Dados Internacionais de Catalogação na Publicação (CIP)

G967l Guisoli, Felipe Brandão.

The languages of spacetime / Felipe Brandão Guisoli. – 2022.
102f. : il.

Orientador: Nelson de Oliveira Yokomizo.

Dissertação (mestrado) – Universidade Federal de Minas Gerais,
Departamento de Física.
Bibliografia: f. 95.

1. Relatividade geral. 2. Gravitação. I. Título. II. Yokomizo, Nelson de

Oliveira. III. Universidade Federal de Minas Gerais, Departamento de Física.

CDU – 530.12 (043)

Ficha catalográfica elaborada por Romário Martins Ribeiro – CRB6 3595

Biblioteca Professor Manoel Lopes de Siqueira – Departamento de Física - UFMG
1/22/23, 7:23 PM SEI - Documento para Assinatura
Processo: Documento:

23072.274488/2022-85 1981708

UNIVERSIDADE FEDERAL DE MINAS GERAIS

INSTITUTO DE CIÊNCIAS EXATAS
PROGRAMA DE PÓS-GRADUAÇÃO EM FÍSICA

ATA DE DEFESA DE DISSERTAÇÃO

ATA DA SESSÃO DE ARGUIÇÃO DA 688ª DISSERTAÇÃO DO PROGRAMA DE PÓS-

GRADUAÇÃO EM FÍSICA, DEFENDIDA POR FELIPE BRANDÃO GUISOLI, orientado pelo
professor Nelson de Oliveira Yokomizo, para obtenção do grau de MESTRE EM FÍSICA. Às 14 horas de
vinte de dezembro de 2022, reuniu-se a Comissão Examinadora, composta pelos professores Nelson de
Oliveira Yokomizo (Orientador - Departamento de Física/UFMG), Mario Sergio Carvalho Mazzoni
(Departamento de Física/UFMG) e Gláuber Carvalho Dorsch (Departamento de Física/UFMG), para dar
cumprimento ao Artigo 37 do Regimento Geral da UFMG, submetendo o bacharel FELIPE BRANDÃO
GUISOLI à arguição de seu trabalho de dissertação, que recebeu o título de “The Languages of
SpaceTime”. O candidato fez uma exposição oral de seu trabalho durante aproximadamente 50 minutos.
Após esta, os membros da comissão prosseguiram com a sua arguição e apresentaram seus pareceres
individuais sobre o trabalho, concluindo pela aprovação do candidato.

Belo Horizonte, 20 de dezembro de 2022.

Prof. Nelson de Oliveira Yokomizo

Orientador do estudante
Departamento de Física/UFMG

Prof. Mario Sergio Carvalho Mazzoni

Departamento de Física/UFMG

Prof. Gláuber Carvalho Dorsch

Departamento de Física/UFMG

Candidato: Felipe Brandão Guisoli

Documento assinado eletronicamente por Gláuber Carvalho Dorsch, Professor do Magistério

Superior, em 21/12/2022, às 16:54, conforme horário oficial de Brasília, com fundamento no art. 5º
do Decreto nº 10.543, de 13 de novembro de 2020.

Documento assinado eletronicamente por Mario Sergio de Carvalho Mazzoni, Membro, em

21/12/2022, às 18:20, conforme horário oficial de Brasília, com fundamento no art. 5º do Decreto nº
10 543 de 13 de novembro de 2020
[Link] 1/2
1/22/23, 7:23 PM SEI - Documento para Assinatura
10.543, de 13 de novembro de 2020.

Documento assinado eletronicamente por Felipe Brandão Guisoli, Usuário Externo, em 22/12/2022,
às 09:49, conforme horário oficial de Brasília, com fundamento no art. 5º do Decreto nº 10.543, de 13
de novembro de 2020.

Documento assinado eletronicamente por Nelson de Oliveira Yokomizo, Professor do Magistério

Superior, em 22/12/2022, às 15:24, conforme horário oficial de Brasília, com fundamento no art. 5º
do Decreto nº 10.543, de 13 de novembro de 2020.

A autenticidade deste documento pode ser conferida no site

[Link]
acao=documento_conferir&id_orgao_acesso_externo=0, informando o código verificador 1981708 e o
código CRC 62E81DA0.

Referência: Processo nº 23072.274488/2022-85 SEI nº 1981708

[Link] 2/2
 Agradecimentos

Agradeço primeiramente a Deus, pela oportunidade de estar vivo e poder contemplar

a beleza escondida na natureza.
Agradeço ao meu orientador, professor Nelson de Oliveira Yokomizo, não só pela
sua enorme disposição e facilidade em transmitir o conhecimento, cujas discussões impor-
tantíssimas muito me engrandeceram e abriram novos horizontes, mas também pela sua
paciência e amizade. Não tenho dúvidas de que tive o melhor orientador que pude, você
com certeza se tornou uma referência de pesquisador e professor para mim.
Agradeço também à minha família. Primeiro à minha mãe Márcia e meu pai
Cláudio, que me educaram e me proporcionaram a condição e a autonomia de seguir um
caminho voltado ao intelecto e ao conhecimento. Tudo que vivo hoje é fruto das condições
que vocês me proporcionaram, muito obrigado. Agradeço também à minha irmã, Danielle,
por sempre acreditar no meu potencial e me incentivar. É muito bom saber que você está
sempre na torcida.
Agradeço à minha esposa Carol, que faz a vida infinitamente mais divertida.
Obrigado por me apoiar e me incentivar nas melhores e piores circunstâncias. Obrigado
por me escutar tantas vezes falando de física e de matemática. É um privilégio poder
dividir a vida com você.
Aos colegas do Grupo de Física Teórica da UFMG, estudantes, mestrandos e
doutorandos, cujas discussões e entusiamo no estudo da física e da matemática sempre me
lembravam do motivo de eu ter escolhido estudar esses dois assuntos.
Agradeço ao meu amigo e professor Wilio Torres, cujas aulas e conversas enquanto
eu ainda estava no ensino médio me fascinaram e me incentivaram a seguir o caminho da
Física. É um prazer poder dividir o fascínio do estudo da natureza com você.
Aos membros da equipe do Universo Narrado, que dividem comigo a missão e visão
de uma educação que transforma e que ajuda as pessoas a se tornarem mais inteligentes.
Agradeço, ainda, aos professores da UFMG por ajudarem a iluminar a jornada da
Física e à CNPQ pelo apoio financeiro.
Todas as coisas cujos valores podem ser disputado no cuspe à distância servem para a
poesia.
As coisas que não levam a nada têm grande importância.
(Manoel de Barros)
 Resumo
A teoria da relatividade geral surge em 1915, e seu palco matemático é uma variedade
quadridimensional Lorentziana. Nosso objetivo será explorar diferentes linguagens e formu-
lações da teoria, com diferentes parâmetros atuando como variáveis dinâmicas. Iniciaremos
com a formulação original desenvolvida por Einstein e passaremos então para a formulação
lagrangeana da teoria, desenvolvida primeiramente por Hilbert. Desenvolveremos então
duas formulações hamiltonianas da gravitação, baseadas em uma folheação (3+1) do
espaço-tempo. A primeira será feita com o uso da métrica tridimensional como variável
dinâmica. Tal formalismo é conhecido como formalismo ADM da relatividade geral e
possibilita a construção de uma Hamiltoniana para a teoria em termos de vínculos e
multiplicadores de Lagrange. Por fim, analisaremos a formulação hamiltoniana baseada na
ação de Holst, em termos das variáveis de Ashtekar. Tal formulação é um dos caminhos
possíveis para a quantização do campo gravitacional, na abordagem conhecida como
gravitação de laços.

Palavras-chave: Relatividade Geral, Gravitação, Formalismo Hamiltoniano, Variáveis de

Ashtekar, Gravitação Quântica de Laços
 Abstract
The theory of general relativity emerges in 1915, and its mathematical stage is a four-
dimensional Lorentzian manifold. Our goal will be to explore different languages and
formulations of the theory, with different quantities playing the role of dynamical variables.
We will start with the original formulation of the theory developed by Einstein, and then
pass to the Lagrangian formulation, first developed by Hilbert. Then we will develop two
Hamiltonian formulations of gravity, based on a (3+1) foliation of spacetime. The first will
be done with the three-dimensional spatial metric as dynamical variable. Such a formalism
is known as the ADM formalism of general relativity and allows for the construction of a
Hamiltonian for the theory in terms of constraints and Lagrange multipliers. Finally, we
will analyze the Hamiltonian formalism based on the Holst action, in terms of Ashtekar
variables. This formulation provides a possible path for the canonical quantization of the
gravitational field, in the approach known as loop quantum gravity.

Keywords: General Relativity, Gravitation, Hamiltonian Formalism, Ashtekar Variables,

Loop Quantum Gravity
CHAPTER 1. INTRODUCTION 13

Figure 1.1: The square of the sum

that will lead us to build a broader view of the big picture.

This is our goal: to explore the different languages of spacetime. Therefore, in this work, we
will explore some different ways to write the formalism of GR, with different variables playing
the dynamical role. We will start with the first formalism developed by Einstein, then we will go
to the Lagrangian formalism, done first by Hilbert, and finally we will end up in some different
Hamiltonian formulations of the theory, one in terms of the metric, which is known as the ADM
formalism, and another one in terms of some other variables — the Ashtekar variables — which
is one possible path to quantize the theory of gravity.
The present work is organized as follows. In Chapter 2 we will set the mathematical
background and conventions that will be used along the dissertation. The essence will be
differential geometry, tensor calculus and the establishment of some notations. In Chapter 3
we will recover the main concepts of the formulation of GR done by Einstein and the main
geometrical objects in this theory, like the metric, the Riemann curvature tensor and the
connection. We will also discuss Einstein’s equivalence principle, the base upon which GR lays.
Furthermore, we are going to do here the Lagrangian formulation of GR, via Einstein-Hilbert
action and also via the Palatini action, where we will let the connection play a dynamical
role in the theory. In Chapter 4 we will review the formalism and structure of constrained
Hamiltonian systems, and then we will apply it to the Hamiltonian formulation of GR — the
ADM formalism. We will build this formalism in detail, via the foliation of spacetime, where it
will be split in spatial slices evolving through time. This (3+1) split will allow us to tell the
history of spacetime as the time evolution of these spatial slices. With these, we will be able to
build an action for gravity in therms of a 3-metric hab in those slices — this will be the main
dynamical variable of our formalism. We will end up with a constrained Hamiltonian system,
and the symmetries of spacetime will be expressed as constraints in the Hamiltonian.
In chapter 5 we will construct the tetrad formalism, where we will trade the metric gµ‹ for
the local orthonormal frame eIµ as the dynamical variable. We will build all the formalism in
terms of these new variables using Cartan’s structural equations of differential geometry, ending
with the Holst action in forms notation, which will be used for the next formalism. Finally, in
chapter 6 we are going to mix up the two previous formalisms to build the Ashtekar formulation
of GR, which is essentially the construction of the Hamiltonian formalism using triads as the
dynamical variables in place of the metric. We will do the same (3+1) split, where the spatial
part of the tetrad eIµ will be ÁIµ — the triad, our main dynamical variable. This formalism,
developed by Ashtekar in the mid 1980s, consists in rewriting the theory of GR in terms of
some variables that made the theory resemble the theories of particle physics, which allowed the
importation of techniques from particle physics to the quantization of gravity. This approach is
CHAPTER 1. INTRODUCTION 14

known as loop quantum gravity.

Finally, we will introduce present our conclusions and discuss future developments in chapter
7.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 16

Continuous function and homeomorphism

The notion of continuos functions is allowed by the use of a topology. Suppose we have
the two topological spaces X and Y , and the function f : X æ Y . This function is said to be
continuous if, for any open set U œ Y , its inverse image f ≠1 (U ) œ X is also an open set in X,
as is shown in figure 2.1.

Figure 2.1: Continuous function from X to Y .

If the map f is a continuous and bijective between two topological spaces, whose invese is
also continuous, then f is called a homeomorphism.

2.1.3 The manifold

Our idea here is to cover a space with patches that are locally just as Rn .
We say that a collection of open sets Ua covers a topological space X if their union is all of
X.
For an open set U œ X we define a chart to be a continuous function Ï : U æ Rn with a
continuous inverse (where this inverse has its domain in Ï(U ) œ Rn , just as figure 2.2 shows.)

Figure 2.2: Charts.

The idea is that, as long as we work in the chart Ï we can pretend we are in Rn , just as the
Earth looks perfectly flat if we do not go too far. Suppose, for example, we have a function
f : U æ R. We can turn it into a function from Rn æ R using f ¶ Ï≠1 , as figure 2.3 shows.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 17

Figure 2.3: Turning functions in U in functions in Rn .

Definition
A topological n-dimensional manifold M is a topological space such that every point has a
neighbourhood U homeomorphic to an open subset in Rn .
The manifold M is differentiable if the transition function Ï≠1 – ¶ Ï— is smooth where it is
defined.
Some topological manifolds — the differentiable ones — M can be represented as a union of
finite set of coordinate charts U , and the set of coordinate charts Âu that cover M is called an
atlas on M .
The idea is that every point in a differentiable manifold lives in some open subset U– that
looks like Rn , and that we can tell if any function on the manifold is smooth by looking at
transition functions between charts. If there is a function f : M æ R and one uses a chart
Ï– : U– æ Rn , then we say that f is smooth if

– : R æ R
f ¶ Ï≠1 n

is smooth.
But one could instead use a chart Ï— : U— æ Rn . In this case, consider V = U– fi U— the
overlap of the two charts, the grey area represented in figure 2.4. The representation of f in
this chart is
— : R æ R.
f ¶ Ï≠1 n

This function should also be smooth, for the smoothness of a function does not depend on the
chart we use.
But for that to be true, we need
Ï– ¶ Ï≠1
—

to be smooth, since 1 2 1 2
— = f ¶ Ï–
f ¶ Ï≠1 ¶ Ï– ¶ Ï≠1
≠1
— .
From now on, when we mention any manifold, we will always be referring to a smooth
manifold, as defined above.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 18

Figure 2.4: A manifold M .

Diffeomorphism
An isomorphism is a structure-preserving mapping between two structures of the same type
that can be reversed by an inverse mapping. An homeomorphism, as previously defined, is an
isomorphism of topological spaces. A diffeomorphism is a homeomorphism that preserves a
differential structure.

Definition 2.1.1. Given two manifolds M and N , a differentiable map f : M æ N is called

a diffeomorphism if it is a bijection and if its inverse f ≠1 : N æ M is also differentiable.

A map f from a manifold M to another manifold N can be built if one has the maps
g : M ‘æ R and h : N ‘æ R, as shown in figure 2.3, by the composition h≠1 ¶ g.
If there exists a diffeomorphism f between M and N we say that these two manifolds are
diffeomorphic. We will consider the spacetime as a 4-dimensional differentiable manifold.

2.2 Vectors
2.2.1 Introduction
One can think of a vector field in a manifold as a field of arrows, tangent to the space in each
point, as it is in Rn . If we have a direction, we can differentiate a function f in that direction.
The partial derivative of f in the direction of a vector v is, in Rn :

vf = v · Òf = v µ ˆµ f,
where we are thinking of the vector v as something whose purpose is to get a function f and
spit out another function, which is the partial derivative of f in the v direction, that’s why we
wrote it as vf (something like v is an operator acting on f ).
If we look at the first and last member we have vf = v µ ˆµ f , which holds for every function
f , so one may think that we can say that a vector field v can be written as

v = v µ ˆµ , (2.1)
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 19

which says that the vector field v can be expanded in the basis ˆµ .
Here, v is the vector field, while v µ ˆµ is something that acts on a function and give its partial
derivative. For us, a vector field on a manifold will be exactly that: entities whose main purpose
is to differentiate functions.

Figure 2.5: Vector Field.

Definition
A vector field v on a manifold M is a function from C Π(M ) to C Π(M ) satisfying:

• v(f + g) = v(f ) + v(g)

• v(–f ) = –v(f )

• v(f g) = v(f )g + f v(g)

for – œ R and f, g œ C Œ (M ). Here, C Œ (M ) stands for the set of all complex functions infinitely
differentiable in M , as usual.
So it is an object that acts linearly on functions and obeys the Leibniz rule. If we denote
by V (M ) the set of all vector fields in a manifold M one can show that this is indeed a vector
space, as expected.

2.2.2 Tangent Vectors

One can visualize a vector field v in a manifold M as assigning an arrow to each point P in
the tangent space of the manifold. The tangent vector at each point P in M is the vector vp
living in the tangent plane at P, as showed in figure 2.6
We can differentiate the function f in the direction of the vector field v, represented by vf ,
and evaluate it in the point p œ M . We will call this vp f — the tangent vector in the point P.
So we have
vp : C Œ (M ) æ R , vp (f ) = v(f )(p)
where the last line means the partial derivative of f in the direction of the vector field v evaluated
at point P.
It follows immediately that
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 20

Figure 2.6: Tangent Space.

• vp (f + g) = vp (f ) + vp (g)
• vp (–f ) = –vp (f )
• vp (f g) = vp (f )g + f vp (g)
and we call the tangent vector vp at point P the function C Œ æ R that satisfies these 3
properties.

We call Tp (M ), the tangent space at P, the set of all tangent vectors at p œ M . The tangent
space is indeed a vector space, with the sum of tangent vectors and the multiplication by a
scalar defined in the natural way
• (vp + Êp )(f ) = vp (f ) + Êp (f )
• (–vp )(f ) = –vp (f )

2.2.3 Lie Bracket

We define the Lie Bracket of two vectors v and w as
[v, w] = vw ≠ wv, (2.2)
which is just a short notation for
[v, w](f ) = v(w(f )) ≠ w(v(f )).
So, if v and w are vector fields, the Lie bracket [v, w] is also a vector field, since its entry is
a function f œ C Œ (M ) and it spits out another function in C Œ (M ).
For the basis vector ˆµ and ˆ‹ the Lie bracket is evidently zero, which follows from the
commutation of partial derivatives:
ˆµ ˆ‹ = ˆ‹ ˆµ .
Geometrically this can be thought as flowing a little bit in the ˆµ direction and then a little
bit in the ˆ‹ direction. If we invert the order we end up in the same place, at least in flat space.
For general vector fields this is not necessarily true, and the Lie bracket measures the
difference between these who tracks, the failure of the two vector fields to commute, as shown
in figure 2.7.
The Lie derivative Lw v of a vector v in the direction along w is defined as
Lw v = [w, v], (2.3)
which is the derivative of v along the flow [16] generated by w.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 21

Figure 2.7: Lie Bracket.

2.3 Differential Forms

2.3.1 Introduction
The initial idea is to generalize the notion of the gradient of a function to functions on
arbitrary manifolds.
For a function f on Rn we have its gradient expressed by Òf . Here, we will define an
operator d — the exterior derivative — and its action on a function f defined on an arbitrary
manifold M will be expressed as df , and it will generalize the idea of the gradient in a first
approach.
In Rn , the directional derivative of f in the direction of the vector v is just the dot product
between v and the gradient of f :

Òf · v = vf, (2.4)
where in the last step we have written the vector v in the vector basis ˆµ :
v = v µ ˆµ ,
hence
ˆf
vf = v µ ˆµ f = v µ = Òf · v.
ˆxµ
So, we are after an object df that keeps track of the derivative of f in all directions in
the manifold M , just as the gradient does. In Rn , the gradient of f is a vector field, and the
directional derivative is calculated via a dot product, as shown before. But, taking dot products
involves a choice of metric, and manifolds, in general, do not come pre-equipped with it. So, we
will leave the choice of a metric to a further development. Hence it would be nice if, in a first
approach, the df which will generalize the gradient was not a vector field, so that it would not
be necessary to take a dot product in order to extract the directional derivative information.
We will call our df here a 1-form, and it will have the same properties as the gradient does,
so to speak, for each input vector v the operator df · v = vf spits out a scalar function, which is
the directional derivative of f in the direction of v.
So, our df , when fed with a vector v œ V (M ) (the tangent vector space in a manifold M )
will spit out a function g œ C Œ (M ), and it will do it in a linear way, such as the gradient does:
df · (v + u) = df · v + df · u ,
df · (gv) = g (df · v) , (2.5)
for g œ C Œ (Rn ).
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 22

Definition 2.3.1. A 1-form Ê on a manifold M is a map from V (M ) to C Œ (M ) that is linear

over C Π(M ).
So, a 1-form Ê receives a vector v from V (M ) and spits out a function Ê(v) such that:

Ê(v + u) = Ê(v) + Ê(u) ,

Ê(gv) = gÊ(v) , (2.6)
where g œ C Œ (M ). We represent the space of all 1-forms in a manifold M by 1
(M ).

The exterior derivative as the generalization of the gradient

A simple example of a 1-form is, for any smooth function on M , the 1-form defined by:
df (v) = vf, (2.7)
which is just a slick way to write the directional derivative, as observed in (2.4). We can see
that this is really a 1-form by checking linearity:
df (v + u) = (v + u)f = vf + uf = df (v) + df (u),
and
df (gv) = (gv)f = g(vf ) = gdf (v).
The 1-form df is called the differential of f , or the exterior derivative of f .

Composition
The addition of two 1-forms Ê and µ and multiplication by a scalar (function) g is defined
via
(Ê + µ)(v) = Ê(v) + µ(v) (2.8)
and
(gÊ)(v) = gÊ(v) . (2.9)

2.3.2 The Tangent and Cotanget spaces

Let us see what the exterior derivative is in any manifold, working in local coordinates.
From equation (2.7) we can conclude that the 1-forms dxµ form, at each point P , a local basis
of 1-forms in Tpú (M ) — the dual space of Tp (M ) — because, when we feed the 1-form dxµ with
a basis vector of the tangent space ˆ‹ we get
ˆxµ
dxµ (ˆ‹ ) = = ”‹µ .
ˆx‹
So, if ˆµ is a basis of the tangent space on a manifold M and the action of the 1-forms
dx on that basis gives the Kronecker delta, then dx‹ is also a basis in the cotangent space.
‹

Therefore any 1-form Ê œ 1 (M ) can be expanded and written in a unique form as

Ê = ʵ dxµ , (2.10)
with
ʵ = Ê(ˆµ ).
To see that this is the case, we just need to verify that the action of Ê and ʵ dxµ on a vector
v are the same:
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 23

1. Ê(v) = Ê(v ‹ ˆ‹ ) = v ‹ Ê‹
2. ʵ dxµ (v) = ʵ dxµ (v ‹ ˆ‹ ) = v ‹ ʵ (ˆ‹ dxµ ) = v ‹ ʵ (”‹µ ) = Ê‹ v ‹

which proves the statement.

