Spinors and Cosmology in Einstein-Cartan theory

Karim Van Aelst

Supervisor : João Magueijo

September 16th , 2016

Submitted in partial fulfilment of the requirements for the degree of Master of Science of
Imperial College London
Contents

Introduction 3

1 Mathematical framework for spinors 5

1.1 Spinor bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Fiber bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1.1 Definition and examples . . . . . . . . . . . . . . . . . 6
1.1.1.2 Vector fields . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.1.3 Structure group . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Principal bundle . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.2.2 Frame bundle . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.3 Associated bundle . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1.3.1 Construction . . . . . . . . . . . . . . . . . . . . . . . 14
1.1.3.2 Spin structure and spinors . . . . . . . . . . . . . . . . 15
1.2 Spin connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.1 Principal connection . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.1.1 Vertical bundle . . . . . . . . . . . . . . . . . . . . . . 17
1.2.1.2 Equivariant horizontal distribution . . . . . . . . . . . 19
1.2.2 Connection 1-form . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.2.2.1 Vector-valued differential forms . . . . . . . . . . . . . 21
1.2.2.2 Fundamental fields . . . . . . . . . . . . . . . . . . . . 22
1.2.3 Connections in associated bundles . . . . . . . . . . . . . . . . . 24
1.2.3.1 Gauge potential . . . . . . . . . . . . . . . . . . . . . . 24
1.2.3.2 Covariant derivative . . . . . . . . . . . . . . . . . . . 26
1.3 Connection-related objects . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.3.1 Exterior covariant derivative . . . . . . . . . . . . . . . . . . . . 29
1.3.1.1 Bundle-valued differential forms . . . . . . . . . . . . . 29
1.3.1.2 The operator d∇ . . . . . . . . . . . . . . . . . . . . . 30
1.3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.3.2.2 Structure equation . . . . . . . . . . . . . . . . . . . . 32
1.3.2.3 Bianchi identity . . . . . . . . . . . . . . . . . . . . . . 34
1.3.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.3.3.2 Structure equation . . . . . . . . . . . . . . . . . . . . 35
1.3.3.3 Bianchi identity . . . . . . . . . . . . . . . . . . . . . . 36

1
2 Einstein-Cartan gravity 38
2.1 Prerequisite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.1 Spin(3, 1) connection . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.2 Metric connection and contortion . . . . . . . . . . . . . . . . . 42
2.1.3 Vierbein postulate . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2 Field equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2.1 First-order formalism . . . . . . . . . . . . . . . . . . . . . . . . 46
2.2.2 Asymmetric Einstein’s equation . . . . . . . . . . . . . . . . . . 49
2.2.3 Torsion and spin density tensor . . . . . . . . . . . . . . . . . . 51
2.3 Spinor source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.1 Fermion currents . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.2 Four-fermion interaction . . . . . . . . . . . . . . . . . . . . . . 57
2.3.3 Field equations and Dirac equation . . . . . . . . . . . . . . . . 59

3 Cosmology with spinors 62

3.1 Spinor driven inflation . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.1.1 Dirac equation in FRW . . . . . . . . . . . . . . . . . . . . . . . 64
3.1.2 Perturbation expansion and thermal fluctuations . . . . . . . . . 66
3.2 Bouncing universe from spinors . . . . . . . . . . . . . . . . . . . . . . 68
3.2.1 Friedmann equations with torsion . . . . . . . . . . . . . . . . . 69
3.2.2 Bounce, flatness and horizon . . . . . . . . . . . . . . . . . . . . 70

Conclusion 75

References 78

2
Introduction

One-hundred-year-old Einstein’s theory of general relativity is a fully operative the-

ory of gravity. Using the framework of Riemannian geometry, it describes phenomena
such as the bending of light and gravitational redshift, which Newtonian mechanics or
special relativity cannot explain. It also provides tools to formalise and tackle various
fundamental issues of cosmology, such as the horizon and flatness problems. Ultimately,
it allows for the existence of extreme gravity-related objects such as black holes and
gravity waves.
However, general relativity does not comply with the principles of quantum physics,
although the latter unifies the three other fundamental interactions within the standard
model of particle physics. Of course, many theories such as loop quantum gravity, causal
set theory or string theory aim at providing a quantum description of gravity. None of
these have turned out to be conclusive so far.
Yet, it is still possible to deal with gravity together with quantum physics in certain
situations. It is the case of quantum field theory in curved spacetime, which studies the
dynamics of quantum fields defined over a spacetime curved by its content. Although
it is far from being a quantum theory of gravity, it provides striking results such as
Hawking radiation and an interpretation for the cosmological microwave background.
In quantum field theory in curved spacetime, gravity is still described by Einstein field
equations, for which energy-momentum tensor is derived from the fields that are defined
over the considered spacetime.
In this regard, Einstein-Cartan theory of gravity goes a step further than quantum
field theory in curved spacetime. In fact, although it is also far from unifying gravity
to quantum physics, it extends Einstein’s description of gravity by considering the
quantum concept of spin as a source for gravity, on the same terms as the energy-
momentum tensor. As a result, despite the flagrant incompatibility of general relativity
and quantum principles, Einstein-Cartan theory has a very original feature in that it
establishes a quantum notion at the very root of general relativity. It is achieved in the
following way.
Based on both physical and philosophical convictions, Einstein’s general relativity
assumes that spacetime is a Lorentzian manifold equipped with the unique torsion-free
metric connection, that is the Levi-Civita connection, which is entirely determined by
the metric. Based on papers of Élie Cartan in the 1920’s and Dennis Sciama and Tom
Kibble in the 1960’s, it was suggested that the connection might be allowed to have
torsion. Such statement results in asymmetric Ricci and Einstein’s tensors, and makes
way for the spin density tensor, describing the intrinsic angular momentum of matter,
to influence the curvature. These new aspects in the equations of gravitation lead to
several original phenomena with interesting cosmological predictions.
Naturally, the next step is to review the types of field which may now take action

3
in Einstein-Cartan gravity via their spin, in addition to their energy and momentum.
Since the fundamental role of the Lorentz group was made clear with the advent of
special relativity, Lorentz transformations are required to be symmetries of any action
considered in theoretical physics. As a result, any field involved in a theory must stand
in a certain representation of the Lorentz group. Among these representations, the 4-
vector and the 4-component spinor fields are the representations used to describe most
of the elementary particles. These representations bear a non-zero spin, hence their
importance for Einstein-Cartan gravity.
Most of the time yet, only scalar fields, which have spin zero, are considered in
quantum field theory in curved spacetime. This is partly due to the fact that they are
sufficient to provide satisfying answers. So is the inflaton, commonly used to drive infla-
tion. Vector fields may also be implicated in cosmological issues, and may for instance
be used to drive inflation as well [1]. As to spinor fields, they are more rarely used in
cosmology, despite playing a major role in the standard model of particle physics. Thus,
our main goal here will be to present some of the effects of spinor fields in cosmology,
which are particularly interesting in the context of Einstein-Cartan theory of gravity.
Specifically, this dissertation first aims at comprehending the framework required
to use spinors and study their effects in the extended version of general relativity that
constitutes Einstein-Cartan theory. The second chapter then covers the proper Eistein-
Cartan theory, in order to familiarise with the extended Einstein’s field equation and the
physical bound relating spin density to torsion. In particular, the last part of the chapter
focuses on the formulation of Einstein-Cartan theory in presence of a spinor field. This
prepares for the concrete implications of spinors in cosmology that are addressed in the
last chapter.
The presentation merely draws on basics of differential geometry such as the notions
of tangent space at a point, push forward of an application between manifolds, differ-
ential forms, Lie groups and Lie algebras, along with the principles of general relativity
and quantum field theory.

4
Chapter 1

Mathematical framework for spinors

The Einstein-Cartan theory of gravity promotes the concepts of spin and torsion
to a whole other level in describing gravity-related effects. We mentioned that we will
concentrate on the effects of spin on gravity when it is beared by spinors. Accordingly,
this first chapter addresses the mathematical concepts needed to rigorously manipulate
spinor fields over general spacetimes. In fact, defining spinors over a given spacetime is
far from being as straightforward as it is in Minkowski space, which is the background
used in the standard model to wield spinors.
More precisely, the concept of fiber bundle needs to be introduced first. This mathe-
matical structure is of significant interest in modern physics since it provides the frame-
work necessary to properly define and analyse the dynamics of any type of field over a
manifold. It underlies both the gauge theories of standard model and general relativ-
ity, and hence Einstein-Cartan theory. This then leads us to construct spinor bundles,
which are particular bundles essential to consider spinors on general spacetimes.
Naturally, the next step is to develop a general tool that evaluates the dynamics of
spinors, that is an application which properly analyses the variations of spinor fields.
To this end, we present a general definition of connections and covariant derivatives,
which are familiar from general relativity and the standard model.
Ultimately, dealing with connections always leads to introduce two closely related
objects : torsion and curvature. We present a general review of these two objects since
they are central objects of Einstein-Cartan theory.

1.1 Spinor bundles

As mentioned above, the most general structures allowing for the existence of spinors
are spinor bundles. These are a particular type of fiber bundles, which are presented
now.

1.1.1 Fiber bundles

Intuitively, a fiber bundle may be first pictured as a generalisation of the Cartesian
product of two spaces. The standard Cartesian product E of two spaces B and F is
defined as the set of all pairs made up of a first element picked in B and a second
element picked in F . A naive way to extend this is to allow the second space F to
depend on the element that is first picked in B. That is, when picking an element in a

5
fiber bundle, one first picks an element b ∈ B, which defines the space Fb in which the
second element is picked. This defines a generalised Cartesian product whose second
factor is parametrised by the element picked in the first factor of the product. These
ideas are now properly formulated in what follows.

1.1.1.1 Definition and examples

In a very general way, the projection onto one of the factors A or B of a standard
Cartesian product A × B will always be denoted as :


prA : A×B −→ A .
(a, b) 7−→ a
We may now formally define a fiber bundle.

Definition. Let E and B be two smooth manifolds, and p : E → B be a smooth map.

The triplet (E, B, p) defines a fiber bundle if and only if for any b ∈ B, there exists
an open subset Ub of B containing b, and a manifold Fb , such that there exists a smooth
diffeomorphism
φb : p−1 (Ub ) −→ Ub × Fb
such that

prUb ◦ φb = p|p−1 (U ) .
b

In other words, ∀b ∈ B, we have the following commutative diagram.

φb
p−1 (Ub ) ⊂ E Ub × Fb

p
prUb
Ub ⊂ B

When the previous conditions hold, the following terminology applies :

• the manifold E is called the total space ;

• the manifold B is the base ;

• p is the projection of the bundle ;

• ∀b ∈ B, Eb := p−1 ({b}) is the fiber over b.

When the manifolds Fb do not actually depend on the elements b of the base B, so
that they are all identical to a manifold F , the quadruplet (E, B, p, F ) is referred to as
a fiber bundle with fiber F .

6
 Cartesian product Fiber bundle
Product space E = B × F Total space E
First factor B Base B
Projection
 on first factor Projection
 of the fiber bundle
prB : E −→ B p: E −→ B
(b, f ) 7−→ b e ∈ p−1 ({b}) 7−→ b
Second factor F Base-parametrised fibers Fb

Table 1.1 – Analogy between Cartesian products and fiber bundles.

One deduces from the definition that, above a point b of the base, the total space
is locally diffeomorphic to the Cartesian product of the base and a second space Fb
which depends on the element b that was first picked in the base. In other terms, a fiber
bundle is a space which resembles a Cartesian product only locally. One could call this
a base-dependent Cartesian product, since the space that stands for the second factor
of the Cartesian product is parametrised by the base. Table 1.1 sums up the various
extensions that are made when fiber bundles are presented as generalised Cartesian
products.

Since we introduced fiber bundles as parametrised Cartesian products, usual Carte-

sian products should be recovered as a particular cases of fiber bundles. Indeed, for any
two spaces A and B, (A × B, A, prA , B) is a fiber bundle. A fiber bundle that turns out
to be of this form is called trivial.
However, it should be clear that, although the manifolds Fb may all be identical to
a manifold F , it does not mean that the total space is globally diffeomorphic to the
standard Cartesian product B × F . The fact that the diffeomorphisms φb are only local
leaves great freedom in the global structure of the total space. That is, local constraints
still allow the total space to have a shape very different from a bare Cartesian product.
For instance, two fiber bundles (E, B, p, F ) and (E 0 , B, p0 , F ) with same base and same
fiber may however have very different total spaces E and E 0 .
Figure 1.1 shows that the Möbius strip illustrates this clearly : it is a fiber bundle
with base the unit circle S 1 and with fiber an interval I. It is therefore locally diffeo-
morphic to the Cartesian product S 1 × I, which is a cylinder. The Möbius strip and
the cylinder are both fiber bundles with base S 1 and fiber I, but they are obviously
not identical globally. This proves that the global structure of a bundle is not entirely
defined by its base and fiber. Actually, the global shape of a bundle is rather charac-
terised by a particularly important group acting on the bundle, called the structure
group, which is introduced in the end of this section.

Ultimately, let us introduce vector bundles, which are the most abundant type of
fiber bundles in physics. It has been mentioned that all the fibers over the points of the
base of a bundle may all be identical to one fixed manifold called the fiber. We consider
such a situation to define vector bundles.
Definition. Let (E, B, p, F ) be a fiber bundle with fiber F . It defines a vector bundle if
and only if :
• F is a vector space ;

7
 (a) Möbius strip.[2] (b) Cylinder.[3]

Figure 1.1 – The Möbius strip is not a cylinder.

• ∀b ∈ B, the fiber Eb := p−1 ({b}) over b is a vector space ;

• ∀b ∈ B, ∀c ∈ Ub , prF ◦ φb|E : Ec −→ F is an isomorphism between vector spaces.

c

In order to give an example, we remind that the tangent space Tb B in a point b of a

manifold B is the vector space of the equivalence classes of tangent curves at b. Then,
let us define the following set :
[
T B := Tb B .
b∈B

An element in E is a vector of the tangent space at a precise point of the manifold,

so that the vector can be mapped to this particular point. This naturally defines a map
on E, which we will always denote πB . Then, (T B, B, πB ) is a vector bundle, called
the tangent bundle1 , which is omnipresent in the theory of bundles and in physics in
general.

1.1.1.2 Vector fields

Let us now picture how the use of fiber bundles naturally arises in physics. Most
of the time, one or several types of tensors and spinors are specified at any point of a
given spacetime. So are the cases of the metric tensor or the curvature tensor, which
set a tensor of order 2 and 4 respectively in any spacetime point. In order for such a
field to be rigorously defined, the space in which the field takes values needs to sit in
every spacetime point. In the language of fiber bundles, spacetime plays the role of the
base, and the fiber over any spacetime point is the proper space in which the field takes
values. Thus, fiber bundles allow us to set the appropriate space over any point of the
base.
1
In the absence of ambiguity, a fiber bundle (E, B, p) is simply designated by its total space E. In
particular, the tangent bundle to a manifold M is often referred to as T M .

8
 In this context, it is straightforward to give a general definition of vector2 fields over
a general spacetime. This is achieved through the notion of section of a fiber bundle
which emerges as follows. Over any point of the base of a fiber bundle, there are infinitely
many elements, which are the elements of the fiber over the considered point. Picking
one particular element in the fiber over every element of the base naturally defines a
map over the base, which is called a section.
By definition, the projection of a fiber bundle is such that, for any element b of the
base, all the elements of the fiber over b are mapped to b itself. As was just said, the
element picked in b by a given section precisely is in the fiber over b. Therefore, if we
apply the projection of the bundle to any section, we get the identity map on the base.
This is why a section is formally defined as follows.
Definition. A section s of a fiber bundle (E, B, p) is a right inverse of the projection
p of the bundle, that is s : B −→ E is such that p ◦ s = IdB . The set of all the sections
of a fiber bundle (E, B, p) is denoted Γ(E).
In a most general way, vector fields are then defined as the sections of vector bun-
dles. Such a definition does meet our intuitive expectations about a vector field over
spacetime, since considering a section of a vector bundle exactly amounts to picking a
vector in every point of the manifold.

1.1.1.3 Structure group

We mentioned earlier that two fiber bundles with same fiber may yet be very differ-
ent. Especially, the local diffeomorphims φb do not put strong constraints on the way
each fiber is “tied” to its neighbouring fibers. For instance, the fibers may twist from one
another as it is the case for the Möbius strip, or remain perfectly parallel to one another
as it is the case of the cylinder. This is an important feature of any fiber bundle, since
the way the fibers “glue” together is responsible of the global shape of the fiber bundle.
It is governed by what are called the transition functions and the structure group, with
the latter playing an essential role in modern physics as we shall see. Let us start by
considering a fiber bundle (E, B, p, F ). Using the definition of a fiber bundle, the set

O := {Ub , b ∈ B}

is an open cover of B.
Let us take U and V in O such that U ∩V 6= ∅, and consider the local diffeomorphims
φU and φV associated to U and V in the definition of a fiber bundle. We know that

∀e ∈ p−1 (U ∩ V ), prB ◦ φU (e) = prB ◦ φV (e) = p(e) ∈ B.

In other terms, the application

prB ◦ φV ◦ φ−1
U : U ∩ V × F 7−→ U ∩ V

coincides with prB|U ∩V ×F .

2
Here, the term “vector” merely designates an element of a vector space, and does not specifically
refer to the defining representation of the Lorentz Group for instance. Since they are elements of
different vector spaces respectively, tensors of any order and all types of spinors are all vectors in this
view.

9
 Then, for any b ∈ U ∩ V , the diffeomorphism φV ◦ φ−1
U maps {b} × F to itself in a
bijective manner. That is,

∀b ∈ U ∩ V, ∀f ∈ F, ∃!f 0 ∈ F such that φV ◦ φ−1 0

U (b, f ) = (b, f ) .

The element f 0 depends in a unique way on the choice of the subsets U and V , and
the elements b in U ∩ V and f in F . Therefore, we get the following natural map :


tbU V : F −→ F .
f 7−→ f 0 = prF ◦ φV ◦ φ−1
U (b, f )

Since the diffeomorphism φV ◦ φ−1 U maps {b} × F to itself in a bijective manner,

tbU V is also a diffeomorphism, that is tbU V ∈ Dif f (F ).
These remarks now enable us to define the transition functions and the structure
group of a fiber bundle.

Definition. For all pairs of subsets U and V of B such that U ∩ V 6= ∅, the transition
function from U to V is defined as


tU V : U ∩V −→ Dif f (F ) .
b 7−→ tbU V

Let us consider the group G of all the applications of Dif f (F ) reached by the tran-
sition functions :
[
G := Im(tU V ) .
U,V ⊂B,
U ∩V 6=∅

It is a subgroup of Dif f (F ), called the structure group of the fiber bundle.

For instance, the structure group of the trivial bundle is clearly {Id}. More generally,
the larger the structure group is, the more free the fibers are in the way they bind
together. The importance of the transition functions and the structure group is strongly
established by the following theorem.

Theorem (Fiber bundle construction theorem). Let (E, B, p, F ) be a fiber bundle with
fiber F . Let {Uα } be an open cover of B and {tUαi Uαj } be the associated set of transition
functions. Then, the total space E may be constructed from B, F and {tUαi Uαj }.

This theorem clearly states that the structure group carries the information of the
whole structure of the bundle since it contains all the transition functions, which dictate
how the fibers shall tie to one another in order to rebuild the total space.
In physics, the structure group actually corresponds to what is commonly called
the gauge group of a theory. For instance, SU (3) is the gauge group relative to the
theory of the strong interaction of the standard model, while SU (2) × U (1) refers to
the electroweak interaction. Through the gauge transformations, the gauge group acts
on the fields of a theory in any spacetime point. For this to be possible, all the fields

10
of the theory, that is the sections of all the vector bundles involved in the theory, must
belong to a representation space of the gauge group. This exactly requires the fibers to
be representation spaces of the gauge group. This is why most of the fiber bundles used
in physics are such that their fibers are representation spaces of their structure group.
In the standard model for instance, the quarks are in the triplet representation of
the gauge group SU (3), whereas the leptons are in the singlet representation ; the Higgs
is in the doublet representation of the gauge group SU (2), whereas the gluons are in
the singlet representation. In addition, Spin(3, 1) also acts on the fields of the standard
model. This is owing to the fact that Spin(3, 1) is the universal cover of SO(3, 1) which
is required to be a symmetry group of any relativistic action. For this group, the fields
of the standard model are either in the scalar representation, the vector representation,
or the spinor representation.
We are particularly focusing on the last representation, which we want to manipulate
not only on Minkowski spacetime but on any spacetime. In the language of bundles now
at our disposal, this means that we need :
• a vector bundle with spacetime as a base ;

• the fiber over any spacetime point to be a representation space of the spin group ;

• the structure group to be the spin group in order to perform gauge transforma-
tions.
Thus, a section of such a vector bundle would properly define a spinor field. But
building such kind of a bundle is far from being easy to achieve, since there are strict
constraints on the construction, as we shall see in the next few sections.

1.1.2 Principal bundle

In order to build the kind of bundle specified above, it is absolutely necessary to start
with considering principal bundles. These are fundamental cases of bundles whose fibers
are diffeomorphic to their structure group. Then only will it be possible to construct
vector bundles interesting for physics, that is those with representation spaces as fibers.

1.1.2.1 Definition
A principal bundle is a particular case of fiber bundle defined as follows.
Definition. Let G be a Lie group. A principal G-bundle is a fiber bundle (P, B, p) such
that there is a simply transitive right action of G on P that is fiber-preserving. This
means that each fiber of P is mapped to itself under the action of G.
With this definition come two very interesting results :

• all the fibers of a principal G-bundle are diffeomorphic to the group G itself ;

• the structure group of a principal G-bundle is the group G itself.

These features are summarized in figure 1.2. It is also easier to picture these points
when considering the particular examples of principal bundles that are called frame
bundles.

11
 Figure 1.2 – Tangent bundle of a principal bundle. [4]

1.1.2.2 Frame bundle

Let us consider a vector bundle (E, B, p, V ) with fiber a d-dimensional vector space
V . This means that the fiber over any point of B is a vector space isomorphic to V . For
any b ∈ B, we will use the isomorphism between V and the fiber Eb in order to identify
GL(Eb ) to GL(V ). Therefore, we will consider that the elements of GL(V ) directly act
upon the elements of Eb , without making reference to the underlying isomorphism.
As a vector space, the fiber Eb admits infinitely many bases. Let us denote Fb E the
set of all the bases of Eb . Then, the following set
[
F E := Fb E
b∈B

and the following map


pF E : FE −→ B
fb ∈ Fb E 7−→ b

are such that (F E, B, pF E ) is a fiber bundle. It is called the frame bundle associated to
the vector bundle (E, B, p, V ).
Let us show that it is actually a principal GL(V )-bundle. Any basis f ∈ Fb E is a set
of d independent vectors which we denote {fi }. Any element g ∈ GL(V ) is equivalent
to an invertible d-by-d matrix (g ij ). Any such isomorphism g maps any basis f of Eb
to another basis of Eb . This allows to consider the following right action of GL(V ) on
Fb E :

Fb E × GL(V ) −→ Fb E
.
(f, g) 7−→ f · g := {g ij fi }

12
 Basic linear algebra provides the following assertion :

∀(f, f 0 ) ∈ (Fb E)2 , ∃! g ∈ GL(V ), f 0 = f · g .

