Introduction LQ G

Covariant Loop Quantum Gravity
An elementary introduction to Quantum Gravity and Spinfoam Theory
CARLO ROVELLI AND FRANCESCA VIDOTTO
iii
Cover: Martin Klimas, Jimi Hendrix, House Burning Down.
Contents
page ix
Preface
Part I Foundations
iv
1 Spacetime as a quantum object

1.1 The problem
1.2 The end of space and time
1.3 Geometry quantized
1.4 Physical consequences of the existence of the Planck scale
1.4.1 Discreteness: scaling is finite
1.4.2 Fuzziness: disappearance of classical space and time
1.5 Graphs, loops and quantum Faraday lines
1.6 The landscape
1.7 Complements
1.7.1 SU(2) representations and spinors
1.7.2 Pauli matrices
1.7.3 Eigenvalues of the volume
3
3
6
9
18
18
19
20
22
24
24
27
29
2 Physics without time

2.1 Hamilton function
2.1.1 Boundary terms
2.2 Transition amplitude
2.2.1 Transition amplitude as an integral over paths
2.2.2 General properties of the transition amplitude
2.3 General covariant form of mechanics
2.3.1 Hamilton function of a general covariant system
2.3.2 Partial observables
2.3.3 Classical physics without time
2.4 Quantum physics without time
2.4.1 Observability in quantum gravity
2.4.2 Boundary formalism
2.4.3 Relational quanta, relational space
2.5 Complements
2.5.1 Example of timeless system
2.5.2 Symplectic structure and Hamilton function
31
31
35
36
37
40
42
45
46
47
48
51
52
54
55
55
57
Contents
3 Gravity
3.1 Einsteins formulation
3.2 Tetrads and fermions
3.2.1 An important sign
3.2.2 First-order formulation
3.3 Holst action and Barbero-Immirzi coupling constant
3.3.1 Linear simplicity constraint
3.3.2 Boundary term
3.4 Hamiltonian general relativity
3.4.1 ADM variables
3.4.2 What does this mean? Dynamics
3.4.3 Ashtekar connection and triads
3.5 Euclidean general relativity in three spacetime dimensions
3.6 Complements
3.6.1 Working with general covariant field theory
3.6.2 Problems
59
59
60
63
64
65
67
68
69
69
71
73
76
78
78
81
4 Classical discretization
4.1 Lattice QCD
4.1.1 Hamiltonian lattice theory
4.2 Discretization of covariant systems
4.3 Regge calculus
4.4 Discretization of general relativity on a two-complex
4.5 Complements
4.5.1 Holonomy
4.5.2 Problems
82
82
84
85
87
91
98
98
99
Part II 3d theory
5 3d Euclidean theory
5.1 Quantization strategy
5.2 Quantum kinematics: Hilbert space
5.2.1 Length quantization
5.2.2 Spin networks
5.3 Quantum dynamics: Transition amplitudes
5.3.1 Properties of the amplitude
5.3.2 Ponzano-Regge model
5.4 Complements
5.4.1 Elementary harmonic analysis
5.4.2 Alternative form of the transition amplitude
5.4.3 Poisson brackets
5.4.4 Perimeter of the Universe
101
103
103
104
105
106
110
114
115
118
118
119
121
122
vi
Contents
6 Bubbles and cosmological constant

6.1 Vertex amplitude as gauge-invariant identity
6.2 Bubbles and spikes
6.3 Turaev-Viro amplitude
6.3.1 Cosmological constant
6.4 Complements
6.4.1 Few notes on SU(2)q
Part III The real world
123
123
125
129
131
133
133
135
7 The real world: 4D Lorentzian theory