One can then see the 1-forms as actually dual vectors. Just as a vector field v at M gives a
tangent vector vp at each point P of M , we can assign a cotangent vector Êp at each point P of
M . The space of all cotangent vectors at P, as mentioned before, is called Tpú M . The cotangent
vector Ê at P is rigorously defined to be a linear map from the tangent space Tp M to R.
So, if we have a vector field v on M , we can define the cotangent vector field as

Ê(v) = ʵ dxµ v µ ˆµ = ʵ v µ , (2.11)

which is indeed a map Ê(v) : Tp M ‘æ R.

This really means that the 1-forms are the dual vectors of v. This is so since the dual vector
space of V is the space V ú of all linear functionals Ê : V ‘æ R. Hence, the cotangent space Tpú M
is the dual vector space of Tp M .
It is important to note that, if we have a linear map f from one vector space V to another
W
f : V ‘æ W,
we can automatically get a map f ú , the dual of f , from W ú to V ú

f ú : W ú ‘æ V ú ,

that is defined by
(f ú Ê)(v) = Ê(f (v)). (2.12)
For this we call the cotangent vectors covariant: linear maps between vector spaces gives rise
to maps between their duals that go backwards. This is the convention used in [5], probably
because this objects transforms with the same Jacobian matrix of the linear transformation
itself, while tangent vectors transforms with its inverse, hence, they are called contravariant.
We will develop more on this shortly.
So, if „ is a linear map between the tangent spaces at two different points P and Q in M

„ : Tp M ‘æ Tq N,

the dual map goes the other way

„ú : Tqú N ‘æ Tpú M.
We call „ú Ê the pullback of Ê by „.
In coordinates this means that the 1-forms, when we do a coordinate transformation, will
transform with the inverse of the matrix that transform the coordinates of the vectors.
For instance, let the vector v be expressed in two different coordinate systems xµ and xÕ‹

v = v µ ˆµ = v Õ‹ ˆ‹Õ . (2.13)

The object v is naturally the same, but its components v µ or v Õ‹ are not, since they depend
on the choice of basis ˆµ or ˆ‹Õ where the components are written.
ˆxµ
Since ˆ‹Õ = ˆµ , then, in (2.13):
ˆxÕ‹
ˆxÕ‹ µ
v Õ‹ = v , (2.14)
ˆxµ
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 24

and the components of the vector transforms with the inverse of the Jacobian matrix of the
change of coordinates. Objects that behave this way live in the tangent bundle T M and are
called contravariant.
However, for a 1-form Ê, its components ʵ and Ê‹Õ in the two coordinates systems are
related via
Ê = ʵ dxµ = Ê‹Õ dxÕ‹ , (2.15)
ˆxÕ‹ µ
and since dxÕ‹ = dx we can see that the components of Ê are related by
ˆxµ
ˆxµ
Ê‹Õ = ʵ , (2.16)
ˆxÕ‹
which states that they transform with the Jacobian matrix of the change of coordinates. Objects
that behave this way lives in the cotangent bundle T ú M and are called covariant.

A little more on the exterior derivative

We defined df in such a way that when fed with a vector v it spits out the directional
derivative of v, if we are in Rn . But we also know that v = v µ ˆµ , so:

df (v) = v µ ˆµ f,
but df = fµ dxµ , then

df (v) = fµ dxµ (v ‹ ˆ‹ ) = v ‹ fµ ”‹µ = v ‹ f‹ ,

hence, comparing with the first one, we have fµ = ˆµ f and then

df = ˆµ f dxµ . (2.17)
Therefore, the exterior derivative of scalar function is just its gradient in Rn .

2.3.3 Wedge product and p-forms

In order to generalize the cross product in R3 , which is anticommutative, we define the
wedge product · of 1-forms Ê and µ as

Ê · µ = ≠µ · Ê. (2.18)

We can actually define the differential forms on M , denoted by (M ), to be the algebra

generated by 1 (M ) with the relations in equation (2.18).
The 0-forms, 1 (M ), are the functions, and we define the wedge product of a a function
with a differential form to be the ordinary product: f · Ê = f Ê.
The elements that are a linear combination of a product of p 1-forms are called p-forms, and
the space of all p-forms in M is p (M ). Of course, the space of all differential forms in M is
then the direct sum of the subspaces:
n
(M ) = p
(M ).
p

The 1-forms are given by ʵ dxµ , with the coefficients ʵ being functions.
2-forms look like
1
ʵ‹ dxµ · dx‹ ,
2
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 25

where the factor of 1/2 is inserted since dxµ · dx‹ = ≠dx‹ · dxµ . The (dxµ · dx‹ ) term is the
basis of 2-forms.
In general, a p-form looks like
1
ʵ‹...· dxµ · dx‹ · · · · · dx· ,
p!
where the product of p 1-forms dxµ · dx‹ · · · · · dx· is the basis of all p-forms.

2.3.4 The exterior derivative

We can then extend the definition of the exterior derivate d to generalize the gradient, the
divergence and the curl in any dimensions. The exterior derivative is defined to be the operator
d
d : p (M ) ‘æ p+1 (M ) (2.19)
satisfying
• d: 0
(M ) ‘æ 1
(M ) agrees with the previous definition

• d(Ê + µ) = dÊ + dµ and d(cÊ) = cdÊ for all Ê, µ œ (M ) and c œ R

• d(Ê · µ) = dÊ · µ + (≠1)p Ê · dµ for all Ê œ p

(M ) and µ œ (M )

• d(dÊ) = 0 for all Ê œ (M )

For instance, if we have a 1-form Ê, its exterior derivative is

dÊ = d(ʵ dxµ ) = dʵ · dxµ ≠ ʵ · d(dxµ ) = d(ʵ ) · dxµ ,

but df = ˆ‹ f dx‹ , so

dÊ = (ˆ‹ ʵ )dx‹ · dxµ , (2.20)

which is a 2-form.
The third property is the Leibniz rule graded, which is necessary since the product of
differential forms is anticommutative, and then, passing through p 1-forms we gain a sign of ≠1
at each step.
The last property can be demonstrated, as we now show. Recovering equation (2.20):

d(dÊ) = d((ˆ‹ ʵ )dx‹ · dxµ )

= (ˆ· ˆ‹ ʵ )dx· · dx‹ · dxµ
= 0, (2.21)

since ˆ· ˆ‹ is symmetric in [‹, · ] but dx· · dx‹ is antisymmetric in the same indices, which means
that d(dÊ) = ≠d(dÊ) and hence it vanishes.
The exterior derivative generalizes all vector derivatives in 3D. For instance, one can easily
show that
• Gradient: d : 0
(R3 ) ‘æ 1
(R3 )

• Curl: d : 1
(R3 ) ‘æ 2
(R3 )

• Divergence: d : 2
(R3 ) ‘æ 3
(R3 )
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 26

The identity d2 = 0 then contains the two identities of vector calculus

Ò ◊ (Òf ) = 0,

and
Ò · (Ò ◊ v) = 0,
and has profound consequences in physics.

2.3.5 The Hodge Star operator

In the particular case of R3 there is something missing to really conclude that, for example,
the exterior derivative reduces to the curl. In coordinates, take the two 1-forms Ê = Êx dx +
Êy dy + Êz dz and µ = µx dx + µy dy + µz dz and their wedge product:

Ê · µ = (Êx µy ≠ Êy µx )dx · dy + (Êy µz ≠ Êz µy )dy · dz + (Êz µx ≠ Êx µz )dz · dx. (2.22)

If we define a linear map ú to turn elements of 2

(M ) in elements of 1
(M ) such that

ú : dx · dy ‘æ dz

ú : dy · dz ‘æ dx
ú : dz · dx ‘æ dy
then we could really see that equation (2.22) would reduce to the curl, as expected. Note
that defining this operator — the star or Hodge operator — in that way is incorporating the
right-hand rule, since we could just as well have defined

ú : dy · dx ‘æ dz

ú : dz · dy ‘æ dx
ú : dx · dz ‘æ dy
which would imply in adopting a left-hand rule.
More generally, we define the Hodge star operator in a n-dimensional manifold M

ú: p
(M ) ‘æ n≠p
(M ), (2.23)

to be the unique linear map from p-forms to (n ≠ p)-forms such that, for all Ê, µ œ p
(M ),

Ê · úµ = ÈÊ, µÍvol, (2.24)

where ÈÊ, µÍ is the inner product of the forms, which is defined using the metric tensor as will
be discussed shortly, and vol is the volume form:
Ò
vol = det(gµ‹ )dx1 · dx2 · · · · dxn ,

where gµ‹ is the metric.

The definition in (2.24) simply implies a choice of orientation, since the existence of a volume
form states that the manifold is orientable, and the choice of orientation — right-handed or
left-handed — is what it is needed to make the map unique, as previously discussed in the 3D
case.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 27

2.4 Tensors
2.4.1 Definition
Having defined vectors—the geometrical objects living in T M whose basis in local coordinates
are ˆµ —we can define a new object constructed by composing those with the p-forms—the
objects in the dual space T ú M with the dual basis dxµ . Those objects are called tensors, and
we define the bundle of (r, s) tensors to be the tensor product of r copies of T M and s copies of
T úM :
ú
T
¸
M ¢ T M ˚˙¢ · · · ¢ T M˝ ¢ T
¸
M ¢ T ú M˚˙¢ · · · ¢ T ú M˝ .
r s

An object living in this space is an (r, s) tensor. The (0, 0) tensor are scalar fields. In local
coordinates, any (r, s) tensor is just a linear combination of
–
ˆµ ¢ ˆ‹ ¢ · · · ¢ ˆ‡ ¢ dx
¸
¢ dx— ˚˙
¢ · · · ¢ dx“˝ .
¸ ˚˙ ˝
r s

Therefore, a tensor T can be written in components in this basis as

T = T–—...“
µ‹...‡
ˆµ ¢ ˆ‹ ¢ · · · ¢ ˆ‡ ¢ dx
¸
–
¢ dx— ˚˙
¢ · · · ¢ dx“˝,
¸ ˚˙ ˝
r s

where T–—...“
µ‹...‡
are the components of the tensor in this basis, having r upper indices and s lower
indices. This object, when we change coordinates, will transform r times in a covariant way
and s times in a contravariant way. Hence, the components T̃ of the tensor T in a different
coordinate system will be related to the components in the first coordinate system by
µ‹...‡
T̃–—...“ = T◊„...Ê
· ›...” µ
·
‹
› ... ‡
” ( ) –(
≠1 ◊
)—
≠1 „
...( ) “,
≠1 Ê
¸ ˚˙ ˝¸ ˚˙ ˝
r s

where is the Jacobian matrix of the coordinate transformation and ≠1 its inverse, as expected
since vectors transform with and 1-forms with its inverse. One can do the same thing using
any basis eµ of vector fields and its dual basis eµ of 1-forms.
One way to think about the (r, s) tensor T is as a functional that accepts r 1-forms and s
vector fields as inputs and outputs a function on M in a manner that is C Π(M )-linear in each
input.

2.4.2 Metric tensor

A metric g is a (0, 2) tensor that is

• Symmetric: g(v, w) = g(w, v)

• Nondegenerate: if g(v, w) = 0 for all w then v = 0

The metric is the object that allows one to measure distances, angles and hence establishes
the dot product in the manifold. For instance, in Minkoski spacetime the dot product of vectors
v and w is
÷(v, w) = v · w = ≠v 0 w0 + v 1 w1 + v 2 w2 + v 3 w3 = ÷µ‹ v µ w‹ ,
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 28

where ÷µ‹ is the Minkowski metric given by

Q R
≠1 0 0 0
c 0 1 0 0d
÷µ‹ =c
c
d
d (2.25)
a 0 0 1 0b
0 0 0 1

and we will adopt this convetion where Minkowski spacetime has signature (3,1). The signature
(m, n) of a metric tensor is the number of positive and negative eigenvalues of the symmetric
tensor ÷µ‹ written in a basis where it is diagonal. Hence, if there are m positive eingenvalues
and n negative ones, one say that this metric has signature (m, n).
A metric g on a manifold M assigns to each point P œ M a metric gp on the tangent space
Tp M . This is the object used to take inner products of tangent vectors v and w at P

g(v, w) = gµ‹ v µ w‹

and of 1-forms Ê and µ in the dual space Tpú M :

ÈÊ, µÍ = g –— Ê– µ— .

In local coordinates the components gµ‹ of the metric g are given by

gµ‹ = g(ˆµ , ˆ‹ ), (2.26)

and one can use the metric to calculate infinitesimal distances

ds2 = gµ‹ dxµ dx‹ . (2.27)

2.4.3 Covariant derivative

In curved spaces, as we change to a point of coordinates xµ to a nearby point xµ + dxµ not
only the coordinates of a vector v change but also, in general, the basis vectors also change. So,
when one take a derivative of a vector v = v ‹ e‹ written in the basis eµ :

Òµ v = ˆµ (v ‹ e‹ )
= (ˆµ v ‹ )e‹ + v ‹ (ˆµ e‹ )
= (ˆµ v ‹ )e‹ + v ‹ k
µ‹ ek

= (ˆµ v k + µ‹ v )ek ,
k ‹
(2.28)

where we defined ˆµ e‹ := kµ‹ ek . The symbol kµ‹ tracks how the basis vectors eµ changes from
point to point and it is called the connection, since it allow one to connect a vector in one point
to another.
There are a lot of ways to make this connection. There is, however, a unique connection
that satisfies

• Metric compatibility: Òg = 0.

• Torsion free: for any vector fields v and w we have the Lie bracket [v, w] = Òv w ≠ Òw v =
Lv w vanishing.
CHAPTER 2. MANIFOLDS, TOPOLOGY AND DIFFERENTIAL GEOMETRY 29

This connection is called the Levi-Civita connection, and will allow us to take derivatives of
any geometrical object in arbitrary spaces. For instance, for a 1-form Ê = ʵ dxµ :

Òfl Ê = ˆfl ʵ ≠ k
flµ Êk , (2.29)

and, for a rank (r, s) tensor T we have

Òfl T = ˆfl T–—...“
µ‹...‡
+ µ k‹...‡
flk T–—...“ + ‹ µk...‡
flk T–—...“ + ··· + ‡ µ‹...k
flk T–—...“ ≠ k µ‹...‡
fl– Tk—...“ ≠ k µ‹...‡
fl— T–k...“ ≠ k µ‹...‡
fl“ T–—...k .
¸ ˚˙ ˝ ¸ ˚˙ ˝
r s
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 31

masses. Newton’s law of gravitation states that the gravitational force between bodies of mass
m and M is given by
mG M
F =G 2 .
r
On the other hand, Newton’s second law states that the dynamics of m is governed by the
equation of motion
F = mI ẍ.
If the inertial mass mI is equivalent to the gravitational mass mG , then the dynamics of bodies
due to a gravitational field will be independent of the body itself:
mG M
F = mI ẍ = G ,
r2
M
=∆ ẍ = G 2 .
r
Hence, the acceleration of bodies due to the effect of gravity will be the same for all bodies
and there are trajectories in spacetime that dictate how bodies will move if they are under
the effect of gravity. These trajectories are a property of that region of spacetime and do not
depend on the free falling body.

Figure 3.1: Observer A sees the apple free falling, but observer B, who is also in a free fall, does
not feel the effect of gravity. For him, the apple is fluctuating over his hand.

This idea has huge consequences, such as the possibility of changing coordinates to cancel the
effect of gravity, as we will briefly show. Consider observer A, which is in a uniform gravitational
field g, studying the movement of particle C, of mass m. He then writes the equation of motion
for that particle
mẍA = mg = FA , (3.1)
where FA stands for the net force acting on particle C in the frame of reference A.
Now consider the coordinate transformation
1
xB = xA ≠ gt2
2
tB = tA = t
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 32

which, when plugged in equation (3.1), leads us to

FA = mg = mẍA
3 4
d 1
= m 2 xB + gt2
dt 2
= mẍB + mg,
and, finally, the dynamics of particle C in the non inertial reference frame B is given by the
equation of motion
mẍB = FA ≠ mg = 0 = FB . (3.2)
Hence, the two observers write the same physical law, i.e. F = mẍ, the only difference is that
A feels a uniform gravitational field and B does not. The observers do not agree on the forces
acting on the body, but they agree on the physical law which describes its dynamics. We have
the gravitational force being canceled by inertial forces. The B frame of reference represents
a free falling observer: he does not feel the effect of gravity, although, using the equivalence
principle, he can write the same physical law to describe the dynamics in his point of view.
We have seen that the equivalence of the inertial and gravitational masses leads us to the
equivalence between gravity and acceleration: one can annihilate the effect of gravity using
acceleration, or also create the effect of gravity by accelerating.
This will not be true for Earth’s gravitational field, for instance, since it is not a uniform
gravitational field. However, in a small enough region of space and for very small intervals
of time, one can approximate the field of the Earth by a uniform gravitational field. Hence,
the equivalence principle states that in a small enough region of spacetime no experiment
can tell us whether we are in a gravitational field or in an accelerated frame of reference.
Therefore, it is always possible to build a local inertial frame or reference, satisfying the laws
of special relativity. We have then established the relation between metric and gravity: the
absence of gravity corresponds to the flat spacetime metric, the Minkowski metric ÷µ‹ such
that ds2 = ÷µ‹ dxµ dx‹ = ≠dt2 + dx2 + dy 2 + dz 2 . However, in the presence of a non uniform
gravitational field we need the metric gµ‹ since here it is not possible to find coordinates such
that the metric tensor reduces to the Minkowski metric, except in a infinitesimal neighborhood
of a certain point, where ds2 = gµ‹ dxµ dx‹ . Quoting Einstein [7]:
For infinitely small four-dimensional regions the theory of relativity in the restricted
sense is appropriate, if the coordinates are suitably chosen.
This connection between metric and gravity will lead us to Einstein’s field equation very
shortly.

3.3 The classical formulation of GR in four steps

3.3.1 Equation of motion
According to the ideas previously developed we can always find a local coordinate system › –
such that the equation of motion of a particle free falling reduces to
d2 › –
= 0, (3.3)
d· 2
i.e. the effect of gravity is locally canceled via this coordinate transformation. Here, · stands
for proper time, which is the time elapsed in a reference frame where the space interval between
the two events is zero, i.e., the two events have the same spatial coordinates.
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 33

We are trying to relate the equation of motion in the local inertial coordinates › – to the
reference frame in coordinates xµ who is feeling the effects of gravity. Hence, we can rewrite
equation (3.3) as
A B
d2 › – d ˆ› – ˆxµ
=
d· 2 d· ˆxµ ˆ·
ˆ› – d2 xµ dxµ ˆ 2 › –
= +
ˆxµ d· 2 d· ˆxµ ˆ·
– 2 µ
ˆ› d x dxµ dx‹ ˆ 2 › –
= +
ˆxµ d· 2 d· d· ˆxµ ˆx‹
– 2 µ
ˆ› d x ˆx fl
ˆxfl ˆ 2 › – dxµ dx‹
= +
ˆxµ d· 2 ˆ› – ˆ› – ˆxµ ˆx‹ d· d·
d2 xfl µ
fl dx dx
‹
= + , (3.4)
d· 2 µ‹
d· d·
where we have defined the Christoffel symbol as

ˆxfl ˆ 2 › –
fl
µ‹ := , (3.5)
ˆ› – ˆxµ ˆx‹
ˆxfl
and, in the fourth line we multiplied both sides of the equation by , which does not change
ˆ› –
– fl
ˆ› ˆx
the left-hand side since it is equal to zero. We have also used = ”µfl .
ˆxµ ˆ› –
Therefore, from (3.3) and (3.4) we get the geodesics equation

d2 xfl fl dx
µ
dx‹
+ = 0, (3.6)
d· 2 µ‹
d· d·
which gives the curves in spacetime xfl (· ) that describe the trajectories of bodies moving under
effect of gravity. These curves are called geodesics. They are a property of the geometry of
spacetime and do not depend of the particle in motion, as previously discussed.
Taking equation (3.5) as the definition of the Christoffel symbol — the connection — one
can show that its relation to the metric is given by
1
fl
= g fl‹ (ˆµ g‹⁄ + ˆ⁄ gµ‹ ≠ ˆ‹ gµ⁄ ) . (3.7)
µ⁄
2

3.3.2 Newtonian limit

Our new theory needs to be reduced into Newtonian theory in non relativistic limits. This
will set up some conditions that some components of the metric tensor must satisfy. The classical
limit will imply the following conditions:

• The particle will be moving in low speed comparing with the speed of light:
dx dt
π . (3.8)
d· d·

• The gravitational field will be stationary:

ˆ· gµ‹ = 0. (3.9)
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 34

• The gravitational field is weak. Hence, we can introduce the tensor hµ‹ which represents a
low deviation of the spacetime metric gµ‹ from the Minkowski metric ÷µ‹ :

gµ‹ ¥ ÷µ‹ + hµ‹ . (3.10)

The first condition allows us to reduce the equation of motion (3.6) by neglecting some of
its components:
A B2
d2 xfl dx0
=≠ fl
00 ,
d· 2 d·

and, from equation (3.7) we get

1
fl
00 = g fl‹ (ˆ0 g‹0 + ˆ0 g⁄0 ≠ ˆ‹ g00 )
2
1
= ≠ g fl‹ ˆ‹ g00 ,
2
where we have used the third condition to annihilate the time derivatives. Then, we get, for the
equation of motion: A B2
d2 xfl 1 fl‹ dx0
= g ˆ‹ g00 ,
d· 2 2 d·
however, using the weak gravitational field condition we are led to
A B2
d2 xfl 1 dx0
= (≠hfl‹
+ ÷ fl‹
)ˆ (h00 + ÷ 00 )
d· 2 2
‹
d·
A B2
1 dx0
¥ ≠ ÷ fl‹ ˆ‹ h00 ,
2 d·

and, hence, the equation of motion is

A B2
d2 xfl 1 fl‹ dt
≠ ÷ ˆ‹ h00 = 0.
d· 2 2 d·

For fl = 0 the the second term vanishes since ÷ 0‹ ˆ‹ h00 = ≠”0‹ ˆ‹ h00 = ˆ0 h00 = 0, because of
d2 x0
the stationary gravitational field condition. Hence, the equation states that = 0. Therefore
d· 2
dt
= constant.
d·
Now, for fl = i = 1, 2, 3 we get
A B2
d2 xi 1 i‹ dt
≠ ÷ ˆ‹ h00 = 0,
d· 2 2 d·
A B2
dt
which, dividing by , which is just a constant, as we previously showed, leads us to
d·

d2 xi 1
= ˆi h00 .
dt 2 2
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 35

But, looking to Newton’s second law for a gravitational force written in terms of the
gravitational potential „ we have
d2 xi
= ≠Ò„,
dt2
which allow us to make the identification

h00 = ≠2„ + c,

where c is a real constant. But, since at infinity the metric must become the Minkowski metric,
we have, at infinity, g00 = ÷00 + h00 = ÷00 and, therefore, the constant c must be zero, since the
potential „ already vanishes at infinity.
Then we get an equation saying that the time-time component of the metric tensor must
satisfy:

g00 = ÷00 + h00 = ≠ (1 + 2„) . (3.11)

We are looking for an equation of motion that describes gravity, here represented by the
metric gµ‹ . In the classical approach, gravity, or the gravity potential „, is given by Poisson’s
equation
Ò2 „ = 4fiGfl, (3.12)
where fl is the mass density which is generating the gravitational field. In the language of
differential geometry and tensors, the presence of mass will be carried in the energy momentum
tensor Tµ‹ . This tensor will represent the flux of the momentum pµ through the surface where
x‹ is constant. Hence, T 00 will be the flux of energy — p0 — through time – x0 – which is
the energy density in the reference frame where the system is at rest. The T 0j element is the
density of momentum in j direction, and the T ij component will be the flux of the momentum
component in the i direction per unit of time (force) flowing through a surface oriented in the
direction of j, and so on.
Conservation laws can then be written as

ˆµ T µ‹ = 0,

which will be the conservation of energy for ‹ = 0 and the conservation of momentum in the i
direction for ‹ = i.
Therefore, we are looking here for an equation of the form

Ò2 g00 = ≠8fiGT00 .