This exactly proves that the action of GL(V ) on Fb E is simply transitive.

Ultimately, let us consider the action on F E that restricts fiber-wise to that previ-
ously defined :

F × GL(V ) −→ F
.
(f ∈ Fb , g) 7−→ f · g ∈ Fb
By construction, this action is fiber-preserving and simply transitive. Therefore, F E
is a principal GL(V )-bundle. As stated earlier, this implies that GL(V ) is the structure
group of F E, and that any fiber Fb E is diffeomorphic to GL(V ).
We can picture this last point more explicitly by taking a section of F E. This
means that ∀b ∈ B, we choose a basis fb of Eb . Again, basic linear algebra provides the
following assertion :

∀f ∈ Fb E, ∃! gf ∈ GL(V ), f = fb · g .

This shows that there exists a one-to-one correspondence between the elements of
Fb E and those of GL(V ). Thus, the fiber over a point b ∈ B contains all the bases at
point b, and the right action of GL(V ) enables to change from a given basis to any
another basis at the same point b.
It is then always possible to build a principal GL(V )-bundle from any vector bundle
with fiber V : one only needs to consider the corresponding frame bundle as we just
did. Actually, instead of considering all the bases related to one another by an element
of GL(V ), we may only consider a set of bases related by elements picked in a fixed
subgroup G of GL(V ). This will define a principal G-bundle. Therefore, it is always
possible to build a principal G-bundle from any vector bundle with fiber V , where G is
a subgroup of GL(V ).
The frame bundle will be a useful version of the concept of principal bundle whenever
we will consider associated bundles.

1.1.3 Associated bundle

Our goal is to properly specify the scheme leading to the definition and the manip-
ulation of spinors over a spacetime which stands for the base of a vector bundle. To do
so, we mentioned that it is a requisite to build vector bundles such that :

• the structure group is the spin group ;

• the fibers are isomorphic to a representation space of the spin group.

Such a general construction is presented now, and allows to measure the essential role
played by principal bundles in physics.

13
1.1.3.1 Construction
Let G be the group that is needed to act as the gauge group of a given theory, and
let V be a representation space of G. This means that there exists a linear left action
of G on V :

G × V −→ V
.
(g, v) 7−→ g · v
The goal of this section is to build a vector bundle with fiber V and structure
group G. The systematic method to achieve this is to start from a principal G-bundle
(P, B, p, G), and appropriately combine it with the representation space V in order to
build a bundle with the right properties.
We remind that by definition, there exists a right action of G on P :

P × G −→ P
.
(p, g) 7−→ p · g
Since V needs to be the fiber of the vector bundle, a naive guess would be to consider
the trivial bundle B × V . However, as we mentioned earlier, the structure group of the
trivial bundle is {Id}, which would be wrong in the current case. The fibers need to be
tied together another way, so that the transition functions all take value in G.
The way to achieve this still starts with a Cartesian product, but it is P × V rather
than B × V . Then, the right way to attach the fibers together is to define the bundle
as a quotient space of P × V with respect to the following equivalence relation :

(a, v) ∼ (b, w) ⇐⇒ ∃g ∈ G, (b, w) = (a · g, g −1 · v) .

Let us denote this quotient space E := (P × V )/ ∼.

In addition, we remark that, if (a, v) and (b, w) are two representatives of the same
equivalence class, then p(a) = p(b). This is due to the fact that there exists g ∈ G such
that b = a · g. Since the action of G on P is fiber-preserving, a · g necessarily belongs
to the same fiber as a. This exactly means that a and b are in the same fiber of P ,
so that they obviously project onto the same element of B through p. Therefore, the
projection p : P → B may be extended to a map pE : E → B in the following way :
for any equivalence class e ∈ E, we pick any of its representative (a, v) and we set
pE (e) := p(a). Thanks to the previous remark, this map is well defined.

Definition. (E, B, pE ) defines a vector bundle with fiber V and structure group G,
which is called the associated bundle to (P, B, p, G) through the action of G on V .

At first sight, the equivalence relation used to define the total space E does not
seem obvious in any manner. Actually, there is a very intuitive way to picture this
equivalence relation as naturally arising from considerations that are very familiar from
linear algebra. This image makes use of the notion of frame bundle which was introduced
earlier.
Since V is a representation space of G, then G is isomorphic to a subgroup of GL(V ),
which we denote G0 . Let us consider the frame bundle F E built from E where all the
bases of a given fiber are related via an element of G0 . It is a principal G0 -bundle,
and hence a principal G-bundle since G identifies to G0 according to the isomorphism
mentioned above.

14
 In this situation, ∀b ∈ B, the fibers Pb and Fb E have exactly same structure. This
allows to identify P to the frame bundle F E. This point is really important and will
be very useful in future considerations.
Similarly, let us identify V to Kn , where K = R or C whether V is a real or complex
vector space respectively. In this sense, we may think of the elements of V merely as
n-tuples.
With this in mind, an element (a, v) ∈ P × V is pictured as a basis a associated
with a n-tuple v. Interestingly, this is exactly how one refers to the elements of a vector
space when they are described in terms of components : once a basis of the vector
space is picked, the vectors may be identified with their components with respect to
this particular basis.
Now, let us consider another basis b related to a through the action of the iso-
morphism g ∈ G, that is b = a · g. In the language of linear algebra, we would say
that g is the matrix that encodes the change of basis from a to b. In this situation,
we know from linear algebra that the vector that has components v in the basis a
has coordinates v 0 = g −1 · v in the new basis b. We may then refer to this vector as
(b, v 0 ) = (a · g, v 0 = g −1 · v) as well as (a, v). Therefore, any element of P × V related to
(a, v) through the equivalence relation used to define an associated bundle refers to the
same vector as (a, v).
In fact, any vector in the fibers of the associated bundle was defined as the equiv-
alence class made of all the equivalent descriptions of this vector in terms of a basis
combined with a n-tuple. With this perspective, the construction of an associated bun-
dle seems very natural. Moreover, the importance of the concept of principal G-bundle
is now clear, since it allows to systematically build out any vector bundle such that
its fiber is a representation space of G and its structure group is G. In particular, this
makes it possible to act on the associated bundle with G, which is fundamental in the
gauge theories.

1.1.3.2 Spin structure and spinors

Based on the previous construction, it always seems possible to build the exact
space needed over any point of a manifold. This would mean that any mathematical
object could live over this point and be acted upon by the desired gauge group. It is
not quite true, particularly when dealing with spinors. Actually, one needs to be really
cautious when trying to consider spinors on spacetime. In fact, the existence of spinors
on a manifold is subject to a precise condition, which is the possibility to define what
is called a spin structure over the manifold.
More explicitly, let us assume that one wants to consider a spinor representation of
a general spin group Spin(n, p) with representation space V . The proper procedure to
do it is :

1. consider a principal Spin(n, p)-bundle with base B, which we denote Spin(B) ;

2. define the associated bundle Spin(B, V ) built as a quotient space of Spin(B) × V

with respect to the equivalence relation mentioned in the previous section.

A bundle such as Spin(B, V ) is called a spinor bundle, since any of its sections
precisely defines the desired type of spinor in any spacetime point. Yet, before reaching

15
the second step of the procedure, there is an important constraint in getting the right
principal bundle Spin(B). This is the stage where spin structures intervene.
The group Spin(n, p) is the double cover of SO(n, p). In order to build a principal
Spin(n, p)-bundle with base B, denoted Spin(B), one actually needs to start from
a principal SO(n, p)-bundle, denoted SO(B). The latter may then induce Spin(B)
through a particular structuring map c. This leads to the concept of spin structure as
follows.

Definition. Let pSO and pSpin be the projections of SO(B) and Spin(B) respectively.
A spin structure on B is a triplet (SO(B), Spin(B), c) where

c : Spin(B) −→ SO(B)

is such that

• pSO ◦ c = pSpin ;

• it restricts fiber-wise to the double cover of SO(n, p) by Spin(n, p).

In particular, we have the following commutative diagram.

c
Spin(B) SO(B)

pSpin
pSO
B

Such structure is fundamental since the map c gives the way to induce the transition
functions of Spin(B) from those of SO(B). The fiber bundle construction theorem
showed how the transition functions determine the entire structure of a bundle, which
is why a spin structure is so important. This is also why it may get complicated, or even
impossible, to find a spin structure. In fact, a set of functions needs to satisfy stringent
conditions in order to be the set of the transition functions of a bundle. Therefore, the
application c must preserve these conditions since it is expected to induce the transition
functions of Spin(B) from those of SO(B). If it does not, then the induced functions
will not have the properties needed for Spin(B) to be a principal Spin(n, p)-bundle.
This is why such a sophisticated application c may not be easily determined, and may
actually not exist.
These constraints are formalised into a theorem [5] involving elaborated notions
of topology such as the Stiefel-Whitney classes of a bundle. We will not go into such
details. Instead, we will only say that the theorem states that the existence of a spin
structure over a manifold depends on certain topological properties of the considered
manifold. Naively, if B has a simple topology, there is no difficulty in building a spin
structure, and hence considering spinors. But things are different when B is not trivial.
As a result, a spin structure does not always exist on a given manifold, and hence a
spacetime may not admit spinors at all. Of course, in the remaining of the dissertation,
we will implicitly assume that the considered spacetimes admit a spin structure, in
order to go further into the theory of spinor bundles.

16
 To summarize, this section presented the systematic path to follow in order to prop-
erly consider spinors on any spacetime : firstly, principal bundles need to be introduced
as a fundamental type of fiber bundle ; then, when a spin structure may be defined
over a given spacetime, it is possible to consider a principal Spin(n, p)-bundle over this
spacetime ; this finally allows to construct spinor bundles as associated bundles to the
principal Spin(n, p)-bundle, and then define spinor fields as sections of spinor bundles.
However, along this path, it is remarkable to note that spinors cannot exist on certain
manifolds.

1.2 Spin connection

Spinors are central objects of the standard model, which is defined in Minkowski
spacetime M4 . Yet, as is well known, general relativity claims that any source of energy
and momentum, such as matter, radiation, or even a cosmological constant, would make
the shape of the Universe different from M4 . The previous sections showed that defining
spinors on such a spacetime requires new mathematical structures which are not needed
when working in M4 . In the same way, studying the spacetime variations of a field is
more complicated on a general manifold than on M4 . It requires new tools, such as a
covariant derivative, which is familiar from general relativity. Yet, in general relativity,
the covariant derivative is only defined for tensor fields. Of course here, we will need it
to be able to measure the variations of a spinor field, that is evaluate the dynamics of
a section of a spinor bundle.
In particular, a difficulty that appears when a covariant derivative is needed, is that
there are infinitely many ways to build it. This is due to the fact that, on a bundle, there
is no natural or privileged way to define it. As a result, defining a covariant derivative
is partly arbitrary since it amounts to set what will be considered as a variation or not
from then on. For instance, in a given point, a field may have a vanishing covariant
derivative while it has non-zero variations for another covariant derivative. Eventually,
it is the task of physicists to determine the right covariant derivative, in order for the
theory to agree with experimental results.

1.2.1 Principal connection

Again, principal bundles will be of fundamental interest in this section. Formalising
what is called a connection on a principal bundle is an essential step toward specify-
ing covariant derivatives for spinors. It is easily understandable : in the same manner
that associated bundles are built out from a principal bundle, covariant derivatives in
associated bundles are induced from a principal connection. Prescribing a principal con-
nection makes use of the concepts of vertical and horizontal subbundles, which are now
introduced.

1.2.1.1 Vertical bundle

The total space E of a fiber bundle (E, B, p) is a manifold itself, so that it stands
for the base of its tangent bundle (T E, E, πE ). The push-forward of the projection of
(E, B, p) is p∗ : T E → T B, which gives a linear approximation of p.

17
 Intuitively, the tangent space to E allows to perform infinitesimal displacements
within E. More precisely, for e ∈ E, a tangent vector t ∈ Te E has its origin in e and
points toward a neighbouring point e0 ∈ E such that “e0 = e + t”. Then, p∗ (t) indicates
the resulting displacement in B. That is p∗ (t) is the tangent vector to B in p(e) such
that “p(e0 ) = p(e) + p∗ (t)”. If e0 is not in the same fiber as e, then p(e0 ) 6= p(e) so that
p∗ (t) 6= 0. But if e0 and e belong to the same fiber, then

p∗ (t) ∈ ker p∗|T .

eE

The space ker p∗|T E is a vector subspace of Te E. It is called the vertical subspace
e
in e and is usually denoted Ve E. It contains all the tangent vector in e which induce
infinitesimal displacement within the fiber to which e belongs. In this regard, the vector
subspace is pictured as the subspace made of all the vectors “tangent to the fiber‘”.
The following set [
V E := Ve E
e∈E

defines the total space of a vector bundle with base E, which is called the vertical
bundle.
Let us state two interesting facts about the concept of vertical subspace.

1. In any point e of the total space of a vector bundle, the vertical subspace is
isomorphic to the fiber to which it is “tangent”, that is the fiber to which e belongs.

2. The Lie algebra g of a Lie group G is defined as Ti G where i is the identity element
of G. Using the left-invariant vector fields on G, one shows that

∀g ∈ G, Tg G ' g .

In addition, it was already mentioned that the fibers of a principal G-bundle

(P, B, p, G) are diffeomorphic to G. Then, for e ∈ P , the “tangent” space Ve P to
the fiber Pp(e) := p−1 ({p(e)}) which contains e will be isomorphic to g.

Finally, let (P, B, p, G) be a principal G-bundle, and (E, B, p0 , V ) be an associated

bundle, where V is a representation space of G. We may choose to see (P, B, p, G) as a
frame bundle associated to (E, B, p0 , V ). The fiber of this frame bundle over any b ∈ B
is a set of bases such that the bases are related by the elements of G seen as a subgroup
of GL(V ).
In this view, a displacement within a fiber corresponds to shifting from a basis at
a point b ∈ B to another basis at the same point b. In order to use a terminology
closer to that used in classical mechanics, we may call such a displacement a “rotation”,
since it only linearly transforms bases to one another without moving their origin b.
Also, changing from a given fiber of the frame bundle to another fiber, without doing
a rotation before or after the change of fiber, may be called a “translation”, since it
corresponds to changing the origin of the considered bases without rotating. A general
displacement in the frame bundle is therefore a combination of a rotation and a transla-
tion. These naive denominations are rigorously formalised in the next few sections, due
to their bond to the concept of connection, but it is interesting to get some intuitive
pictures already.

18
 Figure 1.3 – Vertical and horizontal subbundles. [4]

1.2.1.2 Equivariant horizontal distribution

There is a clear, unique way to define the vertical bundle of a given fiber bundle
(E, B, p). In any point e ∈ E, the vertical space is a subspace of Te E which allows to
perform infinitesimal displacement only within the fiber p−1 ({p(e)}). It is then tempting
to look for a space which would play an analogous role for infinitesimal displacements
between neighbouring fibers. Intuitively a general displacement in the frame bundle is
a combination of a rotation and a translation, that is a purely vertical displacement
associated with a change of fiber. There is yet no tool to extract the decomposition of
a given displacement in terms of its purely rotational part and its translational part.
In the frame bundle picture of a principal bundle, this means that, as soon as we
change from a fiber over b to a fiber over b0 , there is no way to tell if the basis in b was
merely translated to b0 parallel to itself, or if it also had to be rotated to match the
basis in b0 . This lack of information is due to the nature of fiber bundles which is such
that there is no obvious and natural way to define such a decomposition.
Therefore, in order to get an explicit decomposition of displacements, one has to
artificially prescribe which displacements will be considered as pure translations between
neighbouring fibers. This amounts to picking a vector space He E for any e ∈ E such
that
Te E = Ve E ⊕ He E .
Such a vector subspace of Te E is naturally called the horizontal subspace in e, and
the set [
HE := He E
e∈E
defines the total space of a vector bundle with base E, which is called the horizontal
bundle. The situation is pictured in figure 1.3.
Contrary to the definition of the vertical bundle, the choice of the horizontal bundle
is clearly arbitrary, but it finally gives meaning to the notion of purely translational
displacement between neighbouring fibers. In the frame bundle picture of a principal
bundle, it is now possible to say that, if the displacement from a basis to another basis
is described by a vector in HE, then it is a pure translation, and no rotation had to be
operated.

19
 Although there are infinitely many admissible choices of horizontal bundles, one
may prefer to set certain constraints on the distribution of horizontal spaces, such as
the property of equivariance. As we shall see, the latter provides the horizontal bundle
a coherent structure relevant for the construction of principal connections.

Definition. Let (P, B, p, G) be a principal G-bundle. By definition of a principal bundle,

∀g ∈ G, there exists a map


Rg : P −→ P
e 7−→ e·g
whose push-forward is such that

∀e ∈ P, Rg∗e : Te P −→ Te · g P .

A distribution HP of horizontal subspaces of (P, B, p, G) is said G-equivariant if

and only if
∀e ∈ P, Rg∗e (He P ) = He · g P .

Therefore, a G-equivariant horizontal bundle is such that, if a displacement vector

is horizontal in a point e ∈ P , then for any g ∈ G, its push-forward through Rg∗ must
be horizontal as well in e · g. The property of G-equivariance thus provides a horizontal
bundle with a strongly consistent structure.
We remind that this section focuses on a way to define variations in a fiber bundle.
Intuitively, we want to prescribe that translational displacements, that is horizontal
tangent vectors, correspond to null variation. In the frame bundle picture, if two neigh-
bouring bases are separated by a translational displacement, it means that one is ob-
tained by merely translating the other parallel to itself, without operating any rotation.
This matches with what one would reasonably expect as a definition of a uniform dis-
tribution of bases on a manifold : all neighbouring bases of the distribution are related
by pure translations. Then, what will be considered as a variation when considering two
neighbouring bases will be the pure vertical part of the displacement vector separating
the two bases, that is the rotational part.
Since the tangent space T P may now be written as a direct sum of a vertical subspace
and a horizontal subspace in any point, it is easy to remove the horizontal part of any
tangent vector in order to extract its vertical part. In any point, this is achieved by the
projection onto the vertical subspace along the horizontal subspace. Such a projection
defines what is generally called a principal connection.

Definition. A principal connection on (P, B, p, G) is a map C : T P → T P such that

• C2 = C ;

• ∀e ∈ E, C(Te E) = Ve E ;

• C is G-equivariant, that is ∀g ∈ G, Rg∗ ◦ C = C ◦ Rg∗ .

A principal connection is equivalent to a G-equivariant horizontal bundle. On one

hand, given a G-equivariant horizontal bundle, one naturally defines a principal con-
nection as was done above. On the other hand, given a principal connection C, the

20
set ker C|Te E defines a horizontal subspace in any e ∈ P , and these horizontal sub-
spaces induce a G-equivariant horizontal distribution.

These various notions allow us to compare elements in different fibers although there
is no natural way to do this with the bare structure of fiber bundles. This task is rather
arbitrarily achieved by defining horizontal distributions, which allow to identify two
neighbouring elements of the total space if they are separated by a horizontal vector.
As a result, when moving from an element to another, it is considered that there is
variation if the vertical part of the displacement vector is non vanishing, that is if there
is a concrete variation within the fibers.
In particular, let us consider a section of a frame bundle. This sets one basis in
any point of the base. This exactly matches the familiar idea of a mobile frame on a
spacetime. If the principal connection on this distribution of bases is non-vanishing, it
means that rotations are needed in addition to translations in order to change from
certain bases to some others sitting other points.
In order to coincide with familiar terminology, if the bases of a neighbourhood are
all separated by translations, we may say that the distribution of bases is uniform in
this neighbourhood. A particular choice of horizontal bundle may make a distribution
of bases look uniform in certain points, while another would not.

1.2.2 Connection 1-form

Our objective is to define a way to evaluate the variations of a spinor field, which
is a section of a spinor bundle. To this end, the notion of principal connection, which
is originally defined on a principal bundle, is exploited to define variations on all the
associated bundles of the principal bundle.
The first step is to define the connection 1-form, which relates the principal connec-
tion very closely through the concept of fundamental fields. We will first need a brief
introduction to the notion of vector-valued differential form. It is a natural extension
of differential forms which allows a form to be vector-valued rather than only scalar-
valued, as it is usually the case. It will also be particularly useful when we will discuss
the concept of covariant exterior derivative further on.

1.2.2.1 Vector-valued differential forms

A differential form ω of degree p ∈ N provides an alternating form ωm of degree p
in any point m of a manifold M . The form ωm takes p tangent vectors of Tm M as
arguments, and is scalar-valued (the scalars being elements of R). The set of all the
(scalar-valued) differential forms of degree p on M is denoted Ωp (M ). From this, the
vector-valued differential forms are then defined as follows.

Definition. Let V be a vector space. The set Ωp (M, V ) of the V -valued differential
forms is defined as the following tensor product :

Ωp (M, V ) := Ωp (M ) ⊗ V .

More precisely, an element ω ∈ Ωp (M, V ) may always be decomposed as

ω = ω r ⊗ r ,

21
where ω r ∈ Ωp (M ) and {r } is a basis of V . Then, ∀m ∈ M , ωm takes p tangent vectors
(t1 , ..., tp ) ∈ (Tm M )p as arguments. The quantities ωm r
(t1 , ..., tp ) are scalars, which will
give the decomposition of ωm (t1 , ..., tp ) on the basis {r } of V :
r
ωm (t1 , ..., tp ) = ωm (t1 , ..., tp ) r .

Writing any ω ∈ Ωp (M, V ) as ω = ω r ⊗ r allows to naturally extend the wedge

product with an ordinary differential form ξ as follows :

∀ω ∈ Ωp (M, V ), ∀ξ ∈ Ω(M ), ω ∧ ξ := (ω r ∧ ξ) ⊗ r .

Therefore, if ξ ∈ Ωq (M ), then ω r ∧ ξ ∈ Ωp+q (M ), so that ω ∧ ξ ∈ Ωp+q (M, V ).

Similarly, the decomposition ω = ω r ⊗ r allows to extend the familiar exterior
derivative d defined on Ωp (M ) for any p. We temporarily denote this extension dV ,
which is defined on Ωp (M, V ) for any p.

