geodesic with initial point p and initial velocity v and denote it by γv .
In the remainder of this section, we again consider Riemannian manifolds (M, g) endowed with its
Levi-Civita connection (although the theory is not exclusively applicable to the Riemannian setting).
Despite the fact that maximal geodesics are unique, we can relate one another with proportional initial
velocities due to the rescaling lemma [21, Lemma 5.8]. Moreover, consider the following definition.
Definition 2.65. Let γv be the unique maximal geodesic satisfying γv (0) = p ∈ M and γv0 (0) = v ∈ Tp M .
The exponential map at p is the map expp : Tp M → M defined by
expp (v) := γv (1). (2.52)
♥
Remark 2.66. A complete manifold or geodesically complete manifold 11 is a Riemannian manifold for
which every maximal geodesic is defined for all t ∈ R. Equivalently, the exponential map at p has all of
Tp M as its domain. From now on, we always assume that a Riemannian manifold (M, g) is geodesically
complete. F
Clearly, since for 0 ∈ Tp M we have γ0 (t) = p ∈ M for all t ∈ I, we have that expp (0) = γ0 (1) = p.
Geometrically, expp (v) is the point in M obtained by following the geodesic that passes through p for a
unit time in the direction of v [24, Section 3.2]. This map is in fact smooth (see [20, Proposition 5.7])
and for each p ∈ M its differential is the identity map of Tp M [21, Proposition 5.19]. Therefore, the
differential is invertible, such that the inverse function theorem (see [22, Thoerem 2.8.7]) guarantees the
existence of a star-shaped 12 neighbourhood V of 0 ∈ Tp M and a neighbourhood U about p ∈ M such
that expp : Tp M ⊇ V → U ⊆ M is a diffeomorphism.
 We call the neighbourhood
 U of p a normal neighbourhood of p. Now, an orthonormal basis
∂
∂xi | i ∈ {1, . . . , n} for Tp M induces a basis isomorphism B : Rn → Tp M by B(r1 , . . . , rn ) =
p
∂
ri ∂x i . For a normal neighbourhood U = expp (V ) of p, one could then combine this isomorphism
p
with the exponential map to obtain a (unique, see [21, Proposition 5.23]) smooth coordinate map ϕ =

−1
(x1 , . . . , xn ) := B −1 ◦ expp V : U → Rn . We call these coordinates (Riemannian) normal coordinates
centred at p.
The normal coordinates will be very useful in the next chapter due to the following proposition.
11 Due to the Hopf-Rinow theorem, saying that M is geodesically complete is equivalent to saying M is complete as a

metric space [21, Theorem 6.19].

12 An open subset V containing the origin is called star-shaped if V also contains the line segments from 0 to q for any

q ∈V.

25
2.6 Curvature

Proposition 2.67. Let (M, g) be a Riemannian manifold and let (U, (xi )) be any normal coordinate
chart centered at p ∈ M . Then

(i) in coordinates we have that p = (0, . . . , 0);

(ii) the components of the metric at p are gij (p) = δij ;

(iii) the Christoffel symbols vanish at p, i.e. Γkij (p) = 0;

∂
(iv) the first partial derivatives of the metric vanish at p, i.e. g (p).
∂xk ij

Proof. See [21, Proposition 5.24].

Remark 2.68. Since the Christoffel symbols vanish at p in a normal coordinate system, from equa-
∂
tion (2.33) we see that then ∇i ∂x j = 0. In the light of Lemma 2.52, this implies that for τ ∈ Tsr (M )
p
we have13
   
∂ ∂ ∂ ∂ ∂
(∇k τp ) dxj1 , . . . , dxjr , i1 ,..., = τp dx
j1
, . . . , dxjr , i1 ,..., .
p p ∂x p ∂xis p ∂xk p p ∂x p ∂xis p

2.6 Curvature
With the Levi-Civita connection at hands, we are ready to examine curvature. Informally, we can say that
curvature measures the distance between a Riemannian manifold and Euclidean space. If this distance
at a point is non-zero, that means the space at this point is curved. In this sense, curvature enables us
to express the extent to which a space is curved.

Definition 2.69. Suppose that (M, g) is a n-dimensional Riemannian manifold equipped with its Levi-
Civita connection ∇. The Riemann curvature endomorphism is the map R : X(M ) × X(M ) × X(M ) →
X(M ) defined by
R(X, Y )Z := ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z. (2.53)
♥

Since the Levi-Civita connection is symmetric, it follows from equations (2.48) and (2.43) that

R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z
= ∇X ∇Y Z − ∇Y ∇X Z − ∇(∇X Y −∇Y X) Z
= ∇2X,Y Z − ∇2Y,X Z (2.54)

By using Lemma 2.34, one could derive that R is a (1, 3)-tensor field [21, Proposition 7.3]. Moreover,
using our smooth local frame and coframe for a coordinate neighbourhood U ⊆ M at a point p ∈ M with
local coordinates (xi ), the curvature endomorphism can be expressed as the tensor

l ∂
Rp = Rijk (p)dxi ⊗ dxj ⊗ dxk ⊗ , (2.55)
p p p ∂xl p
l
with the coefficients Rijk given by
 
∂ ∂ ∂ l ∂
Rp i
, j = Rijk (p) . (2.56)
∂x p ∂x p ∂xk p ∂xl p
13 See Section B for a justification of this result.

26
2 DIFFERENTIAL GEOMETRY

Proposition 2.70. If U, (xi ) is a smooth local coordinate chart of a Riemannian manifold M , then
we can express the coefficients of the Riemann curvature endomorphism as
l ∂ l ∂ l
Rijk = Γ − Γ + Γm l m l
jk Γim − Γik Γjm . (2.57)
∂xi jk ∂xj ik
♦
Proof. In terms of smooth coordinate vector fields, we use equation (2.33) to directly compute

l ∂ ∂ ∂ ∂
Rijk = ∇i ∇j k − ∇j ∇i k − ∇[i,j] k
∂xl  ∂x  ∂x  ∂x
l ∂ l ∂
= ∇i Γjk l − ∇j Γik l
∂x ∂x
∂ l ∂ ∂ l ∂ ∂ ∂
= Γ
i jk ∂xl
− Γ
j ik ∂xl
+ Γmjk ∇i m
− Γmik ∇j m
∂x
 ∂x ∂x  ∂x
∂ l ∂ l ∂
= Γ − Γ + Γm l m l
jk Γim − Γik Γjm .
∂xi jk ∂xj ik ∂xl

Example 2.71. Recall the Christoffel symbols for S2 of Example 2.61. The nonzero coefficients of the
Riemann curvature endomorphism are given by
∂
θ
Γθϕϕ + Γθϕϕ Γϕ


Rϕθϕ = ϕθ
∂θ
2
= sin θ − cos θ + cos2 θ
2

= sin2 θ,
and (by similar computations)
ϕ ϕ
θ
Rϕϕθ = − sin2 θ, Rθϕθ = −1, Rθθϕ = 1.
♣
Similarly to Lemma 2.52 where we established how a connection acts on tensor fields, we are interested
in how the curvature endomorphism acts on a tensor bundle Tsr (M ).
Lemma 2.72. Let R be the Riemann curvature endomorphism on the tensor bundle Tsr (M ). If τ ∈
Tsr (M ), ω 1 , . . . , ω r ∈ X∗ (M ) and Z1 , . . . , Zs ∈ X(M ), then
(R(X, Y )τ ) ω 1 , . . . , ω r , Z1 , . . . , Zs =R(X, Y ) τ ω 1 , . . . , ω r , Z1 , . . . , Zs

Xr
τ ω 1 , . . . , R(X, Y )ω i , . . . , ω r , Z1 , . . . , Zk


−
i=1 (2.58)
Xs
τ ω 1 , . . . , ω r , Z1 , . . . , R(X, Y )Zj , . . . , Zs .


−
j=1

♠
Proof. The proof follows from first showing that R obeys the product rule with respect to tensor products
and commutes with all contractions. By using these facts and following similar argumentation as in the
proof of Lemma 2.52, one obtains the desired equality. For more detail, see [18, Proposition 2.43].
Definition 2.73. The Riemann curvature tensor is the (0, 4)-tensor field Rm obtained from the (1, 3)-
Riemann curvature endomorphism by lowering its last index. That is, R[ = Rm : X(M ) × X(M ) ×
X(M ) × X(M ) → C ∞ (M ), defined by
Rm(X, Y, Z, W ) := g (R(X, Y )Z, W ) . (2.59)
♥

27
2.6 Curvature

Remark 2.74. Throughout literature, the Riemann curvature tensor is sometimes just defined as in
Definition 2.69. The curvature tensor, as defined above, is used when convenient. F
In terms of smooth local coordinates (xi ), the Riemann curvature tensor assigns a tensor to each point
p ∈ M as
Rmp = Rmijkl (p)dxi ⊗ dxj ⊗ dxk ⊗ dxl , (2.60)
p p p p
m
with the coefficients Rmijkl = glm Rijk .
Therefore, similarly to equation (2.57), we can express these
coefficients using smooth local coordinates (xi ) as
 
∂ m ∂ m p m p m
Rmijkl = glm Γ − Γ + Γ Γ − Γik jp .
Γ (2.61)
∂xi jk ∂xj ik jk ip

What makes the Riemann curvature tensor specifically valuable, is that it enables one to define
curvature intrinsically. This means that it can be measured for Riemannian manifolds that are not
necessarily embedded in an ambient Euclidean space. The Riemann curvature tensor assigns a tensor
to each point of a manifold and formulates the extent to which the Riemannian metric is not locally
isometric to the Euclidean metric. Since Euclidean space is flat, the curvature is zero. Therefore, we say
that a Riemannian manifold is flat if

∇X ∇Y Z − ∇Y ∇X Z = ∇[X,Y ] Z, ∀X, Y, Z ∈ X(M ). (2.62)

Just as the Levi-Civita connection, the Riemann curvature tensor is invariant under isometries [18,
Theorem 4.1].
Proposition 2.75. The Riemann curvature tensor Rm is invariant under local isometries. That is, if
f, g̃) is a local isometry, then ϕ∗ Rm
ϕ : (M, g) → (M g = Rm. ♦
Proof. The proof follows from a direct computation using Definitions 2.69 and 2.73 and Proposition 2.60.
Let X, Y, Z, W ∈ X(M ), then
 
ϕ∗ Rm
g (X, Y, Z, W ) = Rmg (ϕ∗ X, ϕ∗ Y, ϕ∗ Z, ϕ∗ W )

= g̃ (R(ϕ∗ X, ϕ∗ Y )ϕ∗ Z, ϕ∗ W )
 
= g̃ ∇ e ϕ X∇
∗
e ϕ Y (ϕ∗ Z) − ∇
∗
eϕ Y ∇
∗
e ϕ X (ϕ∗ Z) − ∇
∗
e [ϕ X,ϕ Y ] (ϕ∗ Z) , ϕ∗ W
∗ ∗



= g̃ ϕ∗ ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z , ϕ∗ W
= (ϕ∗ g̃) (R(X, Y )Z, W )
= Rm(X, Y, Z, W ).

Example 2.76. The coefficients of the Riemann curvature tensor on S2 are easily obtained from the
curvature endomorphism coefficients:
ϕ ϕ
Rmθϕθϕ = gϕϕ Rθϕθ = − sin2 θ, Rmθθϕϕ = gϕϕ Rθθϕ = sin2 θ,
θ
Rmϕθϕθ = gθθ Rϕθϕ = sin2 θ, θ
Rmϕϕθθ = gθθ Rϕϕθ = − sin2 θ.

♣
The Riemann curvature tensor satisfies multiple symmetries which often come out useful.
Lemma 2.77. Let (M, g) be a Riemannian manifold. The (0, 4)-Riemann curvature tensor of g satisfies
the following symmetries for all W, X, Y, Z ∈ X(M ):
(i) Rm(W, X, Y, Z) = −Rm(X, W, Y, Z);
(ii) Rm(W, X, Y, Z) = −Rm(W, X, Z, Y );

28
2 DIFFERENTIAL GEOMETRY

(iii) Rm(W, X, Y, Z) = −Rm(X, W, Y, Z);

(iv) Rm(W, X, Y, Z)+Rm(X, Y, W, Z)+Rm(Y, W, X, Z) = 0. This is also referred to as the first Bianchi
identity.
In addition, the total covariant derivative of the Riemann curvature tensor satisfies the second Bianchi
identity:

∇Rm(X, Y, Z, V, W ) + ∇Rm(X, Y, V, W, Z) + ∇Rm(X, Y, W, Z, V ) = 0, ∀X, Y, Z, V, W ∈ X(M ).

(2.63)
♠
Proof. The proofs of these symmetries are not of particular interest, and we refer to e.g. [21, Proposition
7.12] for the details.
The penultimate tensor that we will look at in this section is the Ricci curvature tensor. Its geometric
meaning will be thoroughly investigated in the next section. For now, we will leave it with the definition.
Although it does not carry all the information of the curvature endomorphism and tensor, it is a much
simpler object.
Definition 2.78. The Ricci curvature or Ricci tensor denoted by Ric is the covariant (0, 2)-tensor field
defined as the (1, 1)-trace of the curvature endomorphism. That is, the map Ric : X(M ) × X(M ) →
C ∞ (M ), defined by
Ric(X, Y ) := tr11 (Z 7→ R(Z, X)Y ) . (2.64)
♥
For smooth local coordinates (xi ) about p ∈ M , we have

Ricp = Ricij (p)dxi ⊗ dxj . (2.65)

p p

The coefficients of the Ricci tensor can be conveniently obtained from the Riemann curvature endomor-
phism and tensor:
k
Ricij = Rkij = g km Rmkijm . (2.66)
Proposition 2.79. The Ricci curvature Ric is invariant under local isometries. That is, if ϕ : (M, g) →
f, g̃) is a local isometry, then ϕ∗ Ric
(M g = Ric. ♦
Proof. The proof follows from Proposition 2.75. Observe that
   
∗g ∂ ∂ ∂ ∂
ϕ Ric , = Ric ϕ∗ i , ϕ∗ j
g
∂xi ∂xj ∂x ∂x
 
−1 k m g

∂ ∂ ∂ ∂
=gg ϕ∗ dx , ϕ∗ dx Rm ϕ∗ k , ϕ∗ i , ϕ∗ j , ϕ∗ m
∂x ∂x ∂x ∂x
 
∂ ∂ ∂ ∂
= ϕ∗ gg
−1 dxk , dxm ϕ∗ Rm


g , , ,
∂xk ∂xi ∂xj ∂xm
 
−1 k m

∂ ∂ ∂ ∂
=g dx , dx Rm , , ,
∂xk ∂xi ∂xj ∂xm
 
∂ ∂
= Ric i
, j .
∂x ∂x

Example 2.80. The Ricci coefficients for S2 are found by applying (i) of Lemma 2.77 on equation (2.66):
1
Ricθθ = g ϕϕ Rϕθθϕ = −g ϕϕ Rθϕθϕ = sin2 θ = 1, Ricϕϕ = g θθ Rθϕϕθ = −g θθ Rϕθϕθ = sin2 θ.
sin2 θ
♣

29
2.6 Curvature

Finally, we define the scalar curvature as the trace with respect to g of the Ricci curvature.