This is actually a special case, written in a reference frame where the particles are at low
speed. We could write the equation in a more general way as

Gµ‹ = 8fiGTµ‹ , (3.13)

where the tensor Gµ‹ must have, at most, second order derivatives of the metric tensor, since
we need to recover Poissons’s equation (3.12) in the classical limit. But then: who is this tensor
Gµ‹ that we are looking for?
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 36

3.3.3 Riemann Curvature Tensor

One way to identify the presence of curvature in a certain surface is by the non-commutativity
of the covariant derivatives. Of course, in flat space this is zero, i.e. [ˆµ , ˆ‹ ] = 0.
However, in curved space [Òµ , Ò‹ ] is not necessarily zero, and the deviation of this relation
from zero will be due to the curvature of space — the intrinsic curvature of a surface is actually
defined as the failure of this equation to vanish. Let us then evaluate the expression [Òµ , Ò‹ ]
acting in a certain vector V :

[Òµ , Ò‹ ]Vfl = [ˆµ (Ò‹ Vfl ) ≠ µ‹ (Ò‡ Vfl )

‡
≠ µfl (Ò‹ Vk )],
k

where the antisymmetrizator in the right handside is remembering us to antisymmetrize the

expression in [µ, ‹] at the end.
The second term in the right handside will vanish since ‡·· is symmetric in its two lower
indices. Developing the equation a bit more will lead us to

[Òµ , Ò‹ ]Vfl = [ˆµ (ˆ‹ Vfl ≠ ‹fl V‡ ) ≠

‡
µfl (ˆ‹ Vk ≠ ‹k V‡ )]
k ‡

= [ˆµ ˆ‹ Vfl ≠ ˆµ ( ‡‹fl V‡ ) ≠ kµfl ˆ‹ Vk + kµfl ‡‹k V‡ ],

and here the first term in the last line vanishes since it is symmetric in [µ, ‹]. The second term
will dismember in ≠ˆµ ( ‡‹fl V‡ ) = ≠V‡ ˆµ ‡‹fl ≠ ‡‹fl ˆµ V‡ = ≠V‡ ˆµ ‡‹fl ≠ k‹fl ˆµ Vk , which, when
plugged back in the expression above will give

[Òµ , Ò‹ ]Vfl = [≠V‡ ˆµ ‡

‹fl ≠ k
‹fl ˆµ Vk ≠ k
µfl ˆ‹ Vk + µfl ‹k V‡ ]
k ‡
1 2
= [≠V‡ ˆµ ‡
‹fl ≠ k
fl‹ ˆµ Vk + k
µfl ˆ‹ Vk + µfl ‹k V‡ ].
k ‡

Now note that the term in parenthesis is symmetric in [µ, ‹], therefore it vanishes due the
antisymmetrization. Finally, the equation is reduced to

[Òµ , Ò‹ ]Vfl = [≠V‡ ˆµ ‡

‹fl
k
+
µfl ‹k V‡ ]
‡

= [≠ˆµ ‡‹fl + kµfl ‡‹k ]V‡

= [ˆµ ‡‹fl ≠ kµfl ‡‹k ]V‡
= (ˆµ ‡‹fl ≠ ˆ‹ ‡µfl + k‹fl ‡µk ≠ µfl ‹k )V‡ .
k ‡
(3.14)

The curvature is then given by

R‡flµ‹ := ˆµ ‡
‹fl ≠ ˆ‹ ‡
µfl + k ‡
‹fl µk ≠ k ‡
µfl ‹k . (3.15)

Equation (3.15) defines the Riemann tensor, a (1,3) tensor that carries the information about
the curvature of the space. One can raise or lower indices of this tensor as with any other using
the metric:
R· flµ‹ = g· ‡ R‡flµ‹ .
One can also define the Ricci tensor Rfl‹ by the contraction

Rfl‹ = g · µ R· flµ‹ = g · µ (g· ‡ R‡flµ‹ ) = ”‡µ R‡flµ‹ = Rµflµ‹ ,

and also the curvature scalar R by the total contraction of indices

R = g µ‹ Rµ‹ .
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 37

3.3.4 Bianchi’s identity

Back to the discussion that ended the section Newtonian limit, to write the equation for
gravity in a manifestly covariant way we were looking for a tensor Gµ‹ that had, at most, second
order time derivatives of the metric tensor. One could think of the tensor Gµ‹ as being the
curvature tensor, let us say the Ricci tensor Rµ‹ , for instance. The idea is that the energy
momentum tensor, on the right-hand side of (3.13), is the object that generates curvature, which
must appear on the left-hand side of the equation.
However, from the conservation of energy and momentum, we must have

Òµ T µ‹ = 0, (3.16)

where the conservation of energy is the equation for µ = 0 and the conservation of the 3-
momentum is satisfied for µ = 1, 2, 3.
But the derivative of the Ricci tensor — and also of the Riemann tensor — is not zero,
so Rµ‹ = 8fiGTµ‹ , although it has the element which generates curvature on one side and the
curvature itself on the other, can not be the equation we are looking for. (Historically speaking,
Einstein and Grossmann dismissed it because they were unable to recover Newtonian physics in
the weak field limit[12], as discussed in section 3.2.2).
On the other hand, one can contract Bianchi’s identity

Ò⁄ R–—µ‹ + Ò‹ R–—⁄µ + Òµ R–—‹⁄ = 0,

to get to 3 4
1
Òµ Rµ‹ ≠ Rg µ‹ = 0. (3.17)
2
Now, we have built a symmetric tensor
1
Gµ‹ = Rµ‹ ≠ Rgµ‹ , (3.18)
2
which is related to the curvature, has its covariant derivative vanishing and it is of second order,
since the curvature has at most second order derivatives of gµ‹ .

3.3.5 Einstein’s field equations

Equation (3.13) was built in a covariant way from the Newtonian limit as a restriction.
Our goal was to find the left-hand side that would make the physics hold. We could do some
attempts of finding some tensors Gµ‹ that satisfies certain properties, however, we have already
built the tensor Gµ‹ that we need.
In order to write an equation for gravity that will contain the equivalence principle (equation
(3.4)), reduce to Newtonian gravity in the classical limit (equation (3.12), will be manifestly
covariant and will, beyond satisfying the conservation laws (equation (3.17)), also contain the
physical idea that the presence of mass — Tµ‹ — is responsible for the curvature — Rµ‹ — of
spacetime, then our tensor Gµ‹ that does that is the Einstein tensor, given by (3.18). He have
then built the Einstein field equation
1
Rµ‹ ≠ Rgµ‹ = 8fiGTµ‹ . (3.19)
2
Since the metric is compatible with the covariant derivative Òµ , a more general equation for
gravity could be built by adding any term in the left-handside proportional do the metric, since
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 38

this would not affect the conservation laws. So, a more general equation would then be
1
Rµ‹ ≠ Rgµ‹ + gµ‹ = 8fiGTµ‹ . (3.20)
2
The parameter is called the cosmological constant, and it was added by Einstein to
give solutions for a static cosmological model. This was done before Hubble’s work about the
expansion of universe.
One can note that the addition of this new term will make appear, in the classical limit,
beyond the gravitational Newtonian force, a repulsive force proportional to and to the distance.
So, for small , this repulsive term would be relevant only for great distances.
This term, although denied by Einstein and considered by him as an error, proved to be
very important later on. It is responsible for the dark energy and it is the term that explained
the accelerated expansion of the universe later detected in 1990.
However, since our work will not enter deep into cosmology, equation (3.19) will be our main
subject.

3.4 The Lagrangian Formulation of GR

It is always possible to obtain the same evolution equation via the Lagrangian formalism,
using variational calculus. Depending on what the dynamical variables in play are, one can
write different Lagrangians for gravity. The idea of trying a variational approach came from
Paul Bernays, a student of David Hilbert [12].

3.4.1 The Einstein-Hilbert action

Here we consider the metric gµ‹ as our only dynamical variable. The field equations for
gravity are extracted from the Einstein-Hilbert action:
1 ⁄ Ô
S[g] = R ≠gd4 x
16fiG ⁄
1 Ô
= g µ‹ Rµ‹ ≠gd4 x. (3.21)
16fiG
”S
Since gµ‹ is our only variable, the dynamics of gravity comes from setting = 0. From
”gµ‹
(3.21):
1 ⁄ 4 Ë Ô Ô Ô È
”S = d x R(” ≠g) + ≠gg µ‹ (”Rµ‹ ) + ≠gRµ‹ (”g µ‹ ) .
16fiG
We can break this integral in three terms (here we omited some constants)
s Ô
• ”S1 = d4 xR(” ≠g).
s Ô
• ”S2 = d4 x ≠gg µ‹ (”Rµ‹ ).
s Ô
• ”S3 = d4 x ≠gRµ‹ (”g µ‹ ).
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 39

For the first one we use ”(det M ) = det(M )M≠1

ij ”Mji , which leads us to

Ô 1
”( ≠g) = Ô gg µ‹ ”g‹µ
2 ≠g
1Ô
= ≠gg µ‹ ”gµ‹
2
1Ô
=≠ ≠ggµ‹ ”g µ‹ , (3.22)
2
where, in the last step, we used ”(gµ‹ g µ‹ ) = g µ‹ ”gµ‹ + gµ‹ ”g µ‹ = 0. Hence, we are left with
⁄ ; <
1Ô
”S1 = d4 x ≠R ≠ggµ‹ ”g µ‹ . (3.23)
2
For the second term we will use the Palatini identity, which states that

”Rflµ‡‹ = Ò‡ ” fl
µ‹ ≠ Ò‹ ” fl
µ‡ .

So, for the Ricci tensor we get

”Rµ‹ = ”Rflµfl‹ = Òfl ” fl
µ‹ ≠ Ò‹ ” fl
µfl . (3.24)

Therefore, the corresponding contribution to the variation of the action is written as

⁄
Ô
”S2 = d4 x ≠gg µ‹ (Òfl ” fl
µ‹ ≠ Ò‹ ” fl
µfl )
⁄
Ô
= d4 x ≠g(Òfl g µ‹ ” fl
µ‹ ≠ Ò‹ g µ‹ ” fl
µfl )
⁄
Ô
= d4 x ≠gÒ‡ (g µ‹ ” ‡
µ‹ ≠ g µ‡ ” fl
µfl )
⁄
Ô
= d4 x ≠gÒ‡ Ê ‡ , (3.25)

which is just a boundary term, that vanishes if ” flµ‹ vanishes at infinity.

The third term is already written in terms of the variation of the metric ”g µ‹ . Hence, taking
back the constants, we are left with
1
”S = (”S1 + ”S2 + ”S3 )
16fiG ⁄ ; <
1 Ô 4 1
= ≠gd x ≠R gµ‹ + Rµ‹ ”g µ‹ .
16fiG 2
Then, setting ”S/”g µ‹ = 0 gives us Einstein equation in vacuum
1
Rµ‹ ≠ Rgµ‹ = 0.
2
If we consider matter, then the action would be, except for some constants, S = SE.H. + SM ,
where SE.H. stands for the Einstein-Hilbert action previously developed, and SM for the action
related to matter. Then, setting
”S
Ô = 0,
≠g”g µ‹
we are led to
3 4
1 1 1 ”SM
Rµ‹ ≠ Rgµ‹ + Ô = 0, (3.26)
16fiG 2 ≠g ”g µ‹
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 40

which gives us the equation of motion

1
Rµ‹ ≠ Rgµ‹ = 8fiGTµ‹ ,
2
if one defines the energy momentum tensor as
1 ”SM
Tµ‹ := ≠ Ô . (3.27)
2 ≠g ”g µ‹

3.4.2 The Palatini action

In the Palatini approach we consider that the connection can also play a dynamical role.
So we write the Palatini action
⁄
Ô
S[g, ] = R ≠gd4 x
⁄
Ô
= g µ‹ Rµ‹ ( ) ≠gd4 x. (3.28)

The curvature is completely determined by the connection, therefore it is not affected by

variations of the metric. So, as we previously did, varying this action with respect to the metric
and setting ”S/”g µ‹ = 0 will lead us to
1
Rµ‹ ( ) ≠ R( )gµ‹ = 0,
2
which is just the Einstein field equation.
However, for this method to be equivalent to the Einstein-Hilbert one we need the connection
to be compatible with the metric.
So, for the second equation of motion, we vary the action with respect to the connection
and set ”S/” flµ‹ = 0. We will then get
⁄
Ô
”S = d4 x ≠gg µ‹ (”Rµ‹ )
⁄
Ô
= d4 x ≠gg µ‹ (Òfl ” flµ‹ ≠ Ò‹ ” flµfl )
⁄
Ô
= d4 x ≠g(g µ‹ Òfl ” flµ‹ ≠ g µ‡ Ò‡ ” flµ‹ ”fl‹ )
⁄
Ô
= ≠ d4 x ≠g(Òfl g µ‹ ≠ ”fl‹ Ò‡ g µ‡ )” flµ‹ , (3.29)

where we used the Palatini identity (3.24) in the second line and, in the last one, we did an
integration by parts and neglected the boundary term.
Assuming that the connection is symmetric in [µ, ‹], the variation will vanish if the sym-
metrization of the integrand vanishes:

Òfl g µ‹ + Òfl g ‹µ ≠ ”fl‹ Ò‡ g µ‡ ≠ ”flµ Ò‡ g ‹‡ = 0

2Òfl g µ‹ ≠ ”fl‹ Ò‡ g µ‡ ≠ ”flµ Ò‡ g ‹‡ = 0, (3.30)

and, contracting with ”µfl we get

Òµ g µ‹ ≠ Ò‡ g ‹‡ ≠ Ò‡ g ‹‡ = 0,
CHAPTER 3. THE TRADITIONAL FORMULATION OF GR 41

which, renaming dummy indices, leads us to Ò‡ g ‹‡ = 0. When plugged in (3.30) we get

Òfl g µ‹ = 0, (3.31)

which states that the covariant derivative Ò with respect to the connection gives a null
derivative of the spacetime metric, i.e., the metric is compatible with the connection.
So, the first equation of motion gives us Einstein field equation and the second equation of
motion states that our connection, previously placed as a dynamical variable, is fixed to be the
Levi Civita connection.
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 43

The reduced Bianchi identity states that

Òa Gab = 0 , (4.2)

and, developing (4.2) we have

Òµ Gµ‹ = ˆµ Gµ‹ ≠ ‡ µ
µ‹ G‡ + µ k
µk G‹ = 0.

Opening ˆµ Gµ‹ as ˆ0 G0‹ + ˆi Gi‹ for i spatial we get

ˆ0 G0‹ = ≠ˆi Gi‹ + ‡ µ

µ‹ G‡ ≠ µ k
µk G‹ , (4.3)

and, since
1
‡
= g ‡‹ (ˆ⁄ gµ‹ + ˆµ g⁄‹ ≠ ˆ‹ gµ⁄ ) ,
µ⁄
2
the right-hand side of (4.3) has, at least, second order time derivatives, given that the Christoffel
symbols have at most first order time derivatives and the Einstein’s tensor contains first
derivatives of those symbols. Hence, from the left-hand side of the equation we can infer that
G0‹ has at most first order time derivatives. Therefore, since we have other equations with
second order time derivatives, those four equations for G0‹ are not evolution equations: they are
constraints that the initial data must satisfy. From symmetry, the Einstein’s field equations are
a set of ten partial differential equations, of which only six are time evolution equations. The
equations G0µ = 8fiGTµ0 relate initial values of fields instead of determining how fields evolve.
If we proceed with the computation we can see that only spatial components of the metric
gab appear with their second order time derivatives. The other components do not play the
same dynamical role as gab . The g00 and g0a equations will be the constraints — they will play
the role of the lapse function and the shift vector, as we will see later.

4.2.1 The Lagrangian formalism

For a system with n (finite) degrees of freedom its action is
⁄
S[q i (t)] = L(q i , q̇ i )dt, (4.4)

for i = 1, 2, 3, ..., n. From the least action principle, one can get the Euler Lagrange equations
by setting ”S = 0: A B
d ˆL ˆL
i
≠ i = 0. (4.5)
dt ˆ q̇ ˆq
By the chain rule, one can expand the time derivative as

d ˆ dq i ˆ dq̇ i
= i + i
dt ˆq dt ˆ q̇ dt
and, plugging this in (4.5) one gets
A B A B
ˆ2L ˆ2L ˆL
j i
q̈ j
+ j i
q̇ j ≠ i = 0. (4.6)
ˆ q̇ ˆ q̇ ˆq ˆ q̇ ˆq

If we define the first term as

ˆ2L
Wij := , (4.7)
ˆ q̇ j ˆ q̇ i
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 44

then, equation (4.6) is written as

A B
ˆ2L ˆL
Wij q̈ +
j
j i
q̇ j ≠ i = 0. (4.8)
ˆq ˆ q̇ ˆq
If the matrix Wij is non-degenerate, then one can invert (4.8) to obtain an explicit equation
for q̈ j : A B
ˆ 2 L j ˆL
q̈ = Wij ≠ j i q̇ + i .
j ≠1
(4.9)
ˆq ˆ q̇ ˆq
However, if Wij is singular, then det(Wij ) = 0 and equation (4.8) cannot be inverted. In that
case, q̈ j can not be uniquely determined by positions and velocities, and the system is said to
be constrained, which we will detail better soon.

4.2.2 The Hamiltonian formulation

In this formulation constraints can arise in a similar way as happened in the Lagrangian
formulation.
The starting point is to define the canonical momenta as
ˆL
pi := . (4.10)
ˆ q̇ i
Equation (4.7) can then be rewritten as
ˆpi
Wij = . (4.11)
ˆ q̇ j
If W is nonsingular we can obtain in (4.11) the q̇ Õ s in terms of q Õ s and pÕ s, and then (4.10)
will indeed provide n independent variables — the pÕi s. However, if W is singular, there is no
unique solution of the momenta definition equation expressing the velocities in terms of the
canonical coordinates q i and conjugate momenta pj . In this case, there exists certain relations
Âs (q i , pj ) connecting the momentum variables:
Âs (q i , pj ) = 0. (4.12)
The q Õ s and pÕ s — the dynamical variables of the system — are connected by the primary
constraints, given by (4.12).

Figure 4.1: The constrained phase space

The map (q i , q̇ i ) ‘æ (q i , pj ), when there are no constraints, is a one-to-one map. In the

presence of constraints, it maps the unrestricted space (q i , q̇ i ) to the surface of primary constraint
Âs (q i , pj ) = 0 on the phase space, as shown in figure 4.1. We will name this constrained surface
as C from now on.
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 45

Hamiltonian equations
Let us consider the usual Legendre transformation

H = q̇ i pi (q, q̇) ≠ L(q, q̇), (4.13)

on the unconstrained manifold (q i , q̇ i ).