Definition. For ω = ω r ⊗ r ∈ Ωp (M, V ), the exterior derivative is defined as

dV (ω) := d(ω r ) ⊗ r ,

where, in the right-hand side term, d designates the familiar exterior derivative defined
over the scalar-valued differential forms.

It is worth noting that, since ω r ∈ Ωp (M ), then d(ω r ) ∈ Ωp+1 (M ), so that dV ω ∈

Ωp+1 (M ) ⊗ V = Ωp+1 (M, V ). This does generalise the fact that the usual exterior
derivative satisfies d(Ωp (M )) ⊂ Ωp+1 (M ).
The extended exterior derivative dV naturally inherits the usual properties of d :

• dV is R-linear ;

• dV is an anti-derivation, that is :

∀ω ∈ Ωp (M, V ), ∀ξ ∈ Ω(M ), d(ω ∧ ξ) = d(ω) ∧ ξ + (−1)p ω ∧ d(ξ) ;

• d2V = 0.

For these reasons, and the fact that d is recovered as a particular case of dV when
V = R, dV will be denoted d as well.
The denomination vector-valued differential form is then justified since ω acts much
like a standard differential form, except in any point of the manifold M , it will take
values in the vector space V .

1.2.2.2 Fundamental fields

We mentioned that, in any point of a principal G-bundle (P, B, p, G), the vertical
subspace is isomorphic to the Lie algebra g of G. Let us introduce a map that allows
to make explicit such relation between g and Ve E for some e ∈ P . The right action of
G on P

P × G −→ P
(p, g) 7−→ p · g

22
allows to define the following map for any e ∈ P :


Le : G −→ P .
g 7−→ p·g
In particular, its push-forward Le∗ : T G → T P in the identity element i of G
restricts to
Le∗ : Ti G = g −→ Te · i P = Te P .
Let us picture more concretely how this push-forward relates the elements of g to
the elements of Te P . Consider an element J ∈ g, seen as an element of Ti G. It is a
displacement vector on G with origin in i, and it is pointing at an element g ∈ G in the
neighbourhood of i. Then, Le∗ (J) is the displacement vector from e to e · g. In particular,
e · g is in the same fiber of e since the action on a principal bundle is fiber-preserving.
Therefore, the displacement vector Le∗ (J) is necessarily vertical. More precisely, the
push-forward Le∗ is an isomorphism of vector spaces from g onto Ve P . This allows us
to define the notion of fundamental fields.

Definition. Let us take a set {Jr } of infinitesimal generators of G, that is a basis of

g. Let us denote {jr (e) := Le∗ (Jr )} the corresponding basis of Ve P obtained through the
push-forward Le∗ .
The vertical vector fields over P


jr : P −→ VP
e 7−→ jr (e)

are called the fundamental fields associated to the generators {Jr }.

Let us now introduce the connection form, whose action will then be written in
terms of fundamental fields.

Definition. For any e ∈ P , the principal connection C : T P → T P is such that

Im C|Te P = Ve P .

In addition, we may define an inverse of Le∗ on Ve P , which we simply denote L−1

e∗ .
We may then naturally define

ωe := L−1
e∗ ◦ C|Te P .

This defines a field of 1-forms on P with values in g, called the connection form
ω : e ∈ P 7→ ωe .

Therefore, the connection form is a g-valued differential form on P of degree 1, that

is ω ∈ Ω1 (P, g).
We may use the fundamental fields to make clearer the effects of ω. For any point
e ∈ P , ωe acts upon any tangent vector t ∈ Te P . Since Te P = Ve P ⊕ He P , there exists
a unique decomposition of t :

t = v + h, v ∈ Ve P, h ∈ He P .

23
 A set of fundamental fields {jr } associated to a set of generators {Jr } provides a
basis of Ve P , which is {jr (e)}. Then, v decomposes on this basis as v = v r jr , so that
t = v r jr + h. As a result, we find
ωe (v) = L−1 r
e∗ ( v C(jr ) + C(h) )
= L−1 r
e∗ ( v jr + 0 )
= v r L−1
e∗ (jr )
r
= v Jr .
In particular, ker ωe = He P , which reminds that the connection form also depends
on the arbitrary choice of the equivariant horizontal bundle HP .

1.2.3 Connections in associated bundles

All the notions introduced so far were only meant to allow for the general definition
of a connection on an associated bundle. Such a connection is expected to be an operator
evaluating the variations of the fields over a given spacetime. In particular, this notion
will enable us to specify the spin connection used in Einstein-Cartan theory in the next
chapter.

1.2.3.1 Gauge potential

The isomorphism between g and Ve P is essential since it allows to write all the
important tools that we will need in terms of infinitesimal generators. The first tool
now introduced is the gauge potential, which is ubiquitous in the gauge theories of the
standard model for instance. It is a new step toward building an operator measuring
the variations of the fields over a general spacetime. More precisely, the gauge potential
allows to evaluate the variations of a mobile frame defined over any associated bundle
of the principal G-bundle (P, B, p, G).
Let us start with the proper principal bundle itself and consider one of its section :


e: B −→ P .
b 7−→ e(b) ∈ Pb := p−1 ({b})
Analysing the variations of a section is asking the following question : being in a
point b ∈ B, the value of the section is e(b) ; then, what happens to the section when
moving from b to a neighbouring point b0 ?
As stated earlier, moving from a point b to a point b0 of the base B is easily formalised
with the help of the tangent bundle T B. Then, moving from b to b0 =“b + z” via the
displacement vector z ∈ Tb B induces a displacement vector from e(b) to e(b0 ), which is
given by e∗b (z) ∈ Te(b) P . At this stage, it is naturally tempting to apply the connection
form to e∗b (z), since the connection form is precisely made to extract the variations
contained in the elements of T P .
Definition. Let us consider a section e ∈ Γ(P ). The gauge potential A ∈ Ω1 (B, g)
relative to e is defined as :


A: TB −→ g .
z ∈ Tb B 7−→ ωe(b) (e∗b (z))

24
 We may always decompose it as

A = Ar ⊗ Jr ,

where Ar ∈ Ω1 (B) and {Jr } is a basis of g.

Let {eI } be a basis of Ω1 (B). We may thus decompose any Ar as

Ar = AI r eI ,

where AI r ∈ F(B), that is a function on B. Finally, we write

A = AI r eI ⊗ Jr .

Let us picture what information the gauge potential provides. Firstly, let us remind
that the horizontal bundle identifies elements of neighbouring fibers separated by purely
translational displacements. This allows to compare elements of different fibers and
deduce variations from these comparisons. Secondly, for z ∈ Tb B, A(z) ∈ g points from
the identity i of G toward the element g ∈ G such that e(b) · g is horizontally separated
from e(“b + z”). And a horizontal separation is considered as an absence of variation.
Therefore, A(z) really contains the information about the pure variation between b and
“b + z”.

Let us adapt the definition of the gauge potential to any associated bundle (E, B, p, V )
of (P, B, p, G). This is achieved by making use of the frame bundle F E associated to
(E, B, p, V ). The vector space V is a representation space of G. This means that G iden-
tifies with a subgroup G0 ⊂ GL(V ) through a group homomorphism ρ : G → GL(V ).
From the theory of Lie groups, the Lie group representation ρ induces a Lie algebra
representation ρ̂ : g → gl(V ), so that g identifies with g0 .
We remind that, for b ∈ B, any basis f ∈ Fb E may be identified with an element
e ∈ Pb . This allows to picture the variations in terms of rotations and translations of
bases. In this view, if two neighbouring bases are separated by a horizontal displacement
vector, then it is considered that there is no variation when changing from one to the
other. Then, the horizontal bundle identifies the neighbouring bases that are considered
to be simple parallel translations of each other.
Let us consider a mobile frame f ∈ Γ(F E). Let e ∈ Γ(P ) be the underlying section
of P such that, ∀b ∈ B, f (b) is the basis naturally identified to e(b) when we identify F E
to P . Let b and b0 be two neighbouring points in B separated by a displacement vector
z ∈ Tb B. In order to get f (b0 ) from f (b), we generally need a rotation and a translation.
In this situation, a natural extension of the gauge potential to F E would extract the
information about the rotation part from z. Such an extension will be denoted A0 .
According to previous considerations, the original gauge potential A is such that
A(z) ∈ g points at the element g ∈ G such that e(b) · g is horizontally separated from
e(b0 ). Then, ρ̂(A(z)) ∈ g0 merely points at the element g 0 ∈ G0 such that f (b) · g 0 is only
separated from e(b0 ) by a translation. In other terms, ρ̂(A(z)) points at the rotation of
G0 needed to be performed on the basis f (b) in order to get a basis in b that is parallel
to the basis in b0 . Therefore, ρ̂(A(z)) plays the exact same role as the original gauge
potential, only adapted for the associated bundle E.

Definition. The gauge potential A0 relative to f is the element of Ω1 (B, g0 ) defined as :

25
 A0 : g0

TB −→ .
z ∈ Tb B 7−→ ρ̂(A(z))

In particular, the expression of the gauge potential A0 in the basis {eI ⊗ Jr0 } of
Ω1 (B, g0 ) is read from the following computation :

∀z ∈ T B, A0 (z) = ρ̂( AI r [eI ⊗ Jr ](z) )

= ρ̂( AI r eI (z) Jr )
= ρ̂( AI r z I Jr )
= AI r z I ρ̂(Jr )
= AI r eI (z)Jr0
= AI r [eI ⊗ Jr0 ](z) ,

where {Jr0 := ρ̂(Jr )} is the basis of g0 induced from the basis {Jr } of g.

Therefore, A0 = AI r eI ⊗Jr0 has the exact same expression as A, where the elements Jr
are simply replaced by their image through ρ̂. In particular, the fields AI r ∈ F(B) are
identical to those appearing in the original gauge potential. These fields are essential
in the standard model since they embody the vector bosons such as the gluons of the
strong interaction, or the photon of quantum electrodynamics.
It is worth stressing that the gauge potential A depends on the choice of the sec-
tion e. Such a choice is commonly called a gauge choice, since fixing a gauge in a gauge
theory actually corresponds to choosing a basis in any spacetime point. If we change
the section e, then the transformation law for the fields AI r has the exact form of the
usual gauge transformations of the standard model.

1.2.3.2 Covariant derivative

First of all, let us stress out that we will use the following rule for the choice of the
indices from now on :

• α, β, γ, , refer to the fibers, so that they run from 1 to dim V ;

• r refers to the Lie algebra, and runs from 1 to dim g0 ;

• I, J, K, L, M refer to the base B, and then its tangent spaces ; conventionally,

these run from 0 to dim B − 1 rather than from 1 to dim B ;

• µ, ν, ρ, σ, τ also refer to the base and its tangent spaces, but we reserve them for
objects relating to a coordinate chart, such as the coordinate basis associated to
a coordinate chart ; they obviously run from 0 to dim B − 1 as well.

Let us now consider a vector bundle (E, B, p, V ) associated to a principal G-bundle

(P, B, p, G). Let f be a mobile frame of (E, B, p, V ). The gauge potential A0 relative
to f extracts the variations of f in terms of infinitesimal generators. Yet, the ultimate
goal is to evaluate the variations of a vector field w ∈ Γ(E). Then, we will naturally
aim at generalising the following situation.

26
 In d-dimensional Euclidean space Ed , the variations of a vector field w may be
obtained by taking the derivatives of w component-wise. This defines the Jacobian ma-
trix J[w] relative to w in any point b ∈ Ed . If z is a d-dimensional vector, then Jz [w](b) :=
J[w](b) · z exactly corresponds to the variation of the field w at point b in the direction z
of Ed , since it satisfies

w(b + z) = w(b) + J[w](b) · z + o(||z||) .

In particular, Jz [w] is a vector field of the same nature as w, which is just the linear
approximation of the vector difference between w(b + z) and w(b)
Therefore, we want to define an operator ∇ which generalises the Jacobian opera-
tor J for any associated bundle. The operator ∇ will naturally have two arguments :
the field w ∈ Γ(E) that we want to analyse, and the field z ∈ Γ(T B) which provides
the direction in which we want to analyse w in every point b ∈ B. The image should be
a vector field of the same nature as w, that is ∇z w ∈ Γ(E). Even more natural is the
fact that ∇ will be linear in the direction field z and that it will reproduce the Leibniz
rule on the second argument if the field is multiplied by a scalar function h ∈ F(B).
Such operator is obviously known as a covariant derivative.
Thanks to the previous requirements on the covariant derivative, setting its values
on a finite number of relevant fields defines it entirely. More exactly, we will only need
to define the variations of basis elements in the different independent directions of B.
This is where the gauge potential A0 will be of utmost interest.
In any point b ∈ B, the mobile frame provides a basis f (b) of Eb . Let us denote fα (b)
the basis vectors in b such that f (b) = {fα (b)}. Likewise, a mobile frame {eI } of T B
provides a set of basis vectors {eI (b)} of Tb B, while the set of dual basis vector fields {eI }
provides a basis of Ω1 (B).
We now try to set the quantities ∇eI fα in any point b ∈ B using the following
arguments. We remind that A0 (eI (b)) points at the element g 0 ∈ G0 such that f 0 (b) :=
f (b) · g 0 is parallel to the neighbouring basis in the direction eI (b). Using the exponential
map defined in the context of Lie groups, we have :

g 0 = exp[A0 (eI (b))] .

If we Taylor expand the exponential acting on any vector basis fα (b), we get

g 0 [fα (b)] = IdEb [fα (b)] + A0 (eI (b))[fα (b)] + o(||fα (b)||)
= fα (b) + A0 (eI (b))[fα (b)] + o(||fα (b)||) .

Yet, g 0 (fα (b)) = fα0 (b) so that A0 (eI (b))[fα (b)] is simply the linear approximation
of the vector difference between fα0 (b) and fα (b). Since f 0 (b) identifies with the neigh-
bouring basis in the direction eI (b), then fα0 (b) identifies with the αth vector of the
neighbouring basis. Therefore, A0 (eI (b))[fα (b)] exactly corresponds to the vector varia-
tion between fα (b) and the vector it becomes when moving in the direction eI (b). This
is exactly the object we were looking for, so that we set

∀b ∈ B, ∇eI fα (b) := A0 (eI (b))[fα (b)] .

Now that the variations of the basis vector fields fα are defined, the variations of
any field naturally come : a field is constant if and only if it is constant with respect

27
to any uniform mobile frame. In particular, if it is constant with respect to a non
uniform mobile frame, then its variation will be those of the mobile frame. These ideas
are already implemented in the covariant derivative since they are exactly formalised
by the Leibniz rule. Then, to extract the variations of a field, one only has to project
the field onto the basis vector fields and apply the Leibniz rule : for w ∈ Γ(E) and
z ∈ Γ(T B), we have w = wα fα and z = z I eI , where the wα ’s and the z I ’s are scalar
functions on B. Then,
∇z w = z I ∇eI (wα fα )
= z I (eI [wα ]fα + wα ∇eI fα ) ,
where the last expression is merely obtained by applying the Leibniz rule. Every term
of this expansion of the covariant derivative is now explicitly defined. In particular, we
remind that eI [wα ] denotes the directional derivative of the scalar function wα in the
direction eI .
To be more precise, let us pick a local coordinate chart {xµ }. We will denote {eµ }
the associate coordinate basis. This might be confusing since, if we explicitly write e1 ,
we cannot know if we are referring to the first basis vector field of {eI } or the first
vector of {eµ }. To avoid this confusion, we should for instance denote the first basis
{hI } rather than {eI }. However, referring to both bases with the letter e is the choice
that is always made in the literature devoted to Einstein-Cartan theory. Since one rarely
manipulates any isolated basis vector with an explicit number as index, the confusion
is avoided thanks to the fact that the indices are always µ, ν,... for the coordinate basis
and I, J,... for the other one. We will follow this convention from now on.
We may change from one basis to another according to a matrix eI µ such that
eI = eI µ eµ .
Then, the following relation holds :
eI [wα ] = eI µ ∂µ [wα ] .
However, instead of the term ∇eI fα , we usually manipulate what are called the
connection coefficients, which are derived as follows. Let us denote {f α (b)} the dual
basis of {fα (b)}. Since g0 ⊂ gl(V ), any element Jr0 acts linearly on Eb ' V . We thus
write Jr0 as any endomorphism of Eb :
Jr0 = (Jr0 )αβ fα (b) ⊗ f β (b) .
The gauge potential may then be rewritten as
A0 = AI r eI ⊗ Jr0
= AI r eI ⊗ [(Jr0 )αβ fα ⊗ f β ]
= AI αβ eI ⊗ fα ⊗ f β ,
where AI αβ := AI r (Jr0 )αβ are the connection coefficients.
Therefore, we derive
∇eI fα = A0 (eI (b))[fα ]
= AJ γ  eJ (eI ) f  (fα ) fγ
= AJ γ  δIJ δα fγ
= AI γ α fγ ,

28
so that :

∇z w = z I (eI [wγ ]fγ + wα AI γ α fγ )

= z I (eI [wγ ] + AI γ α wα )fγ .

In a coordinate basis {eµ }, we recover the usual expression familiar from general
relativity and the standard model :

∇z w = z µ (∂µ [wγ ] + Aµγ α wα )fγ .

We thus know how to explicitly build the operator needed to analyse the dynamics
of the sections of any associated bundle. As one knows from the standard model, it is
also a particularly useful tool when one wants to construct a gauge-invariant action.
In the next chapter, we will consider spinor bundles associated to a principal Spin(3, 1)-
bundle. Their fibers will be isomorphic to the representation space whose elements are
the 4-component Dirac spinors. In this situation, we will apply the previous procedure
to explicitly build the appropriate covariant derivative for spinor fields, also known as
spin connection.

1.3 Connection-related objects

In the next chapter, we will exploit the various notions introduced before in the
context of Einstein-Cartan theory of gravity. For this purpose, we will need three addi-
tional operators which are based on the covariant derivative. The first one is the exterior
covariant derivative which generalises the covariant derivative in the same manner that
exterior derivative generalises the differential of functions. The second one, which is
based on the exterior covariant derivative, is the curvature, very familiar from general
relativity and the gauge theories. Lastly, when the tangent bundle may be seen as an
associated bundle, the notion of torsion may be defined, and it is central in Einstein-
Cartan theory.

1.3.1 Exterior covariant derivative

In order to introduce the exterior covariant derivative, we first briefly define bundle-
valued differential forms, which generalise one step further the vector-valued differential
forms.

1.3.1.1 Bundle-valued differential forms

Definition. Let (E, B, p, V ) be a vector bundle. The set Ωp (B, E) of the E-valued dif-
ferential forms is defined as the following tensor product :

Ωp (B, E) := Ωp (B) ⊗ Γ(E) .

Let f = {fα } be a mobile frame. An element ω ∈ Ωp (B, E) may thus always be

decomposed as
ω = ω α ⊗ fα , ω α ∈ Ωp (B) .

29
 For b ∈ B, ωb takes p tangent vectors (t1 , ..., tp ) ∈ (Tb B)p as arguments. The quan-
tities ωbα (t1 , ..., tp ) are scalars, which will give the decomposition of ωb (t1 , ..., tp ) on the
basis {fα (b)} of Eb :
α
ωm (t1 , ..., tp ) = ωm (t1 , ..., tp ) fα .
The denomination bundle-valued differential form is then justified since ω acts much
like a standard differential form, except in any point of the base, it will take values in
the fiber of the bundle over this point.
We may again extend the familiar exterior derivative to Ωp (B, E) by merely applying
the usual exterior derivative to the left factors ω α of the tensor products. We may also
denote this extension d again, since it inherits all usual properties of d. And we may
recover the notion of vector-valued differential form if E is the trivial bundle B × V .
For instance, a vector field of (E, B, p, V ) is a E-valued differential form of degree 0,
i.e. Ω0 (B, E) = Γ(E). Moreover, for a vector field w ∈ Γ(E), the application

∇w : Γ(T B) −→ Γ(E)
z 7−→ ∇z w
is a E-valued differential form of degree 1. Therefore, the covariant derivative defines
the following interesting map which is at the basis of the exterior covariant derivative :

∇: Ω0 (B, E) −→ Ω1 (B, E) .
w 7−→ ∇w

1.3.1.2 The operator d∇

We remind that the usual exterior derivative d is defined as the unique R-linear
operator on Ω(B) such that :
• it coincides with the usual differential of functions on Ω0 (B) ;
• it is an anti-derivation, that is :
∀λ ∈ Ωp (B), ∀ξ ∈ Ω(B), d(λ ∧ ξ) = d(λ) ∧ ξ + (−1)p λ ∧ d(ξ) ;
• d(Ωp (B)) ⊂ Ωp+1 (B) ;
• d2 = 0.
The analogy is obvious with the exterior covariant derivative which is defined as
follows.
Definition. Let ∇ be the covariant derivative used on a vector bundle (E, B, p, V ). The
exterior covariant derivative d∇ is the unique R-linear operator on Ω(B, E) such that :
• it coincides with the usual covariant derivative on Ω0 (B) as introduced above ;
• it is an anti-derivation, that is :
∀λ ∈ Ωp (B, E), ∀ξ ∈ Ω(B), d∇ (λ ∧ ξ) = d∇ (λ) ∧ ξ + (−1)p λ ∧ d(ξ) ;
• d∇ (Ωp (B, E)) ⊂ Ωp+1 (B, E).
We note that there is no requirement analogous to the fact that the square of the
usual exterior derivative vanishes. In fact, (d∇ )2 does not vanish in general. This fact
actually allows for the existence of the curvature as we shall see now.

30
1.3.2 Curvature
Curvature is abundantly used in both the standard model and general relativity, and
hence Einstein-Cartan theory. This is owing to the fact that it is an important feature
of the connection defined on a given bundle. Let us explicitly derive the expressions and
properties of this object.