Definition 2.81. The scalar curvature S : C ∞ (M ) → R is the map defined by

S := trg (Ric). (2.67)

Notice that by (2.26), we have trg (Ric) = g ij Ricij .

Example 2.82. The scalar curvature for S2 is given by
1
S = g ij Ricij = 1 + sin2 θ = 2.
sin2 θ
♣

30
3 RICCI FLOW

3 Ricci flow
Informally, one could think of Ricci flow as the process of stretching or contacting the metric depending
on the Ricci curvature. If the curvature is negative, the metric is stretched, and for positive curvature the
metric is contracted. Moreover, the stronger the curvature, i.e. the ‘more negative’ or ‘more positive’, the
faster the stretching or contracting occurs. The stretching and contracting can be thought of as a change
in the distance between each pair of points of a manifold. In this way, Ricci flow has shown to be a useful
tool for ‘improving’ or ‘smoothing’ the geometry of a manifold whilst preserving any present symmetries.
However, in this process, Ricci flow is likely to develop singularities in finite time. Perelman developed a
technique called ‘surgery’ to properly deal with these singularities, which helped him to prove the famous
Poincaré conjecture in his series of papers [4, 5, 6].
Despite the occurrence of singularities, Ricci flow is nevertheless a useful tool to understand the
topology of a manifold. Although we will not focus on performing surgery on manifolds with singularities,
we will investigate how Ricci flow can be used to ‘improve’ a manifold’s geometry and discuss the short-
time existence and uniqueness of a solution.
Mathematically, Ricci flow allows the metric g of a Riemannian manifold to evolve under the PDE
∂
g(t) = −2Ric(g(t)), g(0) = g0 . (3.1)
∂t
From now on, we are therefore not interested in a single Riemannian manifold (M, g), but rather in
a family of closed manifolds (M, g(t)) with t ∈ [0, ] ⊂ R [18, Section 4.1]. This chapter begins with
providing the necessary intuition for what it actually means to take time derivatives of the metric.
Furthermore, we discuss various important properties that we will use to eventually prove short-time
existence and uniqueness. To start off, let us solve the Ricci flow on S2 .
Example 3.1. In our previous examples, we examined the various curvature tensors on S2 with radius
r = 1. Now, since we consider a family of metrics dependent of time, we consider a radius r dependent
of time. The matrix representation on Tp S2r(t) is then
 
1 0
g(t) = r2 (t) =: r2 (t)g0 . (3.2)
0 sin2 θ
One could derive that the coefficients for the Ricci tensor remain unchanged, and are hence independent
of the radius such that Ric = g0 . The Ricci flow for S2 is hence given by
∂ 2
r (t)g0 = −2g0 , g(0) = r(0)g0 = g0 , (3.3)
∂t
where we assume that the initial radius is r(0) = 1. By taking the time derivatives on the left hand side,
this becomes
∂
2ṙrg0 + r2 (t) g0 = 2ṙrg0 + 0,
∂t
and (3.3) becomes

2ṙrg0 = −2g0 =⇒ ṙr = −1

Z r Z 1
=⇒ rdr = − 1dt
1 0
2
r (t) 1
=⇒ − = −t
2 √2
=⇒ r(t) = 1 − 2t.

Hence, we see that the solution to the Ricci flow is given by g(t) = (1 − 2t)g0 . So S2 shrinks to a
point as t → 21 . In fact, one could show that any unit sphere (Sn , g0 ) shrinks to a point since then
g(t) = (1 − 2(n − 1)t)g0 [18, Section 3.1.1.1].
♣

31
3.1 Time derivatives and deformations of geometric quantities

Figure 3: The sphere S2 shrinks to a point as t → 12 . In the figure, the sphere is given at t = 0 , t = 5
18
and t = 94 .

Remark 3.2. In the above example, note that we are now dealing with manifolds without a fixed volume.
To preserve the volume, one should consider normalized Ricci flow as in e.g. [28, Thereom 1.1-1.3] or [29,
Theorem 1.2]. Since the scalar curvature equals twice the Gaussian curvature K and Ric = Kg in two
dimensions14 , the normalized Ricci flow is then given by

∂
g(t) = (s − S(t))g(t), g(0) = g0 , (3.4)
∂t

where s denotes the average scalar curvature of S. Explicitly, s is given by15

4πχ(M )
s= .
vol(M )

Proving (short-time) existence and uniqueness of solutions to (normalized) Ricci flow on compact two-
dimensional Riemannian manifolds is significantly easier (see e.g. [30]). This is because in two dimensions,
the deformation of the metric is conformal16 , and a solution can be written g(t) = ev(t) g0 , with v :
M × [0, ) → R. If the dimension n ≥ 3, then the Ricci flow becomes a weakly parabolic equation. In
Section 3.3, we proof short-time existence and uniqueness for dimension n ≥ 3. F

3.1 Time derivatives and deformations of geometric quantities

To begin with, it is necessary to establish how one takes time derivatives of the metric and Levi-Civita
connection. Also, we are interested in how the various curvature tensors vary under the evolving metric,
i.e. not necessarily under Ricci flow. This section is mainly based upon [18, Chapter 4], if not mentioned
otherwise. As stated before, we are now more interested in a closed Riemannian manifold M equipped
with a one-parameter family of metrics t 7→ g(t), assuming these exists for some time interval [0, ].
We can consider the elements g(t)
as sections of the smooth rank-2 positive definite symmetric bundle
Σ2 (T ∗ M ), i.e. g(t) ∈ Γ Σ2 (T ∗ M ) [18, Section 4.1.1]. Then we define the derivate of g(t) as follows:

Definition 3.3. Consider the one-parameter family of smooth metrics g = g(t) ∈ Γ Σ2 (T ∗ M ) parametrised


∂
by ‘time’ t. We define the time derivative of the metric as the map ∂t g : X(M )×X(M ) → C ∞ (M ) defined
by
 
∂ ∂
g (X, Y ) := g(X, Y ), (3.5)
∂t ∂t
for any pair of time independent vector fields X, Y ∈ X(M ). ♥
14 Seealso Lemma A.3 in Section A.
15 Here, χ(M ) denotes the Euler characteristic of M , which can be obtained by integrating the Scalar or Gaussian
curvature. For more information we refer to the Gauss-Bonnet theorem for surfaces [21, Chapter 9].
16 Two metrics g and g are said to be conformal if there exists a positive f ∈ C ∞ (M ) such that g = f g [21, Page 59].
1 2 1 2

32
3 RICCI FLOW

Thus, it appears that ∂t ∂

g(X, Y ) is the time derivative of the smooth function g(X, Y ) ∈ C ∞ (M ),
which implies that in local coordinates equation (2.21) becomes
∂
g = ġij (t)dxi dxj . (3.6)
∂t
In accordance with Proposition 2.59, in which we state that the Levi-Civita connection can locally be
written in terms of the metric, ∇ = ∇(t) is time dependent as well.
∂
Definition 3.4. The time derivative of the Levi-Civita connection ∇ is the map ∂t ∇ : X(M ) × X(M ) →
X(M ) defined by  
∂ ∂
∇ (X, Y ) := ∇X Y, (3.7)
∂t ∂t
for any pair of time independent vector fields X, Y ∈ X(M ). ♥
Interestingly, although ∇ is by definition not tensorial since it satisfies the product rule, its time deriva-
tive is tensorial as showed in [18, Lemma 4.3] (By checking the conditions of the tensor characterization
∂
lemma, see Lemma 2.34). Furthermore, we can differentiate the Christoffel symbols Γkij = dxk ∇ ∂ i ∂x j of
∂x
∗ ∞
∇ by considering them as a map Γ : X(M )×X(M )×X (M ) → C (M ) defined by Γ(X, Y, ω) = ω(∇X Y ).
∂
Definition 3.5. The time derivative of the Christoffel symbols of ∇ is the map ∂t Γ : X(M ) × X(M ) ×
X∗ (M ) → C ∞ (M ) defined by  
∂ ∂
Γ (X, Y, ω) := Γ(X, Y, ω), (3.8)
∂t ∂t
for any pair of time independent smooth vector fields X, Y ∈ X(M ) and time independent smooth covector
field ω ∈ X∗ (M ). ♥
∂
Following similar argumentation as for the time derivative of ∇, one could also show that ∂t Γ is
tensorial. Now that we have established how to take time derivatives of the metric, Levi-Civita connection
and the Christoffel symbols, we can describe how the various geometric quantities evolve under the metric.
To begin with, we consider the metric inverse g ij .
Lemma 3.6. Suppose that g(t) is a smooth one-parameter family of metrics on a Riemannian manifold
M . Then17
∂ ij ∂
g = −g ik g jl gkl . (3.9)
∂t ∂t
♠
Proof. Note that gkl g lj = δkj and therefore that
   
∂ lj ∂ lj
gkl g + gkl g = 0.
∂t ∂t
Multiplying by g ik yields
       
ik ∂ lj ik ∂ lj ik jl ∂ ∂ lj
0=g gkl g + g gkl g =g g gkl + δil g ,
∂t ∂t ∂t ∂t
∂ ij ∂



and hence that ∂t g = −g ik g jl ∂t gkl .
Subsequently, we can describe the evolution of the Christoffel symbols of the Levi-Civita connection.
Proposition 3.7. Suppose that g(t) is a smooth one-parameter family of metrics on a Riemannian
manifold. Then the Christoffel symbols of the Levi-Civita connection evolve by
 
∂ k 1 kl ∂ ∂ ∂
Γ = g ∇i gjl + ∇j gil − ∇l gij . (3.10)
∂t ij 2 ∂t ∂t ∂t
♦
  
∂ ∂ ∂
17 Note that due to Definition 3.3, we have g
∂t kl
= ∂t
g , ∂
∂xk ∂xl
.

33
3.1 Time derivatives and deformations of geometric quantities

Proof. Recall the coordinate expression of the Christoffel symbols from (2.49). If we choose normal
coordinates with respect to g(t0 ), i.e. at some time t = t0 , about a point p ∈ M , we compute
   
∂ k 1 ∂ kl ∂ ∂ ∂ 1 kl ∂ ∂ ∂ ∂
Γ (p) = g (p) gjl + gil − gij + g (p) gjl + gil − gij
∂t ij 2 ∂t ∂xi ∂xj ∂xl p 2 ∂t ∂xi ∂xj ∂xl p
 
1 kl ∂ ∂ ∂ ∂ ∂ ∂
= 0 + g (p) gjl (p) + gil (p) − gij (p) ,
2 ∂xi ∂t ∂xj ∂t ∂xl ∂t

where the first term vanishes due to Proposition 2.67 (iv). In view of Remark 2.68, we then have
 
∂ k 1 kl ∂ ∂ ∂
Γ (p) = g (p) ∇i gjl (p) + ∇j gil (p) − ∇l gij (p) ,
∂t ij 2 ∂t ∂t ∂t

from which the desired equality then follows. Since both sides of this equation represent tensors fields in
coordinate expressions, the identity holds for any coordinate system.

In what follows, we derive expressions for the evolution of the Riemann curvature endomorphism and
tensor, and the Ricci curvature.

Proposition 3.8. Suppose that g(t) is a smooth one-parameter family of metric on a Riemannian
manifold M . Then the Riemannian curvature endomorphism R evolves by
 
∂ l 1 ∂ ∂ ∂ ∂ p ∂ p ∂
Rijk = g lm ∇2i,m gjk + ∇2j,k gim − ∇2i,k gjm − ∇2j,m gik − Rijk gpm − Rijm gpk .
∂t 2 ∂t ∂t ∂t ∂t ∂t ∂t
(3.11)
♦

Proof. Recall (2.57) and choose normal coordinates (xi ) with respect to g(t0 ) centred about a point
p ∈ M . We have
       
∂ l ∂ ∂ l ∂ ∂ l ∂ m l m ∂ l
R (p) = i Γ (p) − Γ (p) + Γ (p) Γim (p) + Γjk (p) Γ (p)
∂t ijk ∂x ∂t jk ∂xj ∂t ik ∂t jk ∂t im
   
∂ m ∂ l
− Γik (p) Γljm (p) − Γm
ik (p) Γ (p)
∂t ∂t jm
   
∂ ∂ l ∂ ∂ l
= i Γ (p) − Γ (p) ,
∂x ∂t jk ∂xj ∂t ik

due to the fact that in normal coordinates Γcab (p) = 0. Proposition 3.7 then implies that
    
∂ ∂ l ∂ 1 lm ∂ ∂ ∂
Γ (p) = i g (p) ∇j gkm (p) + ∇k gjm (p) − ∇m gjk (p)
∂xi ∂t jk ∂x 2 ∂t ∂t ∂t
 
1 ∂ lm ∂ ∂ ∂
= g (p) ∇j gkm (p) + ∇k gjm (p) − ∇m gjk (p)
2 ∂xi ∂t ∂t ∂t
  
1 lm ∂ ∂ ∂ ∂
+ g (p) ∇j gkm (p) + ∇k gjm (p) − ∇m gjk (p)
2 ∂xi ∂t ∂t ∂t
 
1 lm 2 ∂ 2 ∂ 2 ∂
= g (p) ∇i,j gkm (p) + ∇i,k gjm (p) − ∇i,m gjk (p) ,
2 ∂t ∂t ∂t

where the last step follows from Proposition 2.67 (iii) and Remark 2.68. Hence, we have
 
∂ l 1 ∂ ∂ ∂
Rijk (p) = g lm (p) ∇2i,j gkm (p) + ∇2i,k gjm (p) − ∇2i,m gjk (p)
∂t 2 ∂t ∂t ∂t
  (3.12)
1 lm ∂ ∂ ∂
− g (p) ∇2j,i gkm (p) + ∇2j,k gim (p) − ∇2j,m gik (p) .
2 ∂t ∂t ∂t

34
3 RICCI FLOW

Notice that equation (2.54) and consequently (2.56) combined with Lemma 2.72 give us that
    
∂ ∂ ∂ ∂ ∂ ∂ ∂
∇2i,j gkm − ∇2j,i gkm = R , g ,
∂t ∂t ∂xi ∂xj ∂t ∂xk ∂xm
     
∂ ∂ ∂ ∂ ∂ ∂ ∂
=R , gkm − g R , ,
∂xi ∂xj ∂t ∂xi ∂xj ∂xk ∂xm
   
∂ ∂ ∂ ∂ ∂
− g , R ,
∂t ∂xk ∂xi ∂xj ∂xm
p ∂ p ∂
=0 − Rijk gpm − Rijm gpk , (3.13)
∂t ∂t
2 2
∂ ∂
gkm = ∇i ∇j gkm −∇j ∇i gkm −∇[i,j] gkm = ∂x∂i ∂xj gkm − ∂x∂j ∂xi gkm −0 = 0. The identity


since R ∂x i , ∂xj

then follows by combining equations (3.12) and (3.13).