If H is a Hamiltonian of the system we need to be able to express it in terms of q i and pj ,
and not only in terms of q i and q̇ i . However, in the constrained case, equation (4.10) cannot be
inverted, so, we cannot express all of the q̇ Õ s in terms of pÕ s, which may lead us to conclude that
it is not possible to write such a function as H(q, p(q, q̇)) in the phase space.
Still, the function H(q i , pj ) is well defined, as we can easily see. From equation (4.13):

”H = ” q̇ i pi + q̇ i ”pi ≠ ”L(q i , q̇ i )
ˆL ˆL
= ” q̇ i pi + q̇ i ”pi ≠ i ”q i ≠ i ” q̇ i
ˆq ˆ q̇
ˆL
= ” q̇ i pi + q̇ i ”pi ≠ i ”q i ≠ pi ” q̇ i
ˆq
ˆL
= ≠ i ”q i + q̇ i ”pi
ˆq
= ≠ṗi ”q i + q̇ i ”pi
ˆH ˆH
= i ”q i + ”pi .
ˆq ˆpi
In the fourth line one can easily see that the variation ”H depends only on the variations of the
momenta pi and the position q i , not on the velocities q̇ i .
Equating the last two lines we get
A B A B
ˆH ˆH
i
+ ṗi ”q i + ≠ q̇ i ”pi = 0. (4.14)
ˆq ˆpi

For any variation ti = (”q i , ”pi ) tangent to the primary constraint surface, the equation above
shows that the vector A B
ˆH ˆH
V := + ṗi , ≠ q̇ i
(4.15)
ˆq i ˆpi
is normal to the surface, since ti Vi = 0
A basis of normal vectors to C is
A B
ˆÂs ˆÂs
vs = grad(Âs ) = , . (4.16)
ˆq i ˆpi
Then, for some functions ⁄ on the surface of primary constraints, we have

V = ⁄s v s . (4.17)

Finally, with equations (4.15), (4.16) and (4.17) one can get the equations of motion:
ˆH s ˆÂs
ṗi = ≠ + ⁄ , (4.18)
ˆq i ˆq i
ˆH ˆÂs
q̇ i = ≠ ⁄s . (4.19)
ˆpi ˆpi
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 46

Comparing the last equations with the Hamilton’s equations of motion, those can be rewritten
as
ˆ(H ≠ ⁄s Âs ) ˆ⁄s
ṗi = ≠ ≠ Â s , (4.20)
ˆq i ˆq i
ˆ(H ≠ ⁄s Âs ) ˆ⁄s
q̇ i = + Âs , (4.21)
ˆpi ˆpi
where we can define the total Hamiltonian of the system as

Htotal = H ≠ ⁄s Âs . (4.22)

We can rewrite the Hamilton’s equation in terms of the total Hamiltonian:

ˆHtotal
ṗi ¥ ≠ , (4.23)
ˆq i
ˆHtotal
q̇ i ¥ . (4.24)
ˆpi
Here, we introduce the weak equality symbol ¥, denoting an equality valid only in the constrained
surface.
The value of the total Hamiltonian does not change on the surface of primary constraints by
adding primary constraints and is independent of the ⁄s . However, the evolution of the system
depends on derivatives of the Âs , which might not be zero, and then the evolution depends
on the ⁄s . To see the role of the ⁄s on the evolution the mathematical theory of constraints,
described in terms of the Poisson structure, is very useful.

4.2.3 Poisson Brackets

In canonical coordinates (q i , pj ) on the phase space, the Poisson bracket of the functions
f (q, p) and g(q, p) is given by
n
A B
ÿ ˆf ˆg ˆg ˆf
{f, g} := i
≠ i . (4.25)
i=1 ˆq ˆpi ˆq ˆpi

It satisfies the following properties:

1. It is antisymmetric:
{f, g} = ≠ {g, f } .

2. It is linear in both entries:

{f1 + f2 , g} = {f1 , g} + {f2 , g} ,

{g, f1 + f2 } = {g, f1 } + {g, f2 } .

3. It obeys the Leibniz law:

{f1 · f2 , g} = f1 {f2 , g} + f2 {f1 , g} .

4. It satisfies the Jacobi identity:

{f, {g, h}} + {g, {h, f }} + {h, {f, g}} = 0.

CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 47

Using the Poisson bracket we can rewrite the Hamilton’s equations of motion 4.24 as:
ṗi ¥ {pi , Htotal } (4.26)
Ó Ô
q̇ i ¥ q i , Htotal , (4.27)

which actually is valid for any function F (q, p) on the phase space, as is easily seen:
dF
= i q˙i +
ˆF ˆF
Ḟ = ṗi
dt ˆq ˆpi
A B A B
ˆF ˆHtotal ˆF ˆHtotal
¥ i + ≠
ˆq ˆpi ˆpi ˆq i
= {F, Htotal } . (4.28)
The total Hamiltonian then generates the dynamical flow of the variables of the phase space in
time.
Because the primary constraints Âs are originated directly from the definition of the canonical
momenta, they need to hold during all the evolution of the system. This means that the evolution
of the system must be contained in the surface of primary constraint Âs . These are called
consistency conditions, expressed by
Â̇s ¥ {Âs , Htotal } = 0. (4.29)
These conditions can add new constraints to the evolution of the system, known as secondary
constraints. Those constraints must also satisfy the consistency conditions, which can lead to a
new generation of constraints. This process goes on until no more constraints are generated.
Opening equation (4.29) we get:
Ó Ô
Â̇s ¥ {Âs , Htotal } = Âs , H ≠ ⁄k Âk
Ó Ô
= {Âs , H} ≠ Âs , ⁄k Âk
Ó Ô
= {Âs , H} ≠ ⁄k {Âs , Âk } ≠ Âk Âs , ⁄k
¥ {Âs , H} ≠ ⁄k {Âs , Âk }
= {Âs , H} ≠ ⁄k Csk = 0, (4.30)
where we have defined
Csk := {Âs , Âk } .
If Csk is non singular the structure of the constraint system is uniquely determined: one can
solve for the ⁄k via
⁄k = Csk≠1
{Âs , H} .
In this case, no further constraints arise and we can fulfill the consistency condition. However, if
the matrix Csk is singular, we cannot determine all the ⁄k . In that case, equation (4.30) implies
the secondary constraints aforementioned. Those follow from the equations of motion, not from
the definition of the momenta as the primary constraints.

4.2.4 Gauge Transformations

Since the Hamiltonian generates the evolution of the system, we can define, as stated in
(4.28), the Hamiltonian vector field Xf associated to any function f as
Xf = {·, f } .
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 48

We call a constraint Âk first class with respect to all constraints if its Hamiltonian vector
field is everywhere tangent to the constraint surface C. That is, for all constraints Âk on the
constraint surface C we must have
{Âs , Âk } = 0,
and we call it second class if that Poisson bracket is nonvanishing on the constraint surface.
First class constraints generate gauge transformations, as we now show.
Consider all constraints, and consider also an arbitrary dynamical variable F , then define
the transformation
F (q, p) ‘æ F (q, p) + {F, ‘Âk } , (4.31)
where ‘ is a control parameter arbitrarily small. Due to the consistency conditions, this
transformation does not affect the Hamiltonian
H(q, p) ‘æ H(q, p) + {H, ‘Âk } ¥ H.
That is, the transformation takes solutions of the equations of motion and constraints into new
solutions. This is a gauge transformation, and that is why constraints are generators of gauge
transformations. Solutions that are related by gauge transformations are then treated as the
same solution.
Any particular choice for the total Hamiltonian will result in equations of motion written
in a specific gauge. But since the theory is invariant under gauge transformations generated
by constraints, the choice of a total Hamiltonian does not matter, and all sets of equations of
motion obtained for different gauges are equivalent.

4.3 Spacetime 3+1 decomposition

4.3.1 Introduction
When we are interested in studying the evolution of the spacetime, something strange
immediately appears: it evolves with respect to what parameter?
GR treats space and time on the same footing, which is not what happens in Hamiltonian
formulations. Spacetime does not evolve in time, it just is. However, we can interpret the
spacetime as the evolution of the 3D space. For that, we will need to do a (3+1) decomposition,
choosing an arbitrary parameter as time t, and considering that spacetime is the evolution of
spatial slices fixed for each t with respect to this time parameter.
This will be necessary because when we write down the Hamiltonian formalism it gives us
the evolution of the system with respect to time, which is not absolute in GR. So, one needs
to choose an arbitrary function to play the role of time and do this decomposition in order to
write the Hamiltonian formalism for GR.
We assume the existence of a foliation of spacetime in terms of space-like 3 dimensional
surfaces S of the spacetime manifold M . Thus, we consider the Lorentzian manifold M to be
diffeomorphic to R ◊ S.
There are lots of ways to build a diffeomorphism
„ : M ‘æ R ◊ S,
which means that time is not absolute in GR. There are different ways of defining a coordinate
t on the manifold to play the role of time, which we will discuss later on. For now, assume that
t œ M is a slice of M for t = constant for some time coordinate t. This can always be done in
globally hyperbolic manifold [4].
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 49

4.3.2 Geometry of Hypersurfaces

Consider a spatial slice t0 in a foliated spacetime manifold R ◊ S. This can be considered
as a constraint surface such as Ât0 = t ≠ t0 = 0.
The spacetime is just the history of the space t with respect to t. In any instant t the
spacetime is described as the immersion of t in the manifold M , as shown in figure 4.2, where
the dashed line is the integral curve of the time vector field — which will be precisely defined
later — joining the same point in the surface along its evolution.

Figure 4.2: The 3+1 decomposition

The surface is called a slice. The foliation is such that

1. t1 fl t2 = ?, if t1 ”= t2 .
t
2. t t = M.
In this way, every point of spacetime belongs to a unique slice. Any embedding that satisfies
this relations is a valid foliation, which reminds us that the foliation is not unique [4].
We can assign to each point of a slice t a time-like vector orthogonal to the surface at
that point. That enables us to define, for a given foliation, a time-like normal vector field na ,
normalized such that
g(n, n) = na na = ≠1, (4.32)
and the negative sign shows that this vector is time-like, as we wanted.
The foliation allows us to decompose all vectors in components parallel and perpendicular
to the spatial slice t . This can be done via the projection operator [2]:
PÎ : T M ‘æ TÎ M (4.33)
x ‘æ x + g(n, x)n = x + nb x n ,
a a a a b a
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 50

Figure 4.3: The spatial slice

and the orthogonal operator

P‹ : T M ‘æ T‹ M (4.34)
xa ‘æ ≠g(n, x)na = ≠nb xb na ,

as shown in the Figure 4.4.

Figure 4.4: The projections on the slice

These projections allow us to break any geometrical object X (a vector or tensor) in its
tangential (œ TÎ M ) and perpendicular (œ T‹ M ) parts:

X = (PÎ X) + (P‹ X).

For the dual space T ú (M ) the action of those operators is similar. The action of the projection
operator on a 1-form Ê , for instance, is:

PÎ : T ú M ‘æ TÎú M (4.35)
Êa ‘æ Êa + g(n, Ê)na = Êa + n Êb na .
b
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 51

For a rank (r, s) tensor T , the projection operator acts as follows

PÎ (T )(v 1 , v 2 , ..., v r , Ê1 , Ê2 , ..., Ês ) := T (PÎ v 1 , PÎ v 2 , ..., PÎ v r , PÎ Ê1 , PÎ Ê2 , ..., PÎ Ês ), (4.36)
where v k œ T (M ) and Êk œ T ú M . Therefore, the projection operator acting on a tensor is the
same as the tensor acting on the projections of its entries. The same is true for the orthogonal
operator.

4.3.3 Metric decomposition

Since we can apply the projection operator to any geometrical object, let us do this for the
metric. The part of the metric that is tangential to the slice t is called the induced metric,
and we will denote it by h, so:
h = PÎ g, (4.37)
or, in components:
PÎ g(X, Y ) = g(PÎ X, PÎ Y )
= gab (X a + nd Xd na , Y b + nd Yd nb )
= gab X a Y b + gab X a nd Yd nb + gab Y b nd Xd na + gab nc Xc na nd Yd nb
= gab X a Y b + Xb nb nd Yd + Ya na nd Xd ≠ nc Xc nd Yd
= gab X a Y b + X b nb Y d nd + Y a na X d nd ≠ nc Xc nd Yd
= (gab + na nb )X a Y b
:= hab X a Y b . (4.38)
One can note that
1. The metric hab lives in t:

na hab = na (gab + na nb ) = na gab + na na nb = nb ≠ nb = 0. (4.39)

2. Let sa be a vector tangent to t, then

hab sa = (gab + na nb )sa = gab sa + sa na nb .
But sa na = 0 since they are orthogonal, hence:
hab sa = gab sa . (4.40)

So, when applied to vectors tangent to t, the induced metric hab gives the same geometry
as gab .
One can then use the induced metric hab to describe projections of any geometrical object.
In coordinates, for a rank (m, n) tensor, one gets:
(PÎ T )ab11...b
...am
n
= hac11 . . . hacmm hdb11 . . . hdbnn Tdc11...d
...cm
n
. (4.41)
To study the dynamics of the canonical formulation, we consider the induced 3-metric hab as
a time-dependent 3-dimensional tensor field evolving on a family of manifolds t . Then, the
time dependent field hab will be the configuration variables of canonical gravity.
However, in order to do this, we have to define a time evolution vector field ta that specifies
the directions of time derivatives, since one will need to take time derivatives of the induced
metric or any other vector fields.
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 52

4.3.4 Time derivatives

If the spacetime is the history of the evolution of the slices t, how can one say how a field
in t , let us say, hab , evolves?

Figure 4.5: The time vector field

If one has just two slices in the foliation, it is impossible to say how a field defined on them
changes, unless we can uniquely associate a point on one slice to a point on the other one. The
vector field that connects a point in one slice to its correspondent point in another one is the
time evolution vector field ta , whose integral curves are shown in the right part of figure (4.5).
To ensure that this vector field agrees with the concept of time it is required that
ta Òa t = 1, (4.42)
which states that the change of t in the direction of the time evolution vector field t is just the a

unity.
It is assumed that the spatial coordinates xb are held fixed:
ta Òa xb = 0, (4.43)
so that
ˆ
ta Òa := . (4.44)
ˆt
By introducing the shift vector N a
N a := PÎ ta = hab tb (4.45)
and the lapse function N , which is the amount of the vector field ta in the direction orthogonal
to t :
N na := ta ≠ hab tb , (4.46)
and by acting with na on both sides of equation (4.46) one gets
N na na = na ta ≠ na hab tb = na ta .
Since na na = ≠1, we get
N = ≠na ta . (4.47)
The time evolution vector field t can then be written in its normal and tangential parts
a

with respect to the surface t :

ta = N na + N a . (4.48)
And now, with the projection operators, the definition of a time derivative of any tensor
field is also possible:
1 2 1 2
Ṫba11...b
...am
n
:= PÎ Lt Tba11...b
...am
n
= hac11 . . . hacmm hdb11 . . . hdbnn Lt Tdc11...d
...cm
n
. (4.49)
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 53

Figure 4.6: The components of the time vector field

4.3.5 Metric decomposition

From (4.48), the normal vector field can be written as
1
na = (ta ≠ N a ) , (4.50)
N
which allows us to write the inverse spacetime metric as
1
g ab = hab ≠ na nb = hab ≠ 2 (ta ≠ N a )(tb ≠ N b ). (4.51)
N
We can then invert this matrix and get the line element
ds2 = gab dxa dxb = ≠N 2 dt2 + hab (dxa + N a dt)(dxb + N b dt). (4.52)
We have then decomposed the metric in ten independent terms: the lapse function N and
the three components of the shift vector N a ; and six independent terms hab . The idea is to
express any geometrical property in terms of these variables: N, N a and hab .
It will be also useful to express the determinant g of the metric gab in terms of the determinant
h of the induced metric hab , since it appears in the Einstein-Hilbert action. One can do this as
follows: from equation (4.51) we can see that g 00 = ≠1/N 2 . Then we can use the relation
Cij
(A≠1 )ij = ,
det A
where (A≠1 )ij is the element of the i-th row and j-th column of the inverse matrix of A, and
Cij is the correspondent cofactor matrix, i.e. the determinant of the minor matrix obtaining by
eliminating the i-th row and j-th column from the matrix A. Then we have
1
g 00 = ≠ 2
N
C00
=
det(gab )
det(hcd )
= ,
det(gab )
from which we conclude that
g = det(gab ) = ≠N 2 det(hcd ) = ≠N 2 h . (4.53)
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 54

4.3.6 Intrinsic and Extrinsic Geometry

The induced metric hab allows us to define a unique covariant derivative metric compatible
in t . If we represent it by Da , the metric compatibility requires — beyond the torsion free
condition — that
Da hbc = 0. (4.54)
One can show that this covariant derivative Da compatible with the induced metric hab is
just the parallel part of Òa , i.e.
Da := PÎ Òa . (4.55)
This is proven as follows:
Da hbc = PÎ [Òa hbc ] = PÎ [Òa (gbc + nb nc )]
= PÎ [Òa (nb nc )]
= PÎ [nc Òa nb + nb Òa nc )]
= PÎ (nc )PÎ (Òa nb ) + PÎ (nb )PÎ (Òa nc )
=0, (4.56)
where in the second line we used the compatibility of Òa with the metric gab and in the last line
the fact that PÎ (nb ) = 0.
This covariant derivative Da can be seen as the projection in t of the derivative Òa by the
induced metric hab :
Dc Tba11...b
...am
n
:= (hac11 . . . hacmm hdb11 . . . hdbnn )hfc Òf Tdc11...d
...cm
n
. (4.57)
Definition 4.3.1 (Intrinsic Curvature). Given the three dimensional covariant derivative Da ,
we can define the intrinsic-curvature tensor 3 Rabcd as for any other covariant derivative:
3
Rabcd Êd = Da Db Êc ≠ Db Da Êc (4.58)
for any spatial 1-form Êc , i.e., Êa na = 0.
With this definition, one can obtain the Ricci tensor 3 Rab and the Ricci scalar 3 R by the
usual contractions.
The intrinsic geometry refers only to ( , hab ). But because is spatial, we cannot talk about
the evolution of the system using only parameters intrinsic to the manifold.

A geometrical object — the extrinsic curvature Kab — will naturally arise when we try to
make the induced metric evolve:
Ln hab = nc Òc hab + hac Òb nc + hbc Òa nc
= nc Òc (gab + na nb ) + Òb na + Òa nb
= nc Òc (na nb ) + Òb na + Òa nb
= n c na Òc n b + nc n b Òc n a + Òb n a + Òa n b
= (gac + na nc )Òc nb + (gbc + nb nc )Òc na
= hca Òc nb + hcb Òc na
= Kab + Kba , (4.59)
where the object Kab appears in the context of the evolution of the induced metric hab . Also,
in the third line we developed hac Òb nc = (gac + na nc )Òb nc = Òb na + na nc Òb nc = Òb na , since
nc Òb nc = 0 as shown in equation (4.61).
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 55

Definition 4.3.2 (Extrinsic Curvature). Given any normal vector na to the surface , the
extrinsic-curvature tensor is a spatial tensor on defined by

Kab := Da nb = hca hdb Òc nd . (4.60)

We could also omit the first projector hdb on the definition because

Kab := Da nb = hca hdb Òc nd

= hca (gbd + nd nb )Òc nd
= hca gbd Òc nd + hca nb nd Òc nd
= hca Òc nb ,

since, in the third line, nd Òc nd = 0. This is easy to see, since

1 1
nd Òc nd = (nd Òc nd + nd Òc nd ) = Òc (nd nd ) = 0. (4.61)
2 2
Another way of thinking about the extrinsic curvature tensor is as the normal component of
the derivative of v with respect to u, for u and v spatial:

K(u, v) = ≠g(Òu v, n). (4.62)

This notion is captured when one splits the derivative Òu v in its normal and tangential parts

Òu v = ≠g(Òu v, n)n + (Òu v + g(Òu v, n)n),

where the first term represents the normal part of it and the second one the tangential part. So,
when we parallel transport v, who lives in , in the direction of u, which also lives in , the
emergence of a normal component in this parallel transport measures exactly the curvature in
that region.
This way of thinking agrees with our previous definition of Kab , since, from the that definition,
we had

Kab ua v b = (Da nb )ua v b

= hca (Òc nb )ua v b
= (Òc nb )uc v b , (4.63)

and, from the notion now placed, we have

K(u, v) = ≠g(Òu v, n)
= ≠gab (Òu v a )nb
= ≠(uc Òc v a )na
= uc (Òc na )v a
= (Òc nb )uc v b , (4.64)

where in the third line we used the metric to lower the index of nb and in the fourth line we
used the fact that both v and u are spatial, then Òc (v a na ) = 0, then na Òc v a = ≠v a Òc na . In
the last line we only renamed a dummy index so it agrees with equation (4.63).
With this view, the tensor K measures how much the surface is curved in the way it
sits in M , because it says how much a vector tangent to will fail to be tangent if parallel
transported using the Levi-Civita connection Ò on M .
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 56

In components, we have
K(u, v) = Kij ui v j
in local coordinates, where
Kij = K(ˆi , ˆj ).
From this point of view it is easy to see that this tensor is symmetric, since
Kij ≠ Kji = K(ˆi , ˆj ) ≠ K(ˆj , ˆi )
= ≠g(Òi ˆj , n) + g(Òj ˆi , n)
= ≠g(Òi ˆj ≠ Òj ˆi , n)
= ≠g([ˆi , ˆj ], n)
= ≠g(0, n)
= 0. (4.65)
The extrinsic-curvature tensor has some important properties:

1. It is symmetric:
Kab = Kba , (4.66)
as shown right above.
2. As developed in equation (4.59) and using the property above, we get that the extrinsic
curvature tensor is half of the Lie derivative of the intrinsic metric along the unit normal:
1
Kab = Ln hab . (4.67)
2
3. The extrinsic curvature tensor can be related to the intrinsic curvature hab , the shift vector
N a and the lapse function N via
1 1 2
Kab = ḣab ≠ Da Nb ≠ Db Na , (4.68)
2N
which can be proven as follows:
1
Kab = Ln hab
2
1
= [nc Òc hab + hac Òb nc + hbc Òa nc ]
2
1
= [N nc Òc hab + hac Òb (N nc ) + hbc Òa (N nc )]
2N
1
= [(tc ≠ N c )Òc hab + hac Òb (tc ≠ N c ) + hbc Òa (tc ≠ N c )]
2N
1
= Lt≠N hab
2N
1 d c
= h h Lt≠N hcd
2N a b
1 d c
= h h [Lt hcd ≠ LN hcd ]
2N a b
1 1 d c 2
= ha hb Lt hcd ≠ hda hcb LN hcd
2N
1 1 ˙ b ≠ Db N a ,
2
= hab ≠ Da N
2N
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 57

where, from the third to the fourth line we used equation (4.48), and in the sixth line we
just smuggled in the induced metric to get the spatial part of the calculation, since Kab
is purely spatial. In the last line we just used the definition of the time derivative of a
tensor given by (4.49) and used the fact that the shift vector is spatial, then
LN hab = PÎ [N c Òc hab + hac Òb N c + hbc Òa N c ]
= PÎ [N c Òc hab ] + PÎ [hac Òb N c + hbc Òa N c ]
= N c PÎ [Òc hab ] + PÎ [(gac + na nc )Òb N c + (gbc + nb nc )Òa N c ]
= N c Dc hab + PÎ [Òb Na + Òa Nb ]
= 0 + Da Nb + Db Na
= Da Nb + Db Na .