7.1 Classical discretization
7.2 Quantum states of gravity
7.2.1 Y map
7.2.2 Spin networks in the physical theory
7.2.3 Quanta of space
7.3 Transition amplitudes
7.3.1 Continuum limit
7.3.2 Relation with QED and QCD
7.4 Full theory
7.4.1 Face-amplitude, wedge-amplitude and the kernel P
7.4.2 Cosmological constant and IR finiteness
7.4.3 Variants
7.5 Complements
7.5.1 Summary of the theory
7.5.2 Computing with spin networks
7.5.3 Spectrum of the volume
7.5.4 Unitary representation of the Lorentz group and the Y map
137
137
140
141
143
145
147
150
152
153
154
155
156
158
158
160
163
166
8 Classical limit
8.1 Intrinsic coherent states
8.1.1 Tetrahedron geometry and SU(2) coherent states
8.1.2 Livine-Speziale coherent intertwiners
8.1.3 Thin and thick wedges and time oriented tetrahedra
8.2 Spinors and their magic
8.2.1 Spinors, vectors and bivectors
8.2.2 Coherent states and spinors
8.2.3 Representations of SU(2) and SL(2,C) on functions of
spinors and Y map
8.3 Classical limit of the vertex amplitude
8.3.1 Transition amplitude in terms of coherent states
8.3.2 Classical limit versus continuum limit
8.4 Extrinsic coherent states
170
170
171
175
176
177
178
180
181
183
184
188
191
Contents
vii
9 Matter
9.1 Fermions
9.2 Yang-Mills fields
Part IV Physical applications
196
196
202
205
207
207
209
211
214
219
220
220
220
221
222
10 Black holes
10.1 Bekenstein-Hawking entropy
10.2 Local thermodynamics and Frodden-Ghosh-Perez energy
10.3 Kinematical derivation of the entropy
10.4 Dynamical derivation of the entropy
10.4.1 Entanglement entropy and area fluctuations
10.5 Complements
10.5.1 Accelerated observers in Minkowski and Schwarzshild
10.5.2 Tolman law and thermal time
10.5.3 Algebraic quantum theory
10.5.4 KMS and thermometers
10.5.5 General covariant statistical mechanics and quantum
gravity
223
11 Cosmology
11.1 Classical cosmology
11.2 Canonical Loop Quantum Cosmology
11.3 Spinfoam Cosmology
11.3.1 Homogeneus and isotropic geometry
11.3.2 Vertex expansion
11.3.3 Large-spin expansion
11.4 Maximal acceleration
11.5 Physical predictions?
226
226
229
231
232
233
234
236
237
12 Scattering
12.1 n-point functions in general covariant theories
12.2 Graviton propagator
238
238
242
13 Final remarks
13.1 Brief historical note
13.2 What is missing
247
247
248
Index
252
References
Index
252
265
Preface
To ours teachers and all those who teach children to question our
knowledge, learn through collaboration and the joy of discovery.
This book is an introduction to loop quantum gravity (LQG) focusing on its covariant formulation. The book has grown from a series of lectures given by Carlo
Rovelli and Eugenio Bianchi at Perimeter Institute during April 2012 and a course
given by Rovelli in Marseille in the winter 2013. The book is introductory, and assumes only some basic knowledge of general relativity, quantum mechanics and
quantum field theory. It is simpler and far more readable than the loop quantum
gravity text Quantum Gravity [Rovelli (2004)], and the advanced and condensed
Zakopane lectures [Rovelli (2011)], but it covers, and in facts focuses, on the momentous advances in the covariant theory developed in the last few years, which
have lead to finite transition amplitudes and were only foreshadowed in [Rovelli
(2004)].
There is a rich literature on LQG, to which we refer for all the topics not covered
in this book. On quantum gravity in general, Claus Kiefer has a recent general introduction [Kiefer (2007)]. Ashtekar and Petkov are editing a Springer Handbook
of Spacetime [Ashtekar and Petkov (2013)], with numerous useful contributions,
including John Engle article on spinfoams.
A fine book with much useful background material is John Baez and Javier Munians [Baez and Munian (1994)]. By John Baez, see also [Baez (1994b)], with many
ideas and a nice introduction to the subject. An undergraduate level introduction
to LQG is provided by Rodolfo Gambini and Jorge Pullin [Gambini and Pullin
(2010)]. A punctilious and comprehensive text on the canonical formulation of the
theory, rich in mathematical details, is Thomas Thiemanns [Thiemann (2007)]. The
very early form of the theory and the first ideas giving rise to it can be found in
the 1991 book by Abhay Ashtekar [Ashtekar (n.d.)].
A good recent reference is the collection of the proceedings of the 3rd Zakopane
school on loop quantum gravity, organized by Jerzy Lewandowski [Barrett et al.
(2011a)]. It contains an introduction to LQG by Abhay Ashtekar [Ashtekar (2011)],
Rovellis Zakopane lectures [Rovelli (2011)], the introduction by Kristina Giesel
and Hanno Sahlmann to the canonical theory, and John Barrett et al review on the
semiclassical approximation to the spinfoam dynamics [Barrett et al. (2011b)]. We
also recommend Alejandro Perez spinfoam review [Perez (2012)], which is complementary to this book in several way. Finally, we recommend the Hall Haggards
viii
Preface
ix
thesis, online [Haggard (2011)], for a careful and useful introduction and reference
for the mathematics of spin networks.
We are very grateful to Klaas Landsman, Gabriele Stagno, Marco Finocchiaro,
Hal Haggard, Tim Kittel, Thomas Krajewski, Cedrick Miranda Mello, Aldo Riello,
Tapio Salminem and, come sempre, Leonard Cottrell, for careful reading the notes,
corrections and clarifications. Several tutorials have been prepared by David Kubiznak and Jonathan Ziprick for the students of the International Perimeter Scholars: Andrzej, Grisha, Lance, Lucas, Mark, Pavel, Brenda, Jacob, Linging, Robert,
Rosa, thanks also to them!
Cassis, November 4th, 2013
Carlo Rovelli and Francesca Vidotto
Part I
FOUNDATIONS
Spacetime as a quantum object

This book introduces the reader to a theory of quantum gravity. The theory is
covariant loop quantum gravity (covariant LQG). It is a theory that has grown
historically via a long indirect path, briefly summarized at the end of this chapter.
The book does not follows the historical path. Rather, it is pedagogical, taking the
reader through the steps needed to learn the theory.
The theory is still tentative for two reasons. First, some questions about its consistency remain open; these will be discussed later in the book. Second, a scientific
theory must pass the test of experience before becoming a reliable description of
a domain of the world; no direct empirical corroboration of the theory is available
yet. The book is written in the hope that some of you, our readers, will be able to
fill these gaps.
This first chapter clarifies what is the problem addressed by the theory and gives
a simple and sketchy derivation of the core physical content of the theory, including its general consequences.
1.1 The problem

After the detection at CERN of a particle that appears to match the expected properties of the Higgs [ATLAS Collaboration (2012); CMS Collaboration (2012)], the
demarcation line separating what we know about the elementary physical world
from what we do not know is now traced in a particularly clear-cut way. What we
know is encapsulated into three major theories:
- Quantum mechanics, which is the general theoretical framework for describing
dynamics.
- The SU(3)SU(2)U(1) standard model of particle physics, which describes all matter we have so far observed directly, with its non-gravitational interactions.
- General relativity (GR), which describes gravity, space and time.
In spite of the decades-long continuous expectation of violations of these theories,
in spite of the initial implausibility of many of their predictions (long-distance entanglement, fundamental scalar particles, expansion of the universe, black holes...),
and in spite of the bad press suffered by the standard model, often put down as an
incoherent patchwork, so far Nature has steadily continued to say Yes to all predictions of these theories and No to all predictions of alternative theories (proton
decay, signatures of extra-dimensions, supersymmetric particles, new short range
3
forces, black holes at LHC...). Anything beyond these theories is speculative. It is

good to try and to dream: all good theories were attempts and dreams, before becoming credible. But lots of attempts and dreams go nowhere. The success of the
above package of theories has gone far beyond anybodys expectation, and should
be taken at its face value.
These theories are not the final story about the elementary world, of course.
Among the open problems, three stand out:
- Dark matter.
- Unification.
- Quantum gravity.
These are problems of very different kind.1 The first of these2 is due to converging
elements of empirical evidence indicating that about 85% of the galactic and cosmological matter is likely not to be of the kind described by the standard model.
Many alternative tentative explanations are on the table, so far none convincing
[Bertone (2010)]. The second is the old hope of reducing the number of free parameter and independent elements in our elementary description of Nature. The
third, quantum gravity, is the problem we discuss here. It is not necessarily related
to the first two.
The problem of quantum gravity is simply the fact that the current theories are
not capable of describing the quantum behaviour of the gravitational field. Because of this, we lack a predictive theory capable of describing phenomena where
both gravity and quantum theory play a role. Examples are the center of a black
hole, very early cosmology, the structure of Nature at very short scale, or simply
the scattering amplitude of two neutral particles at small impact parameter and
high energy. See Figure 1.1.
Observational technology has recently began to reach and probe some aspects of
this regime, for instance its Lorentz invariance, and has already empirically ruled
out some tentative theoretical ideas [Liberati and Maccione (2009)]. This is major
advance from a few years ago, when the quantum gravitational domain appeared
completely unreachable by our observation. But for the moment direct empirical
information on this regime is only minimal. This would be a problem if we had
many alternative complete theories of quantum gravity to select from. But we are
not in this situation: we have very few, if any. We are not all in a situation of excessive theoretical freedom: the shortcoming in the set of fundamental laws is strident
and calls for a solution, but consistency with what we know limits dramatically
our freedom, which is good, since freedom is just another word for nothing left to
lose.
The problem is even more serious: our successful theories are based on contradictory hypotheses. A good student following a general-relativity class in the
1
2
To these one can add the problem of the interpretation of quantum mechanics, which is probably of
still another kind.
Not to be confused with the improperly called dark energy mystery, much less of a mystery than
usually advertised [Bianchi and Rovelli (2010a,b)].
The problem
Spinfoam Cosmology
1. Introduction
GR
In QED one can compute the cross-section for the scattering of two electrons at low
energy.
Fock space of photons and electrons
Theory
Feynman rules
QFT
vertex :
Figure 1.1
Example
wave packets
Feynman diagrams expansion
computation scattering of neutral particles in Plank units