Proposition 3.9. Suppose that g(t) is a smooth one-parameter family of metrics on a Riemannian
manifold M . Then the Riemann curvature tensor Rm evolves by
 
∂ 1 2 ∂ 2 ∂ 2 ∂ 2 ∂
Rmijkl = ∇i,l gjk + ∇j,k gil − ∇i,k gjl − ∇j,l gik
∂t 2 ∂t ∂t ∂t ∂t
  (3.14)
1 pq ∂ ∂
+ g Rmijkp gql + Rmijpl gqk .
2 ∂t ∂t
♦
m
Proof. Since Rmijkl = glm Rijk , we have

∂ ∂ m ∂ m
Rmijkl = glm Rijk + glm Rijk .
∂t ∂t ∂t
From Proposition 3.8 we get
  
∂ m 1 ml 2 ∂ 2 ∂ 2 ∂ 2 ∂ p ∂ p ∂
glm Rijk = glm g ∇i,l gjk + ∇j,k gil − ∇i,k gjl − ∇j,l gik − Rijk gpl − Rijl gpk
∂t 2 ∂t ∂t ∂t ∂t ∂t ∂t
   
1 2 ∂ 2 ∂ 2 ∂ 2 ∂ 1 p ∂ p ∂
= ∇i,l gjk + ∇j,k gil − ∇i,k gjl − ∇j,l gik − Rijk gpl − Rijl gpk
2 ∂t ∂t ∂t ∂t 2 ∂t ∂t
   
1 2 ∂ 2 ∂ 2 ∂ 2 ∂ 1 pq ∂ ∂
= ∇i,l gjk + ∇j,k gil − ∇i,k gjl − ∇j,l gik − g Rmijkp gql + Rmijlp gqk .
2 ∂t ∂t ∂t ∂t 2 ∂t ∂t
(3.15)
Furthermore, by using Lemma 2.77 for the last term of (3.15) and since
∂ q ∂
glq Rijk = g pq Rmijkp gql , (3.16)
∂t ∂t
we obtain the desired equality by adding (3.15) with (3.16) .
Proposition 3.10. Suppose that g(t) is a smooth one-parameter family of metrics on a Riemannian
manifold M . Then the Ricci curvature Ric evolves by
 
∂ 1 kl 2 ∂ 2 ∂ 2 ∂ 2 ∂
Ricij = g ∇k,i gjl + ∇k,j gil − ∇k,l gij − ∇i,j gkl (3.17)
∂t 2 ∂t ∂t ∂t ∂t
♦
k
Proof. Since Ricij = Rkij , the desired result can be conveniently obtained from equation (3.12). I.e., we
have
 
∂ ∂ k 1 kl 2 ∂ 2 ∂ 2 ∂ 2 ∂ 2 ∂ 2 ∂
Ricij = Rkij = g ∇k,i gjl + ∇k,j gil − ∇k,l gij − ∇i,k gjl − ∇i,j gkl + ∇i,l gkj .
∂t ∂t 2 ∂t ∂t ∂t ∂t ∂t ∂t

35
3.2 Evolution of curvature under Ricci flow

Now, notice that

 
1 kl 2 ∂ 2 ∂ 1 ∂ 1 ∂
g ∇i,l gkj − ∇i,k gjl = g kl ∇2i,l gkj − g lk ∇2i,l gkj = 0,
2 ∂t ∂t 2 ∂t 2 ∂t

since both the metric and its inverse are symmetric.

3.2 Evolution of curvature under Ricci flow

∂
With the expressions from the preceding section, we can substitute ∂t g = −2Ric to derive expressions
that describe the evolution of geometric quantities under Ricci flow. It turns out that we can simplify
these expression by means of the quadratic curvature tensor. Again, we will primarily follow [18, Chapter
4].

Definition 3.11. The quadratic curvature tensor is the (0, 4)-tensor field B : X(M ) × X(M ) × X(M ) ×
X(M ) → C ∞ (M ) defined by

B(X, Y, Z, W ) := hRm(X, ·, Y, ?), R(W, ·, Z, ?)i . (3.18)

In accordance with Remark 2.39, for the components we get

    
∂ ∂ ∂ ∂
Bijkl = Rm , ·, , ? , Rm , ·, , ?
∂xi ∂xj ∂xk ∂x;
= g pr g qs Rmipjq Rmkrls
= g pr g qs Rmpiqj Rmrksl , (3.19)

with the last step following from Lemma 2.77 (i) and (ii). In addition, we define the Laplacian on a
connection on the tensor bundle.

Definition 3.12. For any tensor τ ∈ Tsr (M ), we define the connection Laplacian as the trace of the
second covariant derivative with respect to g. That is,

∆τ := trg ∇2 τ .


(3.20)

Recall equations (2.26) and (2.42) to see that

   
j1 jr ∂ ∂ 2

j1 jr ∂ ∂
(∆τ ) dx , . . . , dx , i1 , . . . , is = trg ∇ τ dx , . . . , dx , i1 , . . . , is
∂x ∂x ∂x ∂x
 
1,2 −1 2

j1 jr ∂ ∂
= tr1,2 g ⊗ ∇ τ dx , . . . , dx , i1 , . . . , is
∂x ∂x
  
1,2 pq ∂ ∂ 2 j1 jr ∂ ∂
= tr1,2 g ⊗ ⊗∇ τ dx , . . . , dx , i1 , . . . , is
∂xp ∂xq ∂x ∂x
 
∂ ∂
= g pq ∇2p,q τ dxj1 , . . . , dxjr , i1 , . . . , is .


(3.21)
∂x ∂x

Since the Levi-Civita connection is compatible with g, the covariant derivatives commute with metric
contractions. Therefore, we have that (see [18, Proposition 2.48])

∇k Ricij = g pq ∇k Rmpijq and ∇2k,l Ricij = g pq ∇2k,l Rmpijq (3.22)

Now we can examine how the connection Laplacian acts on Rm.

36
3 RICCI FLOW

Proposition 3.13. The Laplacian of the Riemannian curvature tensor Rm satisfies

∆Rmijkl =∇2i,k Ricjl − ∇2j,k Ricil + ∇2j,l Ricik − ∇2i,l Ricjk
(3.23)
− 2 (Bijkl − Bijlk − Biljk + Bikjl ) + g pq (Rmqjkl Ricpi + Rmiqkl Ricpj ) .
♦
Proof. The following proof is in analogy with the proofs of [18, Proposition 4.2] and [1, Lemma 7.2] and
works with normal coordinates centred about an arbitrary p ∈ M . Due to respectively equation (3.21),
the second Bianchi identity (2.63) and (i) of Lemma 2.77, we have that

∆Rmijkl = trg ∇2 Rmijkl

= g pq ∇2p,q Rmijkl

= g pq −∇2p,i Rmjqkl − ∇2p,j Rmqikl

= g pq ∇2p,i Rmqjkl − ∇2p,j Rmqikl .

Next, recall equation (2.54) and notice that by applying the second Bianchi identity again we have
    
2 2 ∂ ∂ ∂ ∂ ∂ ∂
∇p,i Rmqjkl = ∇i,p Rmqjkl + R , Rm , , ,
∂xi ∂xp ∂xq ∂xj ∂xk ∂xl
    
2 2 ∂ ∂ ∂ ∂ ∂ ∂
= ∇i,k Rmjqlp − ∇i,l Rmjqkp + R , Rm , , , ,
∂xi ∂xp ∂xq ∂xj ∂xk ∂xl
and consequently due to (3.22) that
    
pq ∂ ∂ ∂ ∂ ∂ ∂
g ∇2p,i Rmqjkl = ∇2i,k Ricjl − ∇2i,l Ricjk +g pq
R , Rm , , , .
∂xi ∂xp ∂xq ∂xj ∂xk ∂xl
For the third term of the above equation, by Lemma 2.72 we have
     h  ∂ 
pq ∂ ∂ ∂ ∂ ∂ ∂ pq ∂
g R , Rm , , , =g R , (Rmqjkl )
∂xi ∂xp ∂xq ∂xj ∂xk ∂xl ∂xi ∂xp
   
∂ ∂ ∂ ∂ ∂ ∂
− Rm R , , , ,
∂xi ∂xp ∂xq ∂xj ∂xk ∂xl
   i
∂ ∂ ∂ ∂ ∂ ∂
− · · · − Rm , , ,R ,
∂xq ∂xj ∂xk ∂xi ∂xp ∂xl
=g pq 0 − Ripq
n n n n


Rmnjkl − Ripq Rmqnkl − Ripq Rmqjnl − Ripq Rmqjkn
=g pq g mn Rmpiqm Rmnjkl + Rmpijm Rmqnkl


+ Rmpikm Rmqjnl + Rmpilm Rmqjkn .

Furthermore, observe that

p
g pq g mn Rmpiqm Rmnjkl = g mn Rimp Rmnjkl = g mn Rmnjkl Ricim = g pq Rmpjkl Riciq ,

and by using (3.19) and Lemma 2.77

g pq g mn Rmpijm Rmqnkl = g pq g mn (Rmpimj Rmqlnk − Rmpimj Rmqknl ) = Bijlk − Bijkl ,

and lastly that

g pq g mn (Rmpikm Rmqjnl + Rmpilm Rmqjkn ) = g pq g mn (−Rmpimk Rmqjnl + Rmpiml Rmqjnk ) = −Bikjl + Biljk .

Combining all results yields

∇2p,i Rmqjkl = ∇2i,k Rmjqlp − ∇2i,l Rmjqkp − (Bijkl − Bijlk − Biljk + Bikjl ) + g pq Rmpjkl Riciq ,

and a similar computation for ∇2p,j Rmqikl then yields the desired equality.

37
3.3 Short-time existence and uniqueness

Proposition 3.13 makes it easier to derive an equation for the evolution of curvature under Ricci flow.
From the next result, it becomes apparent that the curvature tensor Rm evolves under Ricci flow similar
to as in the heat equation. In the usual heat equation, one namely has ∂u ∂t = ∆u = div grad u for some
u : Rn × I → R. This will be pivotal for our analysis on existence and uniqueness to solutions of the flow.

Theorem 3.14. Suppose that g(t) is a solution of the Ricci flow, then the Riemannian curvature tensor
Rm evolves by

∂
Rmijkl =∆Rmijkl + 2(Bijkl − Bijlk − Biljk + Bikjl )
∂t (3.24)
− g pq (Rmpjkl Ricqi + Rmipkl Ricqj + Rmijkp Ricql + Rmijpl Ricqk ) .

☼
∂
Proof. Using Proposition 3.9, we can substitute ∂t gij = −2Ricij into equation (3.14):

∂
Rmijkl = −∇2i,l Ricjk − ∇2j,k Ricil + ∇2i,k Ricjl + ∇2j,l Ricik − g pq (Rmijkp Ricql + Rmijpl Ricqk ) .
∂t
Combining this with equation (3.23) with the indices k and l switched, gives us that the Laplacian satisfies

∂
∆Rmijlk = − Rmijkl − 2(Bijlk − Bijkl − Bikjl + Biljk ) − g pq (Rmijkp Ricql + Rmijpl Ricqk )
∂t
+ g pq (Rmpjlk Ricqi + Rmiplk Ricqj ) ,

from which the desired result can be naturally derived by using the symmetries of Lemma 2.77.

To conclude this section, we provide an evolution equation for the Ricci curvature under Ricci flow.
Also here, we see that Ricci curvature evolves as a heat-type equation.

Proposition 3.15. Suppose that g(t) is a solution of the Ricci flow, then the Ricci curvature Ric evolves
by
∂
Ricij = ∆Ricij + 2g pq g rs Rmpijr Ricqs − 2g pq Ricip Ricqj . (3.25)
∂t
♦

Proof. The proof is rather technical and we refer to [18, Corollarly 3.18] for the details.

3.3 Short-time existence and uniqueness

The usual heat equation is the prototype of a strongly parabolic partial differential equation. Although
the Ricci flow resembles the heat equation, we will see that after linearising equation (3.1), the PDE is
only weakly parabolic. Therefore, we cannot use the usual existence theory for parabolic PDE’s. Instead,
we will prove existence and uniqueness to a solution of the the Ricci flow analogously to [7]18
Our aim is first to recognize a parabolic PDE. Here, we follow the approaches of [18, 32, 33] and
[34]. Consider two smooth vector bundles E and F over a Riemannian manifold (M, g) and an evolution
equation u : [0, T ] × M → E given by

∂u
= L(u), u(0) = u0 , (3.26)
∂t
with L : Γ(E) → Γ(F ) a linear differential operator of order k ∈ N. That is, it can be written in the form
X ∂u
L(u) := Lα , (3.27)
∂xα
|α|≤k

18 With that we mean conceptually. Notation wise, we more or less follow [18] and [31].

38
3 RICCI FLOW

with Lα ∈ Hom(E, F ) a bundle homomorphism19 and α a multi-index. Next, we define the total symbol
σL : T ∗ M → Hom(E, F ) of L as the bundle homomorphism
X
σL (p, ξ) := Lα ξ α . (3.28)
|α|≤k

Subsequently, the principal symbol σ̂L of L is the defined as the bundle homomorphism of the highest
term only: X
σ̂L (p, ξ) := Lα ξ α . (3.29)
|α|=k

The principal symbol tells us whether an evolution equation is parabolic.