4.3.7 Curvature relations

Using the definitions and properties previously mentioned we can prove the following relations
among the curvature tensors [4].

The Gauss equation

This relation comes from computing the Riemann curvature tensor Ref gh in terms of the
intrinsic curvature 3 Rabcd and the extrinsic curvature Kab :
hea hfb hgc Ref gh = 3 Rabcd + Kac Kbd ≠ Kbc Kad (4.69)

The Codazzi equation

This relation comes from computing the parallel part of the Riemann curvature tensor
contracted with the unitary normal vector PÎ (Rabcd nd ) which equals
PÎ (Rabcd nd ) = hae hbf hcg Rabcd nd = De Kf g ≠ Df Keg . (4.70)

The Ricci equation

This last equation comes from taking the lie derivative Ln along the unit normal na if the
extrinsic curvature Kab :
Rabcd nc nd = ≠Ln Kab ≠ Kac Kbc + D(a ab) + aa ab , (4.71)
where aa is the normal acceleration aa := nc Òc na (with aa na = 0).

We could also use the Ricci equation (4.71) with the relation Rab na nb = Racdd na nb to get
Rab na nb = (Kaa )2 ≠ Kab Kba + Òa v a , (4.72)
where the vector field v a is defined as
v a := ≠na Òc nc + nc Òc na .
Using the Gauss-Codazzi equations with the Ricci equations one can read the Ricci scalar R:
R = 3 R + Kab K ab ≠ (K aa )2 ≠ 2Òa v a . (4.73)
Hence, up to a divergence term, we can decompose the Ricci scalar into a potential term 3 R
and a kinetic term — quadratic in extrinsic curvature. Then, the extrinsic curvature, as shown
in equation (4.68), plays the role of a velocity of the spatial metric hab and is, thus, a candidate
for its momentum when we formulate the GR in terms of canonical variables, as we do next.
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 58

4.4 The ADM formalism

The action of general relativity in metric variables is given, as already presented, by the
Einstein-Hilbert action
1 ⁄ 4 Ò ⁄
SE.H. [g] = d x ≠ det gR := dtLgrav .
16fiG
Using equations (4.53) and (4.73) one can write the Lagragian for gravity as
1 ⁄ 3 Ô 1 2
Lgrav = d xN det h 3 R + Kab K ab ≠ (K aa )2 , (4.74)
16fiG
where the term proportional to Òa v a was left out once it is a boundary term which does not
affect the equations of motion.
From equation (4.68) we can see that the action depends on ḣab because of the Kab term,
but it is independent of time derivatives of the remaining space-time metric components, as
expected, and also of time derivatives of N and N a .
So we may already extract the primary constraints:
”Lgrav
pN (x) = = 0, (4.75)
” Ṅ (x)

and
”Lgrav
pa (x) = = 0. (4.76)
” Ṅ a (x)
The conjugate momenta of the induced metric hab is
”Lgrav
fi ab (x) =
” ḣab (x)
”Lgrav ”Kab
= (4.77)
”Kab ḣab (x)
”Lgrav 1
= , (4.78)
”Kab 2N
where the last line comes from equation (4.68).

So we get Ô
det h 1 ab 2
fi (x) =
ab
K ≠ K cc hab . (4.79)
16fiG
Contracting this relation with hab we get
16fiG ab
Ô fi hab = hab K ab ≠ K cc hab hab
det h
= K aa ≠ 3K cc
= ≠2K aa ,

and it follows that

8fiG a
K aa = ≠ Ô fi a, (4.80)
det h
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 59

which allows us to isolate K ab in equation 4.79:

8fiG
K ab = Ô (2fi ab ≠ fi cc hab ). (4.81)
det h

With this last relation we can express ḣab in (4.68) in terms of its conjugate momenta fi ab :

16fiGN
ḣab = Ô (2fi ab ≠ fi cc hab ) + 2D(a Nb) . (4.82)
det h
Then we can obtain the Hamiltonian through
⁄ 1 2
H(t) = d3 x [fi ab h˙ab ] + ⁄pN + µa pa ≠ L(t), (4.83)

where the ⁄ and µÕ s are the Lagrange multipliers of the constraints.

Using equation (4.82) to write ḣab in terms of its conjugate momenta fi ab we can write (4.83)
as
⁄ C 3 4 Ô D
16fiGN 1 N det h 3
H= d3 x Ô fiab fi ab ≠ (fi aa )2 + 2fi ab Da Nb ≠ R + ⁄pN + µa pa . (4.84)
det h 2 16fiG

Applying the consistency conditions (4.29) to the constraints we get secondary constraints:

0 = p˙N = {pN , Htotal } := ≠Cgrav (hab , fi ab ), (4.85)

0 = p˙a = {pa , Htotal } := ≠Cagrav (hab , fi ab ). (4.86)

And if we work out the Poisson’s brackets above we get [4]

3 4 Ô
16fiGN 1 N det h 3
Cgrav = Ô fiab fi ab ≠ (fi aa )2 ≠ R ¥ 0, (4.87)
det h 2 16fiG
which is called the Hamiltonian constraint.
Working out the second Poisson bracket [4] we get

Cagrav = ≠2Db fiab ¥ 0, (4.88)

which is called the diffeomorphism constraint.

We can now see that, putting these in (4.84), the lapse function N and the shift vector N a
play the role of Lagrange multipliers of the secondary constraints:
⁄
H= d3 x [N Cgrav + N a Cagrav + ⁄pN + µa pa ] + Hˆ , (4.89)

where the last term refers to the Hamiltonian of the boundary term.

We have finally built a Hamiltoninan representation of the dynamics of the spacetime

geometry. The canonical variables here are the induced metric hab and its conjugate momenta
fi ab . With this Hamiltonian it is now possible to study the spacetime dynamics in a canonical
way, using every tool of the Hamiltonian formalism.
CHAPTER 4. THE HAMILTONIAN FORMULATION OF GR 60

4.5 The equations of motion

Let us now obtain the evolutionary part of Einstein’s equations canonically.
The Hamilton equations give Ṅ (x) = ⁄(x) and Ṅ a (x) = µa (x), which means that these
functions can change arbitrally due to reparametrizations. We have also the equations

ḣab = {hab , Hgrav } ,

which gives us back equation (4.82). Finally, we have also the equation of motion
Ó Ô
fi̇ ab = fi ab , Hgrav ,

which, when developed, gives us

Ô 3 4 3 4
N det h 3 ab 1 3 ab 8fiGN ab cd 1 c 2
fi̇ = ≠
ab
R ≠ Rh +Ô h fi ficd ≠ (fi c ) +
16fiG 2 det h 2
3 4 Ô
32fiGN 1 det h 1 2
≠ Ô fi ac ficb ≠ fi ab fi cc + Da Db N ≠ hab Dc Dc N
det h A 2 16fiG
B
Ô ab
fi N c
+ det hDc Ô ≠ 2fi c(a Dc N b) . (4.90)
det h
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 62

Figure 5.2: Internal vs World indices.

5.2 Tetrad formalism

Since the tetrads represent an orthonormal basis, it is required that eI · eJ = ÷IJ , then, using
(5.1) we get

gµ‹ = g(ˆµ , ˆ‹ )
= ˆµ · ˆ‹
= (eIµ eI ) · (eJ‹ eJ )
= eIµ eJ‹ ÷IJ , (5.2)

and we can also rewrite this relation in index free notation as

g = eT ÷ e.

In this way, we can see the tetrad as a similarity transformation that diagonalizes the metric
gµ‹ and scales it to the unit. The ÷ matrix is the euclidean metric if we are talking about 3D
space (then the basis e should be called a triad) or the Minkowski metric if we are talking about
spacetime — where the name tetrad makes more sense.
Taking the determinant of this equation we get

g = ≠e2 , (5.3)

where g stands for the determinant of the spacetime metric gµ‹ and e for the determinant of the
matrix eIµ . The minus sign comes from the determinant of the Minkowski metric.
Hence, the tetrad represents the square root of the metric and has, therefore, all the
information about the geometry of the manifold. We can thus consider the tetrad as the
fundamental description and the metric as a derived concept.
The spacetime indices are contracted with the metric gµ‹ , as usual, and the internal indices
are contracted with the flat spacetime metric ÷IJ , which, consisting of 0s and ±1s is much easier
to deal with than gµ‹ — we will see that this is the whole point of the formalism.
Thinking of e(x) as a square matrix, we can define its inverse eµI such that
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 63

eµI eJµ = ”IJ . (5.4)

It can be sometimes confusing in the literature what does the term tetrads specifically refers
to. Here, when we say tetrad, one can think of the local orthonormal basis vectors eI , or its
dual — the 1-form eI = eIµ dxµ — or also the matrix eIµ containing the coefficients of the linear
transformation.

5.3 Connections via tetrads

If we have a spacetime vector field v µ and we take its derivative in a certain direction we get

(Òfl v)µ = ˆfl v µ + µ k
flk v , (5.5)

where µflk is the Levi-Civita connection.

In a similar way, when we compute the derivative of vectors in the internal space, we expect
something such as
(Da v)I = ˆa v I + ÊaI J v J , (5.6)
where the 1-form ÊaI J is the spin connection. This is of course valid for a vector. For a general
tensor we add a Ê factor for each of its indices, just as in the covariant derivative with the
Levi-Civita connection. So, for a rank (r, s) tensor we would have:

Da T µ1‹...µ r
1 ...‹s
= ˆa T µ1‹...µ r
1 ...‹s
+
+ Ê µ1ak T k...µ
‹1 ...‹s + ... + Ê ak T ‹1 ...‹s +
r µr µ1...k

≠ Ê k a‹1 T µk...‹
1 ...µr
s
+ ... ≠ Ê k a‹s T µ‹11...k
...µr
.

It can be easily seen that the spin connection is a 1-form since, on a curved manifold, when
we move from a point x to a nearby point x + dx it is expected that the local frame will rotate
(in Euclidean space) or Lorentz transform (in Minkowski flat spacetime), thus, an infinitesimal
translation has the effect of rotating the 1-form eI (x) infinitesimally. Hence, if we apply the
exterior derivative d to this 1-form we should get

deI = ≠Ê IJ eJ , (5.7)

for some antisymmetric ÊIJ , since the generators of rotations or Lorentz transformations are
antisymmetric. The minus sign is just a convention. Since eI is a 1-form, deI is a 2-form and

Ê IJ = Ê IµJ dxµ , (5.8)

is also a 1-form.
If we evaluate the covariant derivative of the Minkowski metric we get

Da ÷IJ = ˆa ÷IJ + ÊaKI ÷KJ + ÊaKJ ÷IK

= ÊaKI ÷KJ + ÊaKJ ÷IK
= Êa IJ + Êa JI , (5.9)

and, if the covariant derivative is required to be compatible with ÷ we get

Êa IJ = ≠Êa JI , (5.10)
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 64

which attest the antisymmetry of the spin connection in its internal indices, as said before. This
means that the coefficients of the connection take values in the Lie Algebra of the Lorentz group
of that signature, as developed in appendix A.
We can built the relation between the spin connection Ê on the internal space and the Levi
Civita connection on the manifold. Remember that we can always express a vector v at point
P by a linear combination of the internal basis vectors eI or by the spacetime basis vectors on
the tangent space Tp M , the ˆµÕ s: v = v I eI = v µ ˆµ .
Also, the connection Ê on the internal space induces a connection on the tangent space Tp M
for a given tetrad e. The covariant derivative D̃ on the internal space is defined via

Òv = e≠1 [D̃e(v)]. (5.11)

Developing these two derivatives leads us to

Òfl v = Òfl (v I eI ) =Òfl (v µ ˆµ )
eI (ˆfl v I ) + v K ÊflJ K eJ = (ˆfl v µ + v k flk )ˆµ ,
µ

and, using ˆµ = eIµ eI on the right side and v K = v µ eK

µ we get

eI ˆfl (v µ eIµ ) + v µ eK
µ Êfl K eJ = (ˆfl v + v
J µ k
flk )eµ eI
µ I

eI v µ (ˆfl eIµ ) + eI eIµ (ˆfl v µ ) + eK

µ v Êfl K eJ =
µ J
(ˆfl v µ )eIµ eI + v k µflk eIµ eI
v µ (ˆfl eIµ ) + eK
µ v Êfl K =
µ I
v k µflk eIµ
ˆfl eIµ + eK
µ Êfl K =
I ‹ I
flk e‹ ,

from where we can express the spin connection in terms of the Levi Civita connection:
1 2
Ê‹I J = eIfl ˆ‹ eflJ + eµJ fl
‹µ , (5.12)

which will be useful later.

5.4 Curvature and Torsion via tetrads

If one defines the exterior covariant derivative as DÊ = d+Ê where d is the exterior derivative,
it is possible to extract the curvature in the Cartan formalism. This is how one can take covariant
derivatives of n-forms taking values in the internal space. We will use just D from now on to
denote the exterior covariant derivative with respect to the connection Ê.
Consider a 0-form „I , which has no spacetime index — only internal indices. Then [16]

DIJ „J = d„I + Ê IJ „J . (5.13)

If we calculate
DKI DIJ „J = d(d„K + Ê KJ „J ) + Ê KL (d„L + Ê LJ „J ), (5.14)
the curvature will immediately emerge. The first term gives

dd„I + (dÊ KJ )„J ≠ Ê KJ d„J ,

and the second term gives

Ê KL d„L + Ê KL Ê LJ „J .
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 65

Then, since d2 = 0, the sum gives, maintaining only the operators in the equation:

DKI DIJ = dÊ KJ + Ê KL Ê LJ := F K
J, (5.15)

where F is the curvature 2-form:

F IJ = Fµ‹
IJ
dxµ dx‹ , (5.16)
i.e., this is the curvature of the connection 1-form ʵIJ on the internal space. Since it is an
antisymmetric tensor, its components are easily extracted from (5.15) and are given by
IJ
Fµ‹ = ˆµ Ê‹IJ ≠ ˆ‹ ʵIJ + [ʵ , Ê‹ ]IJ . (5.17)

If we use equation (5.12) in (5.17) we can get the relation between the curvature 2-form F IJ
and the Riemann curvature tensor Rµ‹‡ fl
= ˆµ ‹‡ ≠ ˆ‹ µ‡ + flµ– –‹‡ ≠ fl‹– –µ‡ , which is
fl
Rµ‹‡ = eflI eJ‡ Fµ‹I J . (5.18)

This relation shows that the Riemann curvature tensor Rµ‹‡fl

of the connection Ò is just the
spacetime image of the curvature Fµ‹ J of the spin connection Ê.
I

We can also convert all indices of Fµ‹J

I
to internal indices, which will be very useful to
calculate the Ricci tensor and Ricci scalar. Thus, let us introduce the object

FM NI J = Fµ‹J
I
eµM e‹N . (5.19)

We can get the internal Ricci tensor by the usual contraction

FIJ = FM IMJ , (5.20)

and the Ricci tensor can be built from this via

Rµ‹ = FIJ eIµ eJ‹ . (5.21)

We can also get the Ricci scalar in the usual way:

R = FIJ ÷ IJ . (5.22)

It is also easy to see that this scalar, in the internal structure, is the same as the Ricci scalar on
the manifold:

R = FIJ ÷ IJ
= Rµ‹ eµI e‹J ÷ IJ
= Rµ‹ g µ‹
= R,

where, in the second line, we used equation (5.22) with inverses of eIµ applied.
We also may define the torsion in the local Minkowski space. First, the torsion T in the
tangent bundle is given by
T (v, u) = Òv u ≠ Òu v ≠ [v, u], (5.23)
which is, in coordinates:
fl
Tµ‹ = Òµ ˆ‹ ≠ Ò‹ ˆµ = fl
[µ‹] .
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 66

However, from equation (5.12) we can write this as

fl
Tµ‹ = eflI (ˆµ eI‹ ≠ ˆ‹ eIµ + Ê IµJ eJ‹ ≠ Ê I‹J eJµ )
= eflI (DeI )µ‹ ,
or, in spacetime index free notation
T I = e≠1 (DeI ),
which states that the torsion in the tangent bundle is the pullback of DeI by the inverse of the
tetrad. So, the torsion in the Minkowski bundle is just
T I = DeI = deI + Ê IJ · eJ . (5.24)

5.5 Cartan’s view of Riemannian geometry

Equations (5.7) and (5.15) are called first and second Cartan’s structural equations, re-
spectively. In Cartan’s formalism, Riemannian geometry can be summarized by these two
equations:
deI + ÊK
I
· eK = 0,
J = dÊ J + Ê L · Ê J .
FK (5.25)
K K L

The protocol goes as follows: given a metric, one chooses a basis of tetrads eI satisfying equation
(5.2). Then one can use the first of Cartan’s equations to figure out the spin connection Ê, and,
finally, with the second of Cartan’s equations, one has the curvature 2-form F . This is the
easiest way of computing the Riemann curvature tensor, which can be done with the relations
between F IJ and Rµ‹‡
fl
developed in the previous section.

5.5.1 The 2-sphere

It is instructive to develop an example to see how the formalism works, which we will do for
the 2-sphere. The same can be done for the Schwarzschild metric or any other [9].
For the 2-sphere, writing the line element in spherical coordinates, we have
ds2 = gµ‹ dxµ dx‹ = R2 d◊2 + R2 sen2 ◊ d„2 .
From gµ‹ = ÷IJ eIµ eJ‹ we can immediately choose, for the tetrads:

e1◊ = R
e2„ = R sen◊,
with all the other components vanishing. For the 1-forms, we have eI = eIµ dxµ , so we get

e1 = Rd◊,
e2 = R sen◊d„.
Now, from the first of Cartan’s structural equations we get
de1 + ÊJ1 · eJ = 0
Ê11 e1 + Ê21 e2 = 0
Ê11 Rd◊ = ≠Ê21 R sen◊d„, (5.26)
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 67

and, since Ê11 = 0, it follows that Ê21 is a 1-form proportional to d„.

Again, using the first of Cartan’s structural equations:

de2 + ÊJ2 · eJ = 0
d(R sen◊d„) + Ê12 e1 + Ê22 e2 = 0
R cos ◊(d◊ · d„) + Ê12 Rd◊ + Ê22 R sen◊d„ = 0,
(5.27)

and again, since Ê kk = 0 (it is anti symmetric) we get

Ê12 = cos ◊d„

Ê21 = ≠ cos ◊d„,

with all the other components vanishing.

Now, from the second Cartan’s structural equation:

F IJ = dÊ IJ + Ê IK · Ê KJ ,

we get directly the vanishing components R11 = R22 = 0 and the non-vanishing components:

F 12 = dÊ 12 + Ê 1K · Ê K2
= sen◊(d◊ · d„) + Ê 11 · Ê 12 + Ê 12 · Ê 22
= sen◊(d◊ · d„),

and, similarly, we get F 21 = ≠ sen◊(d◊ · d„).

With these components in hand we can use equation (5.16) to write

F IJ = Fµ‹
I µ ‹
J dx dx ,

so, from the non-vanishing components we get

F 12 = sen(◊)d◊ · d„ = Fµ‹
1 µ
2 dx dx
‹

1
= F◊„ 2 d◊ · d„,

from where we get

1
F◊„ 2 = sen◊,

and, in a similar way,

2
F◊„ 1 = ≠ sen◊.

Now, from (5.18), we can write the components of the Riemann curvature tensor,
fl
Rµ‹‡ = eflI eJ‡ Fµ‹I J
= efl1 e2‡ Fµ‹1 2 + efl2 e1‡ Fµ‹2 1 .

The only non-vanishing component, up to the symmetries of the Riemann tensor, is:
◊
R◊„„ = e◊1 e2„ F◊„1 2 + e◊2 e1„ F◊„2 1
= e◊1 e2„ F◊„1 2
= R≠1 R sen(◊)F◊„1 2
= sen2 ◊.
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 68

Also, from (5.19) and (5.20) we can write the component F11 of the Ricci tensor in internal
indices:

F11 = Fµ◊N1 eµN e◊1

= Fµ◊21 eµ2 R≠1
= F„◊21 e„2 R≠1
= sen(◊)(R sen◊)≠1 R≠1
= 1/R2 .

The F22 component can be extracted in the same way:

F22 = Fµ„N2 eµN e„2

= Fµ„12 eµ1 (R sen◊)≠1
= F◊„12 e◊1 (R sen◊)≠1
= ( sen◊)(R)≠1 (R sen◊)≠1
= 1/R2 ,

with the other components being F12 = F21 = 0.

We can also use (5.21) to write the Ricci tensor in spacetime:

R◊◊ = FIJ eI◊ eJ◊

= F11 e1◊ e1◊ + F22 e2◊ e2◊
= F11 (e1◊ )2
= R≠2 R2
= 1,

and, for the other non-vanishing component:

R„„ = FIJ eI„ eJ„

= F11 e1„ e1„ + F22 e2„ e2„
= F22 (e2„ )2
= R≠2 (R sen◊)2
= sen2 ◊.