1.3.2.1 Definition
The exterior covariant derivative d∇ was built from the covariant derivative ∇ in
analogy with the standard exterior derivative d. However, these two notions of deriva-
tives do not share the exact same properties. We already mentioned that, unlike d,
(d∇ )2 : Ω0 (B, E) → Ω2 (B, E) does not vanish. Moreover, the operator (d∇ )2 has an-
other particularity : it is F(B)-linear although d∇ is not (it is only R-linear). This
proves that it is possible to decompose (d∇ )2 as follows :
(d∇ )2 : w ∈ Γ(E) 7−→ F αβ ⊗ [(fα ⊗ f β )(w)] ,
where the F αβ ’s are fixed elements of Ω2 (B).
Since [(fα ⊗ f β )(w)] is again a vector field of Γ(E), then (d∇ )2 (w) is an element of
Ω2 (B, E). This element linearly depends on w, as expected.
Definition. With the previous expression in mind, the F(B)-linear operator (d∇ )2 de-
fines the curvature and is denoted F .
The F αβ ’s are known as the curvature 2-forms.
We may derive the explicit expression of the 2-forms F αβ from quantities that are
already known. This is achieved by computing the explicit expression of (d∇ )2 . To do
so, we first remind that ∇eI fα = AI γ α fγ and that the covariant derivative is linear in
the first field. Therefore, we may write
∇fα = AI γ α eI ⊗ fγ .
We will use the set of 1-forms {Aαβ } defined by
Aαβ := AI αβ eI ∈ Ω1 (B) ,
so that
∇fβ = Aαβ ⊗ fα .
Then, we have
F (fβ ) = d∇ (d∇ (fβ ))
= d∇ (∇fβ )
= d∇ (Aαβ ⊗ fα ) .
Since fα ∈ Ω0 (B, E), we may write the last expression as d∇ (fα ∧ Aαβ ). This yields
F (fβ ) = d∇ (fα ) ∧ Aαβ + (−1)0 fα ∧ d(Aαβ )
= ∇(fα ) ∧ Aαβ + fα ∧ dAαβ
= Aγ α ⊗ fγ ∧ Aαβ + dAαβ ⊗ fα
= Aγ α ∧ Aαβ ⊗ fγ + dAαβ ⊗ fα
= (dAαβ + Aα ∧ A β ) ⊗ fα .

31
 Yet, we have

F (fβ ) = F αγ ⊗ [(fα ⊗ f γ )(fβ )]

= F αγ ⊗ [δβγ fα ]
= F αβ ⊗ fα ,

so that we identify
F αβ = dAαβ + Aα ∧ Aβ .
As a result, curvature is entirely based on the 1-forms Aαβ , which are simply defined
from the connection coefficients AI αβ .

1.3.2.2 Structure equation

Curvature obeys important properties, which intervene in general relativity and
gauge theories. The first one is the structure equation, which is equivalent to the defi-
nition of F that was given above. This is the reason why the structure equation is often
taken as the definition of curvature, while the relations

F αβ = dAαβ + Aα ∧ A β

are then called the structure equations.

For any w ∈ Γ(E), F (w) ∈ Ω2 (B, E). Then, in any point b ∈ B, F (w) takes two
arguments from Tb B, which produces a vector in Eb . Therefore, if we take two fields
y and z in Γ(T B), then F (w)(y, z) provides a field in Γ(E). The structure equation
claims that this field is such that :

F (w)(y, z) = ∇y ∇z w − ∇z ∇y w − ∇[y,z] w .

From this point of view, F is a map from Γ(E) × Γ(T B) × Γ(T B) to Γ(E), instead
of a map from Γ(E) to Ω2 (B) ⊗ Γ(E). Of course, this precisely yields the definition of
curvature traditionally given in general relativity3 . More precisely, let us consider the
first expression given for the curvature :

∀w ∈ Γ(E), F (w) = F αβ ⊗ [(fα ⊗ f β )(w)] .

Since the F αβ ’s are elements of Ω2 (B), we may decompose them as

1 α
F αβ = F βIJ eI ∧ eJ .
2
The components F αβIJ are scalar functions on B. As the components of differential
2-forms, the set of functions is naturally antisymmetric in the last two indices :

F αβIJ = −F αβJI .
3
In general relativity, curvature is denoted R rather than F , and also referred to as the Riemann
tensor. We will not use the latter denomination since it usually designates the curvature tensor in the
particular case of Riemannian geometry, that is the geometry based on the Levi-Civita connection.
General relativity only considers this particular connection, but since we will focus on metric with
torsion in the next chapter, we will only refer to F (or R) as the curvature tensor.

32
 Therefore, if we see F as a map with three arguments w, y and z rather than one,
we may write it as
1 α
F = F βIJ eI ∧ eJ ⊗ fα ⊗ f β
2
= F αβIJ eI ⊗ eJ ⊗ fα ⊗ f β ,

which is the standard form of the curvature tensor.

In the context of gauge theories, curvature precisely embodies the notion of field
strength as follows. From the structure equation, one quickly recovers the standard
relation between the components of the curvature tensor and the connection components
in a coordinate basis :

F αβµν = ∂µ Aν αβ + Aµαγ Aν γ β − ∂ν Aµαβ − Aν αγ Aµγ β .

Since the connection coefficients are defined as

Aµαβ := Aµk (Jk0 )αβ ,

we may write an expression of F αβµν where the infinitesimal generators appear explic-
itly :
0 γ
F αβµν = ∂µ [Aν k ](Jk0 )αβ − ∂ν [Aµk ](Jk0 )αβ + Aµl (Jl0 )αγ Aν m (Jm ) γ Aµl (Jl0 )γ β
0 α
) β − Aν m (Jm
0 γ
= [∂µ Aν k − ∂ν Aµk ](Jk0 )αβ + Aµl Aν m [(Jl0 )αγ (Jm ) γ (Jl0 )γ β ]
0 α
) β − (Jm
= [∂µ Aν k − ∂ν Aµk ](Jk0 )αβ + Aµl Aν m [(Jl0 · Jm
0 α 0
) β − (Jm · Jl0 )αβ ] ,

where · denotes the matrix products of the generators seen as endomorphisms acting on
the fibers. In the last expression, one clearly identifies the Lie brackets of the generators :

Jl0 · Jm
0 0
− Jm · Jl0 = [Jl0 , Jm
0
] = cklm Jk0 ,

where the quantities cklm are the structure constants corresponding to these generators.
This finally yields

F αβµν = [∂µ Aν k − ∂ν Aµk + cklm Aµl Aν m ](Jk0 )αβ .

Then, the field strength Fµν k used in the gauge theories is simply defined as the
coefficients of the generators in the expression above :

Fµν k := ∂µ Aν k − ∂ν Aµk + cklm Aµl Aν m .

One easily recovers the curvature tensor from the field strength with the following
relation4 :
F = Fµν k dxµ ⊗ dxν ⊗ Jk0 .
4
We remind that {dxµ } is the dual basis of the coordinate basis {eµ }.

33
1.3.2.3 Bianchi identity
The second important relation, very useful in the computations of general relativity,
is the Bianchi identity. It is merely an exterior derivation of the defining relations of
the curvature 2-forms, which are :
F αβ = dAαβ + Aα ∧ Aβ .
Both the F αβ ’s and the Aαβ ’s are scalar-valued differential forms, so that we may
act upon them with the usual exterior derivative d :
dF αβ = d2 Aαβ + dAα ∧ Aβ + (−1)1 Aα ∧ dAβ
= 0 + (F α − Aαγ ∧ Aγ  ) ∧ Aβ − Aα ∧ (F β − Aγ ∧ Aγ β )
= F α ∧ Aβ − Aα ∧ F β ,
which is the Bianchi identity for curvature, usually written as
dF αβ + Aα ∧ F β − F α ∧ Aβ = 0 .

1.3.3 Torsion
Torsion may only arise when the tangent bundle T B of a manifold B may be seen
as an associated bundle to a principal bundle. Although it seems to be a very particular
case, it is actually ubiquitous in physics since all the fields that transform in the vector
representation of the Lorentz group are rigorously defined as sections of the tangent
bundle. In such situation, the connection induced on T B from the connection form
defined on the principal bundle is called a linear connection. The latter closely relates
to torsion, which is essential in Einstein-Cartan theory.

1.3.3.1 Definition
The fibers of the tangent bundle are merely the tangent spaces of the base. Then,
it is worth stressing that the fibers naturally have same indices as the base. Therefore,
indices such as µ and I will now appear in place of α, β, γ and .
Let us consider the application eI ∧ eI . For any b ∈ B, it may be seen as the identity
of Tb B. In particular, it may be seen as a T B-valued 1-form since, in any point of the
base, it maps a tangent vector to another one (itself in this case).
Definition. Let ∇ be the linear connection on T B and d∇ the corresponding exterior
covariant derivative. Then, the torsion 2-form T is merely defined as
T := d∇ (eI ∧ eI ) .
It is important to keep in mind that, since d∇ is entirely defined from the linear
connection ∇, torsion completely depends on the choice of linear connection as well.
Since eI ∈ Ω0 (B, T B) and eI ∈ Ω1 (B), we may use the properties of the exterior
covariant derivative to expand the definition of torsion :
T = d∇ (eI ∧ eI )
= d∇ (eI ) ∧ eI + (−1)0 eI ∧ deI
= ∇(eI ) ∧ eI + eI ∧ deI .

34
 Since eI ∈ Ω0 (B, T B), we have

eI ∧ deI = deI ⊗ eI

and

∇(eI ) ∧ eI = AJ I ⊗ eJ ∧ eI
= AJ I ∧ eI ⊗ eJ
= AI J ∧ eJ ⊗ eI .

This yields
T = (deI + AI J ∧ eJ ) ⊗ eI .
Since T ∈ Ω2 (B, T B), we may decompose it in terms of components as

T = T I ⊗ eI ,

where the T I are elements of Ω2 (B) called the torsion 2-forms.

Therefore, we may also define torsion as the T B-valued 2-form whose components
are :
T I = deI + AI J ∧ eJ .

1.3.3.2 Structure equation

Torsion admits an equivalent definition known as the structure equation for torsion.
Torsion is an element of Ω2 (B, T B). This means that, in any point b ∈ B, it takes two
arguments in Tb B to generate another element of Tb B. Therefore, if we take two fields
y and z in Γ(T B), then T (y, z) is another element of Γ(T B). The structure equation
claims that this field is such that :

T (y, z) = ∇y z − ∇z y − [y, z] .

It is often presented as the first definition of the torsion of a linear connection. In

this form, it is even more obvious that torsion may only be constructed in the context
of linear connection. The second argument of the covariant derivative must be a section
of the considered associated bundle. Therefore, if one wants to consider terms such as
∇y z where z ∈ Γ(T B), it is required that T B itself is the associated bundle.
Since the torsion 2-forms T I are elements of Ω2 (B), we may decompose them as
1
T I = T IJK eJ ∧ eK .
2
From the structure equation, one quickly finds the following relation between the
components of the torsion tensor and the connection components :

T IJK = AJ I K − AK I J − cI JK ,

where the quantities cI JK are the structure constants of the basis {eI }, which generate
the Lie algebra Γ(T B) equipped with the Lie bracket of vector fields. These quantities
are also known as the objects of anholonomy or anholonomy coefficients.

35
1.3.3.3 Bianchi identity
One may also derive a Bianchi identity for torsion by taking the exterior derivative
of the defining relations of torsion :
dT I = d2 eI + dAI J ∧ eJ + (−1)1 AI J ∧ deJ
= 0 + (F IJ − AI K ∧ AK J ) ∧ eJ − AI J ∧ (T J − AJ K ∧ eK )
= F IJ ∧ eJ − AI J ∧ T J ,
which is the Bianchi identity for torsion, usually written as
dT I + AI J ∧ T J = F IJ ∧ eJ .
In the context of Riemannian geometry, used in general relativity, torsion is null.
As a result, the Bianchi identities for curvature and torsion take the following familiar
form5 :

 F IJKL + F ILJK + F IKLJ = 0
.
 ∇M F I I I
JKL + ∇L F JM K + ∇K F JLM = 0

We only defined the covariant derivative for vectors of order 1, that is sections of T B.
However, in these identities, the covariant derivative of the Riemann tensor appears,
although F is a tensor of order 4. Then, let us remind how the covariant derivative of
vectors is extended in a unique way onto the tensors of higher order.
Since the latter are built out of tensor products of vectors and covectors, one only
needs to define the covariant derivative of covectors and impose a Leibniz rule for the
covariant derivative of tensor products. In order to define the covariant derivative of
covectors, the covariant derivative of a contracted tensor is constrained to equal the
contraction of the covariant derivative.
For instance, let us consider the tensor eI0 ⊗ eJ0 , which is of type (1, 1). Since it is
a second-order tensor, its contraction coincides with its trace, which is δI0 J0 . On the
other hand, applying the Leibniz rule to eI0 ⊗ eJ0 yields
∇µ (eI0 ⊗ eJ0 ) = ∇µ (eI0 ) ⊗ eJ0 + eI0 ⊗ ∇µ (eJ0 )
= AµK I0 eK ⊗ eJ0 + eI0 ⊗ Bµ J0L eL ,

where the quantities Bµ J0L are the coefficients of ∇µ eJ0 on the dual basis eL . The
contraction of the first term yields AµJ0 I0 while the contraction of the second term is
Bµ J0I0 .
Since the contraction of eI0 ⊗ eJ0 is δI0 J0 , its covariant derivative is 0, so that we find

0 = AµJ0 I0 + Bµ J0I0 .
Therefore, the covariant derivative of any covector may be deduced from the follow-
ing relation :
∇µ eI = −AµI J eJ .
5
From now on, we may use the shorter notation ∇I instead of ∇eI .

36
 Then, once a tensor of a given type is decomposed in terms of tensor products of
vectors and covectors, the Leibniz rule provides the covariant derivative correspond-
ing to the considered type of tensor. This covariant derivative coincides with the one
that would arise by following the procedure described in the sections devoted to the
construction of connections in associated bundles.
Let us illustrate this with tensors of type (1, 1). Since T B is assumed to be an asso-
ciated bundle to a principal G-bundle P , then all the fibers of T B must be isomorphic
to a representation space V of G. Representation theory guarantees that V ⊗ V ∗ is
also a representation space of G, where V ∗ denotes the dual space of V . Then, we may
construct the associated bundle to P with fiber V ⊗ V ∗ , which is denoted T B ⊗ T B ∗ .
As stated earlier, the mobile frame {eI } of T B is associated to a section e of P . The
mobile frame of T B ⊗ T B ∗ associated to e then is {eI ⊗ eJ }.
From the gauge potential relative to e in P , one defines in a unique manner the
gauge potential associated to {eI ⊗ eJ } in the associated bundle T B ⊗ T B ∗ , as we did
in the section devoted to the gauge potentials. Ultimately, such gauge potential allows
to define the covariant derivative of the sections of T B ⊗ T B ∗ . It coincides with the
covariant derivative defined above as an extension of the covariant derivative of vectors
of T B.

37
Chapter 2

Einstein-Cartan gravity

The previous chapter introduced the notions of torsion, curvature and connection on
a spinor bundle, which are ubiquitous in Einstein-Cartan theory of gravity. More pre-
cisely, the latter generalises Einstein’s general relativity in the sense that the connection
used in the theory is allowed to have torsion. Moreover, the theory predicts that torsion
is generated by the spin of matter. This fact is partly responsible for our interest in the
spinor fields, since they bear non-zero spin. As a result, this chapter aims at writing
the explicit formulation of Einstein-Cartan theory in presence of a spinor field.
To this end, we first apply the abstract notions presented in the first chapter to
build the definitive form of the covariant derivative for spinors that are defined over
a 4-dimensional Lorentzian manifold. Secondly, defining the notions of contortion and
vierbein allows us to address the explicit formulation of Einstein-Cartan theory, in order
to apply it to a spinor field.

2.1 Prerequisite
The first chapter provided us with the concepts and tools needed to properly define
spinor fields on a general spacetime and study their dynamics through the concept of
connection. The latter will obviously be essential for formulating the equation verified
by a spinor field in the context of Einstein-Cartan theory. Therefore, we need to apply
these concepts in the physical case of a spinor bundle with a 4-dimensional base and
the spin group Spin(3, 1) as its structure group. This will eventually yields the explicit
form of the covariant derivative that is relevant for the study of spinor fields.

2.1.1 Spin(3, 1) connection

The spinor fields that we will manipulate from now on are those appearing in the
standard model of particle physics. They were introduced by Dirac as 4-component
objects that belong to a representation space V of the spin group Spin(3, 1). We will
use them as a source of spin in Einstein-Cartan theory.
Let B be a 4-dimensional Lorentzian manifold representing our spacetime. As stated
in the first chapter, we need to assume that B admits a spin structure relative to
Spin(3, 1). This provides a principal Spin(3, 1)-bundle with base B, denoted Spin(B).
Ultimately, this allows to build a spinor bundle S as the associated bundle to Spin(B)
through the action of Spin(3, 1) on V . Then, the spinor fields we want to study are

38
defined as the elements of Γ(S). In this situation, we may apply the theory of connections
presented in the first chapter to explicitly construct a covariant derivative acting on
these spinors.
Although our main focus are the spinor fields, it is impossible to ignore the vector
fields of the tangent bundle T B. The first reason for this is that they will necessarily
come as the first argument of the covariant derivative for spinors once it will be defined.
Another reason is that, formally, momentum and acceleration are tangent vectors to
B as well, and they are ubiquitous in any theory. All these vectors belong to the 4-
component vector representation of Spin(3, 1), which is the defining representation of
the Lorentz group SO(3, 1). As a result, we also assume that the tangent bundle is
an associated bundle of Spin(B), so that we will derive a linear connection for the
4-vectors. Such a connection will closely relate to the connection acting on the spinors
since they will share the same gauge fields, as it was seen in the first chapter.
First of all, we need to choose a set of infinitesimal generators of Spin(3, 1), that is a
basis of its Lie algebra spin(3, 1). As one knows, spin(3, 1) identifies with the Lie algebra
so(3, 1) of the Lorentz group SO(3, 1) since Spin(3, 1) is the universal cover of SO(3, 1).
Therefore, we may as well use a basis of so(3, 1). Writing η = diag(1, −1, −1, −1) for
the Minkowski metric, we remind that the Lorentz group is defined as :

SO(3, 1) = {Λ ∈ GL4 (R), ΛT ηΛ = η} ,

i.e. it is the group of the matrices preserving the Minkowski metric.

From this result, one quickly derives the defining property of so(3, 1) :

so(3, 1) = {λ ∈ M4 (R), λT η + ηλ = 0} .

As a result, one can show that any matrix λ ∈ so(3, 1) may be written as lη where
l is an antisymmetric matrix. In terms of components, λ rewrites as6 :

λI J = ηJK lIK ,

which is equivalent to
lIJ = η KJ λI K ,
where η IJ is the inverse metric.
Therefore, the matrix l will be denoted λ with all indices up since the expression of
lIJ above coincides with the usual way in which one raises the indices of a tensor when
the metric is η. Explicitly, this means that we set λIJ := lIJ .
Any element of so(3, 1) is thus equivalent to an antisymmetric matrix. An antisym-
metric matrix of dimension 4 has 4×3 2
= 6 independent components, which is therefore
the dimension of so(3, 1). This is expected since the Lorentz group needs three genera-
tors for the boosts and three generators for the rotations. Let us use these six generators
and denote them Jr , with r running from 1 to 6.
As mentioned earlier, the spinors that we focus on lie in the representation space V
of Spin(3, 1). Therefore, there exists a group homomorphism ρ : Spin(3, 1) → GL(V )
and an underlying Lie algebra representation ρ̂ : spin(3, 1) → gl(V ). In particular,
spin(3, 1) identifies with a subspace g0 of gl(V ).
6
Since the vectors tangent to B lie in the defining representation of the Lorentz group, the Lorentz
transformations directly act on the tangent spaces of the base, so that their indices will be of the type
I, J, K,...

39
 The next fundamental step is to give the explicit form of the generators Jr0 of g0
induced from the generators Jr through ρ̂. For instance, this means that, when the
Lorentz transformation exp[Jr ] is performed, then the corresponding transformation
acting on spinors is exp[Jr0 ]. Explicitly, this important correspondence is :
1 KJ
Jr0 := ρ̂(Jr ) = η (Jr )I K [γI , γJ ] ,
8
where the gamma operators satisfy the defining relations of the Dirac algebra :

{γI , γJ } = 2 ηIJ IV .

As explained above, we may write η KJ (Jr )I K as (Jr )IJ , where this last matrix is
antisymmetric. Using the antisymmetry, the relation above rewrites as :
1
Jr0 = (Jr )IJ γI · γJ .
4
Now, let us suppose that there is a principal connection on Spin(B). This allows
us to define a gauge potential if we pick a section e in Γ(Spin(B)). Since S is an
associated bundle to Spin(B), the section e identifies with a mobile spinor frame {sα }7
of S. In other terms, {sα } is a section of the spinor frame bundle F S. Its dual basis is
denoted {sα }, and the dual basis {dxµ } of a coordinate basis {eµ } provides a basis of
Ω1 (B). Then, as presented in the first chapter, the gauge potential relative to e may be
decomposed on the basis {dxµ ⊗ Jr } in terms of some components Aµr :

A = Aµr dxµ ⊗ Jr .

Now that the gauge fields Aµr are set, we immediately find the expression of the
gauge potential in S :
AS := Aµr dxµ ⊗ Jr0 .
The generators Jr0 act as any endomorphism on the fibers of S, so that we may
decompose them on the basis {sα ⊗ sβ } as :
1
Jr0 = (Jr0 )αβ sα ⊗ sβ = (Jr )IJ (γI · γJ )αβ sα ⊗ sβ .
4
This yields the corresponding connection coefficients :
1 r
AS µαβ := Aµr (Jr0 )αβ = A (Jr )IJ (γI · γJ )αβ .
4 µ
Using these quantities, we may now define the covariant derivative of spinor fields
by following the same pattern that was presented in the first chapter. It is important
to take a notation different from the symbol ∇, since this one is always reserved for
the covariant derivative of T B, as we will see soon. The covariant derivative for spinors
might be denoted ∇S to indicate that it refers to the spinor bundle S, but it is rather
denoted D in the literature. We will follow this convention from now on.
7
The spinors sα are elements of the fibers of S, so that we must refer to them with the indices α,
β,... by convention. Moreover, the fibers are isomorphic to V which is 4-dimensional, so that α, β,...
run from 1 to 4.

40
 Thus, the covariant derivative of spinors is explicitly deduced from the following
relation :
Deµ sα := AS µγ α sγ .
Then, for a spinor field Ψ = Ψα sα ∈ Γ(S) and a tangent field z = z µ eµ ∈ Γ(T B),
we explicitly have :
Dz Ψ = z µ (∂µ [Ψγ ] + Aµγ α Ψα ) sγ .
Let us now establish the bound relating this connection to the linear connection
defined on T B. Since T B is an associated bundle to Spin(B), the section e identifies
with a mobile frame {eI } of T B. In other terms, {eI } is a section of the frame bundle
F T B. Since the fields of Γ(T B) are in the defining representation of the Lorentz group,
there is no need to look for the equivalent of ρ and ρ̂ in this situation since both are
the identity. Then, the generators Jr decompose on the basis {eI ⊗ eJ } as

Jr = (Jr )I J eI ⊗ eJ ,

where {eI } is the dual basis of {eI }.

The connection coefficients of the linear connection will be denoted ωµ IJ to avoid
confusions. It was presented in the first chapter how they expand in terms of the gauge
fields and the infinitesimal generators :

ωµ IJ := Aµr (Jr )I J .