Regimes for the gravitational
We
want
to perform
the same
exercise
in Loop Gravity.
Though
this exercise,system
we
c = h = G = 1. E is the
energy
in the
center
of mass
reference
and b the
achieve to:
impact parameter (how close to one another come the two particles). At low energy,
test the spinfoam dynamics in a cosmological setting;
effective QFT is sufficient to predict the scattering amplitude. At high energy,
develop approximation schemes;
classical general relativity is generally sufficient. In (at least parts of the) intermediate
ultimately, extract (new) physics from the theory.
region (colored wedge) we do not have any predictive theory.
A viable covariant dynamics for Loop Quantum Gravity (LQG) is a relatively recent
achievement. In 2008 the spinfoam vertex expansion became available [1, 2, 3, 4, 5]. In
2008, this vertex was proven to yield General Relativity the study of its asymptotics gives
in fact Regge Gravity [6], namely a discretization of classical General Relativity.
In order to apply this new results to cosmology, it was soon realized that a different strategy with respect to the one used in Loop Quantum Cosmology (LQC) [7] was
needed. In LQC, quantization is defined in the context of canonical quantum cosmology:
the basic ingredients are the Wheeler-deWitt framework, the use of connection variables,
the quantization of the holonomies, plus one notion imported from the full loop theory:
the existence of a minimal area gap.
The covariant dynamics allows us to make use of the full quantum theory as a starting
point for quantum cosmology. In doing so, a central tool turns out to be the truncation on
a graph. The Hilbert space of LQG is defined as a sum of graph spaces, i.e. Hilbert spaces
on an abstract graph G defined just by the number N of its nodes, the number L of its
links and their adjacency relations. The restriction of the theory to a given graph defines
a truncation of the infinite number of degrees of freedom of the full theory down to a
finite number [8]. A graph with a single node describes a minimal number of degrees of
freedom. A graph with more nodes (for instance n = 2 instead of n = 1 nodes) and more
links, includes more degrees of freedom, that can be used to describe inhomogeneities
and anisotropies [9]. A simple graph can code information about the whole large scale
structure of the universe.
morning and a quantum-field-theory class in the afternoon must think her teachers are chumps, or havent been talking to one another for decades. They teach
two totally different worlds. In the morning, spacetime is curved and everything is
smooth and deterministic. In the afternoon, the world is formed by discrete quanta
jumping over a flat spacetime, governed by global symmetries (Poincare) that the
morning teacher has carefully explained not to be features of our world.
Contradiction between empirically successful theories is not a curse: it is a terrific opportunity. Several of the major jumps ahead in physics have been the result of efforts to resolve precisely such contradictions. Newton discovered universal gravitation by combining Galileos parabolas with Keplers ellipses. Einstein
discovered special relativity to solve the irreconcilable contradiction between
2
mechanics and electrodynamics. Ten years
later, he discovered that spacetime is
curved in an effort to reconcile Newtonian gravitation with special relativity. Notice that these and other major steps in science have been achieved without virtually any new empirical data. Copernicus for instance constructed the heliocentric
model and was able to compute the distances of the planets from the Sun using
only the data in the book of Ptolemy.3
3
This is not in contradiction with the fact that scientific knowledge is grounded on an empirical basis.
First, a theory becomes reliable only after new empirical support. But also the discovery itself of a
new theory is based on an empirical basis even when there are no new data: the empirical basis is
the empirical content of the previous theories. The advance is obtained from the effort of finding the
overall conceptual structure where these can be framed. The scientific enterprise is still finding theories explaining observations, also when new observations are not available. Copernicus and Einstein
where scientists even when they did not make use of new data. (Even Newton, thought obsessed
by getting good and recent data, found universal gravitation essentially by merging Galileo and
Keplers laws.) Their example shows that the common claim that there is no advance in physics
without new data is patently false.
This is precisely the situation with quantum gravity. The scarcity of direct empirical information about the Planck scale is not dramatic: Copernicus, Einstein,
and, to a lesser extent, Newton, have understood something new about the world
without new data just comparing apparently contradictory successful theories.
We are in the same privileged situation. We lack their stature, but we are not excused from trying hard.4
1.2 The end of space and time

The reason for the difficulty, but also the source of the beauty and the fascination
of the problem, is that GR is not just a theory of gravity. It is a modification of
our understanding of the nature of space and time. Einsteins discovery is that
spacetime and the gravitational field are the same physical entity.5 Spacetime is a
manifestation of a physical field. All fields we know exhibit quantum properties
at some scale, therefore we believe space and time to have quantum properties as
well.
We must thus modify our understanding of the nature of space and time, in
order to take these quantum properties into account. The description of spacetime
as a (pseudo-) Riemannian manifold cannot survive quantum gravity. We have to
learn a new language for describing the world: a language which is neither that
of standard field theory on flat spacetime, nor that of Riemannian geometry. We
have to understand what quantum space and what quantum time are. This is the
difficult side of quantum gravity, but also the source of its beauty.
The way this was first understood is enlightening. It all started with a mistake
by Lev Landau. Shortly after Heisenberg introduced his commutation relations
[q, p] = ih
(1.1)
and the ensuing uncertainty relations, the problem on the table was extending
quantum theory to the electromagnetic field. In a 1931 paper with Peierls [Landau
and Peierls (1931)], Landau suggested that once applied to the electromagnetic
field, the uncertainty relation would imply that no component of the field at a
given spacetime point could be measured with arbitrary precision. The intuition
was that an arbitrarily sharp spatiotemporal localization would be in contradiction
with the Heisenberg uncertainty relations.
Niels Bohr guessed immediately, and correctly, that Landau was wrong. To prove
him wrong, he embarked in a research program with Rosenfeld, which led to a
4
5
And we stand on their shoulders.