For the Ricci scalar, using (5.22), we have, finally

2
R = FIJ ÷ IJ = F11 ÷ 11 + F22 ÷ 22 = F11 + F22 = .
R2

5.6 The Palatini action

The Palatini action for GR is just the Einstein-Hilbert action written as a function of the
frame field e and the connection Ê:
1 ⁄ 4 Ò
S[e, Ê] = d x ≠ det g R[Ê]. (5.28)
16fiG
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 69

From equation (5.3) we can write det g in terms of the tetrads. And, from equation (5.18)
we can get the Ricci tensor by the usual contraction:

Rµ‡ = Rµ‹‡
‹
= Fµ‹I J e‹I eJ‡ , (5.29)

hence, the curvature scalar can be written as

R = Rµ‡ g µ‡
= Fµ‹I J e‹I eJ‡ g µ‡
= Fµ‹I J e‹I eJ‡ ÷ M N eµM e‡N
= Fµ‹I J ÷ M J e‹I eµM
= Fµ‹IM e‹I eµM
= Fµ‹IJ e‹I eµJ , (5.30)

where, in the third line we used the contraction eJ‡ e‡N = ”N

J
and in the last line we just renamed
a dummy index. Now we have R[Ê]. So, the Palatini action is
1 ⁄ 4
S[e, Ê] = d x e eµI e‹J Fµ‹
IJ
, (5.31)
16fiG
Ô
where e = ≠ det g.
We now just apply the variational principle to the action (5.31) to get the equations of
motion. First, we vary the action with respect to the tetrad, i.e., we compute ”S assuming
”Ê = 0. This leads us to
1 ⁄ 4 Ë È
”S = d x e (”eµI )e‹J Fµ‹
IJ
+ e eµI (”e‹J )Fµ‹
IJ
+ (”e)eµI e‹J Fµ‹
IJ
16fiG⁄ 5 6
1 1
= d4 x e e‹J Fµ‹
IJ
(”eµI ) ≠ eK eµ ‹ IJ
e F (”e‡
) , (5.32)
8fiG 2 ‡ I J µ‹ K

where the last term is calculated via ”(det A) = det A Tr(A≠1

ji ”Aij ):

”e = e e‡K ”eK
‡ = ≠e e‡ ”eK ,
K ‡
(5.33)

where in the last equality we used ”(eK

‡ eK ) = 0.
‡

Hence, if we set ”S = 0 we get

1
KJ ‹
F‡‹ eJ ≠ eK eµ e‹ F IJ = 0.
2 ‡ I J µ‹
If we act with e· K = ÷KJ eJ· on both sides we get:
1
eJ e· K ≠ (eK
KJ ‹
F‡‹ ‡ e· ÷KJ )eI eJ Fµ‹ = 0,
J µ ‹ IJ
2
where, from (5.2), the term in parenthesis is just g· ‡ , the last part of the second term, from
equation (5.30), is just the Ricci scalar R, and the first term, from equation (5.18), is just the
Ricci tensor R· ‡ . We then derived Einstein’s field equations in vacuum
1
R· ‡ ≠ Rg· ‡ = 0.
2
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 70

We need also to vary (5.31) with respect to Ê, assuming ”e = 0:

1 ⁄ 4
”S = d x e eµI e‹J (”Fµ‹
IJ
), (5.34)
16fiG
from (5.17) we get
IJ
”Fµ‹ = ˆµ (”Ê‹IJ ) ≠ ˆ‹ (”ʵIJ ) + (”ʵK
I
)ʋKJ + ʵK
I
(”Ê‹KJ ) ≠ (”Ê‹K
I
)ʵKJ ≠ (”ʵKJ )ÊK
I
1 2
= 2 ˆ[µ ”Ê‹]
IJ
+ ”Ê[µK
I KJ
Ê‹] + Ê[µ|K|
I KJ
”Ê‹]
1 2 1 2
[I
= 2 ˆ[µ ”Ê‹] + Ê[µ|K| = 2 ˆ[µ ”Ê‹] + 2Ê[µ|K| ”Ê‹]
IJ I KJ J IK IJ |K|J]
”Ê‹] ≠ Ê[µ|K| ”Ê‹] ,

1
where we used A[µ‹] = (Aµ‹ ≠ A‹µ ). Hence, in equation (5.34) we have
2!
1 ⁄ 4 [I
”S = d x e eµI e‹J (ˆ[µ ”Ê‹] + 2Ê[µ|K| ”Ê‹] ). (5.35)
IJ |K|J]
8fiG
The first term can be rewritten as

e eµI e‹J ˆ[µ ”Ê‹]

IJ
= ≠e eµ[I e‹J] ˆ‹ ”Ê µIJ ,

then, integrating by parts we have (neglecting boundary terms)

⁄ ⁄
d4 x e eµI e‹J ˆ[µ ”Ê‹]
IJ
= d4 x ˆ‹ (e eµ[I e‹J] )”Ê µIJ . (5.36)

The second term can be also rewritten as follow

[I
2 e eµI e‹J Ê[µ|K| ”Ê‹] = 2 e eµ[I e‹J] Ê[µK = 2 e eµ[I e‹J] ʵK (”Ê‹KJ ),
|K|J] I |K|J I
”Ê‹]

where, in the last step, we just left out the antisymmetrization [µ, ‹] since the expression is
already antisymmetric.
So, we have, for the second term of the action, renaming some dummy indices
⁄
≠ d4 x(2 e e‹[K eµJ] Ê‹I
K
)”ʵIJ , (5.37)

so, in (5.35) we have:

1 ⁄ 4 1 2
”S = d x ˆ‹ (e eµ[I e‹J] ) ≠ 2 e e‹[K eµJ] Ê‹I
K
”ʵIJ , (5.38)
16fiG
then, if we set ”S/”ʵIJ = 0 we get

ˆ‹ (e eµ[I e‹J] ) ≠ 2 e e‹[K eµJ] Ê‹I

K
= 0.

However, if we took the covariant derivative of the e eµ[I e‹J] term we would get
1 2 1 2
D‹ e eµ[I e‹J] = ˆ‹ e eµ[I e‹J] ≠ Ê‹I
K
e eµ[K e‹J] ≠ Ê‹J
K
e eµ[I e‹K] ,

which is exactly the expression above, whose value is zero:

1 2
D‹ e eµ[I e‹J] = 0. (5.39)
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 71

Since
1
e eµ[I e‹J] = ‘IJKL ‘µ‹–— eK L
– e— ,
4
we have, in (5.39):
1 2 1 Ë È
D‹ e eµ[I e‹J] = ‘IJKL ‘µ‹–— (D‹ eK )e L
+ (D e )e
L K
4 – — ‹ — –

1
= ‘IJKL ‘µ‹–— (D‹ eK
– )e— ,
L
2
where, in the first line the two terms in the parenthesis are equivalent since the expression is
antisymmetric in [K, L] and [–, —], contributing with two factors of minus one to the last term,
making it identical do the first one. We are left with

‘µ‹–— (D‹ eK
– )e— = 0,
L

where the symbol ‘IJKL was removed since it’s action on antisymmetric rank 2 tensors is
invertible. If we act with eflL on both sides we are left with

–] = 0.
D[‹ eK (5.40)

This implies that the torsion is zero, and, since we have a metric compatibility — equation
(5.10) — we know that we are talking about the Levi Civita connection.
This implies that the tetrad is constant with respect to the covariant derivative defined via
the connection Ê. So, just as it happened in the Palatini approach for the variables gµ‹ and ,
the variation of the action with respect to the connection told us that the metric was compatible
with the covariant derivative defined by that connection. Here, the tetrad, playing the role of
the metric, is compatible with the connection Ê.

5.6.1 The covariant notation

We can write the same formalism using the notation of forms, in a coordinate independent
way, which will be useful later.
We can write the Palatini action in this notation as
1 ⁄ 1
S[e, Ê] = ‘IJKL eI · eJ · F KL . (5.41)
16fiG 2
If we open the integrated term in coordinates we will have
3 4
1 1 1 KL –
‘IJKL eI · eJ · F KL = ‘IJKL (eIµ dxµ ) · (eJ‹ dx‹ ) · F–— dx · dx—
2 2 2
1 1 2
= ‘IJKL eIµ eJ‹ F–—KL dxµ · dx‹ · dx– · dx—
4
1
= ‘IJKL ‘µ‹–— eIµ eJ‹ F–—KL d4 x
4
[– —]
= e eK eL F–—KL d4 x
= eRd4 x,

where, in the fourth line we used one of the relations from appendix B, and in the last line we
just removed the antisymmetrization in [–, —] since the expression is already antisymmetric.
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 72

To get the equations of motion we first vary the action with respect to the connection Ê,
which gives us (removing the constants):
⁄
”S = ‘IJKL eI · eJ · (”F KL )
⁄
= ‘IJKL eI · eJ · D(”Ê KL )
⁄ 1 2 ⁄ 1 2
= I
D ‘IJKL e · e · ”Ê J KL
≠ D ‘IJKL eI · eJ · ”Ê KL
⁄ 1 2
= ≠2 ‘IJKL DeI · eJ · ”Ê KL ,

where the boundary term was neglected from the third to the fourth line and we used the
Palatini identity (see appendix C) in the second line.
So, if we set ”S/”Ê = 0 we will get DeI = 0, which states that the connection Ê is torsion
free, as we already knew.
If we now vary the action with respect to the tetrads we will have
⁄
”S = ‘IJKL (”eJ ) · eI · F KL ,

and, setting ”S/”eJ = 0 will lead us to

‘IJKL eI · F KL = 0.
and, when we open this equation in coordinates we get
‘IJKL (eI‡ F KL
µ‹ )dx · dx · dx = 0.
‡ µ ‹
(5.42)
One should note that the only free index in the above equation is L, which then gives us
four equations — one for each value of L = 0, 1, 2,
1 32 — stating that a certain 3-form vanishes.
And, since in a n-dimensional space a p-form has p independent components, our 3-forms will
n
1 2
have 43 = 4 independent components. Equation (5.42) is then grouping 16 different equations,
which may lead one to infer that this is probably Einstein’s field equations — which indeed is,
as we will now show.
Acting with dxfl in (5.42) gives us
‘IJKL (eI‡ F KL
µ‹ )dx · dx · dx · dx = 0.
‡ µ ‹ fl

Since ‘‡µ‹fl d4 x = dx‡ · dxµ · dx‹ · dxfl , and contracting ‘IJKL eI‡ = e e–J e—K e“L ‘‡–—“ , this leads us
to
e–J e—K e“L F KL
µ‹ ‘
‡µ‹fl
‘‡–—“ = 0.
Since J is a free index, one can act with eJ◊ to get e—K e“L F KL
µ‹ ‘
‡µ‹fl
‘‡–—“ = 0. However, e—K e“L F KL
µ‹ =
R µ‹ and
—“
1 2
‘‡µ‹fl ‘‡–—“ = ≠2 ”–[µ ”— ”“fl + ”–[fl ”— ”“‹ + ”–[‹ ”— ”“fl ,
‹] µ] µ]

which gives us 1 2
[µ [fl [‹
µ‹ ”– ”— ”“ + ”– ”— ”“ + ”– ”— ”“ = 0.
‹] µ] µ]
R—“ fl ‹ fl

Therefore,
1
R–fl ≠ R”–fl = 0,
2
fl :=
where R– Rµ– and R := Rµ . Acting with gflµ we then obtain Einstein’s field equations in
µfl µ

vacuum, as expected:
1
Rµ‹ ≠ Rgµ‹ = 0.
2
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 73

5.7 The Holst action

One could think about the dual of the terms in the Palatini action, such as the dual of the
curvature 2-form:
1 IJ
(ıF )IJ
ab = ‘ KL Fab .
KL
2
Varying the action with this term in the same way we did before leads us to the same compatibility
condition, i.e., we are still dealing with the connection that preserves the tetrads.
This allows us to generalize the Palatini action to
1 ⁄ 4
S[e, Ê] = d x e eµI e‹J P IJKL Fµ‹
KL
. (5.43)
16fiG
where the new term is defined as
[I J] 1 IJ
P IJKL := ”K ”L ≠ ‘ . (5.44)
2“ KL
Again, if we vary this action with respect to the connection, we are led to

‘abcd ‘IJKL P KLM N Da (eM

c ed ) = 0,
N

which still remains being the compatibility condition.

Varying the action with respect to the tetrads, as we did in (5.32), will lead to the Einstein’s
field equation with an extra term ‘Iabc RabIL , which vanishes by the symmetries of the Riemann
tensor.
Therefore, for any value of “, we get the same equations of motion from this new action.
The action in (5.43) is called the Holst action, and “ is the Barbero-Immirzi parameter.
This action is a step towards constructing the Ashtekar’s formulation of general relativity.

5.7.1 Forms notation

It is also possible to write this action in the notation of forms, which would lead us to
CA B D
1 ⁄ 1
S= ı+ eI · eJ · F IJ . (5.45)
32fiG “

Varying this action with respect to the connection gives us

CA B D
1 ⁄ 1
”S = ı+ eI · eJ · ”F IJ
32fiG “
CA B D
1 ⁄ 1
= ı+ eI · eJ · D(”Ê IJ )
32fiG “
CA B D
1 ⁄ 1
=≠ ı+ (DeI ) · eJ · ”Ê IJ ,
16fiG “

where we used ”F IJ = D(”Ê IJ ) in the second line and we integrated by parts and neglected the
boundary term in the next line.
From equation (5.24) it can be seen that forcing this variation to vanish leads us to the
torsion free condition:
T I = DeI = 0.
CHAPTER 5. TETRADS FORMALISM AND PALATINI ACTION 74

This can also be taken as the definition of Ê: our connection is the one that can be entirely
determined by the tetrads, given this condition. So, the only independent variable in our theory
is eI . Defining the connection in this way would obviously make it satisfy the equation of
motion — i.e. there is no variation with respect to Ê since it is not an independent variable.
This formulation is called first order, while the formulation where both e and Ê is independent
is called first order.
Finally, varying the action with respect to the tetrad gives us
CA B D
1 ⁄ 1
”S = (”eI ) · eJ · ı+ F IJ .
16fiG “

Hence, setting ”S/”e = 0 leads us to

A B
1
eJ · ı + F IJ = 0. (5.46)
“

The second term vanishes since, from the first Bianchi identity (see appendix C) we have
that eJ · F IJ = D2 eJ = 0, since the connection is torsion free. So, the “ term vanishes on-shell,
i.e., when the torsion is zero.
Hence, we are left only with the first term:
1 2
eJ · ıF IJ = 0, (5.47)

which is just Einstein’s field equations in forms notation, which can be shown in a similar way
as we did in the Palatini section, using coordinate notation.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 76

Figure 6.1: The triad

However, in the triad formalism, there is an additional condition to the usual spacetime
split done in the ADM formalism, which is the split in the internal directions of the tetrad in
Minkowski time and space components.
There are two ways of doing this gauge fixing. The first one is to require that ea0 = nI eaI = na
be the unit normal to the foliation, this is known as the time gauge. Here we assumed nI = ”0I
to be a timelike internal vector field. Internal Lorentz transformations that preserve nI are
reduced to spatial rotations around the fixed direction nI .
Another way of doing it, which will show to be more practical in the calculations we are
going to develop, is to open the tetrads in its spatial and time components and set the time
gauge directly from it.
First, let us consider the 1-forms eI . For the Minkowski time component:

e0 = e0µ dxµ = e00 dx0 + e0a dxa .

Here we set the spatial part to be zero:

e0a = 0, (6.3)

leaving just the time component for this tetrad:

e0 = e0µ dxµ
= e00 dx0 + e0a dxa
= e00 dx0
= N dt, (6.4)

where we defined the lapse function e00 := N . This fixes the Minkowski time in the internal
space: the direction orthogonal to the spatial slice.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 77

Now, for the spatial Minkowski components one can write

ei = eiµ dxµ
= ei0 dx0 + eia dxa
= N i dx0 + eia dxa , (6.5)

where we defined the shift vector N i := ei0 .

With these, the (3+1) split of the tetrad is done: one has the lapse function N , the shift
vector N i and the spatial triads eia :
A B
N Ni
eIµ = . (6.6)
0 eia
In the formalism that we are going to develop, the 3-metric of the ADM formalism is
substituted by the new dynamical variable, the triads, and those contain the same geometrical
information: hab = ”ij Áia Ájb . The ADM formalism variables will then be replaced by

(hab , fi ab ) ‘æ (Áia , Aib ),

where Aib is an SU (2) connection, the canonical conjugate variable to the triad, which will
appear shortly in our development.

6.3 The space-time split

6.3.1 The Holst Action
We will now expand the Holst action in equation (5.45) in order to do the 3+1 split in the
triads variable.
First, let us note that

eI · eJ · F KL = (eIµ dxµ ) · (eJ‹ dx‹ ) · (F KL

fl‡ dx dx )
fl ‡

= eIµ eJ‹ F KL
fl‡ (dx · dx · dx · dx )
µ ‹ fl ‡
1 2
0 1 2 3
= ≠˜‘µ‹fl‡ eIµ eJ‹ F KL
fl‡ dx · dx · dx · dx
3
= ≠˜‘µ‹fl‡ eIµ eJ‹ F KL
fl‡ dt · d x, (6.7)

where, in the third line we used the fact the the term dxµ · dx‹ · dxfl · dx‡ is completely
anti-symmetric in µ‹fl‡, then, it can be written in terms of the Levi Civita symbol. The minus
sign comes from the fact that we will use ‘˜µ‹fl‡ to represent the Levi-Civita symbol and ‘˜µ‹fl‡ is
defined as being the Levi-Civita symbol multiplied by sign(g) = ≠1. We can then plug this into
the action in (5.45):
A B
1⁄ 1
S= ı+ eI · eJ · F IJ
4 “
I J
1⁄ 1 1
= I J
‘IJKL e · e · F KL
+ ÷IK ÷JL e · e · F
I J KL
4 2 “
⁄ ⁄ I J
1 1 1
=≠ dt d3 x ‘˜µ‹fl‡ fl‡ + ÷IK ÷JL eµ e‹ F fl‡ .
‘IJKL eIµ eJ‹ F KL I J KL
(6.8)
4 2 “
CHAPTER 6. ASHTEKAR FORMULATION OF GR 78

6.3.2 The 3+1 split

Now, given the relations in (6.6), we can decompose the terms in the action (6.8) doing the
(3+1) split in the spacetime indices. First, note that
1 2
‘IJKL ‘˜µ‹fl‡ eIµ eJ‹ F KL ˜0abc eI0 eJa F KL
fl‡ = ‘IJKL ‘ bc + ‘
˜a0bc eIa eJ0 F KL
bc + ‘
˜ab0c eIa eJb F KL
0c + ‘
˜abc0 eIa eJb F KL
c0
1 2
= ‘IJKL ‘˜0abc eI0 eJa F KL
bc + ‘
˜a0bc eIa eJ0 F KL
bc + ‘
˜ab0c eIa eJb F KL
0c + (≠˜
‘ab0c )eIa eJb (≠F KL
0c )
1 2
= ‘IJKL 2˜‘0abc eI0 eJa F KL
bc + 2˜
‘ab0c eIa eJb F KL
0c
1 2
= ‘IJKL ‘˜abc 2eI0 eJa F KL
bc + 2ea eb F 0c ,
I J KL
(6.9)

where we defined in the last line ‘˜abc := ‘˜0abc , and the factor of 2 in the first term came from:

‘˜a0bc ‘IJKL eIa eJ0 F KL ‘0abc ‘IJKL eIa eJ0 F KL

bc = ≠˜ bc
= ‘˜0abc ‘JIKL eIa eJ0 F KL
bc
= ≠˜‘0abc ‘JIKL eJa eI0 F KL
bc
= ‘˜0abc ‘IJKL eJa eI0 F KL
bc ,

hence, the second term in the second line of (6.9) is equal to the first. This was for the first
term in (6.8), but the calculation is analogous for the second term with the ÷ parameters.
Therefore, plugging (6.9) in (6.8) we are led to
I J
1⁄ ⁄
1 1
S=≠ dt d3 x ‘˜µ‹fl‡ fl‡ + ÷IK ÷JL eµ e‹ F fl‡
‘IJKL eIµ eJ‹ F KL I J KL
4 2 “
I J
1⁄ ⁄
1 1 1 1
=≠ dt d3 x ‘˜abc 2 ‘IJKL eI0 eJa F KL bc + 2 ‘IJKL ea eb F 0c + 2 ÷IK ÷JL e0 ea F bc + 2 ÷IK ÷JL ea eb F 0c
I J KL I J KL I J KL
4 2 2 “ “
I J
1⁄ ⁄
3 abc 1 1 1 1
=≠ dt d x ‘˜ ‘IJKL e0 ea F bc + ‘IJKL ea eb F 0c + ÷IK ÷JL e0 ea F bc + ÷IK ÷JL ea eb F 0c .
I J KL I J KL I J KL I J KL
2 2 2 “ “
(6.10)

We now do the same (3+1) split in the internal indices. For the first term inside the brackets
in (6.10) we get
1 1 0 i jk 1 i 0 jk 1 1
bc = ‘0ijk e0 ea F bc + ‘i0jk e0 ea F bc +
‘IJKL eI0 eJa F KL ‘ij0k ei0 eja F 0kbc + ‘ijk0 ei0 eja F k0bc
2 2 2 2 2
1 1 1
= ‘0ijk e00 eia F jkbc + ‘0ijk ei0 eja F 0kbc ≠ ‘0ijk ei0 eja F k0bc
2 2 2
1 1 1
= ‘0ijk e00 eia F jkbc + ‘0ijk ei0 eja F 0kbc + ‘0ijk ei0 eja F 0kbc
2 2 2
1
= ‘0ijk e00 eia F jkbc + ‘0ijk ei0 eja F 0kbc
2
1
= ‘ijk N eia F jkbc + ‘ijk N d eid eja F 0kbc , (6.11)
2
where the second term in the first line vanishes because of the time gauge (e0a = 0) and in
the last term we recovered the definitions of the lapse function N = e00 and the shift vector
ei0 = N i = N d eid , and we also defined ‘0ijk := ‘ijk .
CHAPTER 6. ASHTEKAR FORMULATION OF GR 79