Definition 3.16. The partial differential equation in (3.26) is strongly parabolic 20 if L is an elliptic
operator. That is, there exists a real number c > 0 such that for all (non-trivial) (p, ξ) ∈ T ∗ M and
u ∈ Γ(E) we have
hσ̂L (p, ξ)u, ui ≥ c|ξ|2 |u|2 . (3.30)
♥
Σ2+ (T ∗ M )



flow equation, we can regard the Ricci curvature as the operator Ric : Γ
In the Ricci →
Γ Σ2 (T ∗ M ) . In the light of (3.26), notice that in the Ricci flow we have E = Σ2+ (T M ) and F = ∗

Σ2 (T ∗ M ). In other words, the Ricci curvature sends the positive definite symmetric metric g(t) to a
symmetric 2-tensor. However, the operator is not a bundle homomorphism, which means that the Ricci
flow is not a linear PDE. Hence, we wish to linearise the equation. Before, we will do this the Ricci
curvature, we move back to general non-linear PDE’s on Riemannian manifolds.
Definition 3.17. Consider a non-linear differential operator P : Γ(E) → Γ(F ). The linearisation of P
at u0 is the linear operator DP : Γ(E) → Γ(F ) given by
d
DP (v) = P (u(t)) , (3.31)
u0 dt t=0

where u(0) = u0 and u0 (0) = v. ♥

Example 3.18. If we consider the non-linear PDE
∂u ∂ 2 u
 
∂u
= P x, t, u, , , u(0) = u0 (3.32)
∂t ∂x ∂x2
∂u
then we say that (3.32) is a parabolic non-linear PDE if its linearisation ∂t = DP (v) is parabolic in
u0
the light of Definition 3.16. ♣
Lemma 3.19. The linearisation of the Ricci tensor Ric : Γ(Σ2+ (T ∗ M )) → Γ(Σ2 (T ∗ M )) at g0 is given by
   
∂ 1 ∂ ∂ ∂ ∂ ∂
DRic g = g kl ∇2k,i gjl + ∇2k,j gil − ∇2k,l gij − ∇2i,j gkl , g ∈ Γ(Σ2+ (T ∗ M )).
g0 ∂t ij 2 ∂t ∂t ∂t ∂t ∂t
(3.33)
♠
Proof. Let ∂
∂t g∈ Γ(Σ2+ (T ∗ M )) and recall Proposition 3.10 to observe that
 
∂ ∂
DRic g = Ric(g(t))ij
g0 ∂t ij ∂t t=0
 
1 kl 2 ∂ 2 ∂ 2 ∂ 2 ∂
= g ∇k,i gjl + ∇k,j gil − ∇k,l gij − ∇i,j gkl .
2 ∂t ∂t ∂t ∂t

19 Linear maps between fibres, see e.g. [21, Page 261].

20 We will often just say parabolic to mean strongly parabolic.

39
3.3 Short-time existence and uniqueness

∂
Notice that the highest-order derivatives of ∂t g are just the partial derivatives. Therefore, the principal
symbol of the Ricci curvature is
   
∂ 1 kl ∂ ∂ ∂ ∂
σ̂Ric (p, ξ) g = g ξk ξi gjl + ξk ξj gil − ξk ξl gij − ξi ξj gkl , (p, ξ) ∈ T ∗ M. (3.34)
∂t ij 2 ∂t ∂t ∂t ∂t

∂ ∂


However, if we define ∂t gij = ξi ξj , then σ̂Ric (p, ξ) ∂t g ij = 0. In other words, the principal symbol of
the Ricci curvature has non-trivial kernel, and does not satisfy the requirements of Definition 3.16. In
this sense, we say that the flow is only weakly parabolic. This is caused solely by the diffeomorphism
invariance of the Ricci curvature (Recall Proposition 2.79). If ϕ : (M, g) → (M f, g̃) is a time-dependent
∗
diffeomorphism such that ϕ g̃(t) = g(t) and g̃(t) is a solution to the Ricci flow, then
 
∂ ∗ ∂
g(t) = ϕ g̃(t)
∂t ∂t
(3.35)
= ϕ∗ (−2Ric(g̃(t)))
g
= −2Ric(g(t)).

We see that then g(t) is also a solution to the Ricci flow, and therefore that (3.1) is invariant under
the full diffeomorphism group, which is infinite dimensional. If g(t) is any stationary solution to the
Ricci flow, that would imply we could acquire another linear independent solution with the pullback of
any diffeomorphism. However, the solution space of non-linear elliptic differential operators on compact
manifolds is only finite dimensional [35, Theorem 1]. Hence, the Ricci flow is not a parabolic equation,
which means we cannot directly use well known existence and uniqueness theory of such equations.
From (3.35), it also becomes apparent that Ricci flow preserves any present symmetries. Since sym-
metric operations can be regarded as isometric diffeomorphisms, the pullback of a solution to the Ricci
flow by a symmetry still has all the initial information of the topology of the manifold in question [36,
Section 6.1]. Together with the fact that metrics with positive curvature are contracted and metrics with
negative curvature are stretched, the Ricci flow literally ‘smoothens’ a manifold’s geometry.
In [7], DeTurck introduced a technique to express the weakly parabolic Ricci flow as a strongly
parabolic equation, now known as DeTurck’s trick. Existence and uniqueness for the Ricci flow can
therefore still be guaranteed. Before moving on to this result, we state a useful property and one fi-
nal definition. Due to [37, Proposition 2.61 (i)] (which is derived from [25, Proposition 12.32] using
∂
Definition 2.57 and compatibility of g with ∇), for X = X i ∂x i we have

LX gij = ∇i Xj + ∇j Xi . (3.36)

To proof uniqueness of short-time Ricci flow solutions for the upcoming theorem, the following defi-
nition will come out useful.
Definition 3.20. Let f : (M, g) × I → (M
f, g̃) be a smooth map between closed Riemannian manifolds
with I ⊂ R. The harmonic map Laplacian is the map
 
∂
∆g,g̃ f = g ij (∇i f∗ ) . (3.37)
∂xj

♥
We say that f is harmonic if ∆g,g̃ f = 0 and call the equation

∂
f∗ := ∆g,g̃ f, f (0) = f0 (3.38)
∂t

the harmonic map heat flow. In [38], Eels and Sampson proved that if (M
f, g̃) has non-positive sectional
21
curvature , then there exists a solution to the maximal harmonic map heat flow, i.e. with {ft | 0 < t <
21 See e.g. [21, Page 250]

40
3 RICCI FLOW

T ≤ ∞}, such that f converges to a harmonic map. This implies the existence of a harmonic map in each
homotopy class22 . Hartman extended this result in [40] to prove uniqueness of the harmonic map in its
homotopy class. The extend to which the non-positively sectional curvature of M f is actually necessary
was examined by Chang, Ding and Ye in [41]. They concluded that the maximal time of existence cannot
be expected to be infinite without non-positive sectional curvature.
In our case, we only need short-time existence. Therefore, we can continue to our primary result on
the Ricci flow.
Theorem 3.21. If (M, g0 ) is a closed Riemannian manifold, there exists an unique solution g(t), defined
for time t ∈ [0, ], to the Ricci flow such that g(0) = g0 for some  > 0. ☼
Proof. The proof is divided into four steps:
1. Prove that a modification of the Ricci flow, the Ricci-DeTurck flow, is parabolic and hence enjoys
short-time existence and uniqueness.
2. Relate the Ricci flow to the Ricci-DeTurck flow.
3. Start with a solution to the Ricci flow and relate it to a unique solution to the Ricci-DeTurck flow
by means of a reparametrisation by harmonic map heat flow.
4. Conclude that the solution to the Ricci flow must be unique.
STEP 1. Let (M, g(t)) be a family of Riemannian manifolds and let g̃ ∈ {g(t) | t ∈ [0, ]} be fixed, for
some  > 0. Define W ∈ X(M ) by
∂ 
e ijk ∂ ,

W = W i i := g jk Γijk − Γ (3.39)
∂x ∂xi
which is well-defined since the difference of two connections is a tensor. Also, consider the one-form
obtained from W with the lowering musical isomorphism:
 
W [ = Wi dxi := gij g kl Γjkl − Γe j dxi .
kl (3.40)

Let us now define the Ricci-DeTurck flow :

∂
gij = −2Ricij + LW gij , g(0) = g0 . (3.41)
∂t
We will show that (3.41) is a quasilinear23 strongly parabolic equation. Existence and uniqueness then
follows from e.g. [43, Theorem 1.1]. Recall equation (3.36) to observe that
LW gij = ∇i Wj + ∇j Wi
     
= ∇i gjk g lm Γklm − Γ
e klm + ∇j gik g lm Γklm − Γ
e klm .

By using normal coordinates about a point p ∈ M , the covariant derivatives can be interpreted as partial
derivatives, such that linearising the above expression yields
∂ ∂
DLW gij = (LW g(t))ij
g0 ∂t ∂t t=0
  
1 pq kl ∂ ∂ ∂
= gjk g ∇i g ∇p gql + ∇q gpl − ∇l gpq
2 ∂t ∂t ∂t
  
1 ∂ ∂ ∂
+ gik g pq ∇j g kl ∇p gql + ∇q gpl − ∇l gpq
2 ∂t ∂t ∂t
∂
+ (lower-order derivatives of g)
 ∂t 
kl 2 ∂ 2 ∂ 2 ∂ ∂
=g ∇i,k gjl + ∇j,k gil − ∇i,j gkl + (lower-order derivatives of g),
∂t ∂t ∂t ∂t
22 Homotopies are studied in the branch of algebraic topology. A homotopy class is an equivalence class of continuous

functions that can be continuously deformed into one another. For more information, we refer to [39].
23 That is, it is linear with respect to all the highest order derivatives of the unknown function (see e.g. [42, Section 4.4]).

41
3.3 Short-time existence and uniqueness

where the second step follows from Proposition 3.7. Recall Lemma 3.19 to observe that
 
∂ ∂ ∂
D[−2Ric + LW ] g = g kl ∇2k,l gij + (lower-order derivatives of g),
g0 ∂t ij ∂t ∂t

where we used the symmetry of the metric to rearrange the indices of the first two of equation (3.33).
Now, notice that for the principal symbol we get
 
∂ ∂
σ̂−2Ric+LW (p, ξ) g = g kl ξk ξl gij , (p, ξ) ∈ T ∗ (M ),
∂t ij ∂t

which clearly has trivial kernel. We see that the Ricci-DeTurck flow in (3.41) satisfies the requirements
of Definition 3.16 and hence enjoys short-time existence and uniqueness.

STEP 2. Consider a solution g(t) to the Ricci-DeTurck flow and a time dependent vector field W (t)
defined by (3.39). Since there exists a solution to the Ricci-DeTurck flow, this one-parameter family of
vector fields exists for t ∈ [0, ]. Therefore, by e.g. [22, Theorem 3.3.5] or [31, Lemma 3.15], there exists
a one-parameter family of diffeomorphism ϕt : M → M constructed by solving the ODE
∂
ϕt (p) = −Wϕt (p) (t), ϕ0 = idM ,
∂t
which we recognize as the flow of −W . We claim now that ḡ(t) is a solution to the Ricci flow obtained
by pulling back g(t) with the diffeomorphism ϕt :

ḡ(t) := ϕ∗t g(t), t ∈ [0, ].

Notice that since ϕ0 = idM , we have ḡ(0) = g(0) = g0 . Moreover, we have that
∂ ∂
ḡ = ϕ∗t g(t)
∂t ∂t
∂
ϕ∗ g(t + s)


=
∂s s=0 t+s 
∗ ∂ ∂
ϕ∗t+s g(t)


= ϕt g(t) +
∂t ∂s s=0
∂ h
∗ ∗ i
= ϕ∗t −2Ric(g(t)) + LW (t) g(t) + ϕ−1


t ◦ ϕt+s ϕt g(t) , (3.42)
∂s s=0
∗
where we used that ϕ∗t+s = ϕ−1


t ◦ ϕt+s ◦ ϕt . Moreover, recall Definition 2.35 to observe that

∂ h
∗ i
ϕ−1
t ◦ ϕt+s ϕ∗t g(t) = L ϕ∗t g(t). (3.43)
∂ s=0 ∂
∂s (ϕ−1
t ◦ϕs+t )
s=0

In addition, notice that

   
∂ ∂ ∂
ϕ−1 −1
= ϕ−1 = − ϕ−1 W (t) = −ϕ∗t W (t) (3.44)





t ◦ ϕs+t = ϕt ∗
ϕs+t t ∗
ϕt t ∗
∂s s=0 ∂s s=0 ∂t
Combining equations (3.43) and (3.44) with (3.42) then yields
∂
ḡ = −2ϕ∗t Ric(g(t)) + ϕ∗t LW (t) g(t) − Lϕ∗t W (t) ϕ∗t g(t)
∂t
¯
= −2Ric(ḡ(t)), (3.45)

which proves the claim and concludes the existence part of the proof24 .

24 Notice that the final step follows from the fact that the Lie derivate of g is invariant under diffeomorphisms, see for

example [31, Section 2.2].

42
3 RICCI FLOW

STEP 3. Let (M, ḡ(t)) be a family of Riemannian manifolds such that ḡ(t) is a solution to the Ricci flow.
Also, let (M
f, g̃) be a Riemannian manifold with fixed metric g̃ and corresponding Levi-Civita connection
∇. Define ϕ : (M, ḡ(t)) × [0, ] → (M
e f, g̃) is a diffeomorphism25 .
f, g̃) by (3.38) such that ϕ0 : (M, g0 ) → (M
Recall (2.44) and (2.45) to observe that if (x ) denote local coordinates about p ∈ M and (y α ) denote
i

local coordinates about ϕt (p) ∈ M f [45, Section 1.2.2], then

∂
(ϕt )∗ = ∆ḡ,g̃ ϕt
∂t   
∂
= ḡ ij ∇i (ϕt )∗
∂xj
    
∂ ¯i ∂
= ḡ ij ∇e ∂
(ϕt )∗ i (ϕ t )∗ − (ϕ t ) ∗ ∇
∂x ∂xj ∂xj
  α  
∂ϕt ∂ k ∂
= ḡ ij ∇e ∂
(ϕt )∗ i − (ϕ t ) ∗ Γ̄ij
∂x ∂xj ∂y α ∂xk
!
β
ij ∂ 2 ϕα
t ∂ ∂ϕα
t ∂ϕt e γ ∂ k ∂ϕt
α
∂
= ḡ i j α
+ j i
Γ αβ γ
− Γ̄ ij , (3.46)
∂x ∂x ∂y ∂x ∂x ∂y ∂x ∂y α
k

where we used (2.33) in the third and fourth equality. Let us now define g(t) = (ϕ−1 ∗
t ) ḡ(t) on M . We
f
claim that this is a unique solution to the Ricci-DeTurck flow. The uniqueness follows from the fact that
the harmonic map heat flow has a unique solution. Moreover, notice that

∂ ∂
(ϕ−1 ∗


g= t ) ḡ(t)
∂t ∂t
∂ 
−1 ∗


= ϕt+s ḡ(t + s)
∂s s=0  
−1 ∗ ∂ ∂ 
∗ 
ϕ−1


= ϕt ḡ(t) + t+s ḡ(t)
∂t ∂s s=0
∂ h
∗ −1
∗ i
= −2Ric(g(t)) + ϕt ◦ ϕ−1
t+s ϕt ḡ(t)
∂s s=0

∗
= −2Ric(g(t)) + L(ϕt )∗ ∂ ϕ−1 ϕ−1 t ḡ(t)
∂t t

= −2Ric(g(t)) + L(ϕt )∗ ∂ −1 g(t), (3.47)

∂t ϕt

where we use that the Ricci flow is invariant under diffeomorphisms in the fourth step. Also, note that
the fourth and fifth step follow from similar reasoning as to Step 2. It now rests for us to show that the
∂ −1
Lie derivative term in (3.47) is equal to the one in the Ricci-DeTurck flow, i.e. that (ϕt )∗ ∂t ϕt = −W
from (3.39). For that, choose local coordinates (x ) about ϕt (q) ∈ M and (z ) = (x ) ◦ ϕ−1
i −1 i i
t about
α −1
q ∈ M . In addition, let (y ) be local coordinates fixed about ϕt ◦ ϕt (q) ∈ M , then observe that
f f

y ◦ z −1 = y ◦ ϕt ◦ x−1
= y ◦ (ϕt ◦ ϕ−1
t )◦z
−1
.

−1 α
I.e., we see that ḡij = gij and therefore that Γ̄kij = Γkij such that ϕα
t = (ϕt ◦ ϕt ) . This means that

25 In fact, for all t in the domain [0, ] we have that ϕ is a diffeomorphism due to a kind of elliptic regularity. The proof

of this involves technicalities which do not lie within the scope of this research. In this context, it is however dealt with in
[18, Section 3.2] and on general non-linear partial differential equations in e.g., [44, Section 8.3].