In the mathematics of Riemannian geometry one might distinguish the metric field from the manifold and identify spacetime with the second. But in the physics of general relativity this terminology
is misleading, because of the peculiar gauge invariance of the theory. If by spacetime we denote
the manifold, then, using Einstein words, the requirement of general covariance takes away from
space and time the last remnant of physical objectivity [Einstein (1916)]. A detailed discussion of
this point is given in Sec. 2.2 and Sec. 2.3 of [Rovelli (2004)].
The end of space and time
Figure 1.2
The last picture of Matvei Bronstein, the scientist who understood that quantum
gravity affects the nature of spacetime. Matvei was arrested on the night of August 6,
1937. He was thirty. He was executed in a Leningrad prison in February 1938.
classic paper [Bohr and Rosenfeld (1933)] proving that in the quantum theory of
the electromagnetic field the Heisenberg uncertainty relations do not prevent a single component of the field at a spacetime point from being measured with arbitrary precision.
But Landau being Landau, even his mistakes have bite. Landau, indeed, had a
younger friend, Matvei Petrovich Bronstein [Gorelik and Frenkel (1994)], a brilliant young Russian theoretical physicist. Bronstein repeated the Bohr-Rosenfeld
analysis using the gravitational field rather than the electromagnetic field. And
here, surprise, Landaus intuition turned out to be correct [Bronstein (1936b,a)]. If
we do not disregard general relativity, quantum theory does prevent the measurability of the field in an arbitrarily small region.
In August 1937, Matvei Bronstein was arrested in the context of Stalins Great
Purge, he was convicted in a brief trial and executed. His fault was to believe in
communism without stalinism.
Let us give a modern and simplified version of Bronsteins argument, because it
is not just the beginning, it is also the core of quantum gravity.
Say you want to measure some field value at a location x. For this, you have
to mark this location. Say you want to determine it with precision L. Say you do
this by having a particle at x. Since any particle is a quantum particle, there will
be uncertainties x and p associated to position and momentum of the particle.
To have localization determined with precision L, you want x < L, and since
Heisenberg uncertainty gives x > h /p, it follows that p > h /L. The mean
value of p2 is larger than (p)2 , therefore p2 > (h /L)2 . This is a well known
consequence of Heisenberg uncertainty: sharp location requires large momentum;
which is the reason why at CERN high momentum particles are used to investi-

gate small scales. In turn, large momentum implies large energy E. In the relativistic limit, where rest mass is negligible, E cp. Sharp localization requires large
energy.
Now lets add GR. In GR, any form of energy E acts as a gravitational mass
M E/c2 and distorts spacetime around itself. The distortion increases when
energy is concentrated, to the point that a black hole forms when a mass M is
concentrated in a sphere of radius R GM/c2 , where G is the Newton constant.
If we take L arbitrary small, to get a sharper localization, the concentrated energy
will grow to the point where R becomes larger than L. But in this case the region
of size L that we wanted to mark will be hidden beyond a black hole horizon, and
we loose localization. Therefore we can decrease L only up to a minimum value,
which clearly is reached when the horizon radius reaches L, that is when R = L.
Combining the relations above, we obtain that the minimal size where we can
localize a quantum particle without having it hidden by its own horizon, is
L=
EG
pG
h G
MG
= 4 = 3 = 3.
c2
c
Lc
c
(1.2)
Solving this for L, we find that it is not possible to localize anything with a precision better than the length
r
h G
1033 cm,
(1.3)
LPlanck =
c3
which is called the Planck length. Well above this length scale, we can treat spacetime as a smooth space. Below, it makes no sense to talk about distance. What
happens at this scale is that the quantum fluctuations of the gravitational field,
namely the metric, become wide, and spacetime can no longer be viewed as a
smooth manifold: anything smaller than LPlanck is hidden inside its own miniblack hole.
This simple derivation is obtained by extrapolating semiclassical physics. But
the conclusion is correct, and characterizes the physics of quantum spacetime.
In Bronsteins words: Without a deep revision of classical notions it seems
hardly possible to extend the quantum theory of gravity also to [the short-distance]
domain. [Bronstein (1936b)]. Bronsteins result forces us to take seriously the connection between gravity and geometry. It shows that the Bohr-Rosenfeld argument, according to which quantum fields can be defined in arbitrary small regions
of space, fails in the presence of gravity. Therefore we cannot treat the quantum
gravitational field simply as a quantum field in space. The smooth metric geometry of physical space, which is the ground needed to define a standard quantum
field, is itself affected by quantum theory. What we need is a genuine quantum
theory of geometry.
This implies that the conventional intuition provided by quantum field theory
fails for quantum gravity. The worldview where quantum fields are defined over
spacetime is the common world-picture in quantum field theory, but it needs to
abandoned for quantum gravity. We need a genuinely new way of doing physics,
Geometry quantized
Boundary spacetime
where space and time come after, and not before, the quantum states. Space and
time are semiclassical approximations to quantum configurations. The quantum
states are not quantum states on spacetime. They are quantum states of spacetime.
This is what loop quantum gravity provides.
Spacetime
Conventional quantum field theorist
Figure 1.3
Post-Maldacena string theorist
Genuine quantum-gravity physicist
Pre-general-relativistic physics is conceived on spacetime. The recent developments of

string theory, with bulk physics described in terms of a boundary theory, are a step
towards the same direction. Genuine full quantum gravity requires no spacetime at all.
1.3 Geometry quantized