Now, for the second term in parenthesis in equation (6.10) we have

1 1 0 i jk 1 i 0 jk 1 i j 0k 1
0c = ‘0ijk ea eb F 0c + ‘i0jk ea eb F 0c + ‘ij0k ea eb F 0c + ‘ijk0 ea eb F 0c
i j k0
‘IJKL eIa eJb F KL
2 2 2 2 2
1 1
= ‘ij0k eia ejb F 0k0c + ‘ijk0 eia ejb F k00c
2 2
1 1 1 2
= ‘0ijk eia ejb F 0k0c ≠ ‘0ijk eia ejb ≠F 0k0c
2 2
i j 0k
= ‘ijk ea eb F 0c , (6.12)

where the first two terms in the first line vanishes due to the time gauge (e0a = 0) and in the
last term we also used ‘0ijk := ‘ijk .
For the last two terms in (6.10) we will open the indices in space and time using the metric
relations above: Y
_
_
_÷00 ÷00 = 1
_
_
]÷ ÷ = ≠”
00 ij
÷IK ÷JL = _ (6.13)
ij
_
_÷ ij ÷00 = ≠”ij
_
_
[
÷ik ÷jl = ”ik ”jl
Then, for the third term in (6.10) we get
1 1 Ë È
bc =
÷IK ÷JL eI0 eJa F KL ÷00 ÷00 e00 e0a F 00bc + ÷00 ÷jl e00 eja F 0lbc + ÷00 ÷ik ei0 e0a F k0bc + ÷ik ÷jl ei0 eja F klbc
“ “
1 Ë È
= e00 e0a F 00bc ≠ ”jl e00 eja F 0lbc ≠ ”ik ei0 e0a F k0bc + ”ik ”jl ei0 eja F klbc
“
1 Ë È
= ≠”jl e00 eja F 0lbc ≠ ”ik ei0 e0a F k0bc + ”ik ”jl ei0 eja F klbc
“
1 Ë È
= ”ik ”jl N d eid eja F klbc ≠ ”jl N eja F 0lbc , (6.14)
“
where the first term in the second line vanishes since the curvature 2-form is anti-symmetric
(F µµ = 0), in the third line the second term also vanishes due to the time gauge, and in the last
line we plugged in the definitions of N and N i = ei0 = N d eid .
Finally, for the fourth term in parenthesis in equation (6.10) we get
1 1Ë È
0c =
÷IK ÷JL eIa eJb F KL ÷00 ÷00 e0a e0b F 000c + ÷00 ÷jl e0a ejb F 0l0c + ÷ik ÷00 eia e0b F k00c + ÷ik ÷jl eia ejb F kl0c
“ “
1 Ë 0 0 00 È
= ea eb F 0c ≠ ”jl e0a ejb F 0l0c ≠ ”ik eia e0b F k00c + ”ik ”jl eia ejb F kl0c
“
1Ë È
= ≠”jl e0a ejb F 0l0c ≠ ”ik eia e0b F k00c + ”ik ”jl eia ejb F kl0c
“
1
= ”ik ”jl eia ejb F kl0c , (6.15)
“
where the first two terms in the third line vanishes due to the time gauge.
Now, plugging (6.11), (6.12), (6.14) and (6.15) in the action (6.10) leads us to
I J
1⁄ ⁄
1 1 1 1
S=≠ dt d3 x ‘˜abc bc + ‘IJKL ea eb F 0c + ÷IK ÷JL e0 ea F bc + ÷IK ÷JL ea eb F 0c
‘IJKL eI0 eJa F KL I J KL I J KL I J KL
2 2 2 “ “
1 ⁄ ⁄
=≠ dt d3 x ‘˜abc {} , (6.16)
2“
CHAPTER 6. ASHTEKAR FORMULATION OF GR 80

where the term in parenthesis is

3 4
1 1 2
{} = “‘ijk N eia F jkbc + “‘ijk N d eid eja F 0kbc + “‘ijk eia ejb F 0k0c
1
2 2 1 2
+ ”ik ”jl N d eid eja F klbc ≠ ”jl N eja F 0lbc + ”ik ”jl eia ejb F kl0c , (6.17)

which, by grouping terms proportional to N and N d , can be written as

1 2
{} = ”ik ”jl eia ejb F kl0c + “‘ijk eia ejb F 0k0c +
1 2
+ N d ”ik ”jl eid eja F klbc + “‘ijk eid eja F 0kbc +
3 4
j 0l 1
≠ N ”jl ea F bc ≠ “‘ijk ea F bc .i jk
(6.18)
2
Now, with (6.18) in (6.16) we can break the integral in three terms:
1⁄ ⁄
S= dt d3 x(L1 + L2 + L3 ), (6.19)
“
where we have
1 1 2
L1 = ≠ ‘˜abc eia ejb ”ik ”jl F kl0c + “‘ijk F 0k0c , (6.20)
2
1 1 2
L2 = ≠ N d ‘˜abc eid eja ”ik ”jl F klbc + “‘ijk F 0kbc , (6.21)
2
3 4
1 1
L3 = N eia ‘˜abc ”ik F 0kbc ≠ “‘ijk F jkbc . (6.22)
2 2
We will treat those terms separately since each of them will be responsible for a different
constraint in our Hamiltonian formalism, as we will be soon developing.

6.4 The Ashtekar-Barbero Variables

Having defined the triads let us now do some minor modifications in those to define the
Ashtekar-Barbero variables, which are the variables in terms of which we will write the action
and the Hamiltonian.

6.4.1 Densitized Triad

The densitized triad Ẽia is defined as

Ẽia := det (e) eai , (6.23)

where det (e) stands for the determinant of eia .

There are some useful identities relating the densitized triad that will be useful along the
development. We will prove some of them here.
First, let us manipulate the determinant identity for a 3-dimensional matrix e:

‘ijk eia ejb ekc = det (e) ‘˜abc .

One can multiply it by eal and use eia eal = ”li to get to

‘ljk ejb ekc = det (e)eal ‘˜abc = Ẽla ‘˜abc .

CHAPTER 6. ASHTEKAR FORMULATION OF GR 81

Contracting now with ‘˜bcd and using ‘˜abc ‘˜bcd = 2”ad we are left with
‘˜bcd ‘ljk ejb ekc = 2Ẽld ,
and, renaming some dummy indices, we get
1
Ẽia = ‘˜abc ‘ijk ejb ekc . (6.24)
2
One can also show that
‘ijk ‘˜abc Ẽjb Ẽkc
eia = . (6.25)
2 det(e)
For that, let us compute eia eal :
‘ijk ‘˜abc Ẽjb Ẽkc a
eia eal = e
2 det(e) l
det(e) ijk
= ‘ ‘˜abc ebj eck eal
2
det(e) ijk
= ‘ ‘ljk det(e≠1 )
2
det(e) i
= 2”l det(e)≠1
2
= ”li ,
where in the second line we just used (6.23). In the third line we used the equation for the 3-
dimensional determinant (see Appendix B) and in the fourth line we used det(A≠1 ) = (det A)≠1 .
Some other useful identities that follow from those definitions are
‘˜abc eja = ebp ecq ‘jpq det e, (6.26)
and
‘ijk Ẽ b Ẽ c
‘˜abc eia = Ô j k , (6.27)
det Ẽ
which follows from (6.25). Those identities will be useful along further development.

6.4.2 Ashtekar-Barbero Connection

One can use the spatial components of the (3+1) split on the spin connection ʵij to define
a new connection on the spatial slice.
We start by defining the extrinsic curvature Kai via
Kai := Êai0 = ≠Êa0i .
Secondly, let us remember that the spin connection is anti-symmetric in its internal indices:
ʵij = ≠ʵji . Hence, it is a 2-form on the internal space. Taking its Hodge dual we obtain the
dual spin connection ia :
1 i
a := ≠ ‘ jk Êa ,
i jk
2
or, inverting the equation:
Êajk = ≠‘ijk ia .
With those elements in hand we can define the Ashtekar-Barbero connection Aia :
Aia := i
a + “Kai , (6.28)
where “ is the Barbero-Irimizi parameter. This object is a 1-form in space, not spacetime, since
we have done the (3+1) split in defining those quantities.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 82

6.5 The Curvature Terms

We will now do the (3+1) split in the curvature terms that appear in equations (6.20), (6.21)
and (6.22). Let us remind the definition of the curvature 2-form
1 IJ µ ‹
F IJ = Fµ‹ dx dx ,
2
whose components are given by
1 IJ
Fµ‹ = ˆ[µ Ê‹]IJ + ÷KL Ê[µIK Ê‹]LJ .
2
Doing the (3+1) split in the spacetime indices:
1 IJ
F = ˆ[0 Êc]IJ + ÷KL Ê[0IK Êc]LJ , (6.29)
2 0c
1 IJ
F = ˆ[b Êc]IJ + ÷KL Ê[bIK Êc]LJ . (6.30)
2 bc
With those, we can decompose it in the internal space, using Ê 00 = 0, ÷00 = ≠1 and
÷ij = ”ij . Looking into equations (6.20), (6.21) and (6.22) we see that we have four different
terms involving the curvature. Let us write down in the (3+1) internal split each of those terms.
For the first term we get
1 0k
F = ˆ[0 Êc]0k + ÷ml Ê[00m Êc]kl
2 0c
= ≠ˆ[0 Kc]k ≠ ”ml K[0m Êc]kl
= ≠ˆ[0 Kc]k + ”ml ‘lk q K[0m q
c]

= ≠ˆ[0 Kc]k ≠ ‘k pq K[0P Q

c] , (6.31)
where we used Êai0 = Kai in the second line and in the third we used Êaij = ≠‘ijk a.
k

The second curvature term which appears in the integral is

1 kl
F = ˆ[0 Êc]kl + ÷00 Ê[0k0 Êc]0l + ÷mn Ê[0km Êc]nl
2 0c
= ˆ[0 Êc]kl + K[0k Kc]l + ”mn Ê[0km Êc]nl
= ≠‘kl p ˆ[0 p
c] + K[0k Kc]l + ”mn ‘km s ‘nl r s r
[0 c]

= ≠‘kl p ˆ[0 c]p + K[0k Kc]l ≠ k

[0
l
c] . (6.32)
For the third curvature term we have
1 k
F = ˆ[b Êc]0k + ÷ml Ê[b0m Êc]lk
2 bc
= ≠ˆ[b Kc]k ≠ ”ml K[bm Êc]lk
= ≠ˆ[b Kc]k + ”ml ‘lk q K[bm q
c]

= ≠ˆ[b Kc]k ≠ ‘k pq K[bp q

c] , (6.33)
and, finally, the last curvature term appearing in the integral is
1 kl
F = ˆ[b Êc]kl + ÷00 Ê[bk0 Êc]0l + ÷mn Ê[bkm Êc]nl
2 bc
= ≠‘kl p ˆ[b c]p + K[bk Kc]l + ”mn Ê[bkm Êc]ln
= ≠‘lk p ˆ[b p
c] + K[bk Kc]l + ”mn ‘km p ‘ln q p q
[b c]

= ≠‘kl p ˆ[b c]p + K[bk Kc]l ≠ k

[b
l
c] . (6.34)
CHAPTER 6. ASHTEKAR FORMULATION OF GR 83

6.6 The Ashtekar action and the equations of motion

Now, inserting equations (6.31), (6.32), (6.33) and (6.34) in (6.20), (6.21) and (6.22) we get,
finally, the partial terms of the action completely decomposed in the (3+1) split — in internal
and spacetime indices. As said before, each of these three terms, when developed, will lead us
to constraints in the system, as we will now show.

6.6.1 L1 : The first term and the Gauss constraint

From equation (6.20) we have
1 1 2
L1 = ≠ ‘˜abc eia ejb ”ik ”jl F kl0c + “‘ijk F 0k0c =
2 1 2
= ‘˜abc eia ejb ”ik ”jl ‘kl p ˆ[0 c]p ≠ K[0k Kc]l + k
[0
l
c] +
1 2
+ ‘˜abc eia ejb “‘ijk ˆ[0 Kc]k + ‘k pq K[0p q
c] .

1
Opening the anti-symmetrizers and using Ẽai = ‘˜abc ‘ijk eia ejb we get
2
1 Ë È
L1 = ‘ijm Ẽmc
”ik ”jl ‘kl p (ˆ0 cp ≠ ˆc 0p ) ≠ K0k Kcl + Kck K0l + 0
k l
≠ k
0
l
+
2 Ë È
c c

+ Ẽkc “ ˆ0 Kck ≠ ˆc K0k + ‘k pq (K0p cq ≠ Kcp 0q ) ,

and using ”ik ”jl ‘ijm ‘kl p = 2”pm we are left with

1 c m 1 k l 2
L1 = Ẽpc ˆ0 p
+ Ẽm
≠ Ẽpc ˆc ‘ kl 0 c ≠ ck 0l ≠ K0k Kcl + Kck K0l +
p
0
c
Ë 2 È
+ “ Ẽk ˆ0 Kc ≠ ˆc K0k + ‘k pq (K0p cq ≠ Kcp 0q ) .
c k

Now remember from equation (6.19) that the L1 term is being integrated in space and time.
Hence, integrating by parts we get, neglecting boundary terms:
⁄ ⁄
≠ Ẽpc ˆc p0 = p c
0 ˆc Ẽp
⁄ ⁄
≠ “ Ẽkc ˆc K0k = “K0k ˆc Ẽkc .

Then, we have
1 2
L1 = Ẽpc ˆ0 p
c + p c
0 ˆc Ẽp + Ẽm
c m
‘ kl 0
k
c
l
≠ K0k Kcl +
+ “ Ẽkc ˆ0 Kck + “K0k ˆc Ẽkc + “ Ẽkc ‘k pq (K0p c
q
≠ Kcp 0 ),
q

where in the parentheses of the first line we used the fact that ‘m kl is anti-symmetric in [k, l]
CHAPTER 6. ASHTEKAR FORMULATION OF GR 84

1
and that Akl = (Akl ≠ Alk ). Grouping now some terms and relabeling some indices we get
2
1 2 1 2 1 2 1 2
L1 = Ẽkc ˆ0 k
c + “ Ẽkc ˆ0 Kck + k c
0 ˆc Ẽk + “K0k ˆc Ẽkc + Ẽkc ‘kij i
0
j
c + “Kcj ≠ Ẽkc ‘kij K0i Kcj ≠ “ j
c =
1 2 1 2 1 2 1 2
= Ẽkc ˆ0 k
c + “Kck + ˆc Ẽkc k
0 + “K0k + Ẽkc ‘kij i
0
j
c + “Kcj ≠ Ẽkc ‘kij K0i Kcj ≠ “ j
c =
I A BJ
1 2 Ó 1 2Ô 1 j
= Ẽkc ˆ0 k
c + “Kck + i
0 ˆc Ẽic + Ẽkc ‘kij j
c + “Kcj + “K0i ˆc Ẽic ≠ Ẽkc ‘kij K ≠ j
=
“ c c

1 2 Ó 1 Ó 1 2Ô
1 2Ô
= Ẽkc ˆ0 k
c + “Kck + i
0 ˆc Ẽic + Ẽkc ‘kij≠ K0i ≠“ 2 ˆc Ẽic + Ẽkc ‘kij “Kcj ≠ “ 2 jc =
j
c + “Kcj
“
1 2 Ó 1 2Ô 1 Ó Ô
= Ẽkc ˆ0 kc + “Kck + i0 ˆc Ẽic + Ẽkc ‘kij jc + “Kcj ≠ K0i ≠“ 2 ˆc Ẽic + Ẽkc ‘kij “Kcj ≠ Ẽkc ‘kij “ 2 jc =
“
1 2 Ó 1 2Ô 1 Ó 1 2Ô
= Ẽkc ˆ0 kc + “Kck + i0 ˆc Ẽic + Ẽkc ‘kij jc + “Kcj ≠ K0i ˆc Ẽic + Ẽkc ‘kij jc + “Kcj +
“
1 Ó Ô
≠ K0i ≠“ 2 ˆc Ẽic ≠ Ẽkc ‘kij jc ≠ Ẽkc ‘kij “ 2 jc ≠ ˆc Ẽic =
“
A B
1 2 1 i Ó 1 2Ô
= Ẽk ˆ0 c + “Kc + 0 ≠ K0 ˆc Ẽic + Ẽkc ‘kij jc + “Kcj +
c k k i
“
I A B A BJ
1 1
+ K0 ˆc Ẽi “ +
i c
+ Ẽk ‘ ij c “ +
c k j
=
“ “
A B A B
1 2 1 Ó 1 2Ô 1 Ó Ô
= Ẽkc ˆ0 kc + “Kck + i0 ≠ K0i ˆc Ẽic + Ẽkc ‘kij jc + “Kcj + K0i “ + ˆc Ẽic + Ẽkc ‘kij jc ,
“ “

where, from the fifth to the sixth line we added and subtracted the term ˆc Ẽic + Ẽkc ‘kij j
c inside
the last parentheses of the equation.
Now, using
Akc = kc + “Kck , (6.35)
and introducing the quantities A B
1
– :=
i
+ “ K0i , (6.36)
“
1
⁄i := i
0 ≠ K0i , (6.37)
“

Gi := ˆc Ẽic + Ẽkc ‘kij Ajc , (6.38)

we are left with
L1 = Ẽkc ˆ0 Akc + ⁄i Gi + –i (ˆc Ẽic + Ẽkc ‘kij c ).
j
(6.39)
The last term in parentheses is just d E, which vanishes. This happens because the
connection is torsionless, i.e. T = d e = 0, where e is the frame field in form notation. One
could also write the densitized triad in forms notation via
1
E i ab = ‘i jk eja ekb = [e, e]i ab ,
2
hence, it is easy to see that
1
d E = d [e, e] = [d e, e] = [T, e] = 0
2
CHAPTER 6. ASHTEKAR FORMULATION OF GR 85

indeed vanishes.
Hence, the first term is just
L1 = Ẽkc ˆ0 Akc + ⁄i Gi , (6.40)
where Gi is called the Gauss Constraint, which generates SU (2) gauge transformations as we
will discuss later.

6.6.2 L2 : The second term and the Diffeomorphism constraint

From equations (6.33) and (6.34) in (6.21) we get
1 1 2
L2 = ≠ N d ‘˜abc eid eja ”ik ”jl F klbc + “‘ijk F 0kbc =
2 1 2
= N ‘˜ ed ea ”ik ”jl ‘kl
d abc i j
p ˆ[b c] ≠ K[b Kc] + [b
p k l k l
c] +
1 2
+ N d ‘˜abc eid eja “‘ijk ˆ[b Kc]k + ‘kpq K[bp q
c] .

Note that ”ik ”jl ‘kl p = ‘ijp , and, relabelling some dummy indices we can write
1 2
L2 = N d ‘˜abc eid eja ‘ijk ˆ[b k
c] + N d ‘˜abc eid eja ”ik ”jl k
[b
l
c] ≠ K[bk Kc]l +
1 2
+ N d ‘˜abc eid eja “‘ijk ˆ[b Kc]k + “‘ijk ‘kpq K[bp q
c] ,

and, grouping some similar terms we are led to

1 2 1 2
L2 = N d ‘˜abc eid eja ‘ijk ˆ[b k
c] + “ˆ[b Kc]k + N d ‘˜abc eid eja ”ik ”jl k
[b
l
c] ≠ K[bk Kc]l +
+ N d ‘˜abc eid eja “‘ijk ‘kpq K[bp q
c] .

From the definition of the Ashtekar-Barbero connection one can easily write the first term
in parentheses as ˆ[b c]k + “ˆ[b Kc]k = ˆ[b Ac]k . Also, in the last line we can write:

‘ijk ‘kpq = ÷ mk ‘ijk ‘pqm = ÷ mk (”ip ”jq ”km ≠ ”iq ”jp ”mk ) = ”ip ”jq ≠ ”iq ”jp .