43
3.3 Short-time existence and uniqueness

(3.46) gives us

∂
(ϕt )∗ = (∆ḡ,g̃ ϕt )ϕ−1 (q)
∂t ϕ−1
t (q)
t
!
β
∂ 2 ϕα
t ∂ ∂ϕα
t ∂ϕt e γ ∂ ∂ϕα ∂
= ḡ ij
i j α
+ j i
Γαβ γ − Γ̄kij kt
∂x ∂x ∂y ∂x ∂x ∂y ∂x ∂y α
ϕ−1
t (q)
2
(ϕt ◦ ϕ−1 α
∂(ϕt ◦ ϕ−1 α
∂(ϕt ◦ ϕ−1 β
∂(ϕt ◦ ϕ−1 α
 
ij ∂ t ) ∂ t ) t ) eγ ∂ t ) ∂
=g + Γαβ γ − Γkij
∂xi ∂xj ∂y α ∂xj ∂xi ∂y ∂xk ∂y α q

ϕ−1


= ∆g,g̃ (ϕt ◦ t ) q
∂
= (ϕt ◦ ϕ−1
t )∗ . (3.48)
∂t q
In other words, the harmonic map Laplacian is invariant under change of diffeomorphisms. In our specific
case, since ϕt ◦ ϕ−1
t (q) = q, we simply have

∂ −1 
e ijk − Γijk ∂

(ϕt )∗ ϕt = g jk Γ
∂t ∂xi
= −W, (3.49)

which proofs the claim that g(t) is a unique solution to the Ricci-DeTurck flow.

STEP 4. Now, suppose that both ḡ1 (t) and ḡ2 (t) are two solutions to the Ricci flow with ḡ1 (0) = ḡ2 (0).
Let Mf = M and ϕ0 = idM . Using Step 3, we could provide one with solutions to the Ricci-DeTurck
flow g1 (t) and g2 (t) which satisfy g1 (0) = ḡ1 (0) = ḡ2 (0) = g2 (0), since ϕ0 = idM . Consequently, since
solutions to the Ricci-DeTurck flow are unique, we have that g1 (t) = g2 (t) for all t ∈ [0, ]. Therefore,
the time dependent vector field W is also the same for the two solutions. The unique solutions to the
harmonic map heat flow ϕ1,t and ϕ2,t together with (3.49) show us that

∂
ϕi,t (p) = −Wϕi,t (p) (t), ϕi,0 = idM (3.50)
∂t
and hence that ϕ1,t (p) = ϕ1,t (p). We see that therefore ḡ1 = ϕ∗1,t g1 = ϕ∗2,t g2 = ḡ2 , which proves
uniqueness to the Ricci flow.

44
4 ALMOST-RIEMANNIAN GEOMETRY & RICCI FLOW

4 Almost-Riemannian geometry & Ricci flow

In this research, we are particularly interested in what happens when Ricci flow is practised on an
almost-Riemannian structure. This type of structure is an example of a sub-Riemannian structure, just
like Riemannian manifolds. The chapter provides a brief introduction of sub-Riemannian structures
and almost-Riemannian structures, after which we examine the possibilities of evolving Ricci flow on
two-dimensional almost-Riemannian manifolds.

4.1 Definitions & properties

This first part of this section is based upon [46, 47, 27] and [48]. We always assume that M is a n-
dimensional smooth manifold, i.e. not necessarily a Riemannian manifold. At a certain point, we will put
our attention only on two-dimensional smooth manifolds with an almost-Riemannian structure. Then,
we will use results of [49, 17, 50] and [51].
To define a sub-Riemannian structure, we firstly have to broaden our knowledge on Lie brackets
(recall Definition 2.22) and smooth vector fields. The set of all smooth vector fields over M has both the
structure of a real vector space, and of a C ∞ (M )-module26 [22, Proposition 3.1.5]. When equipped with
the Lie bracket, it turns out that X(M ) is also a Lie algebra over R [27, Proposition 3.9].
Definition 4.1. A Lie algebra over a field K is a pair (V, [·, ·]), where V is a vector space over K and
[·, ·] : V × V → V is the Lie bracket that satisfies the following properties:
(i) Bilinearity over K. I.e., for all a, b ∈ K and x, y, z ∈ V , [ax+by, y] = a[x, z]+b[y, z] and [z, ax+by] =
a[z, x] + b[z, y];
(ii) Antisymmetry. I.e., for all x, y ∈ V , [x, y] = −[y, z];
(iii) The Jacobi identity. I.e., for all x, y, z ∈ V , [x, [y, z]] + [z, [x, y]] + [y, [z, x]] = 0.
♥
We also need to define such algebra’s for subsets of X(M ).
Definition 4.2. Let F ⊆ X(M ) be a family of smooth vector fields. The Lie algebra generated by F is
smallest Lie subalgebra of X(M ) containing F. That is,
Lie F := span{[X1 , . . . , [Xj−1 , Xj ]] | Xi ∈ F, j ∈ N}. (4.1)
♥
If for each p ∈ M the evaluation at p of Lie F equals Tp M , then we say that F satisfies the Hörmander
condition. Mathematically, that is
Liep F = {X(p) | X ∈ Lie F} = Tp M. (4.2)
Definition 4.3. Let M be connected smooth n-dimensional manifold. A sub-Riemannian structure on
M is a pair (U, f ) for which the following properties hold:
(i) U is a Euclidean bundle with base space M . An Euclidean bundle is a vector bundle whose fibres
Up , for all p ∈ M , are equipped with a smoothly varying inner product h·, ·ip with respect to p;
(ii) The map f : U → T M is a smooth map that is a morphism27 of vector bundles such that its
restriction to any fibre of U is linear and the following diagram commutes:
f
U TM
π (4.3)
πU
M
26 We
can also say that X(M ) is a module over the ring C ∞ (M ). An algebra can be constructed from a module by
equipping it with a bilinear operation. These objects are algebraic structures studied in abstract algebra. See e.g. [52].
27 Morphisms are studied in category theory, the abstract algebra of functions. Morphisms are also called arrows, and are

structure preserving maps between two objects in a category [53, Section 1.1].

45
4.1 Definitions & properties

where πU : U → M and π : T M → M denote the projections;

(iii) The collection of smooth vector fields D = {f ◦ σ | σ : M → U is a smooth section} satisfies the
Hörmander condition.
♥
We say that the triple (M, U, f ) is a sub-Riemannian manifold.
Example 4.4. If we take U = T M such that f : T M → T M and f (Up ) = Tp M , then (M, U, f ) is a
Riemannian manifold. ♣
We will soon see that almost-Riemannian manifolds are also specific types of sub-Riemannian mani-
folds. More elaborately, they are the prototypes of rank-varying sub-Riemannian manifolds. To see what
that means, consider the following definition.
Definition 4.5. Let (M, U, f ) be a sub-Riemannian manifold. Its distribution is the family of subspaces
{D(p)}p∈M , where
D(p) := {f ◦ σ(p) | f ◦ σ ∈ D} = f (Up ) ⊆ Tp M. (4.4)
♥
The bundle rank of the sub-Riemannian structure is k = rank(U), and r(p) := dim Dp is the rank of
the sub-Riemannian structure at p ∈ M . So, if r(p) is constant, then (U, f ) has constant rank. If this is
not the case, then (U, f ) is called rank-varying. For every p ∈ M , the rank of a sub-Riemannian manifold
(M, U, f ) hence satisfies
r(p) ≤ min{k, n}. (4.5)
In order to define distances between two points, we need the concept of admissible curves and their
lengths.
Definition 4.6. A Lipschitz curve γ : [0, T ] → M is admissible for a sub-Riemannian structure if there
exists a measurable essentially bounded function28 t →
7 u(t), called a control, such that

γ̇(t) = f (γ(t), u(t)), for a.e.29 t ∈ [0, T ]. (4.6)

Moreover, we define the sub-Riemannian length of an admissible curve γ as30

Z T
`(γ) := kγ̇(t)k dt. (4.7)
0

♥
Having obtained a notion of length, we can define distances.
Definition 4.7. The sub-Riemannian distance between two points p, q of a sub-Riemannian manifold
M is
d(p, q) = inf{`(γ) | γ : [0, T ] → M is admissible, γ(0) = p, γ(T ) = q}. (4.8)
♥
In fact, if M is a sub-Riemannian manifold, by the Chow-Rashevskii theorem we have that (M, d) is a
metric space and the topology induced on M is equivalent to the regular manifold topology [46, Theorem
3.31]. This ensures that there exists an admissible curve between any two points in M . Moreover, (M, d)
is complete when there exists an  > 0 such that the closed ball B̄r (p) is compact for every p ∈ M [46,
Proposition 3.47]. One could also define geodesics in this setting. As it turns out, these are characterized
as normal Pontryagin extremal trajectories. For more detail, see [46, Section 4.7].
28 That is, almost equal to a bounded function. This is a concept from measure theory, see for example [54, Section 2.11]
29 That is, almost everywhere. Again, a term from measure theory. It means that only in a subspace of measure zero, a
certain property holds everywhere.
30 See [46, Definition 3.8] for a precise definition of the sub-Riemannian norm.

46
4 ALMOST-RIEMANNIAN GEOMETRY & RICCI FLOW

Definition 4.8. Let (U, f ) and (U, e f˜) be two sub-Riemannian structures on M . The structures are
equivalent as distributions if the following conditions hold:

(i) There exists a Euclidean bundle V and two surjective bundle morphisms p : V → U and p̃ : V → U
e
such that the following diagram commutes;

U
p f

V TM (4.9)

p̃ f˜
U
e

(ii) The projections p, p̃ are compatible with the scalar product. That is, we have31

|u| = min{|v|, p(v) = u}, ∀u ∈ U,

|ũ| = min{|v|, p̃(v) = ũ}, ∀ũ ∈ U.
e

♥
e f˜) are
If (i) and (ii) of Definition 4.8 are satisfied, we say that the two structures (U, f ) and (U,
equivalent (as sub-Riemannian structures).

Definition 4.9. Let (M, U, f ) be a sub-Riemannian manifold. The minimal bundle rank of M is defined
as the infimum of bundle ranks that induce equivalent structures on M . The local minimal bundle rank
at p ∈ M is the minimal bundle rank of the structure restricted to a sufficiently small neighbourhood
Op 3 p. ♥

Almost-Riemannian structures are sub-Riemannian structures such that its local minimum bundle
rank at every point is equal to the dimension of the manifold. For a given sub-Riemannian (U, f ) with a
constant local minimum bundle rank k, we can always find a sub-Riemannian structure (U, e f˜) equivalent
to (U, f ) such that rank U
e = k [46]. Therefore, we can define almost-Riemannian structures as follows:

Definition 4.10. Let M be connected smooth n-dimensional manifold. A n-dimensional almost-Riemannian

structure (n-ARS) on M is a pair (U, f ) for which the following properties hold:

(i) U is a rank n Euclidean bundle with base space M . A Euclidean bundle is a vector bundle whose
fibres Up , for all p ∈ M , are equipped with a smoothly varying inner product h·, ·ip with respect to
p;

(ii) The map f : U → T M is a smooth map that is a morphism of vector bundles such that its
restriction to any fibre of U is linear and the following diagram commutes:

f
U TM
π (4.10)
πU
M

where πU : U → M and π : T M → M denote the projections;

(iii) The collection of smooth vector fields D = {f ◦ σ | σ : M → U is a smooth section} satisfies the
Hörmander condition.

♥
31 Here,
p
we have that |u| = hu, uip for u ∈ Up .

47
4.1 Definitions & properties

Moreover, we introduce the step of the distribution. As we will see, for some points p ∈ M with a
n-ARS, one only needs the span of n vector fields to span the tangent space. For other points, we will
indeed need them to obey the Hörmander condition to span Tp M .
Definition 4.11. The step of the distribution at p ∈ M is the minimal s ∈ N, with s ≥ 1, such that
Ds (p) = Tp M , where D1 := D and Di+1 := Di + [D1 , Di ], for i ≥ 1. ♥
Let us now discuss some of the properties of almost-Riemannian structures.
Definition 4.12. Let Ω ⊂ M . An orthonormal frame on Ω for a n-ARS is the set of vector fields
{X1 , X2 , . . . , Xn } = {f ◦ σ1 , f ◦ σ2 , . . . , f ◦ σn }, (4.11)
where {σ1 , σ2 , . . . , σn } is an orthonormal frame on a local trivialization Ω × Rn of U with respect to h·, ·ip
for all p ∈ Ω. ♥
If such an orthonormal frame exists on M , then we say that the n-ARS is free.
Definition 4.13. The singular set (or singular locus) Z on a n-ARS of M is the set of points p ∈ M
such that r(p) < n. We call these points singular points. ♥
The singular set is what differs between an n-ARS and a Riemannian structure on M . Formally, we
have that a n-ARS is a Riemannian structure on M \ Z [27, Theorem 3.20]. Hence, we refer to p ∈ M \ Z
as Riemannian points.

4.1.1 Two-dimensional almost-Riemannian structures

From now one, we restrict ourselves to 2-ARS and hence a smooth connected two-dimensional manifold
M . Most of the research on almost-Riemannian structures has been on this particular case, which makes
it a suitable choice to see what happens when Ricci flow is evaluated on a surface endowed with a 2-ARS.
For example, while the metric and curvature explode when approaching the singular set, geodesics can
pass through it. In particular, all geodesic are smooth and they coincide with the set of non-trivial normal
Pontryagin extremal trajectories [46, Corollary 9.25, Proposition 9.26]. Specifically, an admissible curve
γ : [0, T ] → M is a geodesic for a 2-ARS if for every sufficiently small non-trivial interval [t1 , t2 ] ⊂ [0, T ],
γ [t1 ,t2 ] is a minimizer of `(·) from (4.7). If M is compact, then any two points of M are connected by a
minimizing geodesic as a consequence of the Chow-Rashevskii theorem [51, Section 2.2].
Before moving on, recall Definition 2.8 and Remark 2.18 for the following definition
Definition 4.14. A 2-ARS (U, f ) over M is said to be oriented if U is oriented. It is said to be fully
oriented if both M and U are oriented. ♥
Remark 4.15. It is in fact possible to define a non-orientable almost-Riemannian structure on orientable
manifolds, and vice-versa [17, Chapter 1]. F
Notice that in the 2-ARS case, we have [55, Section 1.2]
Z = {p ∈ M | dim(D(p)) = 1}. (4.12)
As shown in [46, Section 9.4.1], a 2-ARS on a smooth connected two-dimensional manifold M gener-
ically satisfies the following properties:
(i) The singular set Z is an embedded one-dimensional submanifold of M ;
(ii) The points p ∈ M at which dim(D2 (p)) = 1 are isolated;
(iii) For every p ∈ M we have D3 (p) = Tp M .
We refer to the above three conditions as H0. Assuming that H0 is true, one

| could show that, generically,

|
a (local) orthonormal frame is given by the vector fields X1 (x, y) = 1 0 and X2 (x, y) = 0 f (x, y)
on R2 , with f (x, y) a smooth function [17, Theorem 16]. The following proposition enables us to express
the local behaviour of g and the scalar curvature32 S with these two vector fields [27, Theorem 3.21].
32 In two dimensions, the scalar curvature equals twice the Gaussian curvature [21, Corollary 8.28].