The best guide we have towards quantum gravity is our current quantum theory
and our current gravity theory. We cannot be sure whether the basic physics on
which these theories are grounded still apply at the Planck scale, but the history
of physics teaches that vast extrapolation of good theories often works very well.
The Maxwell equations, discovered with experiments in a small lab, turn out to
be extremely good from nuclear to galactic scale, some 35 orders of magnitudes
away, more than our distance from the Planck scale. General relativity, found at
the Solar system scale, appears to work remarkably well at cosmological scales,
some 20 orders of magnitudes larger, and so on. In science, the best hypothesis,
until something new appears empirically, is that what we know extends.
The problem, therefore, is not to guess what happens at the Planck scale. The
problem is: is there a consistent theory that merges general relativity and quantum
theory? This is the form of thinking that has been extraordinarily productive in the
past. The physics of guessing, the physics of why not trying this? is a waste of
time. No great idea came from the blue sky in the past: good ideas come either
from experiments or from taking seriously the empirically successful theories. Let
us therefore take seriously geometry and the quantum and see, in the simplest
possible terms, what a quantum geometry implies.
General relativity teaches us that geometry is a manifestation of the gravita-
10
tional field. Geometry deals with quantities such as area, volume, length, angles...
These are quantities determined by the gravitational field. Quantum theory teaches
us that fields have quantum properties. The problem of quantum gravity is therefore to understand what are the quantum proprieties of geometrical quantities
such as area, volume, et cetera.
The quantum nature of a physical quantity is manifest in three forms:
i. in the possible discretization (or quantization) of the quantity itself;
ii. the short-scale fuzziness implied by the uncertainty relations;
iii. in the probabilistic nature of its evolution (given by the transition amplitudes).
We focus here on the first two of these (probabilistic evolution in a gravitational
context is discussed in the next chapter), and consider a simple example of how
they can come about, namely how space can become discrete and fuzzy. This example is elementary and is going to leave some points out, but it is illustrative and
it leads to the most characteristic aspect of loop quantum gravity: the existence of
quanta of space.
Lets start by reviewing basic quantum theory in three very elementary examples; then we describe an elementary geometrical object; and finally we see how
the combination of these two languages leads directly to the quanta of space.
Harmonic oscillator
Consider a mass m attached to a spring with
elastic constant k. We describe its motion in
terms of the position q, the velocity v and
m
the momentum p = mv. The energy E =
k
1
1 2
2
2 mv + 2 kq is a positive real number and
is conserved. The quantization postulate
from which the quantum theory follows is the existence of a Hilbert space H where
( p, q) are non-commuting (essentially) self-adjoint operators satisfying [Born and
Jordan (1925)]
//////
//////
//////
[q, p] = ih.
(1.4)
This is the new law of nature [Heisenberg (1925)] from which discretisation
can be computed. These commutation relations imply that the energy operator
p2
2m
+ 2k q2 has discrete spectrum with eigenvalues (E(n) = En (n) )

1
En = h n +
,
(1.5)
2
where = k/m. That is, energy is quantized: it comes in discrete quanta.

Since a free field is a collection of oscillators, one per mode, a quantum field is a
collection of discrete quanta [Einstein (1905a)]. The quanta of the electromagnetic
field are the photons. The quanta of Dirac fields are the particles that make up
ordinary matter. We are interested in the elementary quanta of gravity.
E( p, q) =
Geometry quantized
11
The magic circle: discreteness is kinematics

Consider a particle moving on a circle, subject to a potential V (). Let its position be
an angular variable S1 [0, 2 ] and its
p2
hamiltonian H = 2c + V (), where p = c d/dt is the momentum and c is a constant (with dimensions ML2 ). The quantum behaviour of the particle is described
by the Hilbert space L2 [S1 ] of the square integrable functions () on the circle
and the momentum operator is p = ih d/d. This operator has discrete spectrum,
with eigenvalues
pn = nh,
(1.6)
independently from the potential. We call kinematic the properties of a system that
depend only on its basic variables, such as its coordinate and momenta, and dynamic the properties that depend on the hamiltonian, or, in general, on the evolution. Then it is clear that, in general, discreteness is a kinematical property.6
The discreteness of p is a direct consequence of the fact that is in a compact
domain. (The same happens for a particle in a box.) Notice that [, p] 6= ih because the derivative of the function on the circle diverges at = 0 2: indeed, is a discontinuous function on S1 . Quantization must take into account
the global topology of phase space. One of the many ways to do so is to avoid
using a discontinuous function like and use instead a continuous function like
s = sin() or/and c = cos(). The three observables s, c, p have closed Poisson
brackets {s, c} = 0, { p, s} = c, { p, c} = s correctly represented by the commutators of the operator ihd/d, and the multiplication operators s = sin() and
c = cos(). The last two operators can be combined into the complex operator
h = ei . In this sense, the correct elementary operator of this system is not , but
rather h = ei . (We shall see that for the same reason the correct operator in quantum gravity is not the gravitational connection but rather its exponentiation along
loops. This is the first hint of theloops of LQG.)
Angular momentum
Let ~L = ( L1 , L2 , L3 ) be the angular momentum of a system that can rotate, with components { Li }, with i = 1, 2, 3.The total angular
momentum is L = |~L| = Li Li (summation
on repeated indices always understood unless
stated). Classical mechanics teaches us that ~L is
the generator (in the sense of Poisson brackets) of infinitesimal rotations. Postulating that the corresponding quantum operator is also the generator of rotations in
6
Not so for the discreteness of the energy, as in the previous example, which of course depends on
the form of the hamiltonian.
12

the Hilbert space, we have the quantization law [Born et al. (1926)]
[ Li , L j ] = ihij k Lk ,
(1.7)
where ij k is the totally antisymmetric (Levi-Civita) symbol. SU(2) representation

theory (reviewed in the Complements to this Chapter) then immediately gives the
eigenvalues of L, if the operators ~L satisfy the above commutation relations. These
are
q
1
3
L j = h j( j + 1),
j = 0, , 1, , 2, ...
(1.8)
2
2
That is, total angular momentum is quantized. Notice that the quantization of angular momentum is a purely kinematical prediction of quantum theory: it remains
the same irrespectively of the form of the Hamiltonian, and in particular irrespectively of whether or not angular momentum is conserved. Notice also that, as
for the magic circle, discreteness is a consequence of compact directions in phase
space: here the space of the orientations of the body.
This is all we need from quantum theory. Let us move on to geometry.
Geometry
Pick a simple geometrical object, an elementary portion of space. Say we pick a
small tetrahedron , not necessarily regular.
The geometry of a tetrahedron is characterized by the length of its sides, the area of its
faces, its volume, the dihedral angles at its
edges, the angles at the vertices of its faces, and
so on. These are all local functions of the gravitational field, because geometry is the same
thing as the gravitational field. These geometrical quantities are related to one another. A set
of independent quantities is provided for instance by the six lengths of the sides,
but these are not appropriate for studying quantization, because they are constrained by inequalities. The length of the three sides of a triangle, for instance,
cannot be chosen arbitrarily: they must satisfy the triangle inequalities. Non-trivial
inequalities between dynamical variables, like all global features of phase space,
are generally difficult to implement in quantum theory.
Instead, we choose the four vectors ~L a , a = 1, ..., 4 defined for each triangle a
as 12 of the (outward oriented) vector-product of two edges bounding the triangle.
See Figure 1.4. These four vectors have several nice properties. Elementary geometry shows that they can be equivalently defined in one of the two following ways:
Geometry quantized
13
Figure 1.4
The four vectors ~L a , normals to the faces.
The vectors ~L a are (outgoing) normals to the faces of the tetrahedron and their
norm is equal to the area of the face.
The matrix of the components Lia for a = 1, 2, 3 (notice that only 3 edges are
involved) is L T = 12 (det M ) M1 , where M is the matrix formed by the components of three edges of the tetrahedron that emanate from a common vertex.
Exercise: Show that the definitions above are all equivalent.
The vectors ~L a have the following properties:
They satisfy the closure relation