Then, the covariant derivative for the tangent fields is such that, for two tangent
fields y = y I eI and z = z µ eµ in Γ(T B), we explicitly have :

∇z y = z µ (∂µ [y I ] + ωµ IJ y J ) eI .

Yet, we notice that the quantities Aµr (Jr )I J also appear in the spin connection
coefficients, so that we find a close relation between the spin connection and the linear
connection :
1 r
AS µαβ = A (Jr )IJ (γI · γJ )αβ
4 µ
1
= Aµr η KJ (Jr )I K (γI · γJ )αβ
4
1
= η KJ ωµ IK (γI · γJ )αβ .
4

Due to this tight relation between the linear connection coefficients ωµ IJ and the
spin connection, it is common practice in the literature to only work with the linear
connection. For instance, the covariant derivative of spinors is rather written as :
1 KJ I
Dz Ψ = z µ (∂µ [Ψγ ] + η ωµ K (γI · γJ )γ α Ψα ) sγ .
4

For these reasons, the linear connection coefficients ωµ IJ are also referred to as the
spin connection. This does not cause any confusion since the coefficients AS µαβ actually
never appear in the literature.

41
 Let us make a quick last point. One is used to raising and lowering the indices of
connection coefficients although these are not tensor. This operation is defined as if the
indices were tensor, that is :
ωµ IJ := g KJ ωµ IK ,
where gIJ is the metric tensor on B in the basis {eI }, and g IJ is its inverse. One must be
careful with this notation : although ωµ IK := Aµr (Jr )I K , we still have ωµ IJ 6= Aµr (Jr )IJ
since g KJ (Jr )I K 6= (Jr )IJ in general.
This is only true when the metric g is the Minkowski metric η in the basis {eI }.
In this particular case, the quantity η KJ ωµ IK coincides with ωµ IJ . As a result, ωµ IJ
is antisymmetric in its last two indices, and the covariant derivative of spinors writes
more simply as :
1 IJ
Dz Ψ = z µ (∂µ [Ψγ ] + ω (γI · γJ )γ α Ψα ) sγ .
4 µ

2.1.2 Metric connection and contortion

Since Einstein-Cartan theory makes use of the notion of metric connection, we briefly
review the important features of this kind of connection. Firstly, a metric connection
is a particular type of linear connection. This means that it is a covariant derivative
for the tangent vector fields on B, which are the elements of Γ(T B). In the context of
linear connections, the connection coefficients are denoted Γµρν instead of Aρµν , as it is
the case in general relativity.
As mentioned in the first chapter, there are infinitely many possible connections on
a manifold, due to the arbitrary choice of the horizontal bundle of the principal bundle.
Yet, as one knows, metric connections are more relevant than other connections in
physics since their defining property is

∇g = 0 ,

where g is the metric tensor on spacetime. Such metric connections seem more natural
in regard of the following physical argument : for a metric connection, the inner product
of two vectors is constant when the latter are parallel transported, which is intuitively
a very legitimate requirement.
What is particularly interesting about the metric connections is that they are entirely
defined by the torsion T and the familiar Christoffel symbols. We remind the expression
of the latter in any coordinate basis :
µ 1 µσ
{ρ ν } = g (∂ρ gνσ + ∂ν gσρ − ∂σ gρν ) .
2
Explicitly, the symmetric part of a metric connection is constrained to be
µ µ
Γµ(ρν) = {ρ ν } + T(ρν) ,

where the first and third indices of T were lowered and raised using the metric tensor,
as it is commonly done in tensor calculus.
In addition, we remind that torsion is related to the connection through the following
relation :
T µρν = Γµρν − Γµνρ − cµρν ,

42
 Yet, the structure constants cµρν of a coordinate basis always vanish due to Schwartz’s
theorem. Thus, we find :
1
Γµ[ρν] = T µρν .
2
As a result, a metric connection always have the following form :
µ
Γµρν = {ρ ν } + K µρν ,

where K µρν is the contortion tensor defined as :

µ 1
K µρν := T(ρν) + T µρν
2
1
= (Tρν µ + Tνρ µ + T µρν ) .
2
Therefore, the degrees of freedom of a metric connection only lie in the components
of its contortion tensor.

2.1.3 Vierbein postulate

The equivalence principle leads to postulating that our 4-dimensional spacetime
is locally Minkowski spacetime M4 . This argument integrates into general relativity
by describing spacetime as a 4-dimensional Lorentzian manifold. This allows for the
existence of an orthonormal basis {eI } in any spacetime point, that is a basis in which
the metric is that of M4 . Such a mobile frame is called a tetrad or a vierbein8 . The
vierbein along with the spin connection constitute the two fundamental fields upon
which Einstein-Cartan theory is formulated, which is why we now introduce the various
features of this formalism.
Once a coordinate chart {xµ } is chosen, the vierbein {eI } relates to the coordinate
basis {eµ } according to the following change of basis :

eI = eI µ eµ .

The matrix (eI µ ) will be referred to as the vierbein matrix. In particular, the quadru-
plet (eI0 µ )µ=0..3 are the coordinates of eI0 in the coordinate basis. Therefore, since the
vierbein {eI } is orthonormal, one must find :

gµν eI µ eJ ν = ηIJ ,

where the quantities gµν are the components of the metric in the coordinate basis.
Indices such as µ or I are raised and lowered as usual with the action of the corre-
sponding metric. This allows to define the following quantities :

eI µ := η IJ gµν eJ ν .

These quantities turn out to coincide with the inverse matrix of e = (eI µ )I,µ=0..3 in
the sense that (e−1 )µI = eI µ , so that the following relations hold :
8
In German,“vier” means “four” and “bein” means “leg”, hence this logical denomination for a mobile
frame of a 4-dimensional manifold.

43
 
 eI e ν = δ ν
µ I µ
µ
.
 e e J J
I µ = δI

In particular, we have :

I I µ
 e = e µ dx


eµ = eI µ eI .


 g I J
µν = ηIJ e µ e ν

The last relation is particularly important since it shows that the vierbein formalism
contains all the information about the metric. This is the reason why the latter becomes
only a second-class field in the vierbein formalism, which is used in Einstein-Cartan
theory.
Conventionally, the connection coefficients are denoted ωµ IJ when they refer to the
vierbein, and Γρµν when they refer to the coordinate basis, that is :

 ∇ e = ω I e
µ I µ J J
.
 ∇µ eν = Γρ eρ
µν

Of course, one transforms into the other through a gauge transformation. For in-
stance, we have :

∇µ eν = ∇µ (eI ν eI )
= ∂µ (eI ν ) eI + eI ν ωµ IJ eJ
= [∂µ (eJν ) + eI ν ωµ IJ ] eJ
= [∂µ (eJν ) + eI ν ωµ IJ ] eJ ρ eρ ,

so that we identify :

Γρµν = eJ ρ [ωµ JI eI ν + ∂µ (eJν )]

= eJ ρ ωµ JI (e−1 )ν I + eJ ρ ∂µ (eJν ) ,

where the last expression does have the exact same form as the gauge transformations
familiar from the standard model. Similarly, one finds :

ωµ IJ = eI ρ [Γρµν eJ ν + ∂µ (eJ ρ )]
= (e−1 )ρI Γρµν eJ ν + (e−1 )ρI ∂µ (eJ ρ ) .

Like general relativity, Einstein-Cartan theory assumes the covariant derivative to

be metric compatible. Starting from the expression of the metric in the vierbein, we

44
must have :

0 = ∇µ g = ∇µ (ηIJ eI ⊗ eJ )
= ηIJ (−ωµ IK eK ⊗ eJ − eI ⊗ ωµ JL eL )
= −(ωµJI + ωµIJ ) eI ⊗ eJ .

Therefore, metric compatibility requires antisymmetry of the last two indices of the
connection coefficients in an orthonormal basis. We remark that this coincides with
what was found in the section devoted to building a Spin(3, 1) connection.
Ultimately, let us start from the metric written in the basis {eI ⊗ dxν } so that the
metric writes as :
g = eI ⊗ eJ ν dxν .
Then, metric compatibility rewrites as :

0 = ηIJ ∇µ (eJ ν eI ⊗ dxν )

= [∂µ (eIν ) − eJ ν ωµJI − eIρ Γρµν ] eI ⊗ dxν ,

which is equivalent to :

∀I, ∀ν, ∂µ (eIν ) + eJ ν ωµJI − eIρ Γρµν = 0 ,

where we used the antisymmetry of the last two indices of the spin connection. This
last relation is known as the vierbein postulate [6]. We remark that it coincides with
the gauge transformations above if we first rewrite it as

∀J, ∀ν, eJ τ Γτ µν = ωµ JI eI ν + ∂µ (eJ ν ) .

Then, left multiplying by eJ ρ yields :

eJ ρ [ωµ JI eI ν + ∂µ (eJ ν )] = eJ ρ eJ τ Γτ µν
= δ Jτ Γτ µν
= Γρµν ,

which is the first gauge transformation found above.

2.2 Field equations

The previous sections provided us with the last notions required to properly address
Einstein-Cartan theory of gravity. As a theory of gravity allowing torsion, the Einstein-
Cartan field equations are expected to provide a description of the metric tensor and
the torsion tensor. Although these objects may not be the principal unknown of the
field equations, it is at least required that their behaviour may be deduced from the
solutions to the field equations. Since the latter derive from the least action principle,
the first section introduces the features of an action in Einstein-Cartan theory, such as
the fields that come as the arguments. The second and third parts are then naturally
devoted to the derivation and explicit formulation of the field equations.

45
2.2.1 First-order formalism
Einstein field equations may be derived from the variational principle. In a spacetime
filled with a matter9 field Φ with action Sm [Φ, gµν ], Einstein’s equations arise from
varying the total action SEH [gµν ]+Sm [Φ, gµν ] with respect to the metric gµν . We remind
that, in Planck units, which we will always use, the Einstein-Hilbert action SEH is given
by : Z
1
q
SEH [gµν ] = d4 x −g[gµν ] R[gµν ] ,
16π
where g is the determinant of the metric tensor, and R is the Ricci scalar built from
the Riemann tensor.
The only argument of the Einstein-Hilbert action is the metric gµν . This is expected
since, in Riemannian geometry, which is the framework of general relativity, the metric
alone characterises the whole structure of spacetime and defines the variations of the
fields defined over it. This is of course due to the fact that the Levi-Civita connection
is entirely constructed from the metric, and hence the Riemann tensor, the Ricci tensor
and the Ricci scalar.
In Eintein-Cartan theory, it will be possible to take a metric compatible connection
as well, like the Levi-Civita connection. But, as was pointed out earlier, metric compat-
ibility is far from entirely fixing the connection, due to the various degrees of freedom
of the contortion tensor. Yet, the fundamental characteristic of Einstein-Cartan theory
is to allow non-vanishing torsion, so that the connection coefficients come as an addi-
tional field independent from the metric. Also, it was announced in the previous section
that Einstein-Cartan theory is formulated within the vierbein formalism. In particu-
lar, it was pointed out that the metric is easily recovered from the vierbein matrices.
Therefore, although the metric was the one argument of the Einstein-Hilbert action,
it now loses such primacy in favour of the matrix (eI µ ). Then, as a theory of gravity
with torsion, Einstein-Cartan theory reformulates the Einstein-Hilbert action in terms
of two independent fields, which are the spin connection and the vierbein matrix.
An action that considers the connection as an additional argument independent from
the field describing the metric is known as a Palatini action. Strictly speaking, however,
the Palatini formalism is not the most general action in the sense that it still assumes
the Lagrangian of “matter” to only depend on the metric and not on the connection.
Thus, a Palatini action takes the following form :
Z
1
q
ρ
SP [gµν , Γ µν , Φ] = d4 x −g[gµν ] R[Γρµν ] + Sm [Φ, gµν ] ,
16π
where the Ricci scalar R does only depend on the connection coefficients Γρµν since there
is no longer any relation assumed between the metric and the connection coefficients.
Similarly to the Einstein-Hilbert action, varying the Palatini action with respect to
the metric yields the Einstein’s equations. Yet, these have the additional feature that the
Ricci tensor and the Ricci scalar are expanded in terms of the connection coefficients
instead of the metric. If the energy-momentum tensor is fixed, Einstein’s equations
thus have two unknown quantities, gµν and Γρµν , while gµν was the only unknown in
the Einstein-Hilbert formalism.
9
Here, the term “matter” generically refers to any type of field, be it proper matter, radiation, or
dark energy.

46
 Therefore, a second set of equations is needed to solve the system. It is provided by
the Euler-Lagrange equations relative to the connection. Since the matter action has
no dependence on the connection, varying SP with respect to the connection merely
amounts to varying the Einstein-Hilbert action, in which all the dependence on the
connection is borne by the Ricci scalar. Interestingly, this computation exactly leads to
expressing the connection coefficients as the Christoffel symbols. That is, in a spacetime
filled with a matter field whose behaviour does not depend on the connection, the
connection can only be the Levi-Civita connection.
However, such an action seems to contradict itself in the following sense. A connec-
tion is supposed to be used to study the variations of the fields defined over spacetime.
In order for a field to have dynamics, it is expected that the matter action Sm involves
a kinetic term built from covariant derivatives of the field. This would obviously imply
a dependence of the matter action on the connection coefficients. Therefore, forbidding
the matter action to depend on the connection, as it is the case in the Palatini for-
malism, amounts to forbidding the field to have dynamics defined from the connection
coefficients. In other terms, the Palatini action allows any kind of connection to be
considered in the theory, as long as it is not used for the main purpose a connection is
generally defined for.
Lastly, since the matter action may yet depend on the metric, the dynamics of the
matter field might only be defined by using the metric tensor. That is, since the matter
action is not allowed to make use of the connection coefficient, it needs to build its
own ad hoc connection from the metric tensor only. In this view, the Euler-Lagrange
equations relative to the connection coefficients seem to restore consistency within the
whole theory by requiring the original connection coefficients Γρµν to be also constructed
from the metric tensor only. As said before, these coefficients are actually the Christoffel
symbols.
The next natural step is to finally consider the most general case, in which the matter
action is allowed to make proper use of the connection coefficients. Such an approach
is known as the metric-affine formalism. It is the approach adopted in Einstein-Cartan
theory although, as mentioned above, the metric tensor is replaced by the vierbein
matrix as argument of the action.
The coefficients Γρµν and ωµ IJ equivalently describe the connection, since they are
formulations of the same connection relatively to different bases. One may then equally
change from one to the other through the gauge transformations that were presented
earlier. Then, in order to match the literature, we will use ωµ IJ rather than Γρµν as a
second argument. Thus, a generic action in Einstein-Cartan theory has the following
form :
Z
1
q
µ IJ
SEC [eI , ωµ , Φ] = d x −g[eI µ ] R[eI µ , ωµ IJ ] + Sm [Φ, eI µ , ωµ IJ ] .
4
16π
It is worth noting that, in the Palatini and metric-affine actions, all the derivatives
of the arguments appearing in the action are at most of first order. This directly ensues
from the expression of the curvature tensor, and hence that of the Ricci scalar, which
involve only first-order derivatives of the connection coefficients. This is why these
actions are also known as first-order formalisms. On the contrary, the Christoffel symbols
are defined in terms of first-order derivatives of the metric, so that the Riemann tensor
is built from second-order derivatives of the the metric. Hence, the Einstein-Hilbert

47
action involves second-order derivatives of its argument and is called a second-order
formalism.

In the next sections, we will briefly go through the successive steps of the derivation
of Einstein-Cartan field equations. For this purpose, we need to express the explicit
dependences of the various quantities involved in the action in terms of the arguments
eI µ and ωµ IJ . As mentioned above, one will notice that only first-order derivatives of
the arguments will be involved.
Let us first focus on the Ricci scalar. Firstly, we recovered earlier that the general
expression of the curvature tensor is given by :

F αβµν = ∂µ Aν αβ + Aµαγ Aν γ β − ∂ν Aµαβ − Aν αγ Aµγ β ,

where the quantities Aµαβ are the connection coefficients. We remind that the first two
indices α and β refer to the fibers. In the context of general relativity, the latter are
the tangent spaces which we refer to with the vierbein indices I and J, the curvature
tensor is designated by the letter R rather than F , and the connection coefficients are
ωµ IJ . All the indices of RI Jµν are tensor so that we raise and lower them using the
corresponding metric. As mentioned already, the spin connection with indices raised or
lowered is also defined according to the same procedure, so that we have :

RIJ µν = ∂µ ων IJ + ωµ IK ων KJ − ∂ν ωµ IJ − ων IK ωµ KJ .

The Ricci tensor is obtained by contracting the first index of the curvature tensor
together with the third index. To perform the Einstein summation corresponding to a
contraction, the indices need to refer to the same basis. Since RIJ µν is a tensor, we may
perform a change of basis by merely multiplying by the corresponding vierbein matrix :

RρJ µν = eI ρ RIJ µν .

The Ricci tensor is then obtained as follows :

RJ ν = RµJ µν
= eI µ RIJ µν .

Likewise, the Ricci scalar is the trace of the Ricci tensor, which we obtain as follows :

R = Rν ν
= eJ ν R J ν
= eJ ν eI µ RIJ µν .

We now have the explicit expression of the Ricci scalar entirely in terms of the two
arguments eI µ and ωµ IJ .
Regarding the determinant g, we remind that the metric tensor relates to the vier-
bein matrices as follows :
gµν = ηIJ eI µ eJ ν .
√
This relation will allow to vary −g with respect to the coefficients eI µ .
As for the matter action, we reasonably assume that the corresponding Lagrangian
density Lm does not depend on derivatives of the vierbein coefficients nor on derivatives

48
of the connection coefficients, and that it can depend on derivatives of the matter field
of first order at most. The matter action then rewrites as :
Z
Sm [Φ, eI , ωµ ] = d4 x Lm (Φ, ∂µ Φ, eI µ , ωµ IJ ) .
µ IJ

2.2.2 Asymmetric Einstein’s equation

Varying with respect to the vierbein coefficients eI µ yields the following set of Euler-
Lagrange equations :
√
∂ ∂ −gR
( ρ
− ∂σ ρ
)[ + Lm ] = 0 .
∂eK ∂(∂σ eK ) 16π

Since no term has any dependence in the first derivatives of the vierbein coefficients
∂σ eK ρ , the equation rewrites as :

1 √ ∂R R ∂g µν ∂g ∂Lm
( −g ρ
− √ ρ
) = −8π ,
2 ∂eK 2 −g ∂eK ∂g µν
∂eK ρ

that is :
1 ∂R R ∂g µν ∂g −8π ∂Lm
( ρ
+ ρ
)= √ = 8π T Kρ ,
2 ∂eK 2g ∂eK ∂g µν −g ∂eK ρ
since the general definition of the energy-momentum tensor T in the vierbein formalism
is10 :
−1 ∂Lm
T Kρ := √ .
−g ∂eK ρ
When Lm can simply be expressed in terms of the metric rather than the vierbein,
the latter definition does coincide with the usual definition of the energy-momentum
tensor used in standard general relativity, which is :
−2 ∂Lm
Tµν := √ .
−g ∂g µν

In fact, in this situation, one may simply write the following chain rule :
∂Lm ∂g στ ∂Lm
= .
∂eK ρ ∂eK ρ ∂g στ

Let us then use the relation g µν = η IJ eI µ eJ ν to compute the following quantity,

which we need again later :
∂g µν
ρ
= η IJ (δIK δρµ eJ ν + eI µ δJK δρν )
∂eK
= δρµ eKν + δρν eKµ .
10
We use a curly T to distinguish from the torsion tensor T .

49
 Therefore, we find
∂g στ ∂Lm ∂Lm ∂Lm
ρ στ
= δρσ eKτ στ + δρτ eKσ στ
∂eK ∂g ∂g ∂g
∂Lm ∂Lm
= eKτ ρτ + eKσ σρ
∂g ∂g
∂L m ∂L m
= eKτ ρτ + eKσ ρσ
∂g ∂g
∂Lm
= 2eKτ ρτ ,
∂g
so that
−1 ∂Lm
T Kρ = √ 2eKτ ρτ
−g ∂g
−2 ∂Lm
= eKτ ( √ ).
−g ∂g ρτ

This is equivalent to :
−2 ∂Lm
Tσρ = eKσ T Kρ = eKσ eKτ ( √ )
−g ∂g ρτ
−2 ∂Lm
= δστ ( √ )
−g ∂g ρτ
−2 ∂Lm
=√ ,
−g ∂g ρσ

so that we recover the standard definition of the energy-momentum tensor. However, it

is not always possible to assume that Lm may be expressed in terms of the metric rather
than the vierbein. Actually, the energy-momentum tensor T Kρ will not be symmetric in
general.
Let us now focus on the terms on the left-hand side term of the Euler-Lagrange
equations. We will first need the following useful identity [6] :

∂g
= −g gµν .
∂g µν
This yields
1 R ∂g µν ∂g gµν R µ Kν
ρ µν
=− (δρ e + δρν eKµ )
2 2g ∂eK ∂g 4
R
= − (gρν eKν + gµρ eKµ )
4
1
= − gµρ eKµ R
2
1 K
= − e ρR .
2

50
 Finally, let us compute the following derivative :
1 ∂R 1 ∂
= [e µ e ν RIJ µν ]
2 ∂eK ρ
2 ∂eK ρ I J
1
= (δIK δρµ eJ ν + eI µ δJK δρν ) RIJ µν
2
1
= (eJ ν RKJ ρν + eI µ RIK µρ )
2
1
= (RKν ρν + RK ρ ) .
2
At this stage, we need to anticipate the result of the second set of Euler-Lagrange
equations, obtained from varying the action with respect to the connection. This second
set will allow solutions such that the connection is metric compatible, which is the type
of solution we will always choose. With such solutions, we will recover the fact that the
last two indices of the connection are antisymmetric. The curvature tensor obviously
inherits such antisymmetry on its first two indices, due to its definition in terms of the
connection coefficients. Therefore, we have :

RKν ρν = eI ν RKI ρν = −eI ν RIK ρν = −RνK ρν = RνK νρ = RK ρ ,

so that
1 ∂R
= RK ρ .
2 ∂eK ρ
Eventually, the Euler-Lagrange equations relative to the vierbein writes as :
1 K
RK ρ − e R = 8π T Kρ .
2 ρ
Clearly, such an equation is very close to the original Einstein field equations re-
garding its form. In such situation, one also defines the Einstein’s tensor as :
1 K
GK ρ := RK ρ − e R.
2 ρ
However, each quantity here is more complex than the corresponding term in the
original Einstein’s equations. Firstly, the curvature tensor and the Ricci scalar are no
longer Riemannian since the connection they are built from is allowed to have torsion.
Secondly, the metric is replaced by the vierbein, which contains all the information
about the metric and more. In particular, it is not a symmetric object. This fact is
inherited by the energy-momentum tensor which is no longer symmetric either.