48
4 ALMOST-RIEMANNIAN GEOMETRY & RICCI FLOW

Proposition 4.16. Let (x, y) denote local coordinates on Ω ⊂ M with an orthonormal frame for a
2-ARS on Ω of the form    
1 0
X1 (x, y) = , X2 (x, y) = , (4.13)
0 f (x, y)
with f : Ω → R a smooth function. The singular set is then given by Z = {(x, y) ∈ Ω | f (x, y) = 0} and
on Ω ∩ (M \ Z) we have the following expressions for the metric g and the scalar curvature S:
 2 2
∂f
1 2 −2 ∂x + f ∂∂xf2
2
g = dx + 2 dy , S= (4.14)
f 2f 2
♦
Proof. Observe that X1 (p) and X2 (p) are linearly independent if and only if f (p) = 0. I.e., in that case
we have dim(D(p)) = 1 such that by (4.12) we indeed have Z = {(x, y) ∈ Ω | f (x, y) = 0}. Recall (2.22)
to note that {X1 , X2 } is an orthonormal frame on the Riemannian points if

gp (Xi (p), Xj (p)) = δij .

Clearly, we then must have that g = dx2 + f12 dy 2 . For the scalar curvature, one could first compute
the Christoffel symbols and, subsequently, the coefficients of the Riemann curvature endomorphism. The
expression for S in (4.14) can then be obtained by applying multiple consecutive traces (that is, apply
equations (2.61), (2.66) and (2.67)).
Remark 4.17. This expression for the scalar curvature is especially relevant, since we will analyse the Ricci
flow on almost-Riemannian surfaces in the next section and Ric = Kg = S2 g in these dimensions33 . F
Also under the assumption that HO is true, we can distinguish between three kinds of points for a
2-ARS on M [17, Definition 19].
Definition 4.18. Consider a 2-ARS on M such that H0 holds. We call a point p ∈ M a
(i) Riemannian point if D(p) = Tp M ;
(ii) Grushin point if dim(D(p)) = 1 and D2 (p) = Tp M ;
(iii) tangency point if dim(D(p)) = dim(D2 (p)) = 1 and D3 (p) = Tp M .
♥
Equivalently, if p is a Grushin point, then the distribution D(p) is transversal to T

|p Z. To see this, let
us adopt the assumptions of Proposition 4.16 to notice that then D(p) = span{ 1 0 } and (by working
n | o
out [X1 , X2 ]) that Tp Z = span − ∂f ∂y (p) ∂f
∂x (p) , which is transversal since ∂f
∂x (p) 6= 0 by definition
of a Grushin point. Similarly, If p is a tangency point, then D(p) coincides withnTp Z, i.e. it is o
tangent

| |
∂f
to Z. In that case, we again have D(p) = span{ 1 0 }, but now Tp Z = span − ∂y (p) 0 , since
D2 (p) 6= Tp M for a tangency point [46, Proposition 9.36].

Example 4.19. The Grushin sphere is an example of a free almost-Riemannian structure on S2 with
(U, f ) given by U = S2 × R2 and f (θ, ϕ, u1 , u2 ) = (θ, ϕ, u1 , u2 tan θ) (see also [49, Table 1]). In contrast
to Example 2.47, let us now parametrize S2 by
n π π o
x = sin θ cos ϕ, y = sin θ sin ϕ, z = cos θ, (θ, ϕ) | − < θ < , 0 ≤ ϕ ≤ 2π .
2 2
If we set D = {X1 , X2 } with
   
1 0
X1 (θ, ϕ) = , X2 (θ, ϕ) = ,
0 tan θ
33 For a proof, we refer to Lemma A.3.

49
4.2 A discussion on Ricci flow on two-dimensional almost-Riemannian manifolds

then X1 and X2 span Tp S2 except at points with θ = 0 (See also Figure 4). By Proposition 2.23, we have
 
∂ tan θ ∂ 1 ∂ 0
[X1 , X2 ](θ, ϕ) = = = 1 ,
∂θ ∂ϕ p cos2 θ ∂ϕ p cos2 θ

such that D2 (p) = Tp M for all p ∈ S2 . I.e., D satisfies the Hörmander condition. In particular, we have
Z = {(θ, ϕ) | θ = 0} and the Riemannian metric on S2 \ Z given by

1
g = dθ2 + dϕ2 ,
tan2 θ
due to Proposition 4.16. In addition, for the scalar curvature we have
1

2 2
−2 cos2 θ + 2 tan θ
cos2 θ sin2 θ − 1 1
S= 2 = 2 2 =− 2 .
2 tan θ cos θ sin θ sin θ
♣

Figure 4: The unit sphere S2 with a 2-ARS.

4.2 A discussion on Ricci flow on two-dimensional almost-Riemannian man-

ifolds
In this section, we investigate the possibilities of evolving the Ricci flow on a complete, connected smooth
two-dimensional manifold M endowed with a 2-ARS. One would suppose that we can best focus on
M \ Z, since there the structure is Riemannian. From (4.14), it becomes apparent that, generically, both
the metric and curvature explode when a point p approaches Z. Hence, it seems prudent to consider
the space M := {p ∈ M | d(p, Z) > }, where d(·, ·) denotes the almost-Riemannian distance34 . If we
additionally require M to be compact and oriented, we have the following two results from [17, Lemma
24, Lemma 25]:
34 This is the same as the sub-Riemannian distance from Definition 4.7.

50
4 ALMOST-RIEMANNIAN GEOMETRY & RICCI FLOW

Lemma 4.20. Every compact orientable two-dimensional manifold admits a free 2-ARS satisfying H0
and having no tangency points. ♠

Lemma 4.21. Let M be a compact and oriented. For a 2-ARS on M satisfying H0, the singular set Z
is the union of finitely many curves diffeomorphic to S1 . Moreover, there exists and 0 > 0, such that, for
every 0 <  < 0 , the set M \ M is homeomorphic to Z × [0, 1]. Under the additional assumption that M
contains no tangency points, 0 can be taken in such a way that ∂M is smooth for every 0 <  < 0 . ♠

From now on, we therefore assume M to also be compact and oriented, and choose the free 2-ARS
satisfying H0 on M without tangency point such that we can take 0 > 0, such that ∂M is smooth for
every 0 <  < 0 . Given an orientation on M , let us also consider M + as the subset of M \ Z having
the same orientation as M , and M − as the subset of M \ Z having the opposite orientation35 . We are
interested in the topology of M± := M ± ∩ M . Then, we can examine the evolution of Ricci flow on
(lim→0 M± , g0 ). First of all, notice that M is compact with smooth boundary ∂M . To add to this, we
have that M is non-complete. Recall that when M is compact, any two points of M are connected by
a minimizing geodesic due to the Chow-Rashevskii theorem. If we remove the singular set Z, a geodesic
from a point p ∈ M+ to q ∈ M− can never be achieved. Hence, we are dealing with two non-complete
compact surfaces M± with smooth boundaries ∂M± and possibly multiple connected components.
In the following two paragraphs, we discuss various results of Ricci flow on surfaces with boundary
and non-complete surfaces respectively. Two of those were already mentioned in the introduction of this
thesis. In order to analyse whether they may apply to our case, we ignore non-completeness in the first
paragraph, and ignore the boundary in the second paragraph.

4.2.1 Ricci flow on surfaces with boundary

Theorem 3.21 only applies to compact Riemannian manifolds without boundary, i.e. closed manifolds.
The surfaces that we are dealing with now, have significant different topological characteristics. On a
surface M with boundary, the aim is to impose boundary conditions in such a way, that the Ricci-DeTurck
flow is a parabolic boundary value problem. The first result for Ricci flow on manifolds with boundary is
from [56]. There, the author proofs short-time existence to the flow on compact manifolds with umbilic
boundary and, as a specific case, totally geodesic boundary36 . This result was recently improved in [10],
where the authors do not necessarily start with umbilic boundary, but will become so in positive time.
With umbilic boundary, we mean that the identity IIαβ = λgαβ , with λ a constant and II the second
fundamental form37 , holds on ∂M [56, Definition 1.3]. In [57], similar results have been established for
manifolds with boundary of merely constant mean curvature. For an arbitrary initial metric on a compact
manifold with boundary, short-time existence and uniqueness to the Ricci flow is proved by prescribing
the mean curvature and conformal class in [58]. Existence and uniqueness for normalized Ricci flow (or
related curvature flows) on surfaces with boundary with various curvature constraints and assumptions
are shown in [59, 60, 61] and [62]. Finally, the un-normalized Ricci flow on surfaces with boundary is
briefly dealt with in [63]. Although [56, 10, 57, 58] provide results for manifolds with boundary, it remains
to be shown if they apply to two-dimensional manifolds endowed with a 2-ARS. Since [59, 60, 61, 62, 63]
deal specifically with surfaces, their results are most likely to naturally apply to our case. Therefore, we
will provide a brief analysis on whether this can be achieved. As mentioned before, for the moment we
ignore the non-completeness of M± .
In [59] and [60] the authors require positive scalar curvature. Recall (4.14) to observe that we then
must find a smooth function f : Ω → R that satisfies
2
∂2f

∂f
−2 +f > 0.
∂x ∂x2

35 To be precise, the orientations for M ± are defined by the volume form dAs associated with M \ Z.
36 See Section A.
37 Also, see Definition A.1 in Section A.

51
4.2 A discussion on Ricci flow on two-dimensional almost-Riemannian manifolds

Let us assume that f (x, y) = xα , for some α ∈ R. Then

2
∂2f

∂f
0 < −2 +f = −2α2 x2(α−1) + α(α − 1)x2α−2
∂x ∂x2
= −α2 x2(α−1) − αx2(α−1) =⇒ −1 < α < 0.

But, if we choose −1 < α < 0, then points lie in Z only when x → ∞. Although this choice of f does
not cover all possibilities, it certainly illustrates the difficulties of finding a suitable smooth function such
that we obtain positive scalar curvature on M \ Z.
Hence, let us consider the results from [62]. The author obtains two results: one for an initial metric
with vanishing geodesic curvature on ∂M and constant Gaussian curvature in M , and one for an initial
metric with vanishing Gaussian curvature in M and constant geodesic curvature on ∂M . The latter
situation gives rise to similar difficulties as before. For the former note that (4.14) implies that f : Ω → R
is constant if it satisfies the differential equation
 2
∂f ∂2f
−2 + f 2 = f 2 C, C ∈ R.
∂x ∂x
However, solving this equation does not guarantee that X1 and X2 satisfy the Hörmander condition.
In [61], the author proves existence and uniqueness for all times to the evolution equation

∂ 2
 ∂t g(t) = − α (K(t) − αλ)g(t),
 in M × (0, T ],
∂ 2
∂t g(t) = − β (κ(t) − βλ)g(t), on ∂M × (0, T ], (4.15)

g(0) = g0 , in M,


with α, β ∈ R and by letting A denote the area of the surface and r the line along ∂M ,
2πχ(M )
λ= R R .
α M dA + β ∂M dr

Using the Gauss-Bonnet formula, the author derives that λ lies between χ(M ) and 2χ(M ). The solu-
tion converges exponentially to a metric with constant Gaussian curvature in M̊ and constant geodesic
curvature on ∂M R . The regular
R Gauss-Bonnet theorem on a compact oriented two-dimensional manifold
M asserts that M KdA + ∂M κdr = 2πχ(M ). In [17, Section 5.2], the authors derive a Gauss-Bonnet
like formula for a generic oriented 2-ARS without tangency points. The Gauss-Bonnet theorem can be
applied to the both of the surfaces (M± , g0 ). Hence, it seems promising to consider the evolution equation
(4.15) on both surfaces (M± , g0 ) individually and argue whether the same results can be acquired for the
similar normalized Ricci flow.
Lastly, let us discuss the results of [63]. As mentioned before, [59, 60, 61, 62] all apply to the
normalized Ricci flow. The un-normalized Ricci flow on a surface with boundary that is dealt with in
[63] is given by the following equations38 :

∂
 ∂t g(t) = −S(t)g(t),
 in M × (0, T ],
κ(·, t) = ψ(·, t), on ∂M × (0, T ], (4.16)

g(0) = g0 , in M,


with ψ a real-valued smooth function on ∂M × [0, ∞). One might argue that (4.16) can be applied to the
surfaces (M± , g0 ) to obtain short-time existence and uniqueness results. For the moment we will totally
ignore the presence of the singular set Z. Because, after a deformation of the metrics of M± without a
fixed volume constraint, it is unlikely that we can still consider whole of M as smooth connected manifold.
As mentioned in Remark 3.2, on a surface without boundary, the deformation is conformal such that a
solution is given by g(t) = ev(t) g0 . This result can be extended to a surface with boundary, such that
38 Recall that on a surface the Ricci curvature equals Ric = Kg, and K = 1 S. As opposed to the previous chapter, we
2
now use T to denote the final time to avoid confusion with the epsilon of M .

52
4 ALMOST-RIEMANNIAN GEOMETRY & RICCI FLOW

we obtain a non-linear partial differential equation with Robin boundary conditions and initial condition
v(0) = 1. Short-time existence can indeed be obtained for a smooth g0 with the solution g(t) in the
parabolic Hölder space39 C 2,γ that is smooth except at the corner40 [63, Theorem 2.2]. In conclusion,
when ignoring the non-completeness of M± and singular set Z, it is likely that we can indeed evolve
(4.16) on (M± , g0 ) and obtain short-time existence and uniqueness results. However, the smoothness at
the corner must be taken great care with especially when  → 0.

4.2.2 Ricci flow on non-complete surfaces

On non-complete surfaces, the existence of Ricci flow has been shown in [11] and [64]. Let us start with
stating part of the main result of [11, Theorem 1.3]:

Theorem 4.22. Let (M, g0 ) be a smooth Riemannian surface which need not be complete, and could
have unbounded curvature. Depending on the conformal type, we define T ∈ (0, ∞] by

if (M, g0 ) ∼

1
 8π volg0 M,
 = S2 ,
1
T := 4π volg0 M, if (M, g0 ) ∼
= C or (M, g0 ) ∼
= RP2 , (4.17)

∞, otherwise.