~ :=
C
~La = 0.
(1.9)
a =1
(Keep this in mind, because this equation will reappear all over the book.)
The quantities ~L a determine all other geometrical quantities such as areas, volume, angles between edges and dihedral angles between faces.
All these quantities, that is, the geometry of the tetrahedron, are invariant under
a common SO(3) rotation of the four ~L a . Therefore a tetrahedron is determined
by an equivalence class under rotations of a quadruplet of vectors ~L a s satisfying (1.9).
Check that the resulting number of degrees of freedom is correct.
The area A a of the face a is |~L a |.
The volume V is determined by the (properly oriented) triple product of any
three faces:
V2 =
2 ~
2
2
j
( L ~L2 ) ~L3 = eijk L1i L2 L3k = det L.
9 1
9
9
(1.10)
(Defined in this manner, a negative value of V 2 is simply the indication of a

change of the order of the triple product, that is, an inversion in the orientation.)
14

Exercise: Prove these relations. Hint: choose a tetrahedron determined by a triple
of orthonormal edges, and then argue that the result is general because the formula
is invariant under linear transformations. Derive the 29 factor.
(In Chapter 3 we describe the gravitational field in terms of triads and tetrads.
Let us anticipate here the relation between the vectors ~L a and the triad. If the tetrahedron is small compared to the scale of the local curvature, so that the metric can
be assumed to be locally then
Lia =
1 i
e
2 jk
e j ek .
(1.11)
Equivalently, ~L a can be identified with the flux of the densitized inverse triad field
j
Eia = 21 e abc eijk eb eck , which is Ashtekars electric field, across the face. Since the triad
is the gravitational field, this gives the explicit relation between ~L a and the gravitational field. Here the triad is defined in the 3d hyperplane determined by the
tetrahedron.)
Quantization of the geometry

We have all the ingredients for jumping to quantum gravity. The geometry of a real
physical tetrahedron is determined by the gravitational field, which is a quantum
field. Therefore the normals ~L a are to be described by quantum operators, if we
take the quantum nature of gravity into account. These will obey commutation
relations. The commutation relation can be obtained from the hamiltonian analysis
of GR, by promoting Poisson brackets to operators, in the same manner in which
(1.1) and (1.7) can; but ultimately they are quantization postulates, like (1.1) and
(1.7). Let us therefore just postulate them here. The simplest possibility [Barbieri
(1998)] is to mimic (1.7), namely to write
j
[ Lia , Lb ] = iab lo2 ij k Lka ,
(1.12)
where lo2 is a constant proportional h and with the dimension of an area. These
commutation relations are again realizations of the algebra of SU(2), like in the
case of the rotator, reflecting again the rotational symmetry in the description of
~ defined in (1.9) is precisely
the tetrahedron. This is good: for instance we see that C
the generator of common rotations and therefore the closure condition (1.9) is an
immediate condition of rotational invariance, which is what we want: the geometry is determined by the ~L a up to rotations, which here are gauge. Let us thus fix
(1.12) as the quantization postulate.
The constant lo must be related to the Planck scale LPlanck , which is the only
dimensional constant in quantum gravity. Leaving the exact relation open for the
moment, we pose
lo2 = 8 L2Planck =
h (8G )
,
c3
(1.13)
15
Figure 1.5
Geometry quantized
An image from De triplici minimo et mensura (1591) by Giordano Bruno, showing

discrete elements and referring to Democrituss atomism. Brunos images played an
important role as sources of the modern intuition about the atomic structure of
matter.
where 8 is just there for historical reasons (the coupling constant on the right of
Einsteins equations is not G but 8G) and is a dimensionless parameter (presumably of the order of unity) that fixes the precise scale of the theory. Let us study
the consequences of this quantization law.
Quanta of Area
One consequence of the commutation relations (1.12) is immediate [Rovelli and
Smolin (1995a); Ashtekar and Lewandowski (1997)]: the quantity A a = |~L a | behaves as total angular momentum. As this quantity is the area, it follows immediately that the area of the triangles bounding any tetrahedron is quantized with
eigenvalues
q
1
3
j = 0, , 1, , 2, ...
(1.14)
A = lo2 j( j + 1),
2
2
This is the gist of loop quantum gravity. As we shall see, the result extends to any
surface, not just the area of the triangles bounding a tetrahedron.
Quanta of Volume
Say that the quantum geometry is in a state with area eigenvalues j1 , ..., j4 . The four
vector operators ~L a act on the tensor product H of four representations of SU(2),
with respective spins j1 , ..., j4 . That is, the Hilbert space of the quantum states of
the geometry of the tetrahedron at fixed values of the area of its faces is
H = H j1 H j2 H j3 H j4 .
(1.15)
We have to take also into account the closure equation (1.9), which is a condition
~ is nothing else than
the states must satisfy, if they are to describe a tetrahedron. C
16

the generator of the global (diagonal) action of SU(2) on the four representation
~ = 0) are the states
spaces. The states that solve equation (1.9) (strongly, namely C
that are invariant under this action, namely the states in the subspace
K = InvSU (2) [H j1 H j2 H j3 H j4 ] .
(1.16)
Thus we find that, as anticipated, the physical states are invariant under (common) rotations, because the geometry is defined only by the equivalence classes
of ~L a under rotations. (This connection between the closure equation (1.9) and invariance under rotations is nice and encouraging; it confirms that our quantisation
postulate is reasonable. When we will come back to this in the context of full general relativity, we will see that this gauge is nothing else than the local-rotation
gauge invariance of the tetrad formulation of general relativity.)
Consider now the volume operator V defined by (1.10). This is well defined in
~ namely it is rotationally invariant. Therefore we
K because it commutes with C,
have a well-posed eigenvalue problem for the self-adjoint volume operator on the
Hilbert space K. As this space is finite dimensional, it follows that its eigenvalues are discrete [Rovelli and Smolin (1995a); Ashtekar and Lewandowski (1998)].
Therefore we have the result that the volume has discrete eigenvalues as well. In
other words, there are quanta of volume or quanta of space: the volume of our
tetrahedron can grow only in discrete steps, precisely as the amplitude of a mode
of the electromagnetic field. In the Complement 1.7.3 to this chapter we compute
some eigenvalues of the volume explicitly.
It is important not to confuse this discretization of geometry, namely the fact
that Area and Volume are quantized, with the discretization of space implied by
focusing on a single tetrahedron. The first is the analog of the fact that the Energy
of a mode of the electromagnetic field comes in discrete quanta. It is a quantum
phenomenon. The second is the analog of the fact that it is convenient to decompose a field into discrete modes and study one mode at the time: it is a convenient
isolation of degrees of freedom, completely independent from quantum theory.
Geometry is not discrete because we focused on a tetrahedron: geometry is discrete because Area and Volume of any tetrahedron (in fact, any polyhedron, as we
shall see) take only quantized values. The quantum discretization of geometry is
determined by the spectral properties of Area and Volume.
The astute reader may wonder whether the fact that we have started with a fixed
chunk of space plays a role in the argument. Had we chosen a smaller tetrahedron
to start with, would had we obtained smaller geometric quanta? The answer is
no, and the reason is at the core of the physics of general relativity: there is no notion of size (length, area, volume) independent from the one provided by the gravitational
field itself. The coordinates used in general relativity carry no metrical meaning.
In fact, they carry no physical meaning at all. If we repeat the above calculation
starting from a smaller tetrahedron in coordinate space, we are not dealing with
a physically smaller tetrahedron, only with a different choice of coordinates. This
is apparent in the fact that the coordinates play no role in the derivation. Whatever coordinate tetrahedron we may wish to draw, however small, its physical size
17
Geometry quantized
will be determined by the gravitational field on it, and this is quantized, so that its
physical size will be quantized with the same eigenvalues. Digesting this point is
the first step to understanding quantum gravity. There is no way to cut a minimal
tetrahedron in half, just as there is no way to split the minimal angular momentum
in quantum mechanics. Space itself has a granular structure formed by individual quanta.
The shape of the quanta of space and the fuzziness of the