2.2.3 Torsion and spin density tensor

The variation with respect to the connection coefficients ωµ IJ yields a second set of
Euler-Lagrange equations :
√
∂ ∂ −gR
( LM
− ∂σ LM
)[ + Lm ] = 0 .
∂ωρ ∂(∂σ ωρ ) 16π

51
 Expanding this equation is much less straightforward than it was for the previous
set of Euler-Lagrange equations. Tedious computations eventually lead to the following
relation involving torsion :
ρ
T ρLM + 2 e[L T σM ]σ = 8π S ρLM ,

where S is the spin density tensor, very generally defined as :

1 ∂Lm
S ρLM := √ .
−g ∂ωρ LM

This expression is clearly analogous to the general definition of the energy-momentum

tensor, only adapted to the argument ωρ LM instead of eK ρ . Then, as T is a source for
curvature via the asymmetric Einstein’s equation, so S is the source for torsion via this
the relation above. This is even better pictured when this formula is inverted, in order
to get the explicit expression of the torsion in terms of the spin density tensor.
Since T and S are tensors, we may multiply the equation by the right vierbein
matrices in order to write it with all the indices in the coordinate basis :
ρ
T ρµν + 2 δ[µ T σν]σ = 8π S ρµν .

This allows us to contract the index ρ with the index ν :

8π S ρµρ = T ρµρ + δµρ T σρσ − δρρ T σµσ

= T ρµρ + T σµσ − 4 T σµσ
= −2 T ρµρ .

Therefore, the Euler-Lagrange equations relative to the connection rewrite as :

ρ
T ρµν = 8π (S ρµν + δ[µ S σν]σ ) .

Thus, the torsion is completely set from the spin density tensor of the matter field Φ.
In particular, if the spin density tensor vanishes, so does the torsion. As presented earlier,
the contortion tensor of a metric connection is entirely defined from the torsion tensor.
This means that the spin density defines a unique metric compatible connection, which is
then explicitly expressed in terms of the Christoffel symbols and the spin density tensor.
Since the former are entirely constructed from the metric tensor, we may rewrite them
in terms of the vierbein field. This allows to substitute all the connection coefficients
appearing in the asymmetric Einstein’s equation by an expression involving only the
vierbein field and the spin density.
At this stage, the vierbein field is the only remaining unknown of the asymmet-
ric Einstein’s equation. Solving the equation would eventually provide the metric of
spacetime. The Christoffel symbols would be immediately computed in order to get
the explicit expression of the connection coefficients. In particular, we remind that a
connection is originally defined from the choice of a horizontal bundle in an underlying
principal bundle. We may then say in the language of bundles that, given a spin density
tensor and an energy-momentum tensor defined over a spacetime B, the metric com-
patibility condition defines a Spin(3, 1)-equivariant distribution of horizontal subspaces
on the principal Spin(3, 1)-bundle Spin(B).

52
 The purely mathematical definitions of the energy-momentum tensor and the spin
density tensor used here are hard to interpret physically. For a better understanding of
these quantities, let us quickly mention that in certain simple situations, these objects
may be defined from much more intuitive and physical notions, which justify their
denominations. For instance, we remind that, for a perfect fluid, the energy-momentum
tensor takes a much more physical expression, given by :

Tµν = (ρ + p) uµ uν − p gµν .

From such an expression, the information contained in T in terms of energy and

momentum is much clearer. Likewise, for some sufficiently simple fluids, called spinning
fluids [7], it is possible to express the spin density tensor in such a physical manner.
For such fluids, we have :

 Tµν = Pµ uν


Sρµν = uρ sµν ,

µν = uν Pµ − uµ Pν

 ṡ

where P is the momentum density of the fluid, u is the unit velocity, and s is
an antisymmetric tensor. From such a definition, the spin density tensor in a point
extracts the information about the intrinsic angular momentum of the fluid relative to
this point by removing the classical orbital momentum. Of course, such an intrinsic
angular momentum coincides with the concept of spin, hence the denomination of this
tensor.

Finally, thanks to the Euler-Lagrange equations relative to the connection coeffi-

cients, it becomes possible to rewrite the asymmetric Einstein’s equation in a particu-
larly interesting form [8]. More precisely, the structure of the equation will be even more
similar to that of the original Einstein’s equation, since the left-hand side term will be
the torsion-free, symmetric Einstein’s tensor. This will allow us to concretely identify
the new physical description of gravity provided by Einstein-Cartan theory compared
to the original Einstein’s general relativity.
As we know, a metric compatible connection writes as the sum of the Levi-Civita
connection and the contortion tensor K. Therefore, the curvature tensor R may be
written as the sum of the Riemann tensor11 R̃ entirely defined from the Christoffel
symbols, and a second part depending on K, and hence S. As a result, the Ricci tensor
and the Ricci scalar may also be split into a torsion-free part and a contortion-related
part. Finally, this also allows to decompose the Einstein tensor G into its Riemannian
part G̃ and a remaining part G(K) . As for the right-hand side term of the equation, one
may also decompose the energy-momentum tensor into a purely symmetric, torsion-free
part T̃ and a contortion-related remainder T (K) .
With all the indices lowered, the asymmetric Einstein’s equation then rewrites as :
(K)
(K) Gµν
G̃µν = 8π (T̃µν + Tµν − ).
8π
11
Conventionally, all the quantities independent of torsion, and hence only related to the metric, are
denoted with a tilde.

53
 Since the contortion tensor is now completely determined from the spin density
tensor S, we group together the contortion-related terms within a quantity which may
be rightly considered as a spin-related energy-momentum tensor, which is necessarily
symmetric as well :
(K)
(spin) (K) Gµν
Tµν := Tµν − .
8π
As a result, we have
(ef f )
G̃µν = 8π Tµν ,
where T (ef f ) := T̃ + T (spin) is the final energy-momentum tensor. It acts as the effective
source of the metric since it entirely defines the torsion-free Einstein’s tensor G̃. Then,
regarding the metric tensor, the description of gravity in Einstein-Cartan theory is
almost equivalent to the original, torsion-free theory of general relativity. In fact, the
dynamics of the metric is driven by exactly the same type of equation as the original
Einstein’s equation. In this regard, the only new feature of Einstein-Cartan theory is
that it includes an additional contribution generated by the spin of matter in the source
term of the equation. In particular, if the spin density tensor vanishes, we recover the
exact original Einstein’s equation.
Yet, one must not forget that, if there really is torsion, the parallel transport of the
various types of fields may still be very different from the torsion-free situation. Then,
strictly speaking, Einstein-Cartan gravity cannot be entirely reformulated in terms of
torsion-free concepts.

2.3 Spinor source

Since our main concern are the spinor fields, we will now consider an explicit matter
action for which the argument Φ is a spinor field Ψ. In particular, this will lead to
extracting an interaction potential for the spinor field and deriving the Euler-Lagrange
equations relative to the spinor field, which is the Dirac equation.

2.3.1 Fermion currents

Working with a spinor field on a general spacetime requires to generalise the Dirac
operator γ µ ∂µ of Minkowski spacetime. Since the variations of a spinor field on a given
spacetime are evaluated through the spinor covariant derivative D rather than ∂, the
Dirac operator naturally generalises as γ µ Dµ , as one would expect. Therefore, the matter
action will be the Dirac action SD which takes the following form12 :
√
Z
µ
SD [Ψ, eI , ωµ ] = d4 x −g LD
IJ

√
Z
i
= d4 x −g [ (Ψ̄γ µ Dµ Ψ − Dµ Ψγ µ Ψ) − mΨ̄Ψ ] .
2
12 √
In order to match the conventions, we do not include the volume factor −g in the Dirac La-
grangian density LD , although we did include it in the generic matter Lagrangian in the previous
sections.

54
 Regarding their Lorentz index, the gamma matrices actually belong to the vector
representation of the Lorentz group. This is what allows to consider the matrices γ µ in-
volved in the Dirac action since they are defined from the constant matrices γ I according
to γ µ = eI µ γ I .
Let us make a first remark, which will be useful to compute the Noether currents
of this action and to easily derive the Dirac equation in the last section. The Dirac
(∂)
Lagrangian is initially a function LD of the arguments Ψ, Ψ̄, ∂µ Ψ, ∂µ Ψ and ωµ IJ .
(D)
However, due to its covariant structure, we may see it as a function LD of the arguments
Ψ, Ψ̄, Dµ Ψ and Dµ Ψ only. Since the explicit form of Dµ Ψ is :

1 IJ
[Dµ Ψ]α = ∂µ [Ψα ] + ω [γI , γJ ]αβ Ψβ ,
8 µ
we may properly apply the chain rule to find that :
(∂) (D)
∂LD ∂[Dµ Ψ]β ∂LD
=
∂[∂µ Ψ]α ∂(∂µ Ψα ) ∂[Dµ Ψ]β
(D)
∂LD
= δαβ
∂[Dµ Ψ]β
(D)
∂LD
= ,
∂[Dµ Ψ]α

and
(∂) (D) (D)
∂LD ∂LD ∂[Dµ Ψ]β ∂LD
= +
∂Ψα ∂Ψα ∂Ψα ∂[Dµ Ψ]β
(D) (D)
∂LD 1 ∂LD
= α
+ ωµ IJ [γI , γJ ]β α .
∂Ψ 8 ∂[Dµ Ψ]β

Like the term “covector”, we may use the term “cospinor” to refer to a 1-form acting
on the fibers of the spinor bundle that we are working on. The indices of such an object
are naturally down. Then, similarly to the case of the covariant derivative of a covector,
the covariant derivative of a cospinor χ is :
1 IJ
[Dµ χ]α = ∂µ [χα ] − ω [γI , γJ ]β α χβ .
8 µ
All the previous relations help derive the following useful form of the Euler-Lagrange
equations, particularly adapted to Lagrangians which have an explicit covariant struc-
ture :
(∂) (∂) (D) (D) (D)
∂LD ∂LD ∂LD 1 ∂LD ∂LD
α
− ∂µ ( α
)= α
+ ωµ IJ [γI , γJ ]β α β
− ∂µ ( )
∂Ψ ∂[∂µ Ψ] ∂Ψ 8 ∂[Dµ Ψ] ∂[Dµ Ψ]α
(D) (D)
∂LD ∂LD
= α
− [Dµ ( )]α .
∂Ψ ∂[Dµ Ψ]

To be clear, the last term is the component α of the covariant derivative of the
(D) (D)
∂LD ∂LD
cospinor denoted ∂[Dµ Ψ]
, which has components ∂[Dµ Ψ]γ
.

55
 Obviously, the same arguments apply to Ψ̄ and Dµ Ψ. This yields Euler-Lagrange
equations with the same form, with opposite position of the indices.
(D)
From now on, the Dirac Lagrangian LD will actually rigorously coincide with LD .

Let us also give the explicit form of the conjugate of the covariant derivative. Using
the fact that (γ 0 )−1 (γ µ )† γ 0 = γ µ , one finds :
1 IJ
Dµ Ψ = (∂µ [Ψ̄α ] − ω Ψ̄β [γI , γJ ]β α ) sα .
8 µ
Finally, let us focus on the symmetries and corresponding currents of the Dirac
action. The latter admits two rigid symmetries, whose corresponding Noether currents
will be useful in the next sections and in the last chapter. Firstly, the transformation
Ψ 7→ Ψ0 = eiθ Ψ for any fixed parameter θ, is obviously a U(1) rigid symmetry of the
action.
In particular, the variation of the action must vanish under such a transformation.
For any small θ, let us consider ∆Ψ = Ψ0 − Ψ = iθΨ and ∆Ψ̄ = Ψ̄0 − Ψ̄ = −iθΨ̄. Then,
the Noether current V µ corresponding to the above symmetry is given by :
1 ∂LD ∂LD
Vµ =− ( ∆Ψ + ∆Ψ̄ )
θ ∂Dµ Ψ ∂Dµ Ψ
1 i i
= − ( Ψ̄γ µ (iθΨ) + (−iθΨ̄)(− γ µ Ψ) )
θ 2 2
µ
= Ψ̄γ Ψ .

Secondly, we remind that the operator γ 5 := iγ 0 γ 1 γ 2 γ 3 is Hermitian and anticom-

mute with any γ µ . From the anticommutation relations verified by the gamma operators,
one shows that (γ 5 )2 = IV , where we remind that V is the fiber of the spinor bundle
5
we are working on. This last relation implies that eiγ θ = cos θ IV + i sin θ γ 5 , hence
5 5 5
e−iγ θ γ µ = γ µ eiγ θ . All this eventually shows that the transformation Ψ 7→ Ψ0 = eiγ θ Ψ
for any fixed parameter θ, is a U(1) rigid symmetry of the kinetic term of the action.
(kin)
The latter will be denoted LD . In particular, the variation of the kinetic term must
vanish under such a transformation. Any small θ generates the following variations of
the fields :
∆Ψ = Ψ0 − Ψ = iθγ 5 Ψ

.
∆Ψ̄ = Ψ̄0 − Ψ̄ = (iθγ 5 Ψ)† γ 0 = −iθΨ† γ 5 γ 0 = iθΨ† γ 0 γ 5 = iθΨ̄γ 5

Therefore, the Noether current Aµ of the kinetic term, called the axial current, is
given by :
(kin) (kin)
µ 1 ∂LD ∂L
A = ( ∆Ψ + ∆Ψ̄ D )
θ ∂Dµ Ψ ∂Dµ Ψ
1 i i
= ( Ψ̄γ µ (iθγ 5 Ψ) + (iθΨ̄γ 5 )(− γ µ Ψ) )
θ 2 2
= Ψ̄γ 5 γ µ Ψ .

Although this symmetry of the kinetic term is a symmetry of the whole action only
in the massless case, the axial current will still appear in the following considerations
in spite of a possibly non-zero mass.

56
 As soon as an action involves covariant derivatives, it is expected that expanding
the kinetic term leads to interaction terms. This is very common in the standard model,
since it is from the kinetic terms that emerge the interaction terms between the vector
gauge fields such as the gluons and the spinor fields or the scalar fields. In addition, we
remark that, in the standard model, the kinetic terms are quadratic in the covariant
derivatives, so that their expansions also lead to pure kinetic terms for the gauge fields,
which are themselves quadratic in the gauge fields.
In our situation, the role of the gauge fields is played by the spin connection ωµ IJ .
Since the Dirac Lagrangian is not quadratic in the covariant derivatives, it does not
properly provide a kinetic term for the spin connection. In fact, the dynamics of the
connection is only driven by the gravitational action involving the Ricci scalar, which
thus embodies the role of a kinetic term for the connection. However, the Dirac La-
grangian still provides interaction terms between the connection and the spinor field,
which is quadratic in the spinor field.
Even more interestingly, we remind that the torsion, and hence the contortion of
the connection are entirely set from the spin density tensor. But the spin density tensor
corresponding to the Dirac Lagrangian is actually quadratic in the spinor field, and
so is the contortion tensor eventually. Thus, the interaction term between the connec-
tion and the spinor field turns out to be quartic in the spinor field. This defines a
four-fermion interaction which is derived in the next section. Ultimately, such a quar-
tic interaction generates a non-linear field equation relative to the spinor field. This
equation corresponds to the Dirac equation, which is described in the last section.

2.3.2 Four-fermion interaction

Let us explicitly derive the four-fermion interaction mentioned above. One first
starts by expanding the kinetic term of the Dirac Lagrangian in terms of standard
partial derivatives and connection coefficients :

(kin) i
LD = (Ψ̄γ µ Dµ Ψ − Dµ Ψγ µ Ψ)
2
i 1 1
= (Ψ̄γ µ ∂µ Ψ + ωµ IJ Ψ̄ γ µ [γI , γJ ] Ψ − ∂µ (Ψ̄)γ µ Ψ + ωµ IJ Ψ̄ [γI , γJ ] γ µ Ψ)
2 8 8
i i
= (Ψ̄γ µ ∂µ Ψ − ∂µ (Ψ̄)γ µ Ψ) + ω IJ Ψ̄ {γ µ , [γI , γJ ]} Ψ .
2 16 µ
We obtain the spin density tensor from the last term since it gathers all the depen-
dence of the Dirac Lagrangian in the connection :
√
µ 1 ∂( −gLD )
S IJ := √
−g ∂ωµ IJ
i
= Ψ̄ {γ µ , [γI , γJ ]} Ψ .
16

Let us now use the fact that γ µ = eK µ γ K , along with the following useful identity :

{γK , [γI , γJ ]} = −4i KIJL γ 5 γ L ,

57
where  is the familiar Levi-Civita symbol. The spin density tensor thus rewrites as :
i Kµ
S µIJ = e Ψ̄ {γK , [γI , γJ ]} Ψ
16
1
= eKµ KIJL Ψ̄ γ 5 γ L Ψ
4
1 Kµ
= e KIJL AL ,
4
where we recognised the axial current AL = Ψ̄ γ 5 γ L Ψ.
Let us now remind the expression of torsion in terms of the spin density tensor with
a relevant choice of indices :

T µIJ = 8π (S µIJ + eµ[I S σJ]σ ) .

In our situation, this relation becomes particularly simple since the second term
within the brackets vanishes :

S σJσ = eLσ S σKL

1
= η LP eP σ eM σ KLM N AN
4
1
= η LP δPM KLM N AN
4
1
= η LP KLP N AN = 0 ,
4
where the last expression vanishes due to the contraction of the indices L and P , which
are symmetric for η while they are antisymmetric for .
This results in the following simple expression for torsion :

T µIJ = 8π S µIJ = 2π eKµ KIJL AL .

After a few computations, this yields the contortion tensor, with indices relevant for
the following considerations :
1 I
K IµJ := (T µJ + TµJ I + TJµ I )
2
= π η IK eM µ JKM L AL .

With the previous results in hand, all the dependence of the kinetic term in the
contortion tensor may be substituted by the spinor terms. In addition, we choose to
only expand the non-Riemannian part of the covariant derivative. This will leave the
torsion-free part gathered with the partial derivatives under the Levi-Civita connection
for spinors. The latter is denoted D̃. This yields :

(kin) i i I J
LD = (Ψ̄γ µ D̃µ Ψ − D̃µ Ψγ µ Ψ) + K Ψ̄ {γ µ , [γI , γJ ]} Ψ ,
2 16 µ

58
but, using the previous results, the last term rewrites as :
1
π η IK η JP eM µ P KM L AL eQµ Ψ̄ QIJN γ 5 γ N Ψ
4
π IK JP
= (η η P KM L QIJN )(eM µ eQµ )AL AN
4
π
= − η P J η KI η M Q P KM L JIQN AL AN
4
π
= − ηRL 3! (4 − 3)! (det η)−1 δN R L N
A A
4
3π 3π 2
= ηN L AL AN = A .
2 2
This is the final form of the spinor interaction term. Since the axial current in-
cludes two spinor terms, its norm squared does generate a four-fermion interaction, as
announced earlier. This naturally results in a non-linear field equation relative to the
spinor field, which is derived in the next section.

2.3.3 Field equations and Dirac equation

The Einstein-Cartan field equation relating torsion to the spin density tensor in the
case of a spinor source was derived earlier.
Let us now simply compute the energy-momentum tensor of the spinor field to get
the explicit form of the asymmetric Einstein’s equation. From relations obtained in the
section devoted to the asymmetric Einstein’s equation, one easily derives the following
useful identity : √
∂ −g √
ρ
= − −g eK ρ .
∂eK
Then, starting from the definition of T Kρ , one has :
√
K −1 ∂( −gLD )
T ρ := √
−g ∂eK ρ
√
∂ i µ I µ I −1 ∂( −g)
= [ (Ψ̄eI γ Dµ Ψ − Dµ ΨeI γ Ψ) − mΨ̄Ψ ] + √ LD
∂eK ρ 2 −g ∂eK ρ
= Ψ̄γ K Dρ Ψ − Dρ Ψγ K Ψ − eK ρ LD ,

so that the asymmetric Einstein’s equation in the case of a spinor source writes as :
1 K
RK ρ − e R = Ψ̄γ K Dρ Ψ − Dρ Ψγ K Ψ − eK ρ LD .
2 ρ
Finally, the last field equation is provided by the Euler-Lagrange equations relative
to the spinor field Ψ itself, that is the Dirac equation. Since the only dependence of the
total action in Ψ lies in the Dirac action, one only need to vary the Dirac Lagrangian.
As justified earlier, we may derive the Euler-Lagrange equations as follows :
∂LD ∂LD α
0= − [Dµ ( )]
∂ Ψ̄α ∂[Dµ Ψ]
i
= −mΨα + ([γ µ Dµ Ψ]α + [Dµ (γ µ Ψ)]α ) .
2

59
 Without the indices, this equation simply rewrites as :

(i γ µ Dµ − m)Ψ = 0 .

From this form, one easily gets the equivalent equation on Ψ̄ :

i Dµ Ψγ µ + mΨ = 0 .

Interestingly, this means that if the Dirac equation holds, then the Dirac Lagrangian
cancels in any spacetime point :
i
LD = (Ψ̄γ µ Dµ Ψ − Dµ Ψγ µ Ψ) − mΨ̄Ψ
2
1
= (Ψ̄(mΨ) − (−mΨ̄)Ψ) − mΨ̄Ψ = 0 .
2
As a result, this informs us that all the critical points of the Dirac action with respect
to its argument Ψ take the same value, which is always zero, and they are reached on
the fields satisfying the Dirac equation. For such spinor fields, the energy-momentum
tensor simplifies as follows :

T Kρ = Ψ̄γ K Dρ Ψ − Dρ Ψγ K Ψ ,

so that the Einstein’s equation rewrites as :

1 K
RK ρ − e ρ R = Ψ̄γ K Dρ Ψ − Dρ Ψγ K Ψ .
2
In the Dirac equation obtained above, the non-Riemannian part of the covariant
derivative involves the contortion tensor. The latter contains spinor terms that are
responsible for the nonlinearity of the equation. To see this explicitly, we may rather
derive the Dirac equation with respect to the Riemannian covariant derivative. For this
purpose, we expand the Dirac Lagrangian as follows :
i i I J
LD = (Ψ̄γ µ D̃µ Ψ − D̃µ Ψγ µ Ψ) − mΨ̄Ψ + K Ψ̄ {γ µ , [γI , γJ ]} Ψ
2 16 µ
i 3π 2
= (Ψ̄γ µ D̃µ Ψ − D̃µ Ψγ µ Ψ) − mΨ̄Ψ + A .
2 2
The Dirac equation is thus derived from the following equations :
(D)
∂LD ∂LD
0= − [D̃µ ( )]α
∂ Ψ̄α ∂[D̃µ Ψ]
3π 5 3π i
= −mΨα + [γ γI Ψ]α AI + AI [γ 5 γ I Ψ]α + ([γ µ D̃µ Ψ]α + [D̃µ (γ µ Ψ)]α )
2 2 2
= [(i γ µ D̃µ − m)Ψ]α + 3πAI [γ 5 γ I Ψ]α ,

or, without the indices :

(i γ µ D̃µ − m)Ψ + 3πAI γ 5 γ I Ψ = 0 .