Here, it is implied that (M, g0 ) ∼

= S2 have equivalent conformal type. That is, there exists a diffeomor-
phism f : (M, g0 ) → S such that f and f −1 are holomorphic41 .
2

Then there exists a smooth solution to the Ricci flow g(t) for t ∈ [0, T ] such that

(i) g(0) = g0 ;

(ii) g(t) is instantaneously complete. That is, (M, g(t)) is complete for all t ∈ (0, T ];

(iii) g(t) is maximally stretched. That is, if g̃(t) is any solution on M with g̃(0) ≤ g(0), then g̃(t) ≤ g(t)
for all t ∈ [0, min{T, Te}].

and this flow is unique in the sense that if g̃(t) for t ∈ [0, Te] is another solution to the Ricci flow on M
satisfying (i)-(iii), then Te ≤ T and g̃(t) = g(t) for all t ∈ [0, Te]. ☼

Informally, the authors of [11] show that if the initial manifold is non-complete, a unique solution to
the (un-normalized) Ricci flow can be obtained such that the family of manifolds (M, g(t)) is complete.
It remains to be shown whether there exists almost-Riemannian manifolds M such that (M± , g0 ) has
equivalent conformal type to either S2 , C or RP2 . In the other case, i.e. T = ∞, when ignoring the fact
that M± have smooth boundaries, a solution to the Ricci flow (3.1) can be obtained for (M± , g0 ).
In [64], the main result applies to arbitrary surfaces with upper bounded Gaussian curvature. The
final time of the solution to the flow depends only on the supremum of the Gaussian curvature. Similar
to [11], the initial surface may be non-complete, and (M, g(t)) is complete for t ∈ (0, T ]. We already
concluded that the Scalar curvature (and hence the Gaussian curvature) is likely to be negative. Hence,
an upper bound does not seem to be a very unrealistic requirement, implying that we can also use this
result for our case. A rigorous result on Gaussian curvature upper bounds for a generic 2-ARS is, however,
not yet available.

4.2.3 A conjecture on the Ricci flow on two-dimensional almost-Riemannian manifolds

As it turns out, to our knowledge, there are not yet enough results on Ricci flow to guarantee existence
and uniqueness on non-complete surfaces with boundary. It seems to be the most promising to consider
(4.16) on (M± , g0 ), for a sufficiently small  > 0, and combine [63, Theorem 2.2] with [11, Theorem 1.3]
39 For
a definition, see e.g. [44, Chapter 5].
40 Thisis in fact described in [59] from the same author.
41 See, for example, [65, Page 32] for more information on conformal types. Also, a holomorphic function is a complex

differentiable function.

53
4.2 A discussion on Ricci flow on two-dimensional almost-Riemannian manifolds

or [64, Theorem 1.1]. However, due to the deformations, it will be unlikely that M as a whole can still
be regarded as a smooth surfaces endowed with a 2-ARS. This will already be more realistic when one
considers the normalized Ricci flow on both surfaces, but this requires further analysis on whether the
results of e.g. [61] can be combined with [11]. Nevertheless, let us conclude this thesis with a conjecture
on the un-normalized Ricci flow on (M± , g0 ).

Conjecture. Let (M, g0 ) be a complete, connected, compact and orientable two-dimensional surface
endowed with a generic 2-ARS without tangency points such that (M± , g0 ) are compact non-complete
surfaces with boundary. Consider the evolution equations

∂
 ∂t g(t) = −S(t)g(t),
 in M± × (0, T ],
κ(·, t) = ψ(·, t), on ∂M± × (0, T ], (4.18)
±

g(0) = g0 , in M .


There exists a unique short-time solution g(t) to (4.18) such that (M, g(t)) are complete for all t ∈ (0, T ].

54
5 CONCLUSION

5 Conclusion
In this thesis, we have studied differential geometry and, in particular, Riemannian geometry to eventually
investigate the possible evolution of Ricci flow on two-dimensional almost-Riemannian manifolds.
The thesis seems to be an useful handbook for understanding the basics of Ricci flow. The first
chapter covers all the tools that are required to define Ricci flow and describe its evolution on Riemannian
manifolds of any dimension. This enabled us to provide all proofs with adequate argumentation. Although
we do not propose any new techniques regarding the short-time existence and uniqueness, one could argue
that the proof depicted in this thesis excels through its clarity.
In the final chapter of the thesis, we introduce almost-Riemannian structures as the prototype of
rank-varying sub-Riemannian structures. In two dimensions, they have been shown to generically satisfy
certain properties. We proceeded in studying a generic free 2-ARS without tangency points on a compact
and oriented manifold and considered all the points tat lie more than an epsilon distance away from the
singular set. Under these circumstances, we established the topology on the two surfaces disjoint from
the singular set. These surfaces both consist out of Riemannian points only, but are non-complete and
have smooth boundary. This complicates the evolution of Ricci flow on these surfaces, since no results
are available on surfaces with these kind of topological properties.
The thesis is concluded with an analysis of previous works in which either manifolds with boundary,
or non-complete surfaces were subject to the (normalized) Ricci flow. We investigate whether the results
may apply naturally to our case. The most promising is to consider the un-normalized Ricci flow on
both surfaces individually, and combine the results of [63, Theorem 2.2] with [11, Theorem 1.3] or [64,
Theorem 1.1]. This, however, requires a very precise analysis on the actual methods that were used in all
of the proofs to see if they can be combined at all. Moreover, due to the deformations of the flow, it is not
likely that the manifold as a whole can still be regarded as almost-Riemannian. In addition, to use [64,
Theorem 1.1], one would require curvature upper bounds on almost-Riemannian manifolds. Obtaining
curvature upper bounds for a generic 2-ARS seems to be a realistic and promising point of interest for
further research in this direction, since the curvature is likely to be negative under these circumstances.

55
A A note on Riemannian submanifolds
The following section is based on [21, Chapter 8]. To understand curvature of a Riemannian submanifold
(M, g) of a Riemannian manifold (M f, g̃) one first needs to understand how the Levi-Civita connection
of M is related to that of M . To that end, let us first define the tangential and respectively normal
f
projections given by
π⊥ : T M
f → T M,
M
π⊥ : T M
f → N M.
M
(A.1)

If X ∈ Γ(T M f ), then we write X ⊥ = π ⊥ X and X ⊥ = π ⊥ X. If X, Y ∈ X(M ), we can extend them to

M
f by applying the Levi-Civita connection ∇
vector fields on open neighbourhoods of M e of M
f:
 ⊥  ⊥
∇
e XY = ∇e XY + ∇e XY . (A.2)

Now, recall the normal bundle of (2.29) for the following definition.
Definition A.1. The second fundamental form of M is the map II : X(M ) × X(M ) → Γ(N M ) is defined
by
 ⊥
II(X, Y ) := ∇e XY . (A.3)
♥
The second fundamental form satisfies certain nice properties, as shown in [21, Proposition 8.1]. More
of its geometric meaning can be understood by studying the curvature of curves. Recall the definition of
the acceleration of a smooth curve from Definition 2.63.
Definition A.2. Let M be a Riemannian manifold and γ : I → M a smooth unit-speed curve in M .
The geodesic curvature of γ is the length of the acceleration vector field. That is, the map κ : I → R
defined by
κ(t) := |Dt γ 0 (t)| . (A.4)
♥
It measures how far the curve is from being a geodesic. The extend to which a Riemannian submanifold
curves within its ambient manifold can be expressed by means of the intrinsic curvature and extrinsic
curvature. If γ : I → M is regular curve with M ⊂ M f, then the intrinsic curvature κ is a curve in M , and
its extrinsic curvature κ̃ is a curve in M . The relationship between the two can be computed with the
f
second fundamental form (See [21, Proposition 8.10]). We remark that a submanifold M of M f is called
totally geodesic if every geodesic corresponding to g̃ that is tangent to M at some time t0 , stays in M for
all t ∈ (t0 − , t0 + ).
In the case that M ⊂ M f is a hypersurface, i.e. a submanifold of codimension 1, we can use a related
but slightly different object to the second fundamental form. That is, the scalar second fundamental
form h, where hN : X(M ) × X(M ) → C ∞ (M ) is symmetric covariant (0,2)-tensor field hN ∈ Γ(Σ2 T ∗ M ),
defined by
hN (X, Y ) := hN, II(X, Y )i , N ∈ Γ(N M ). (A.5)
The map sN : X(M ) → X(M ) associated with the scalar second fundamental form is a (1, 1)-tensor field
and referred to as the shape operator of M , which is related to hN through

hsN (X), Y i = hN (X, Y ), X, Y ∈ X(M ). (A.6)

Subsequently, we define the mean curvature as

1 1
H := tr(s) = trg (h), (A.7)
n n
the Gaussian curvature as
det(h)
K := det(s) = , (A.8)
det(g)

56
A A NOTE ON RIEMANNIAN SUBMANIFOLDS

and remark that the coefficients of the Riemann curvature tensor satisfy the Gauss equation [21, Equation
8.21]
Rmijkl = hil hjk − hik hjl . (A.9)
.
From Gauss’s Theorema Egregium it becomes apparent that when M is an embedded two-dimensional
submanifold of R3 , the Gaussian curvature is an intrinsic local isometric invariant of (M, g). To be precise,
under those circumstances the Gaussian curvature equals half the scalar curvature, or, equivalently:
S = 2K [21, Theorem 8.27, Corollary 8.28]. Moreover, for the Ricci curvature we then have the following.

Lemma A.3. Let (M, g) be a two-dimensional Riemannian manifold, then

S
Ricij = gij . (A.10)
2
♠
Proof. By using the Bianchi identities (Lemma 2.77), one could derive that the Riemann curvature tensor
in two-dimensions only has one independent component:

Rm1212 = Rm2121 = −Rm1221 = −Rm2112 .

From (A.9), it becomes apparent that then

Rm1212 = h12 h21 − h11 h22 = − det(h).

Consequently, notice that (A.8) then implies that

Rm1212 = −K det(g),

and hence that

Rmijkl = K(gik gjl − gil gjk ).
S
Recall that Ricij = g kl Rmikjl and that K = 2 to observe that therefore

S
Ricij = g kl Rmikjl = Kg kl (gij gkl − gil gkj ) = gij ,
2
as desired.

57
B Covariant derivative of tensor fields in normal coordinates
We claim that
   
∂ ∂ ∂ ∂ ∂
(∇k τp ) dxj1 , . . . , dxjr , i1 , . . . , is = τp dxj1 , . . . , dxjr , i1 , . . . , is .
p p ∂x p ∂x p ∂xk p p ∂x p ∂x p
(B.1)
To begin with, lemma 2.52 (ii) implies that for τ ∈ Tsr (M ) we have
   
∂ ∂ ∂ ∂ ∂
(∇k τp ) dxj1 , . . . , dxjr , i1 , . . . , is = k τp dxj1 , . . . , dxjr , i1 , . . . , is
p p ∂x p ∂x p ∂x p p ∂x p ∂x p
j r  
X ∂ ∂
− τp dxj1 , . . . , ∇k dxjl , . . . , dxjr , i1 ,...,
j =j
p p p ∂x p ∂xis p
l 1

is  
X ∂ ∂ ∂
− τp dxj1 , . . . , dxjr , i1 , . . . , ∇k ,..., .
im =i1
p p ∂x p ∂xim p ∂xis p

Now, notice that

 
j1 jl jr ∂ ∂
τp dx , . . . , ∇k dx , . . . , dx , , . . . , is
p p p ∂xi1 p ∂x p
 
j1 ···jr ∂ ∂ i1 is ∂ ∂
= τi1 ···is (p) j1 ⊗ · · · ⊗ jr ⊗ dx ⊗ · · · ⊗ dx dxj1 , . . . , ∇k dxjl , . . . , dxjr , i1 , . . . , is
∂x p ∂x p p p p p p ∂x p ∂x p
         
···jr ∂ ∂ ∂ ∂ ∂
= τij11···i (p)dxj1 · · · ∇k dxjl · · · dxjr dxi1 · · · dxis .
s
p ∂xj1 p p ∂xjl p p ∂xjr p p ∂xi1 p p ∂xis p

By lemma 2.40 (i), observe that

     
∂ ∂ ∂
jl
= ∇k dxjl jl
= ∇k δjill − 0 = 0,


∇k dx | − dx ∇k j
p ∂xjl p p ∂xjl p p ∂x l p

and consequently that

 
∂ ∂
τp dxj1 , . . . , ∇k dxjl , . . . , dxjr , i1 ,..., = 0,
p p p ∂x p ∂xis p

for each jl ∈ {j1 , . . . , jr }. In addition,

 
∂ ∂ ∂
τp dxj1 , . . . , dxjr , i1 , . . . , ∇k im , . . . , is
p p ∂x p ∂x p ∂x p
         
···jr ∂ ∂ ∂ ∂ ∂
= τij11···i (p)dx j1
· · · dxjr
dxi1
· · · dxim ∇k im · · · dxis ,
s
p ∂xj1 p p ∂xjr p p ∂xi1 p p ∂x p p ∂xis p
| {z }
=0

such that also  

∂ ∂ ∂
τp dxj1 , . . . , dxjr , i1 , . . . , ∇k ,..., = 0,
p p ∂x p ∂xim p ∂xis p

for each im ∈ {i1 , . . . , js }. This concludes the proof for (B.1).