geometry.
As we shall prove later on, the four areas A a of the four faces and the volume V
form a maximally commuting set of operators in the sense of Dirac. Therefore they
can be diagonalized together and quantum states of the geometry of the tetrahedron are uniquely characterized by their eigenvalues | ja , vi.
Is the shape of such a quantum
state truly a tetrahedron?
The answer is no, for the following reason. The geometry of
a classical tetrahedron is determined by six numbers, for instance, the six lengths of its edges.
(Equivalently, the 4 3 quantities
Lia constrained by the 3 closure
equations, up to 3 rotations.) But
the corresponding quantum numbers that determine the quantum states of the tetrahedron are not six; they are only
five: four areas and one volume.
The situation is exactly analogous to angular momentum, where the classical
system is determined by three numbers, the three components of the angular momentum, but only two quantities (say L2 , Lz ) form a complete set. Because of this
fact, as is well known, a quantum rotator has never a definite angular momentum
~L, and we cannot really think of an electron as a small rotating stone: if L x is sharp,
necessarily Ly is fuzzy, is quantum spread.
For the very same reason, therefore, geometry can never be sharp in the quantum theory, in the same sense in which the three components of angular momentum can never be all sharp. In any real quantum state there will be residual quantum fuzziness of the geometry: it is not possible to have all dihedral angles, all
areas and all lengths sharply determined. Geometry is fuzzy at the Planck scale.
We have found two characteristic features of quantum geometry:
Areas and volumes have discrete eigenvalues.

Geometry is spread quantum mechanically at the Planck scale.
Quantum geometry differs from Riemannian geometry on both these grounds.
18
1.4 Physical consequences of the existence of the

Planck scale
1.4.1 Discreteness: scaling is finite
The existence of the Planck length sets quantum GR aside from standard quantum field theory for two reasons. First, we cannot expect quantum gravity to be
described by a local quantum field theory, in the strict sense of this term [Haag
(1996)]. Local quantum theory requires quantum fields to be described by observables at arbitrarily small regions in a continuous manifold. This is not going to
happen in quantum gravity, because physical regions cannot be arbitrarily small.
Second, the quantum field theories of the standard model are defined in terms
of an infinite renormalization group. The existence of the Planck length indicates
that this is not going to be the case for quantum gravity. Lets see this in more
detail.
When computing transition amplitudes for a field theory using perturbation
methods, infinite quantities appear, due to the effect of modes of the field at arbitrary small wavelengths. Infinities can be removed introducing a cut-off and adjusting the definition of the theory to make it cut-off dependent in such a manner
that physical observables are cut-off independent and match experimental observations. The cut-off can be regarded as a technical trick; alternatively, the quantum field theory can be viewed as effective, useful at energy scales much lower
than the (unknown) natural scale of the physics. Accordingly, care must be taken
for the final amplitudes not to depend on the cut-off. Condensed matter offers a
prototypical example of independence from short-scale: second-order phase transitions. At the critical point of a second-order phase transition, the behavior of the
system becomes scale independent, and large-scale physics is largely independent
from the microscopic dynamics. Conventional quantum field theories are modeled
on condensed matter phase transitions: they are defined using a cut-off, but this
is then taken to infinity, and the theory is defined in such a manner that the final
result remains finite and independent from the details of the cut-off chosen.
This framework has proven effective for describing particle physics, but it is not
likely to work for quantum gravity. The alternative is also indicated by condensed
matter physics: consider a generic matter system not at a critical point. Say, a bar
of iron at room temperature. Its behavior at macroscopic scales is described by
a low-energy theory, characterized by certain physical constants. This behavior
includes wave propagation and finite correlation functions. The high frequency
modes of the bar have an effect on the value of the macroscopic physics, and can
be explored using renormalization group equation describing the dependence of
physical parameters on the scale. But the system is characterized by a physical
and finite cut-off scale: the atomic scale; and there are no modes of the bar beyond
this scale. The bar can be described as a system with a large but finite number of
Physical consequences of the existence of the Planck scale
19
degrees of freedom. The short-distance cut off in the modes is not a mathematical
trick for removing infinities, nor a way for hiding unknown physics: it a genuine
physical feature of the system. Quantum gravity is similar: the Planck scale cut-off
is a genuine physical feature of the system formed by quantum spacetime.7
The existence of a minimal length scale gives quantum gravity universal character, analogous to special relativity and quantum mechanics: Special relativity can
be seen as the discovery of the existence of a maximal local physical velocity, the
speed of light c. Quantum mechanics can be interpreted as the discovery of a minimal action, h , in all physical interactions, or, equivalently, the fact that a compact
region of phase space contains only a finite number of distinguishable (orthogonal) quantum states, and therefore there is a minimal amount of information8 in
the state of a system. Quantum gravity yields the discovery that there is a minimal
length lo at the Planck scale. This leads to a fundamental finiteness and discreteness of the world.
Natural physical units are obtained by measuring speed as the ratio to the maximal speed c, action in multiples of the minimal action h , and lengths in multiples
of the minimal length lo . In these natural units c = h = lo = 1. To avoid confusion with , we shall not use these units in the first part of the book, and rather
use the more conventional Planck units c = h = 8G = 1.
1.4.2 Fuzziness: disappearance of classical space and time