This yields the form of the Dirac equation in which the nonlinear spin-interaction
term is explicit. This nonlinear term is a consequence of the underlying four-fermion

60
interaction of the theory. In the case of vanishing torsion, the nonlinear term van-
ishes and we recover the standard Dirac equation in curved spacetime. In a general
way, Einstein-Cartan theory with vanishing torsion naturally coincides with field the-
ory in curved spacetime. For instance, when torsion vanishes, we already noticed that
the asymmetric Einstein’s equation coincides with the original, torsion-free Einstein’s
equation, which is the equation obeyed by the ambient metric in the context of field
theory in curved spacetime. In such a theory, Dirac equation is linear like the original
equation formulated in Minkowski spacetime.
Yet, even in the case of vanishing torsion, the Dirac equation obviously remains
coupled to the Einstein’s equation (which is generally not linear in the metric or the
vierbein). This is due to the fact that the Riemannian part of the covariant derivative
is constructed from the Christoffel symbols, and these are built from the metric tensor
whose dynamics obeys the Einstein’s equation.
With these various results in hand, Einstein-Cartan theory allows us to study the
effects that spinor fields may have on certain important cosmological phenomena, which
are the topic of the next chapter.

61
Chapter 3

Cosmology with spinors

Among other questions such as determining the true nature of dark matter, modern
cosmology addresses four major issues :

• the flatness problem, which arises from evidences suggesting that our Universe is
very close to be flat ; this implies that it must have been even closer to being flat
in the past, which is a very strange feature ;

• the horizon problem, which arises from the isotropy of the Cosmological Microwave
Background (CMB), which seems in contradiction with the fact that all the regions
of the Universe have not been causally connected for a long part of its history ;

• the inhomogeneities of the CMB, that is the relatively small thermal fluctuations
of the CMB, whose origin needs to be determined ;

• the Big Bang singularity, which corresponds to the fact that the temperature and
the densities of the various fluids filling the Universe all become infinite as the
scale factor reaches zero when going back in time.

The three first issues find explanations thanks to the model of inflation. The latter
is a very short exponential expansion phase of the Universe occurring immediately after
the Big Bang. The central object of this process is the inflaton scalar field, whose slow
roll dynamics allow for such a de Sitter expansion of the Universe. Such a model of
the Universe generally requires at least 30 e-folds of inflation in order to match the
observations of today’s Universe relative to the flatness and horizon problems.
Furthermore, this model explains the origin of the inhomogeneities of the CMB
from the quantum fluctuations of the scalar inflaton. The results of such an approach
are remarkable since they predict a nearly scale-invariant spectrum of the temperature
fluctuations of the CMB, which is strongly consistent with the observations.
In standard general relativity, one may try to drive the inflation by a spinor field
rather than a scalar field, but the results are much less satisfying, since they lead to
a scale-dependent spectrum. However, if this substitution is carried out within the
framework of Einstein-Cartan, then, very interestingly, scale invariance is recovered.
This is the topic of the first part of the chapter.
As mentioned above, inflation may require more than 30 e-folds of expansion to
solve the flatness and horizon problems, sometimes much more, which is a relatively
strong requirement. Therefore, it is a striking fact that, in Einstein-Cartan theory, the

62
cosmological equations generalising the familiar Friedmann equations yield an elegant
solution to the flatness and horizon problems, without assuming any inflation phase.
Furthermore, such a cosmological model does not contain any Big Bang singularity
because it suggests that the expansion of the Universe began with a non-zero scale
factor and was preceded by a contraction phase. This will be the focus of the second
part of the chapter.
Therefore, this chapter aims at providing an insight into the potential impacts of a
spinor field on the cosmological issues listed above, within the framework of Einstein-
Cartan cosmology. Let us remind that, at this stage of the dissertation, the field equa-
tions of Einstein-Cartan theory of gravity in presence of a spinor field have been estab-
lished in the second chapter.
The first equation, obtained from varying the action with respect to the vierbein, is
an asymmetric version of the original Einstein’s equation of general relativity. Though,
as it was presented, it is possible to reformulate it in a symmetric way by splitting the
various terms into their Riemannian part and their torsion-related part. The equation
will be mostly used under this form in this chapter, since it is the most relevant for
applications in cosmology.
The second equation is obtained from varying the action with respect to the spin
connection. Although it is harder to derive than the first equation, it yields a much
simpler equation which allows to reformulate all the quantities made from torsion in
terms of the spin density tensor. As a result, all the cosmological effects of torsion
can actually be understood as induced by spin. It also seems more relevant to relate
cosmological data to an accessible physical quantity such as spin, rather than an abstract
geometrical notion such as torsion.
Ultimately, varying with respect to the Dirac field yields the Dirac equation. The
latter rules the dynamics of the spinor field Ψ, from which are derived the source tensors
T and S.
Expectedly, these three equations lie at the basis of this chapter since they rule the
behaviour of the spinor field that may drive the inflation, along with the dynamics of
the Universe which avoids the Big Bang singularity.

3.1 Spinor driven inflation

Driven by a scalar field, inflation provides answers to the horizon and flatness prob-
lems along with the tiny inhomogeneities of the CMB, in accordance with the observa-
tions. The framework of Einstein-Cartan theory allows to substitute such a scalar field
by a spinor field and recover correct predictions in regards of the characteristics of the
CMB.
In the case of the scalar inflaton, slow roll conditions are assumed in order to main-
tain a de Sitter expansion. Similarly, in order for a Dirac field to drive such an exponen-
tial expansion, we consider its potential V (s) (where s is the quantity Ψ̄Ψ) to be more
general than ms, which was the potential used until now. In this general situation, the
condition equivalent to slow roll is [8] :
 
 d ln V 
 d ln s   1 .
 

63
 This guarantees a quasi-exponential expansion, that is a nearly constant Hubble
parameter H := aȧ . Therefore, the behaviour of the scale factor is such that :

a ∼ a0 eH(t−t0 ) ,

where subscripts 0 denotes values relative to today’s Universe.

In such a context, we will go through the main steps leading to recover the spectrum
of the CMB from the quantum fluctuations of the Dirac field.

3.1.1 Dirac equation in FRW

In order to study the quantum fluctuations of a Dirac field driving inflation, it is
first needed to determine a solution to the equation ruling their classical dynamics.
This equation derives from introducing perturbation into the Dirac equation. More
precisely, the Dirac field is split into a background solution Ψ(b) satisfying the Dirac
equation, and a perturbation δΨ. Similarly to the case of scalar field driven inflation, the
background solution is assumed to be time dependent only, while the fluctuations have
both time and spatial dependences. Since both the total field Ψ = Ψ(b) + δΨ and the
background field Ψ(b) satisfy the Dirac equation, the difference between these provides
the equation ruling the fluctuations. We remark that since the Dirac equation is not
linear in Einstein-Cartan theory, δΨ is not a solution of the Dirac equation.
In order to illustrate this procedure, let us position ourself in the standard context of
a homogeneous, isotropic universe described by the flat Friedmann-Robertson-Walker
(FRW) metric, and let us derive the corresponding form of the Dirac equation. In such
a cosmological model, we may rather use the Cartesian coordinates, so that the metric
is :
ds2 = dt2 − a2 (t) δij dxi dxj .
As a result, a vierbein is provided by the following matrix :

(eI µ ) = diag (1, a−1 , a−1 , a−1 ) ,

since it satisfies the following relation :

gµν eI µ eJ ν = ηIJ .

Since we use a general potential V , the Dirac equation is slightly modified and takes
the following form :
(i γ µ Dµ − V 0 )Ψ = 0 ,
where 0 denotes derivative with respect to the only argument, which is s.
As expected, expanding the torsion-related part simply yields :

(i γ µ D̃µ − V 0 )Ψ + 3πAI γ 5 γ I Ψ = 0 ,

where D̃µ is the Riemannian spin connection. This simply means that it is explicitly
defined as follows :
1 IJ
[D̃µ Ψ]α = ∂µ [Ψα ] + ω̃µ [γI , γJ ]αβ Ψβ ,
8

64
where ω̃µ IJ are the spin connection coefficients corresponding to the Levi-Civita con-
ρ
nection. In other terms, ω̃µ IJ relates to the Christoffel symbols {µ ν } according to the
following gauge transformation :
ρ
ω̃µ IJ = eI ρ [ {µ ν } eJ ν + ∂µ (eJ ρ ) ] .
The fact that the vierbein is diagonal makes the gauge transformations quick to
compute from the Christoffel symbols. For any (i, j, k) ∈ {1, 2, 3}3 , the latter take the
following form in flat FRW metric :
 0 0 i 0 i
 {0 i } = {i 0 } = {0 0 } = {0 0 } = {j k } = 0

0
{i j } = a ȧ δij .
 i
 i ȧ i
{ j 0 } = { 0 j } = a δj
We remind that the Kronecker delta is not a tensor. For instance, δji 6= gjk δ ik . More
precisely, the set of coefficients (δba ), (δ ab ) and (δ ab ) are all numerically equal for any
types of indices a and b. Therefore, the indices of δ are not raised and lowered by
multiplying them by the metric tensor. They are only written up or down in order to
conveniently match with the summation conventions used in a given expression.
For any k ∈ {1, 2, 3}, any (K, L) ∈ {1, 2, 3}2 and any (I, J), the previous results
provide the following spin connection coefficients :
 K  KL
 ω̃k L = ω̃0 IJ = 0  ω̃k = ω̃0 IJ = 0
ω̃ 0 = ȧ δkJ ⇐⇒ ω̃k 0J = ȧ η JM δkM = −ȧ δkJ .
 k IJ I I0 0M I 00 I I
ω̃k 0 = ȧ δk ω̃k = η ω̃k M = η ω̃k 0 = ȧ δk


We remark the antisymmetry of the last two indices of the spin connection. It is due
to the fact that the Levi-Civita connection is metric compatible :
∀(µ, I, J), ω̃µ IJ = −ω̃µ JI .
Let us then substitute these explicit expressions into the first term of the Dirac
equation13 :
γ µ D̃µ Ψ = eMµ γ M D̃µ Ψ
1 IJ
= eM 0 γ M Ψ̇ + eM k γ M [∂k Ψ +
ω̃ γI γJ Ψ]
4 k
3
0 −1 k M 1 X I0
= γ Ψ̇ + a δM γ [∂k Ψ + ω̃ γI γ0 Ψ]
2 I=1 k
3
1 k −1 M X
= γ 0 Ψ̇ + a−1 δM
k
γ M ∂k Ψ + δM a γ ȧ δkI (−γ I ) γ 0 Ψ
2 I=1
3
ȧ X k
= γ 0 Ψ̇ + a−1 δM
k
γ M ∂k Ψ − δ δkI γ M γ I γ 0 Ψ
2a I=1 M
3
H X
= γ 0 Ψ̇ + a−1 δM
k
γ M ∂k Ψ − δM I γ M γ I γ 0 Ψ
2 I=1
3H 0
= γ 0 Ψ̇ + a−1 δM
k
γ M ∂k Ψ + γ Ψ,
2
13
We need to remind that indices such as i, j or k conventionally refer to the coordinate basis, but
they may only run from 1 to 3.

65
where we used the fact (γ I )2 = −IV when I 6= 0.
k
P3 Let us briefly stress out why the term δM γ M ∂k Ψ cannot be replaced either by
M k
M =1 γ ∂M Ψ or by γ ∂k Ψ. Since the index M refers to the vierbein rather than
the coordinate basis, the term ∂M Ψ in the former expression could legitimately be
understood as the directional derivative of Ψ in the direction eM , which is eM µ ∂µ Ψ. On
the other hand, the index k of γ k in the last expression refers to the coordinate basis
and not the vierbein, so that γ k does not equal γ K even if k is numerically equal to K.
We only have γ k = eI k γ I = a−1 δIk γ I , so that γ k = a−1 γ K where k is numerically equal
k
to K. Therefore, there is no other choice than sticking with the expression δM γ M ∂k Ψ,
k
in which δM does not vanish if and only if k is numerically equal to M .
Substituting the expression above into the Dirac equation, and left multiplying the
latter by −iγ 0 finally yields :
3H
Ψ̇ + a−1 δM
k
γ M ∂k Ψ + Ψ + iV 0 γ 0 Ψ − 3iπAI γ 0 γ 5 γ I Ψ = 0 .
2

3.1.2 Perturbation expansion and thermal fluctuations

We may now extract the classical equation satisfied by the perturbation δΨ. We just
need to subtract the Dirac equation satisfied by the background field to the equation
above. Since the background field is only time dependent, the Dirac equation that it
satisfies is obtained by removing all the spatial dependent terms of the equation above.
This yields :
3H (b) (b)
Ψ̇(b) + Ψ + iV 0 γ 0 Ψ(b) − 3iπAI γ 0 γ 5 γ I Ψ(b) = 0 ,
2
(b)
where AI is the axial current constructed from Ψ(b) only.
Then, subtracting this relation to the Dirac equation satisfied by the total field
yields :
3H
δ Ψ̇ + a−1 δM
k
γ 0 γ M ∂k δΨ + δΨ + iV 0 γ 0 δΨ
2
(b)
− 3iπ[ AI γ 0 γ 5 γ I δΨ + δAI γ 0 γ 5 γ I Ψ(b) + δAI γ 0 γ 5 γ I δΨ ] = 0 ,

where δA = A − A(b) contains terms in δΨ and δ Ψ̄.

In order to solve this equation, it is actually relevant to consider the perturbation
equation corresponding to the torsion-free situation :

˙ + a−1 δ k γ 0 γ M ∂ δ Ψ̃ + 3H
δ Ψ̃ M k δ Ψ̃ + iV 0 γ 0 δ Ψ̃ = 0 .
2
The plane-wave solutions of this equation write as us (k, t) eik · x and vs (k, t) e−ik · x ,
where the functions us and vs are explicitly known in terms of the Hankel functions [8].
These plane waves allow to expand a solution δΨ of the full perturbation equation with
torsion as follows :
2
d3 k X
Z
δΨ = [ A(k, t) us (k, t) as (k) eik · x + B(k, t) vs (k, t) bs (k)† e−ik · x ] ,
(2π)3/2 s=1

66
where A and B are scalar functions, and as and bs are operators satisfying the usual
fermion anticommutation relations.
This expansion is essential to interpret the temperature fluctuations observed in
the CMB as consequences of a spinor field driven inflation. In fact, thermodynamics
considerations predict that the temperature Tγ of photons behaves like the inverse of
the scale factor, i.e.14 Tγ ∼ a1 . This implies that the thermal fluctuations are such that
δTγ ∼ − aδa2 , hence δT
Tγ
γ
∼ − δa
a
.
Yet, the behaviour of a relates to that of s = Ψ̄Ψ. To establish this relation, let
us take the conjugate of the Dirac equation satisfied by Ψ(b) . This yields the equation
satisfied by Ψ̄(b) :

˙ (b) + 3H Ψ̄(b) − iV 0 γ 0 Ψ̄(b) + 3iπA(b) Ψ̄(b) γ 0 γ I γ 5 = 0 .

Ψ̄ I
2
This immediately provides the equation satisfied by s(b) = Ψ̄(b) Ψ(b) :
(b)
ṡ(b) + 3Hs(b) + 6iπAI γ 0 γ I γ 5 Ψ(b) = 0 .

As we will see in the following section, the torsion-related effects become negligible
very quickly as the scale factor increases. This is obviously the case during inflation, as
the scale factor increases exponentially. As a result, the last term in the equation above
may be removed in order to determine the general behaviour of s(b) during inflation.
(b) (b)
This immediately yields s(b) ∼ s0 e−3H(t−t0 ) ∼ s0 a30 a−3 . Since δs := s − s(b) does not
(b)
alter the behaviour of s by definition, we have s ∼ s(b) , and hence a ∼ (s0 )1/3 a0 s−1/3 .
Finally, we derive :
δa δ(s−1/3 ) 1 s−4/3 δs
∼ −1/3 ∼ − −1/3 δs ∼ − ,
a s 3s 3s
so that :
δTγ δs
∼ .
Tγ 3s
To sum this up, the quantum fluctuations δΨ of the spinor field Ψ naturally induce
fluctuations of the quantity s, which relates to the scale factor a and hence the temper-
ature Tγ . The CMB provides a picture of the distribution of temperature at the time
of the last scattering (LS). Therefore, different values of δΨ at different positions of the
Universe at this time generate the inhomogeneities of the CMB. In order to analyse
these inhomogeneities, one may compute the following 2-point correlation function :
δTγ δTγ δs δs
h (x) (y) i|LS ∼ h (x) (y) i|LS .
Tγ Tγ 3s 3s

We may use the fact that s(b) is only time dependent, so that it is spatially constant
(b)
at last scattering, with value denoted sLS . Using the fact that s ∼ s(b) , we may simplify
the relation above as :
δTγ δTγ 1
h (x) (y) i|LS ∼ (b) h δs(x) δs(y) i|LS .
Tγ Tγ 9sLS
14
We remind that the equivalence relation ∼ between two functions means that their quotient, when
it is well-defined, tends to 1 asymptotically.

67
 Then, writing δs in terms of the field perturbations δΨ and δ Ψ̄ allows to make
use of the explicit expansion of δΨ above. Such computation provides the expression
of δs, from which one deduces the correlation function. Finally, one may extract the
dimensionless power spectrum ∆T which is defined according to the following relation :
Z 3
δTγ δTγ dk 2
h (x) (y) iLS = ∆ (k) e−ik · (x−y) .
Tγ Tγ k3 T

A very important fact is that the dimensionless power spectrum turns out to be
nearly scale-invariant [8], which is consistent with the observations of the CMB. In the
context of Riemannian general relativity, this feature of the spectrum is also present in
the case of a scalar field driven inflation. However, it is not the case for a Dirac field
driven inflation which has a strong scale-dependent spectrum [9]. This means that the
introduction of torsion in the theory of gravity really cures the inconsistency of the
Dirac field model of inflation with the observations.
More precisely, in the torsion-free theory, the Dirac equation is linear, so that the
perturbation δ Ψ̃ is a solution of the Dirac equation as well. In this situation, δ Ψ̃ may also
be expanded in terms of the plane-wave solutions according to some functions à and B̃.
However, the latter are eventually responsible for the scale-dependence of the spectrum.
On the contrary, the functions A and B appearing in the expansion of δΨ are different
than in the linear case, and they really contain new properties needed by δΨ to solve
the nonlinear Dirac equation. In the end, these new characteristics of the functions A
and B provide the scale invariance of the spectrum obtained in Einstein-Cartan theory.

3.2 Bouncing universe from spinors

Besides providing an interpretation for the inhomogeneities of the CMB, inflation
is a central concept in cosmology because it solves the flatness and horizon problems.
Yet, as was mentioned earlier, Einstein-Cartan cosmology may also explain the flatness
and horizon problems, without assuming an extensive expansion of the Universe such
as inflation. In addition, the cosmological equations lead to a “bounce” of the scale
factor. This allows to avoid the Big Bang singularity that is present in standard gen-
eral relativity, and suggests that the Universe was actually contracting before it began
expanding.
In order to study Einstein-Cartan cosmology, we position ourself in a homogeneous,
isotropic universe described by a general FRW metric :

dr2
ds2 = dt2 − a2 (t) [ + r2 (dθ2 + sin2 θ dφ2 ) ] ,
1 − kr2
where a(t) is the scale factor and k ∈ {−1, 0, 1} is the spatial curvature.
Formulating the Einstein-Cartan equations in FRW will yield the equations ruling
the dynamics of the scale factor. These turn out to be analogous to the usual Friedmann
equations of general relativity. Moreover, comparing the original Friedmann equations
and the Einstein-Cartan cosmological equations will allow to precisely identify the in-
fluence of torsion, and hence spin, on the cosmological dynamics.

68
3.2.1 Friedmann equations with torsion
Let us first remind the symmetrised field equation relative to the vierbein, whose
structure closely relates to that of the original Einstein’s equation :
(ef f )
G̃µν = 8π Tµν .
On one hand, the left geometric term is only related to the Riemannian part of the
connection, that is the Levi-Civita connection, as it is the case of the original Einstein’s
tensor of general relativity. In the context of FRW metric, G̃µν results in having the
same expression as the Einstein’s tensor of general relativity :

2 k2
 G̃00 = 3 ( H + )

 a2
G̃0i = 0 ,

 G̃ = − ( 2aä + ȧ2 + k ) h

ij ij

1
where hij = diag ( 1−kr 2 2
2,r ,r sin2 θ ) is the spatial metric.
On the other hand, the effective density ρ(ef f ) and pressure p(ef f ) are defined from
(ef f )
the effective energy-momentum tensor Tµν in the same way that density and pressure
are defined from the energy-momentum tensor in general relativity :
( (ef f )
ρ(ef f ) := T00
1 (ef f )
.
p(ef f ) := a2 hii
Tii

As expected, the diagonal Einstein’s equations yield two equations with the exact
same form as the Friedmann equations of general relativity :
(
k 8π
H2 + a2
= 3
ρ(ef f )
ä
.
a
= − 4π
3
( ρ(ef f ) + 3p(ef f ) )

Yet, we remind that the effective energy-momentum tensor is defined as :

(ef f ) (spin)
Tµν := T̃µν + Tµν ,
where the torsion-free energy-momentum tensor T̃µν defines the torsion-free density ρ̃
and pressure p̃ that would appear in the original general relativity :
(
ρ̃ := T̃00
1
.
p̃ := a2 hii
T̃ii

Therefore, in order to identify in which ways the Einstein-Cartan theory modifies

the original Friedmann equations, it is interesting to reformulate the new Friedmann
equations in terms of ρ̃ and p̃. Due to the definition of the effective energy-momentum
tensor, the following relations hold :
( (spin)
ρ(ef f ) := ρ̃ + T00
1 (spin)
.
p(ef f ) := p̃ + a2 hii
Tii

69
 From the explicit computation of T (spin) , one eventually finds [10] the modifications
induced by the spin density on the effective density and pressure :
(
3π
ρ(ef f ) := ρ̃ − 2
A2
3π
.
p(ef f ) := p̃ − 2
A2

Comparing these to the expressions of pressure and density in standard general

relativity, adding spin through torsion in cosmology naively ends up in adding a con-
tribution from a fluid with equal negative density ρ(spin) and pressure p(spin) [10] :
3π 2
ρ(spin) = p(spin) = − A <0.
2
3k
Defining the usual (torsion-free) curvature density ρ̃k := − 8πa 2 , the Friedmann

equations of Einstein-Cartan theory finally rewrite as :

(
8π 3π
H2 = 3
( ρ̃ + ρ̃k − 2
A2 )
ä
.
a
= − 4π
3
( ρ̃ + 3p̃ − 6π A2 )

It is worth noting that, if H > 0, the condition for an accelerated expansion is

ρ̃ + 3p̃ < 6π A2 , which replaces the standard general relativistic condition ρ̃ + 3p̃ < 0.
Also, in standard general relativity, the second Friedmann equation is often replaced
by the conservation law ∇ ˜ µ T µν = 0. The latter is equivalently obtained from taking the
time derivative of the first Friedmann equation multiplied by a2 and combining it with
the second Friedmann equation. Similarly, such combination of the above Friedmann
equations with torsion yields the following conservation law :
d 3π 2
[ ρ̃ − A ] + 3H( ρ̃ + p̃ − 3π A2 ) = 0 .
dt 2
Therefore, all the equations of Einstein-Cartan cosmology in metric FRW may all
be recovered from the standard Friedmann equations with the addition of a particular
fluid with equal negative density ρ(spin) and pressure p(spin) . However, it should not be
forgotten that these artificial density and pressure are induced by the spin density of
the fluid made of the particles of the Dirac field, whose torsion-free density and pressure
are ρ̃ and p̃.