58
REFERENCES

References
[1] R. S. Hamilton. “Three-manifolds with positive Ricci curvature”. In: Journal of Differential Ge-
ometry 17.2 (1982). cited By 1364, pp. 255–306. doi: 10.4310/jdg/1214436922. url: https:
//www.scopus.com/inward/record.uri?eid=2- s2.0- 84972513449&doi=10.4310%2fjdg%
2f1214436922&partnerID=40&md5=28f8ee6c75d9a7ef402af061e8bf3409.
[2] W. P. Thurston. “Three dimensional manifolds, Kleinian groups and hyperbolic geometry”. In:
Bulletin (New Series) of the American Mathematical Society 6.3 (1982), pp. 357–381. doi: bams/
1183548782. url: https://doi.org/.
[3] H. Poincaré. Cinquième complément à l’Analysis situs. Dec. 1904. doi: 10.1007/bf03014091. url:
https://doi.org/10.1007/bf03014091.
[4] G. Perelman. “The entropy formula for the Ricci flow and its geometric applications”. In: (2002).
arXiv: math/0211159 [math.DG].
[5] G. Perelman. “Ricci flow with surgery on three-manifolds”. In: (2003). arXiv: math / 0303109
[math.DG].
[6] G. Perelman. “Finite extinction time for the solutions to the Ricci flow on certain three-manifolds”.
In: (2003). arXiv: math/0307245 [math.DG].
[7] D. M. DeTurck. “Deforming metrics in the direction of their Ricci tensors”. In: Journal of Differential
Geometry 18.1 (1983), pp. 157–162. doi: 10.4310/jdg/1214509286. url: https://doi.org/10.
4310/jdg/1214509286.
[8] W. Shi. “Deforming the metric on complete Riemannian manifolds”. In: Journal of Differential
Geometry 30.1 (1989), pp. 223–301. doi: 10.4310/jdg/1214443292. url: https://doi.org/10.
4310/jdg/1214443292.
[9] B. Chen and X. Zhu. “Uniqueness of the Ricci flow on complete noncompact manifolds”. In: Journal
of Differential Geometry 74.1 (2006), pp. 119–154. doi: 10.4310/jdg/1175266184. url: https:
//doi.org/10.4310/jdg/1175266184.
[10] T. A. Chow. Ricci Flow on Manifolds with Boundary with Arbitrary Initial Metric. 2020. arXiv:
2012.04430 [math.DG].
[11] G. Giesen and P. M. Topping. “Existence of Ricci Flows of Incomplete Surfaces”. In: Communi-
cations in Partial Differential Equations 36.10 (Oct. 2011), pp. 1860–1880. issn: 1532-4133. doi:
10.1080/03605302.2011.558555. url: http://dx.doi.org/10.1080/03605302.2011.558555.
[12] M. Lovric, M. Min-Oo, and E. Ruh. “Deforming Transverse Riemannian Metrics of Foliations”. In:
The Asian Journal of Mathematics 4 (Jan. 2000). doi: 10.4310/AJM.2000.v4.n2.a1.
[13] Y. Dong. Eells-Sampson type theorems for subelliptic harmonic maps from sub-Riemannian mani-
folds. 2019. arXiv: 1903.04702 [math.DG].
[14] F. Baudoin and N. Garofalo. Curvature-dimension inequalities and Ricci lower bounds for sub-
Riemannian manifolds with transverse symmetries. 2014. arXiv: 1101.3590 [math.DG].
[15] A. A. Agrachev and P. W. Y. Lee. Bishop and Laplacian Comparison Theorems on Three Dimen-
sional Contact Subriemannian Manifolds with Symmetry. 2013. arXiv: 1105.2206 [math.DG].
[16] V. V. Grušin. “On a class of hypoelliptic operators”. In: Mathematics of the USSR-Sbornik 12 (Oct.
2007), p. 458. doi: 10.1070/SM1970v012n03ABEH000931.
[17] A. A. Agrachev, U. Boscain, and M. Sigalotti. “A gauss-bonnet-like formula on two-dimensional
almost-riemannian manifolds”. In: Discrete and Continuous Dynamical Systems 20.4 (2008). cited
By 40, pp. 801–822. doi: 10.3934/dcds.2008.20.801. url: https://www.scopus.com/inward/
record.uri?eid=2- s2.0- 44849132955&doi=10.3934%2fdcds.2008.20.801&partnerID=40&
md5=3ed29d4b26e3cbc864b9906ab4c40a13.

59
REFERENCES

[18] B. Andrews and C. Hopper. “The ricci flow in riemannian geometry: A complete proof of the
differentiable 1/4-pinching sphere Theorem”. In: Lecture Notes in Mathematics 2011 (2011). cited
By 14, pp. 1–313. doi: 10.1007/978-3-642-16286-2_1. url: https://www.scopus.com/inward/
record.uri?eid=2-s2.0-78249238464&doi=10.1007%2f978-3-642-16286-2_1&partnerID=40&
md5=bd3f1e745f18dd93538f0039c825e9bf.
[19] W. A. Sutherland. Introduction to metric and topological spaces. 2nd ed. Oxford University Press,
2009. isbn: 9780199563081.
[20] J. M. 1950- Lee. Riemannian manifolds: an introduction to curvature. Graduate texts in mathe-
matics 176. New York: Springer, 1997. isbn: 038798271X. url: http://catdir.loc.gov/catdir/
enhancements/fy0815/97014537-t.html.
[21] J. M. 1950- Lee. Introduction to Riemannian manifolds. Cham, Switzerland: Springer, 2018. isbn:
9783319917559. doi: 10.1007/978- 3- 319- 91755- 9. url: https://doi.org/10.1007/978- 3-
319-91755-9.
[22] M. Seri. “Analysis on Manifolds”. 2020. url: https://github.com/mseri/AoM/releases.
[23] J. M. 1956- Lee. Manifolds and differential geometry. Vol. 107. Graduate studies in mathematics.
Providence, R.I.: American Mathematical Society, 2009. isbn: 9780821848159. url: http://www.
ams.org/books/gsm/107/.
[24] M. P. do Carmo. Riemannian geometry / Manfredo do Carmo ; translated by Francis Flaherty.
eng. Mathematics. Theory and applications. Boston: Birkhäuser, 1992. isbn: 0817634908. url:
https://www.springer.com/gp/book/9780817634902.
[25] J. M. 1950- Lee. Introduction to smooth manifolds. New York: Springer, 2012. isbn: 9781441999825.
doi: 10.1007/978-1-4419-9982-5. url: https://doi.org/10.1007/978-1-4419-9982-5.
[26] L. W. Tu. An Introduction to Manifolds. Universitext (1979). Springer, 2011. isbn: 9781441974006.
url: https://www.springer.com/gp/book/9781441973993.
[27] K. Yasaka. “Spectral almost Riemannian geometry and the magnetic Aharonov-bohm efect”. 2020.
url: https://fse.studenttheses.ub.rug.nl/22692/1/bPHYS_2020_YasakaK.pdf.
[28] R. S. Hamilton. “The Ricci flow on surfaces”. In: Mathematics and General Relativity. Vol. 71.
American Mathematical Society, 1988, pp. 237–261. url: https://bookstore.ams.org/conm-71.
[29] B. Chow. “The Ricci flow on the 2-sphere”. In: Journal of Differential Geometry 33.2 (1991). cited
By 200, pp. 325–334. doi: 10.4310/jdg/1214446319. url: https://www.scopus.com/inward/
record.uri?eid=2- s2.0- 84972506625&doi=10.4310%2fjdg%2f1214446319&partnerID=40&
md5=8ccc4f369251329746e2b80930964e11.
[30] J. Isenberg, R. Mazzeo, and N. Sesum. Ricci Flow in Two Dimensions. 2011. arXiv: 1103.4669
[math.DG].
[31] B. Chow and D. Knopf. “The Ricci flow: an introduction”. In: Mathematical surveys and monographs
110 (2004). doi: https://doi.org/http://dx.doi.org/10.1090/surv/110.
[32] P. M. Topping. Lectures on the Ricci Flow. London Mathematical Society Lecture Note Series.
Cambridge University Press, 2006. doi: 10.1017/CBO9780511721465.
[33] O. Iacovlenco. “An introduction to Hamilton’s Ricci flow”. In: (2012). url: https://www.math.
mcgill.ca/gantumur/math581w12/downloads/Ricci.pdf.
[34] S. Johar. “Parabolic PDE’s and Ricci flow”. [email protected]. url: https://zenodo.
org/record/801592/files/Parabolic%20PDEs%20and%20Ricci%20Flow.pdf.
[35] T. Sunada. “The solution spaces of non-linear partial differential equations of elliptic type on com-
pact manifolds”. In: Proceedings of the Japan Academy 49.6 (1973), pp. 385–389. doi: 10.3792/
pja/1195519288. url: https://doi.org/10.3792/pja/1195519288.
[36] R. van der Veen. “Lecture Notes Geometry”. 2020. url: http://www.rolandvdv.nl/G20/Lecture
NotesGeometry20.pdf.

60
REFERENCES

[37] S. Gallot, D. Hulin, and J Lafontaine. Riemannian geometry. Berlin: Springer-Verlag, 1990. isbn:
3540179232.
[38] J. Eells and J. H. Sampson. “Harmonic Mappings of Riemannian Manifolds”. In: American Journal
of Mathematics 86.1 (1964), pp. 109–160. issn: 00029327, 10806377. url: http://www.jstor.org/
stable/2373037.
[39] A. Hatcher. Algebraic topology. Cambridge: Cambridge Univ. Press, 2000. url: https : / / cds .
cern.ch/record/478079.
[40] P. Hartman. “On Homotopic Harmonic Maps”. In: Canadian Journal of Mathematics 19 (1967),
pp. 673–687. doi: 10.4153/CJM-1967-062-6.
[41] K. Chang, W. Y. Ding, and R. Ye. “Finite-time blow-up of the heat flow of harmonic maps from
surfaces”. In: Journal of Differential Geometry 36.2 (1992), pp. 507–515. doi: 10 . 4310 / jdg /
1214448751. url: https://doi.org/10.4310/jdg/1214448751.
[42] P. J. Olver. Introduction to partial differential equations. Cham: Springer, 2014. isbn: 9783319020990.
doi: 10.1007/978-3-319-02099-0. url: https://doi.org/10.1007/978-3-319-02099-0.
[43] C. Mantegazza and L. Martinazzi. “A note on quasilinear parabolic equations on manifolds”. en. In:
Annali della Scuola Normale Superiore di Pisa - Classe di Scienze Ser. 5, 11.4 (2012), pp. 857–874.
url: http://www.numdam.org/item/ASNSP_2012_5_11_4_857_0/.
[44] L. C. Evans. “Partial differential equations”. In: Graduate studies in mathematics 19.2 (2010). url:
https://bookstore.ams.org/gsm-19-r.
[45] Y. Xin. Geometry of harmonic maps. Vol. 23. Springer Science & Business Media, 1996. isbn:
0817638202. url: https://www.springer.com/gp/book/9780817638207.
[46] A. A. Agrachev, D. Barilari, and U. Boscain. A Comprehensive Introduction to Sub-Riemannian
Geometry. Cambridge Studies in Advanced Mathematics. Cambridge University Press, 2019. doi:
10.1017/9781108677325.
[47] A. A. Agrachev, D. Barilari, and L. Rizzi. “Curvature: a variational approach”. In: Memoirs of the
American Mathematical Society 256.1225 (Nov. 2018). issn: 1947-6221. doi: 10.1090/memo/1225.
url: http://dx.doi.org/10.1090/memo/1225.
[48] U. Boscain, G. Charlot, M. Gaye, and P. Mason. “Local properties of almost-Riemannian structures
in dimension 3”. In: Discrete and Continuous Dynamical Systems- Series A 35.9 (2015). cited By
4, pp. 4115–4147. doi: 10.3934/dcds.2015.35.4115. url: https://www.scopus.com/inward/
record.uri?eid=2-s2.0-84926617391&doi=10.3934%2fdcds.2015.35.4115&partnerID=40&
md5=2fc016865b37720ab5761874c0e688e8.
[49] U. Boscain, D. Prandi, and M. Seri. “Spectral analysis and the Aharonov-Bohm effect on certain
almost-Riemannian manifolds”. In: Communications in Partial Differential Equations 41.1 (2016).
cited By 7, pp. 32–50. doi: 10.1080/03605302.2015.1095766. url: https://www.scopus.com/
inward/record.uri?eid=2- s2.0- 84953352326&doi=10.1080%2f03605302.2015.1095766&
partnerID=40&md5=eb24f3f4ede44aaec76521f281327ccc.
[50] A. A. Agrachev, U. Boscain, G. Charlot, R. Ghezzi, and M. Sigalotti. “Two-dimensional almost-
Riemannian structures with tangency points”. In: Annales de l’Institut Henri Poincare (C) Analyse
Non Lineaire 27.3 (2010). cited By 21, pp. 793–807. doi: 10.1016/j.anihpc.2009.11.011. url:
https://www.scopus.com/inward/record.uri?eid=2-s2.0-77951255295&doi=10.1016%2fj.
anihpc.2009.11.011&partnerID=40&md5=e5be4acedf000aab3148786aedc1efc1.
[51] U. Boscain and C. Laurent. “The laplace-beltrami operator in almost-riemannian geometry”. In:
Annales de l’Institut Fourier 63.5 (2013). cited By 17, pp. 1739–1770. doi: 10.5802/aif.2813.
url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-84893906378&doi=10.5802%
2faif.2813&partnerID=40&md5=5a12e830bb1c46c3d9647d662d563e0f.
[52] G. T. Lee. Abstract algebra: an introductory course. Cham: Springer, 2018. isbn: 9783319776491.
doi: 10.1007/978-3-319-77649-1. url: https://doi.org/10.1007/978-3-319-77649-1.

61
REFERENCES

[53] A. Steve. Category Theory. Vol. 2nd ed. Oxford Logic Guides Vol. 49. OUP Oxford, 2010. isbn:
9780199587360. url: http://search.ebscohost.com.proxy-ub.rug.nl/login.aspx?direct=
true&db=nlebk&AN=375073&site=ehost-live&scope=site.
[54] V. I. Bogachev. Measure theory / V.I. Bogachev. eng. Berlin ; Springer, 2007. isbn: 9783540345138.
[55] R. Ghezzi. “Almost-Riemannian Geometry from a Control Theoretical Viewpoint”. MA thesis.
SISSA, 2010. url: http://hdl.handle.net/1963/4705.
[56] Y. Shen. “On Ricci deformation of a Riemannian metric on manifold with boundary”. In: Pacific
Journal of Mathematics 173.1 (1996), pp. 203–221. doi: pjm/1102365851. url: https://doi.org/.
[57] A. Pulemotov. Quasilinear Parabolic Equations and the Ricci Flow on Manifolds with Boundary.
2012. arXiv: 1012.2941 [math.AP].
[58] P. Gianniotis. “The Ricci flow on manifolds with boundary”. In: Journal of Differential Geometry
104.2 (2016), pp. 291–324. doi: 10.4310/jdg/1476367059. url: https://doi.org/10.4310/jdg/
1476367059.
[59] J. C. Cortissoz and A. Murcia. The Ricci flow on surfaces with boundary. 2016. arXiv: 1209.2386
[math.DG].
[60] J. C. Cortissoz and A. Murcia. The Ricci flow on a cylinder. 2016. arXiv: 1604.02132 [math.DG].
[61] S. Brendle. “A family of curvature flows on surfaces with boundary”. In: Mathematische Zeitschrift
241.4 (2002), pp. 829–869. doi: 10.1007/s00209-002-0439-1. url: https://doi.org/10.1007/
s00209-002-0439-1.
[62] S. Brendle. “Curvature flows on surfaces with boundary”. In: Mathematische Annalen 324.3 (2002),
pp. 491–519. doi: 10.1007/s00208-002-0350-4. url: https://doi.org/10.1007/s00208-002-
0350-4.
[63] J. C. Cortissoz. “The Ricci flow on the two ball with a rotationally symmetric metric”. In: Russian
Mathematics 51.12 (2007), pp. 30–51. doi: 10.3103/S1066369X07120031. url: https://doi.org/
10.3103/S1066369X07120031.
[64] P. M. Topping. “Ricci flow compactness via pseudolocality, and flows with incomplete initial met-
rics”. In: Journal of the European Mathematical Society 12.6 (2010). cited By 34, pp. 1429–1451.
doi: 10.4171/JEMS/237. url: https://www.scopus.com/inward/record.uri?eid=2- s2.0-
77956977480&doi=10.4171%2fJEMS%2f237&partnerID=40&md5=9244fdcd71c6856ac6324fa030d
5aa78.
[65] Y. Fang. Conformal types of Riemann surfaces. Proceedings of the Centre for Mathematics and
Its Applications, Australian National University. Centre for Mathematics and Its Applications,
Australian National University, 1996, pp. 32–36. isbn: 9780731524433. url: https://projecteuc
lid.org/proceedings/proceedings-of-the-centre-for-mathematics-and-its-application
s/Lectures-on-Minimal-Surfaces-in-R3/toc/pcma/1416324349.

62