Then we are left with

1 2
L2 = N d ‘˜abc eid eja ‘ijk ˆ[b Akc] + N d ‘˜abc eid eja ”ik ”jl k
[b
l
c] ≠ K[bk Kc]l +
+ N d ‘˜abc eid eja “(”ip ”jq ≠ ”iq ”jp )K[bp q
c] =
1 2
= N d ‘˜abc eid eja ‘ijk ˆ[b Akc] + N d ‘˜abc eid eja ”ik ”jl k
[b
l
c] ≠ K[bk Kc]l +
+ N d ‘˜abc eid eja “(K[bi j
c] ≠ K[bj c] ),
i

and, since the entire expression is being multiplied by ‘˜abc , which is already anti-symmetric in
[bc], we can drop out the anti-symmetrizers:
Ó 1 2 Ô
L2 = N d ‘˜abc eid eja ‘ijk ˆb Akc + ”ik ”jl b
k
c
l
≠ Kbk Kcl + “(Kbi c
j
≠ Kbj c)
i
=
Ó 1 2 Ô
= N d ‘˜abc eid eja ‘ijk ˆb Akc + i j
b c ≠ Kbj Kci + “(Kbi c
j
≠ Kbj c)
i
. (6.41)

However, the curvature 2-form of the Ashtekar-Barbero connection Akc is defined, in index
notation, as
1 k 1
Fbc := ˆ[b Akc] + ‘klm Alb Am
c . (6.42)
2 2
CHAPTER 6. ASHTEKAR FORMULATION OF GR 86

Hence, expanding Akc = k

c + “Kck and contracting both sides of (6.42) with ‘˜abc ‘ijk we get
5 6
1 abc 1
‘ ˜ ‘ijk Fbck = ‘˜abc ‘ijk ˆ[b Akc] + ‘ijk ‘klm ( lb + “Kbl )( mc + “Kc )
m
2 5
2 6
1
= ‘˜abc
‘ijk ˆ[b Ac] + (”il ”jm ≠ ”im ”jl )( b + “Kb )( c + “Kc )
k l l m m

5
2 6
1 2 l m
= ‘˜abc
‘ijk ˆ[b Ac] + (”il ”jm ≠ ”im ”jl )( b c + b “Kc + “Kb c + “ Kb Kc )
k l m l m l m

Ë
2 Ó 1 2 ÔÈ
2 l m
= ‘˜ ‘ijk ˆ[b Ac] + ”il ”jm lb m
abc k
c + “ l
K
b c
m
+ K l m
b c + “ K K
b c ) ,

1
where in the last line we used again the notation for anti-symmetric objects Aµ‹ = (Aµ‹ ≠ A‹µ ).
2
Therefore, subtracting ‘˜abc ”il ”jm (1 + “ 2 )Kbl Kcm from both sides we can write
; <
1 Ë Ó1 2 1 2ÔÈ
‘˜
abc
‘ijk Fbck ≠ ”il ”jm (1 + “ 2 )Kbl Kcm = ‘˜abc ‘ijk ˆ[b Akc] + ”il ”jm l m
≠ Kbl Kcm + “ l m
b Kc + Kbl m
.
2 b c c

Hence, in equation (6.41) we have

; <
1
L2 = N d ‘˜abc eid eja ‘ijk Fbck ≠ ”il ”jm (1 + “ 2 )Kbl Kcm ,
2
and since ‘˜abc eja = ebp ecq ‘jpq det e (equation (6.26)) we have
; <
1
L2 = N d eid ebp ecq |e|‘jpq ‘ijk Fbck ≠ ”il ”jm (1 + “ 2 )Kbl Kcm
;
2 <
1 p q
= N d eid ebp ecq |e| (”k ”i ≠ ”ip ”kq )Fbck ≠ ”il ‘pqm (1 + “ 2 )Kbl Kcm
;
2 <
1
= N |e| eid (ebp eci Fbcp ≠ ebi ecq Fbcq ) ≠ ”il eid ebp ecq ‘pqm (1 + “ 2 )Kbl Kcm
d
2
; Ë <
1 b i c p È
2
= N |e|
d
e (e e )F ≠ (ed ei )eq Fbc ) ≠ ”il ed ep eq ‘ m (1 + “ )Kb Kc
i b c q i b c pq l m
2 p d i bc
; Ë <
1 b p È
2
= N |e|
d
e F ≠ eq Fdc ) ≠ ”il ed ep eq ‘ m (1 + “ )Kb Kc
c q i b c pq l m

Ó
2 p bd Ô
= N d |e| ebp Fbd
p
≠ ”il eid ebp ecq ‘pqm (1 + “ 2 )Kbl Kcm , (6.43)

where we used eid eci = ”dc in the fifth line and Aµ‹ = (Aµ‹ ≠ A‹µ )/2 in the last one.
We can also express this in terms of the densitized triad Ẽia = |e|eai :
Ó Ô
L2 = N a |e| ebp Fba
p
≠ ”il eia ebp ecq ‘pqm (1 + “ 2 )Kbl Kcm
Ó Ô
= ≠N a |e| ebp Fab
p
+ ”il eia ebp ecq ‘pqm (1 + “ 2 )Kbl Kcm
Ó Ô
= ≠N a Ẽpb Fab
p
+ ”il eia ebp (1 + “ 2 )Kbl ‘pqm Kcm Ẽqc . (6.44)

Note that, from (6.38) we can write:

Gi = ˆc Ẽic + Ẽkc ‘kij ( j

c + “Kcj )
= ˆc Ẽic + Ẽkc ‘kij j
c + Ẽkc ‘kij “Kcj
= Ẽkc ‘kij “Kcj ,

since ˆc Ẽic + Ẽkc ‘kij j

c = d E = 0.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 87

1
Therefore we can write that ‘pqm Kcm Ẽqc = ≠ Gp . Hence, in (6.44) we get
“
I A B J
1
L2 = ≠N a p
Ẽpb Fab ≠ ”il eia ebp + “ Kbl Gp .
“

The part with the Gauss constraint Gp is redundant, its content is already covered by L1 .
We are then left with
L2 = ≠N a Ẽpb Fab
p
, (6.45)
or, defining the momentum constraint as

Va := Ẽpb Fab
p
, (6.46)

we can write it as
L2 = ≠N a Va . (6.47)
This is called the vector constraint, which is related to spatial diffeomorphisms, as we will
show and discuss later.

6.6.3 L3 : The third term and the Hamiltonian constraint

Finally we get, for the third term, from equations (6.33) and (6.34) in (6.22):
3 4
1 1
L3 = N eia ‘˜abc ”ik F 0kbc ≠ “‘ijk F jkbc
2 1
2 2
= N ea ‘˜ ”ik ≠ˆ[b Kc] ≠ ‘k pq K[bp c]q +
i abc k

1 1 2
≠ N eia ‘˜abc “‘ijk ≠‘jk p ˆ[b c]p + K[bj Kc]k ≠ j
[b
k
.
2 c]

Using equation (6.27) we are left with

N ‘imn Ẽmb
Ẽnc 1 2
L3 = ≠ Ô ”ij ˆb Kcj + ‘jpq Kbp cq +
det Ẽ
1 N ‘imn Ẽm
b
Ẽnc Ë 1 2È
+ Ô “ ‘ijk ‘jk p ˆb cp + ‘ijk bj ck ≠ Kbj Kck ,
2 det Ẽ
where we dropped the anti-symmetrizers since the expression is already anti-symmetric in [b, c].
Also, since ‘ijk ‘jk
p = 2”ip , and renaming some dummy indices, we get

N ‘imn Ẽmb
Ẽnc 1 2
L3 = ≠ Ô ”ij ˆb Kcj + ‘ijk Kbj ck +
det Ẽ
5 26
N ‘ Ẽ Ẽ c
imn b
1 1
+ Ô m n “ ”ij ˆb cj + ‘ijk bj ck ≠ Kbj Kck ,
det Ẽ 2
and, regrouping some terms:
C A BD
N ‘imn Ẽ b Ẽ c 1
L3 = Ô m n ”ij “ˆb c
j
≠ Kcj +
det Ẽ “
5 26
N ‘ Ẽm Ẽnc
imn b
1 1
≠ Ô ‘ijk Kbj k
+ “ Kbj Kck ≠ j k
.
det Ẽ
c
2 b c
CHAPTER 6. ASHTEKAR FORMULATION OF GR 88

1 j
Now, plugging Kcj = (A ≠ c)
j
into the equation we are led to
“ c
C A BD
N ‘imn Ẽ b Ẽ c 1
L3 = Ô m n ”ij “ˆb cj ≠ 2 (Ajc ≠ jc ) +
det Ẽ “
C I JD
N ‘imn Ẽm
b
Ẽnc 1 j 1 1 j
≠ Ô ‘ijk (Ab ≠ b ) ck + “
j
(A ≠ b )(Ac
j k
≠ c)
k
≠ j k
,
det Ẽ “ 2 “2 b b c

and, developing the equation:

C A B D
N ‘imn Ẽm
b
Ẽnc 1 1
L3 = Ô ˆb “ ci ≠ (Aic ≠ ic ) ≠ ‘ijk (Ajb ≠ jb ) ck +
det Ẽ “ “
C I JD
imn b
N ‘ Ẽm Ẽn c
1 1 j k
≠ Ô ‘ijk (A A ≠ Ab c + b c ≠ b Ac ) ≠ “ b c
j k j k j k j k
=
det Ẽ 2 “ b c
N ‘imn Ẽm
b
Ẽnc Ë 2 È
= Ô “ ˆb ci ≠ ˆb Aic + ˆb ic ≠ ‘ijk (Ajb ≠ jb ) ck +
“ det Ẽ
N ‘imn Ẽm
b
Ẽnc 1Ë È
≠ Ô ‘ijk (Ajb Akc ≠ Ajb kc + jb kc ≠ jb Akc ) ≠ “ 2 bj ck .
“ det Ẽ 2

Note that the term proportional to ‘ijk is

1 1 1 1 1
≠Ajb k
+ k j
≠ Ajb Akc + Ajb k
≠ j k
+ j k
b Ac + “2 j k
c ,
c c b
2 2 c
2 b c
2 2 b

which can be simplified as

1 1 1 1 1
≠ Ajb k
+ k j
≠ Ajb Akc + j k
b Ac + “2 j k
c ,
2 c
2 c b
2 2 2 b

or, finally
1 k j 1 j k 1 2 1 1
≠ AA + “ j k
= (1 + “ 2 ) k j
≠ Ajb Akc .
2 c b 2 b c 2 b c
2 c b
2
Hence, we can write the Lagrangian as

N ‘imn Ẽmb
Ẽnc Ë È
L3 = ≠ Ô ˆb Aic ≠ (1 + “ 2 )(ˆb ic ) +
“ det Ẽ
5 6
N ‘imn Ẽm
b
Ẽnc 1 j k 1 2
≠ Ô ‘ijk Ab Ac ≠ (1 + “ ) c b , k j
“ det Ẽ 2 2

then,
53 4 3 46
N ‘imn Ẽ b Ẽ c 1 1
L3 = ≠ Ô m n ˆb Aic + ‘ijk Ajb Akc ≠ (1 + “ 2 ) ˆb i
+ ‘ijk k j

“ det Ẽ 2 c
2 c b

or, using ‘imn = ” ip ‘pmn :

53 4 3 46
N ‘ mn Ẽ b Ẽ c 1 1
L3 = ≠ iÔ m n ˆb Aic + ‘ijk Ajb Akc ≠ (1 + “ 2 ) ˆb i
+ ‘ijk k j
,
“ det Ẽ 2 c
2 c b
CHAPTER 6. ASHTEKAR FORMULATION OF GR 89

However, the expressions in parentheses are the equations in components for the curvature
2-forms of the connections Aij and ij , respectively:
1 i 1
F bc := ˆb Aic + ‘ijk Ajb Akc , (6.48)
2 2
1 i 1
R := ˆb ic + ‘ijk ck jb .
2 bc 2
Hence, the expression can be written as

N ‘imn Ẽm
b
Ẽnc Ë i È
L3 = ≠ Ô F bc ≠ (1 + “ 2 )Ribc . (6.49)
2“ det Ẽ
If one defines the scalar constraint C as
‘imn Ẽm
b
Ẽnc Ë i È
C := ≠ Ô F bc ≠ (1 + “ 2 )Ribc , (6.50)
2“ det Ẽ
then
L3 = N C, (6.51)
where C is the scalar constraint — or Hamiltonian constraint — and N is just a Lagrangian
multiplier.

6.7 The Hamiltonian as a Linear Combination of Con-

straints
Back to equation (6.19) we can now write the action as
1⁄ ⁄ 1 2
S= dt d3 x Ẽia ˆt Aia + ⁄i Gi + N a Va + N C , (6.52)
“
which is the Ashtekar action for classical gravity. From the first term we can see that the
Ashtekar-Barbero connection Aib and the densitized triad Ẽib are conjugate variables. Here,
⁄i , N a and N are Lagrange multipliers. The following terms — already previously defined —
deserve to be highlighted again for the sake of clarity:

• Gauss constraint: Gi := ˆc Ẽic + Ẽkc ‘kij Ajc

• Vector constraint: Va := Ẽib Fab

i

‘imn Ẽm
b
Ẽnc i
• Hamiltonian constraint: C := ≠ Ô [F bc ≠ (1 + “ 2 )Ribc ]
2“ det Ẽ
We can then get the Hamiltonian
⁄ 1 2
H[Aia ; Ẽia ] = d3 x ⁄i Gi + N a Va + N C , (6.53)

with the first class constraints, which generates the expected gauge freedom: the triad rotations
and spacetime diffeomorphisms, which is discussed in the next section.
If one writes the Hamilton equations that result from this Hamiltonian, one will indeed
reproduce Einstein field equations, as expected.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 90

6.8 Geometrical interpretation of the Constraints

We will now conclude our discussion by developing the geometrical interpretation of the
constraints in the Hamiltonian on Ashtekar’s formulation of General Relativity.

6.8.1 Electromagnetism
Let us first develop the geometrical interpretation of the constraints for a more familiar
theory, classical electromagnetism. The Lagrangian for electromagnetism is
1⁄ 3
L= d xFµ‹ F µ‹ , (6.54)
4
where Fµ‹ = ˆµ A‹ ≠ ˆ‹ Aµ is the electromagnetic tensor and Aµ is the vector potential. It is easy
to see that the zeroth component of the vector potential will not appear with a time derivative,
since Fµ‹ vanishes for µ = ‹. Hence, only spatial derivatives of A0 appear in the Lagrangian,
which means that it does not play a dynamical role. It is, actually, a Lagrange multiplier, as we
will show later.
Hence, being Ab our dynamical variable, its conjugate momentum is given by
”L
fib = , (6.55)
” Ȧb

which, when one takes the functional derivative of (6.54), gives Ẽ b — the electric field. The
canonical variables are then Ab (x) and Ẽ b (x), where a stands for spatial coordinates, as usual:
a = 1, 2 or 3 (remember that A0 is not a dynamical variable).
When taking the functional derivative we get a density, as expected, since the Lagrangian is
a volume integral, hence, its integrand must be a scalar density. When one takes a functional
derivative, the integral disappears and the result must be a density. The Poisson bracket of
Ẽ b (x) and Ab (x) (which is not a density) is again a density, as expected
Ó Ô
Ab (x), Ẽ b (x) = ”ba ” 3 (x ≠ y).

By the usual Legendre transformation one can build the Hamiltonian

⁄ 1 2
H := d3 x Ẽ b (x)Ȧb (x) ≠ L̃

which, when written in terms of the canonical pairs, gives

⁄ 3 4
3 1Ë a È
H= dx E (x)E b (x) + B a (x)B b (x) ”ab ≠ A0 ˆa E a (6.56)
2
1
where B a = ‘abc Fbc is the magnetic field, which is a function of Aa .
2
Working out the equation of motion for fi 0 we get [8]
Ó Ô
fi̇ 0 = fi 0 , H
= ˆa E a , (6.57)

”L
which should vanish, since fi – = = F –0 is zero for – = 0. Hence, its time evolution should
” Ȧ–
also vanish. Therefore we get ˆa E a = 0, which is Gauss’s law without the presence of charges.
CHAPTER 6. ASHTEKAR FORMULATION OF GR 91

This is a constraint, since it implies that we cannot have any E a for a electric field, but only
configurations for which the divergence is zero. We can now see how constraints are generators
of symmetries. Here, it is good to introduce the idea of a smeared constraint
⁄
G(⁄) := d3 x⁄ˆa E a ,

where the parameter ⁄ is an arbitrary smooth and differentiable function of x. Requiring that
the smeared constraint G(⁄) vanishes for all ⁄ is equivalent to requiring that the constraint
itself vanishes at all points of the manifold. This is important to do since we are dealing with
densities and distributions and not with functions itself. Since distributions behave better under
an integral, it will be in most cases easier to deal with the smeared constraint than with the
constraint itself.
Taking the Poisson bracket of the smeared constraint and the Hamiltonian, one finds out
that it vanishes Ó Ô
G(⁄), H = 0,
as expected, which means that the Hamiltonian does not change under the transformation
generated by the constraint; hence, the theory and the physics is unmodified, and this is indeed
a symmetry, as expected.
Taking the Poisson bracket of the smeared constraint with the conjugate variables we get
Ó Ô
G(⁄), E a = 0,

which means that the electric field is unchanged under the transformation generated by the
constraint. And, finally, one can compute that
Ó Ô
G(⁄), Aa = ˆa ⁄,

which means that the vector potential may change by the gradient of a function ⁄(x). We
already knew that the vector potential is defined up to the gradient of a function, which is a
gauge freedom of the theory. Therefore, we see that the constraints give rise to symmetries,
which are revealed in the gauge freedom of the system. In this context, Gauss law is called the
generator of gauge transformations, since it comes up as a constraint and it gives rise to the
gauge freedom that we have in choosing the potential vector Aµ . Moreover, A0 is the Lagrange
multiplier of the constraint ˆa E a , as we mentioned previously.
By computing the time evolution of the canonical variables Aa and E a one can recover the
rest of Maxwell equations
Ó Ô
Ȧa = Aa , H = Ea + ˆa A0
Ó Ô
Ė a = E a , H = ‘abc ˆb Bc

where it is easy to note that the evolution depends on the choice of the Lagrange multiplier A0 ,
which makes the vector potential defined only up to the gradient of a function ⁄(x), which is
the gauge symmetry of Maxwell’s theory.

6.8.2 Gravity
We may now apply the same reasoning to the Hamiltonian in (6.53) in terms of the conjugate
canonical pair Ẽia and Aia .
CHAPTER 6. ASHTEKAR FORMULATION OF GR 92

We can introduce a combination of the vector and Gauss constraints, which we call the
diffeomorphism constraint:
Ca = Va ≠ Aia Gi . (6.58)
Writing the smeared diffeomorphism constraint V as
⁄
˛) =
C(N d3 xN a Ca

and computing the Poisson bracket of this smeared constraint with a function of the canonical
coordinates f (Ẽ, A), one gets Ó Ô
C(N ˛ ), f (Ẽ, A) ≥ L ˛ f, (6.59)
N

which states that the orbit generated by the constraint in phase space is just the Lie derivative
along N a , up to a constant factor that depends on the choice of renormalization. Therefore, this
is called the diffeomorphism constraint, since it generates infinitesimal spatial diffeomorphism
transformations.
One can also smear the Gauss constraint Gi and get
⁄
G(⁄) = d3 x⁄i Gi ,

which generates the infinitesimal gauge transformation

Ó Ô
G(⁄), A ≥ dA ⁄ (6.60)

and also Ó Ô
G(⁄), E ≥ [⁄, E] (6.61)
which are the SU (2) gauge transformations.
CHAPTER 7. CONCLUSION 94

GR, so the decomposition of the metric gab in its spatial part hab will be necessary for one to
define a time parameter in order to talk about the evolution of the system, which is necessary if
one wants to build a Hamiltonian representation of the system.
This allowed us to develop a Hamiltonian formalism for GR — the ADM formalism —
where we could write a Hamiltonian as a sum of constraints multiplied by Lagrange multipliers.
This provided a Hamiltonian representation of the dynamics of the spacetime geometry. The
canonical variables here are the induced metric hab and its conjugate momenta fiab . With this
Hamiltonian it becomes possible to study the spacetime dynamics in a canonical way, using
every tool of the Hamiltonian formalism.
We then developed the tetrads formalism. Here, we replaced the metric gµ‹ as the main
dynamical variable for the tetrads eIµ and eJ‹ . Since Riemannian manifolds are locally flat,
one can always choose an orthonormal basis of vectors {e0 , e1 , e2 , e3 } for each point P on the
manifold M , and that is the starting point of this new formalism. The tetrads contain the
same geometrical information from the manifold as the metric gµ‹ , since they are related via
gµ‹ = eI‹ eJ‹ ÷IJ . Hence, by taking the determinant, we get g = ≠e2 and the tetrad is just the
square root of the metric and it has, therefore, all the information about the geometry of the
manifold. We can thus consider the tetrad as the fundamental description and the metric as a
derived concept.
We then built the two Cartan’s structural equations. The first one was deI + Ê I K eK = 0,
which allowed us to find the spin connection Ê if we had the tetrads e. The second one was
F K J = dÊ K J + Ê K L Ê L J , which allowed us to find the curvature 2-form FJK . So, given a metric,
one chooses a basis of tetrads eI , founds the spin connection and then the curvature 2-form with
both of Cartan’s structural equations.
Since we had the relation between F IJ and the Riemann curvature tensor Rµ‹‡ fl
, we were then
able to translate the Palatini action in terms of the tetrads and the curvature 2-form F. This
action led us to the same conclusions that we arrived before, only now in a different language —
the Einstein field equations and the metric compatibility equation in the notation of differential
forms. Finally, we did a slight modification in this action, by introducing a parameter “, the
Barbero-Immirzi parameter. This did not change the equations of motion, but we were led to a
more general action — the Holst action.
Finally, we developed the Hamiltonian formalism using the Holst action, which led us to the
formulation of GR in terms of some new variables — the Ashtekar-Barbero connection Aia and
the densitized triad Ẽjb . In a way, we mixed what was developed in the last two chapters, since
the main idea here consisted in developing a Hamiltonian formalism after we did the (3+1)-split
of the geometry. However we have used the Holst action instead of the Einstein-Hilbert one,
hence, our variables were in terms of the triads — the spatial part of the tetrads — and not
the 3-metric hab . Although the development was extensive, the steps followed in this part led
us to the construction of a constrained Hamiltonian for GR, which allowed us to extract some
symmetries of the system, the constraints giving rise to gauge transformations.
This puts us one step behind the quantization of gravity. In the Hamiltonian formalism,
one can promote the Poisson brackets to commutators and the conjugate variables to operators,
and, at least in theory, quantize gravity. The path followed for the canonical quantization of
gravity in this approach with the Ashtekar formulation of gravity is known as loop quantum
gravity, which can be studied in a future work.
APPENDIX C. BIANCHI AND PALATINI IDENTITIES 100

we have that
IJ
”Fµ‹ = ˆ[µ ”Ê‹]
IJ
+ Ê[µK
I KJ
”Ê‹] J
≠ Ê[µK IK
”Ê‹]
= D(”Ê IJ )µ‹

or, suppressing the spacetime indices

”F IJ = D(”Ê IJ )
APPENDIX D. TENSOR VS TENSOR DENSITIES 102

This quantity is the same in any coordinate system, which is why it is not a tensor. However
we can relate the tensor density à to the proper tensor A via
1 2Ê
à = |g|1/2 A

where Ê is the weight of the tensor density and g stands for the determinant of the metric.
One can show that
‘˜µ‹‡fl = |g|1/2 ‘µ‹‡fl
and thus the Levi Civita symbol is a tensor of weight +1.
For the indices (IJKL), we have not used the tilde since those indices are of the internal
flat space, and, hence, ‘IJKL is actually a tensor and not a density.
It is also possible to define the 3-dimensional symbol as ‘˜abc := ‘˜0abc , which could also be
done for 2D or 1D. The same could be done for the (IJKL) indices in the internal space.