3.2.2 Bounce, flatness and horizon

Actually, the form of the cosmological equations with torsion are the same if we
consider additional fluids made of particles with vanishing spin density, such as a dust
fluid or a cosmological constant. This is due to the fact that the energy-momentum
tensors of the additional fluids contribute to ρ̃ only, while ρ(spin) remains unchanged.
We may also include radiation such as photons or neutrinos if we assume that their spin
densities are negligible compared to that of the spinor field ; otherwise, ρ(spin) would
contain additional terms to − 3π2
A2 , which is not considered here. In such a situation,
the first Friedmann equation rewrites as :
8π
H2 = ( ρ̃Λ + ρ̃k + ρ̃d + ρ̃r + ρ̃Ψ + ρ(spin) ) ,
3

70
where ρ̃Λ , ρ̃d , ρ̃r and ρ̃Ψ are the (torsion-free) densities corresponding to the cosmological
constant, dust, radiation and the Dirac fluid respectively.
Let us assume that each fluid obeys the usual conservation law along with a barotropic
equation :
(
ρ̃˙ C + 3H( ρ̃C + p̃C ) = 0
, C = Λ, d, r, Ψ.
p̃C = wC ρ̃C

Then, the densities explicitly rewrite in terms of the scale factor :

a0 3(1+wC )
ρ̃C = ρ̃C0 ( ) , C = Λ, d, r, Ψ,
a
where a subscript 0 denotes the value of a quantity measured in today’s Universe.
Regarding the quantity ρ(spin) , one finds [10] from statistical arguments that :
2

A2 ∝ ρ̃Ψ
1+wΨ
.

Using the explicit form of ρ̃Ψ given above, this yields

A2 ∝ a−6 ,

so that
(spin) a0 6
p(spin) = ρ(spin) = ρ0 ( ) ,
a
where we defined
(spin) 3π 2
ρ0 := −
A <0.
2 0
As a result, ρ(spin) satisfies the following conservation law :

ρ̇(spin) + 3H( ρ(spin) + p(spin) ) = 0

Therefore, the conservation law holds for all the quantities involved in the first
Friedmann equation. Owing to linearity, this guarantees that all the expressions above
are consistent with the general conservation law of Einstein-Cartan cosmology that was
obtained in the last section. This general law stipulates that :
d
[ ρ̃ + ρ(spin) ] + 3H( ρ̃ + p̃ + ρ(spin) + p(spin) ) = 0 .
dt
Finally, using the standard values of the coefficient wC for the cosmological constant
(wΛ = −1), dust (wd = 0) and radiation (wΛ = 13 ), the first Friedmann equation rewrites
as :
8π a0 a0 a0 (spin) a0 6
H2 = ( ρ̃Λ0 + ρ̃k0 ( )2 + ρ̃d0 ( )3 + ρ̃r0 ( )4 + ρ̃Ψ + ρ0 ( ) ),
3 a a a a
3k
where ρ̃k0 = − 8πa 2.
0
From this equation, and the fact that the Universe is expanding, come the following
remark : even if no assumption was made about the behaviour of ρ̃Ψ yet, the fact
that ρ(spin) behaves like a−6 shows that the contribution from this term soon becomes

71
negligible compared to ρ̃Λ , or even ρ̃d . It is the case as soon as the scale factor becomes
relatively large, which is quite early in the life of the Universe. This shows that the
effect of torsion in the first Friedmann equation of Einstein-Cartan theory is relevant at
very early times only, when the scale factor is very small. After this period, the Hubble
parameter naturally has the same behaviour as that predicted by general relativity,
since all torsion-related contributions become very small.
More precisely, at very early times, the torsion-related spin density ρ(spin) is among
the dominant contributions. This has a major impact on the evolution of H in the
beginning of the Universe, very different from that predicted by general relativity. In
fact, if we go back in time from today, the Universe contracts, i.e. a decreases. Reaching
the very early times of the Universe, the contributions from the cosmological constant,
curvature and dust surely become negligible owing to the presence of ρ(spin) . As for ρ̃Ψ ,
it is reasonable [10] to assume that the particles of the Dirac field are ultrarelativistic
due to the very high energies of the particles in early Universe. Therefore, they are
assimilated to radiation, that is wΨ = 13 . Then, at early times, one has :
a0 4
ρ̃Ψ = ρ̃Ψ0 ( ) .
a
Let us define
ρ̃0r0 := ρ̃r0 + ρ̃Ψ0 .
Removing the negligible contributions from the cosmological constant, curvature
and dust, the Friedmann equation at early times writes as :
8π 0 a0 4 (spin) a0 6
H2 = ( ρ̃r0 ( ) + ρ0 ( ) ).
3 a a
This relation is remarkable since it suggests that, as we go back in time, the con-
(spin)
traction stops before a = 0. More precisely, the negative sign of ρ0 allows H, and
hence ȧ, to vanish when the scale factor is :
s
(spin)
ρ0
abounce = a0 − 0 >0.
ρ̃r0
(spin)
Observations today provide the values of a0 along with the densities ρ̃0r0 and ρ0
[10], which yield abounce = 9 × 10−6 m. This is of course a very small value, yet much
bigger than the Planck scale.
If one wants to go further back in time, then the scale factor a must remain greater
than or equal to abounce since H 2 is obviously positive. The first possibility is that a
always equalled abounce , so that H = 0. This describes a universe that was static before
a perturbation caused the expansion to start.
The second possibility is that the dynamics of ȧ before abounce is reached is the exact
opposite of that after abounce . As a result, ȧ, and hence H, are odd functions of time if
the origin is set when abounce is reached, and then a and H 2 are even functions of time.
In particular, if we go forward in time, then a is decreasing before reaching abounce . This
means that the Universe is contracting until its scale factor reaches the value of abounce ,
and starts expanding after this value is reached, hence the term “bounce”.
This bounce is a very original and interesting feature of Einstein-Cartan theory
because in Riemannian general relativity, the scale factor is expected to reach 0 when

72
going back in time. This naturally leads the various densities involved in the Friedmann
equation to become infinite, which is difficult to accept as a physical reality of the history
of our Universe. Then, the Einstein-Cartan theory removes the Big Bang singularity at
the origin of the Universe, and hence provides a more satisfying picture of its history.
Yet, without this singular point, the model of bouncing universe does not provide
a clear origin to the history of the Universe : if there was a bounce, then the Universe
was undergoing a contraction phase before the bounce, but nothing in the theory indi-
cates when and how this contraction started. In particular, nothing in the Friedmann
equations suggests that the contraction itself followed a previous expansion phase, hap-
pening as some kind of inverse bounce when the scale factor would have reached a
maximal value. In this sense, the model of bouncing universe should not be confounded
with the model of ekpyrotic universe [11] which it might remind at first sight.

Besides avoiding the Big Bang singularity, Einstein-Cartan theory provides elegant
explanations to the flatness and horizon problems, since they do not assume any inflation
phase.
Regarding the flatness problem, we remind that the critical density ρcrit of the
Universe at time t is the total density a flat universe should have in order to have same
Hubble parameter. Therefore, one merely finds :

ρcrit (t) = ρ(ef f ) (t) + ρ̃k (t) .

From this quantity is defined the total density parameter at any time t :

ρ(ef f ) (t)
Ω(t) := .
ρcrit (t)

This quantity equals 1 if the Universe is flat (k = 0), it is greater than 1 if the
Universe is closed (k = 1), and smaller if the Universe is open ((k = −1). Actually,
observations suggest that our Universe could be closed, with Ω0 = 1.0023 [12], which is
very strangely close to 1.
By definition, one has :
3H 2 (t)
ρcrit (t) = .
8π
Therefore, the total density parameter rewrites as :
k
Ω(t) = 1 + .
ȧ2 (t)

If we neglect the recent dark energy epoch, the scale factor has behaved like tp with
p < 1 since the beginning of the radiation era and through the matter era. Therefore,
ȧ has behaved like a negative power of time since then, so that it has only decreased.
As a result, if the Universe is not flat, Ω(t) has only shifted away more and more from
the value 1 since the radiation era.
However, as we said, the observed value of Ω0 is really close to 1, which implies that
Ω must have been much closer to 1 in the past. For instance, more precise computations
actually require that, at the epoch of nucleosynthesis, |Ω − 1| < 10−16 at least.
Preceding the radiation era with an inflation phase, such as that discussed in the
first section of the chapter, provides a way to generate a sufficiently small value of

73
|Ω − 1| at the beginning of the radiation era. However, this requires the assumption of
an exponential de Sitter expansion of the Universe of more than 20 e-folds in the best
case.
On the contrary, Einstein-Cartan cosmology provides another solution to this prob-
lem without assuming any phase of exponential expansion. To present it, we remind
that, at early times, the Friedmann equation simplifies as :
8π 0 a0 4 (spin) a0 6
H2 = ( ρ̃r0 ( ) + ρ0 ( ) ).
3 a a
This immediately allows to rewrite the total density parameter at early times as :

3 a4 k
Ω := 1 + (spin)
.
8π a40 ( ρ̃0r0 a2 + ρ0 a20 )
(spin)
Then, the values of the densities ρ̃0r0 and ρ0 allow to numerically evaluate Ω for
√
any scale factor a. It turns out that when the scale factor equals 2 abounce , then the
total density parameter is such that |Ω − 1| < 10−63 which is extremely smaller than
10−16 . Therefore,
√ if nucleosynthesis was to stop approximately when the scale factor
equals 2 abounce , then the extremely small value of |Ω0 − 1| measured today would be
naturally predicted by Einstein-Cartan theory.

Let us also briefly mention that the horizon problem finds an elegant solution in
Einstein-Cartan theory as well. This problem arises from the observation that the CMB
is particularly homogeneous, which suggests that all the regions of the Universe might
have been causally connected at some stage. However, if radiation era started imme-
diately after the Big Bang, then the causally connected regions of spacetime represent
only 1◦ of the sky today.
In the scenario of inflation, it is assumed that the whole Universe had enough time
to be in causal contact so that properties such as temperature and density were homo-
geneous in the whole universe. Then, the violent expansion corresponding to inflation
spread all this homogeneous information everywhere in the immense volume that con-
stituted the Universe after inflation. The latter was then made of an immense number
of causally disconnected regions, yet sharing very similar properties. This would thus
lead to the present aspect of the CMB.
In the case of Einstein-Cartan theory, it is similarly assumed the all the regions of
the Universe became causally connected shortly before or after the bounce. Afterwards,
(spin)
the relatively small and negative value of ρ0 generated immense relative velocities
which got greater and greater than c as the Universe expanded [10]. This also resulted in
spreading the common data initially shared by the whole Universe to numerous causally
disconnected regions of the Universe. Such a scenario does not require any exponential
(spin)
expansion and is essentially based on the parameter ρ0 , which is only present in
Einstein-Cartan theory.

74
Conclusion

Einstein-Cartan theory extends general relativity so as to include torsion and spin.

Such generalisation is achieved by reformulating the action of general relativity in terms
of vierbein and spin connection, and allowing these to vary. Among the corresponding
Euler-Lagrange equations, that relative to the spin connection explicitly reformulates
torsion in terms of spin density. This allows to substitute the former by the latter every-
where in the theory. On the other hand, the second Euler-Lagrange equation precisely
generalises the original Einstein’s equation. In fact, reformulating it in a symmetric way
reveals that the new equation is a close analogous of the original Einstein’s equation
in that it merely adds an extra spin-induced term to the effective energy-momentum
tensor.
In presence of a matter field, the total action involves a matter term, which may be
varied with respect to the matter field in order to yield the equation ruling its dynamics.
Since the fundamental feature of Einstein-Cartan theory is the introduction of spin as a
source for gravitation, it is tempting and relevant to consider matter fields with non-zero
spin. Among these are the spin-1/2 spinor fields, spin-1 vector fields, spin-3/2 Rarita-
Schwinger fields and spin-2 second-order tensor fields. Spinor fields are central objects
of the standard model and, although they are not used very widely in cosmology, they
may bring interesting cosmological phenomena through the Einstein-Cartan formalism.
For these reasons, this dissertation focused on these particular fields.
Yet, in order to use spinors in Einstein-Cartan theory, one needs to settle the proper
framework required to consider and analyse spinor fields on almost any spacetime. The
first step in this direction consists in introducing the concept of principal G-bundle,
where G is a fixed Lie group. Principal bundles are fundamental bundles in the sense
that each of them generates a family of bundles called associated bundles. Provided
that there exists a spin structure on spacetime, a spinor bundle is actually built as an
associated bundle to a principal G-bundle, where G is a spin group. If G = Spin(3, 1),
the fiber of the spinor bundle may be chosen to be the representation space whose
elements are exactly the spinors used in the standard model. This provides the exact
framework needed to define Dirac spinor fields as sections of such a spinor bundle.
Dirac spinors are meant to take part in the action of Einstein-Cartan theory. There-
fore, the partial derivatives that appear in the Minkowskian Dirac action have to be
extended to any type of manifold. It is thus needed to prescribe a satisfying way to define
the variations of a spinor field over spacetime. Through the notions of vertical bundle,
connection form and gauge potential, one may finally define the covariant derivative for
spinor fields. This object provides the right extension of partial derivatives in order to
formulate the Dirac equation on any spacetime.
In addition to the covariant derivative, Einstein-Cartan theory makes use of the
notions of curvature and torsion as well. Although the former is already present in

75
general relativity, the latter is not. Torsion is only defined when the tangent bundle of
spacetime is an associated bundle of the principal Spin(3, 1)-bundle that was used to
build the spinor bundle. All these notions, reviewed in the first chapter, also constitute
all the geometrical aspects of the gauge theories, which make permanent use of the
notions of gauge fields and field strength.
With these various objects in hands, the action of Einstein-Cartan theory in presence
of a spinor field can be properly considered for any suitable spacetime. This allows to
compute the energy-momentum and spin density tensors relative to the spinor field.
As mentioned above, it is possible to substituting everywhere in the action torsion by
the spin density. This yields a four-fermion interaction term, which is a very original
feature of Einstein-Cartan theory. This quartic interaction results in a nonlinear Dirac
equation ruling the spinor field.
All these results were introduced in the second chapter in order for the third chapter
to present how Einstein-Cartan theory may handle some major cosmological issues with
the help of a spinor field. The inhomogeneities of the CMB constitute such a cosmo-
logical issue. They are usually understood as consequences of the quantum fluctuations
of a scalar field. This interpretation is made within the framework of standard quan-
tum field theory in curved spacetime, which makes use of the original Einstein’s field
equations. Yet, it turns out that a spinor field may actually fulfil the same function
as the scalar inflaton, under the important condition that the analysis is carried out
within the framework of Einstein-Cartan theory. This condition allows to take the spin-
induced effects into account since they are not present in standard general relativity.
Eventually, this approach succeeds in recovering important properties of the CMB such
as scale invariance.
Besides providing a second type of field able to drive inflation, Einstein-Cartan
theory suggests solutions to the flatness and horizon problems. This combines with the
fact that Einstein-Cartan cosmology leads to a model of universe which was contracting
before reaching a critical non-zero scale factor, from which it started the expansion phase
that the Universe is still undergoing nowadays. These results derive from the equations
corresponding to the Friedmann equations corrected with torsion-related terms, that is
spin-related terms.
In particular, these equations suggest that, in a general isotropic and homogeneous
model including dust, radiation, a cosmological constant and a Dirac field, spin effects
dominate in early ages of the Universe, and vanish very rapidly as the scale factor
increases. This means that, numerically, the changes brought by Einstein-Cartan theory
to the standard Friedmann equations are only relevant at very high densities, which
cannot be found in today’s Universe. Therefore, for a long part of the history of the
Universe, gravity has been extremely close to be Riemannian, even in Einstein-Cartan
theory. This fact prevents from determining which of Riemannian general relativity
or Einstein-Cartan theory provides the most accurate description of gravity, since no
experiment sensitive enough can yet be conceived to detect torsion effects in today’s
Universe.
However, even if this difference between the Riemannian Friedmann equations and
the Friedmann equations with torsion only affects a short period of time, it is sufficient to
explain the flatness and horizon problems from the spin density only, instead of requiring
the numerous e-folds of expansion involved in the standard inflationary models.

Of course, this dissertation covered only a small number of the aspects of Einstein-

76
Cartan theory, and a lot more issues may be discussed about this theory. For instance,
the allowance for torsion in Einstein-Cartan theory of gravity leads to a four-fermion
interaction term which was explicitly derived in the second chapter. Very interestingly,
it was discussed [13] that such an interaction might lead to particle pairs production
when densities are high enough, particularly in the very early Universe.
Also, as Einstein-Cartan theory provides a new description of gravity, it is natural to
wonder about its renormalisability when one tries to quantise it [14]. Actually, similarly
to standard general relativity, Einstein-Cartan theory is not renormalisable [15]. Yet, the
action of Einstein-Cartan theory, which was introduced in the second chapter, may be
extended with an additional term called the Holst term. This term does not modify the
classical equations of motion of Einstein-Cartan theory when the spin density vanishes.
However, the Holst term might be an important element in the path toward a quantum
description of gravity since it lies at the root of Loop Quantum Gravity [16, 17]. In this
sense, Einstein-Cartan theory, which extends general relativity by giving the quantum
concept of spin a much greater role in the description of gravity, may be a step closer
than general relativity toward quantum gravity.

77
References

[1] Alexey Golovnev, Viatcheslav Mukhanov, and Vitaly Vanchurin. Vector Inflation.
JCAP 0806:009, 2008.

[2] Paul Bourke. Möbius (Moebius) strip. http://paulbourke.net/geometry/

mobius/, 1997.

[3] Andreas Rejbrand. The Rejbrand Encyclopædia of Curves and Surfaces. http:
//trecs.se/theory.php, 2012-2016.

[4] Rupert Way. Introduction to connections on principal fibre bundles. March 2010.

[5] A. Borel and F. Hirzebruch. Characteristic Classes and Homogeneous Spaces I.

American Journal of Mathematics, 80:458–538, 1958.

[6] Jeffrey Yepez. Einstein’s vierbein field theory of curved space. arXiv:1106.2037v1,
2008.

[7] Andrzej Trautman. Einstein-Cartan theory. Encyclopedia of Mathematical Physics,

2:189–195, 2006.

[8] Tomoki Watanabe. Dirac-field model of inflation in Einstein-Cartan theory.

arXiv:0902.1392v1, 2009.

[9] C. Armendáriz-Picón and P. B. Greene. Spinors, Inflation, and Non-Singular Cyclic

Cosmologies. General Relativity and Gravitation, 35:1637–1658, 2003.

[10] Nikodem J. Poplawski. Cosmology with torsion : An alternative to cosmic inflation.

Phys. Lett. B694 ; Erratum-ibid. B701, pages 181–185, 2010.

[11] Paul J. Steinhardt and Neil Turok. A Cyclic Model of the Universe. arXiv:hep-
th/0111030, 2001.

[12] D. Larson et al. Seven-Yar Microwave Anisotropy Probe (WMAP) Observations :

Power Spectra and WMAP-Derived Parameters. The Astrophysical Journal Sup-
plement Series, 192(2), 2011.

[13] G. David Kerlick. Cosmology and particle pair production via gravitational spin-
spin interaction in the Einstein-Cartan-Sciama-Kibble theory of gravity. Phys.
Rev. D, 12:3004–3006, 1975.

[14] Stefano Lucat. Cosmological singularity and bounce in Einstein-Cartan-Kibble-

Schiama Gravity. PhD thesis, Utrecht University, 2004.

78
[15] Ilya L. Shapiro and Poliane M. Teixeira. Quantum Einstein-Cartan theory with
the Holst term. Classical and Quantum Gravity, 31(18), 2014.

[16] Marcin Kaźmierczak. Einstein-Cartan gravity with Holst term and fermions. Phys.
Rev. D, 79, 2009.

[17] Kinjal Banerjee. Some Aspects of Holst and Nieh-Yan Terms in General-Relativity
with Torsion. Classical and Quantum Gravity, 27(13), 2010.

[18] Werner Greub, Stephen Halperin, and Ray Vanstone. Connections, curvature, and
cohomology, volume 2. Academic Press, 1973.

[19] Roger Penrose and Wolfgang Rindler. Spinors and space-time, volume 1. Cam-
bridge University Press, 1984.

[20] Mikio Nakahara. Geometry, topology and physics. Institute of Physics Publishing,
2003.

[21] Robert Coquereaux. Espaces fibrés et connexions. http://www.cpt.univ-mrs.

fr/~coque/link-to-book.html, 1997.

[22] F.W. Hehl, P. Von Der Heyde, G.D. Kerlick, and J.M Nester. General Relativity
with Spin and Torsion : Foundations and Prospects. Rev. Mod. Phys., 48:393-416,
1976.

[23] J. Magueijo, T.G. Zlosnik, and Kibble T.W.B. Cosmology with a spin. Phys. Rev.
D, 87:063504, 2013.

[24] Tomoki Watanabe and Mitsuo J. Hayashi. General Relativity with Torsion.
arXiv:gr-qc/0409029v1, 2004.

[25] Nikodem J. Poplawski. Big bounce from spin and torsion. Gen. Relativ. Gravit.,
44:1007–1014, 2012.

[26] Brian P. Dolan. Chiral fermions and torsion in the early Universe. Classical and
Quantum Gravity, 27, 2010.

79