The absence of the conventional notions of space and time at small scales forces
us to rethink the basis of physics. For instance, Hamiltonian mechanics is about
evolution in time, so is the conventional formulation of quantum mechanics, and
so is QFT, where time evolution is captured by the unitary representations of the
Poincare group. All this must change in quantum gravity. Understanding quantum spacetime requires therefore a substantial conceptual revolution. The physics
of quantum gravity is not the physics of the gravitational field in spacetime. It is
the physics of the quantum fields that build up spacetime. The basic ontology of
physics, which has evolved during the last century, simplifies.
According to Descartes, who in this was essentially still following Aristotle, matter, moving in time, was the only component of the physical universe and extension was just a property of matter (res extensa). Newton introduced a description of the world in terms of particles located in space and moving in time. Faraday
and Maxwell showed that this ontology needed to be supplemented by a new entity: the field. Special relativity showed that space and time must be thought of as
aspects of a single entity, spacetime. General relativity showed that spacetime is
itself a field: the gravitational field. Finally, quantum theory showed that particles
7
8
An early quote by Einstein comes to mind: Is not a mathematical trick; it is the way of the atomic
world. Get used to it. Quoted in [Stone (2013)].
Information is used here in the sense of Shannon: number of distinguishable alternatives [Shannon (1948)]. It has no relation to semantics, meaning, significance, consciousness, records, storage,
or mental, cognitive, idealistic or subjectivistic ideas.
20
are quanta of quantum fields. This is summarized in the Table 1.1. Bringing all
these results together implies that, as far as we know today, all that exist in nature
is general-covariant quantum fields.
1.5 Graphs, loops and quantum Faraday lines

Above we have described the quantum geometry of a single grain of physical
space. A region of possibly curved physical space can be described by a set of
interconnected grains of space. These can be represented by a graph, where each
node is a grain of space and the links relate adjacent grains. We shall see below
that this picture emerges naturally from the quantization of the gravitational field.
The quantum states of the theory will have a natural graph structure of this kind.
The Loops in Loop Quantum Gravity refers to the loops formed by closed
sequences of links in such a graph.9 The individual lines in the graph can be
viewed as discrete Faraday lines of the gravitational field. The Faraday lines, or
lines of force, introduced by Michael Faraday, form the initial intuition at the
root of the modern notion of field. LQG grew on the intuition that the quantum
discreteness makes these lines discrete in the quantum theory. This idea was anticipated by Dirac:
9
Historically, the first states constructed had no nodes [Rovelli and Smolin (1988, 1990)].
Table 1.1 The ontology of the world in contemporary physics

Descartes:
Newton:
Matter
Particles
Faraday-Maxwell:
Particles
Special relativity:
Particles
Quantum theory:
General relativity:
Quantum gravity:
Space
Fields
Fields
Quantum-Fields
Particles
Time
Time
Space
Time
Spacetime
Spacetime
General-covariant fields
General-covariant quantum fields
21
Figure 1.6
Graphs, loops and quantum Faraday lines
A set of adjacent quantum polyhedra and the graph they determine
We can assume that when we go to the quantum theory, the lines of force become all discrete and
separate from one another. [...] We have so a model where the basic physical entity is a line of force...
P.A.M. Dirac [Dirac (1956)].
Since the gravitational field is spacetime, its discrete quantum lines of force are
not in space, but rather form the texture of space themselves. This is the physical
intuition of LQG.
Here is a more complete and condensed account of what happens in the theory
(if you find the rest of this subsection incomprehensible, do not worry, just skip
it; it will become clear after studying the book): The existence of fermions shows
that the metric field is not sufficient to describe the gravitational field. Tetrads, recalled in Chapter 3, are needed. This introduces a local Lorentz gauge invariance,
related to the freedom of choosing an independent Lorentz frame at each point of
spacetime. This local gauge invariance implies the existence of a connection field
( x ), which governs the parallel transport between distinct
spacetime points. The
R
path ordered exponential of the connection Ue = Pe e along any curve e in the
manifold, called the holonomy in the jargon of the theory, is a group element
and contains the same information as ( x ). (These quantities are defined in Chapter 3.) A priori, one may take either quantity, ( x ) or Ue , as the variable for the
quantum theory. The first is in the Lie algebra, the second in the Lie group; the
first can be derived from the second by taking the limit of arbitrarily short e, as
Ue = 1l + (e) + O(|e|2 ). However, the Planck scale discreteness that we expect in
quantum gravity breaks the relation between the two. If space is discrete, there is
no meaning in infinitesimal shifts in space, and therefore Ue remains well defined,
while its derivative ( x ) is not. Therefore we are led to forget ( x ) and seek for
a quantization using the group variables Ue instead, as mentioned above for the
quantisation of a variable on a circle. These are called loop variables in the jargon of the theory. The corresponding quantum operators are akin the Wilson loop
22
operators in QCD, which are also exponentials of the connection.10 The fact that
the rotation group is compact is the origin of the discreteness, precisely as in the
case of the particle on a circle discussed above.
To see this more clearly, the analogy with lattice QCD (which we review in Chapter 4) is enlightening. In lattice QCD, one takes a lattice in spacetime with lattice
spacing (the length of the links) a and describes the field in terms of group elements Ue associated to the links of the lattice. The physical theory is recovered
in the limit where a goes to zero, and in this limit the group elements associated
to individual es become all close to the identity. The limit defines the Yang-Mills
connection. In gravity, we can equally start from a discretization with group elements Ue associated to the links. But the length of these links is not an external
parameter to be taken to zero: it is determined by the field itself, because geometry is determined by gravity. Since quantization renders geometry discrete, the
theory does not have a limit where the es become infinitesimal. Therefore there is
no connection ( x ) defined in the quantum theory. A connection is defined only
in the classical limit, where we look at the theory at scales much larger than the
Planck scale, and therefore we can formally take the length of the es to zero.
1.6 The landscape

At present we have two theories that incorporate the ultraviolet finiteness following from the existence of the Planck scale into their foundation, are well developed,
and give well-defined definitions for the transition amplitudes in quantum gravity: these are loop quantum gravity and string theory. String theory has evolved
from attempts to quantize gravity by splitting the gravitational field into a background fixed metric field, used to fix the causal structure of spacetime, and a quantum fluctuating gravitational field h ( x ). The non-renormalizability of the resulting theory has pushed the theorists into a quest for a larger renormalizable or
finite theory, following the path indicated by the weak interaction. The quest has
wandered through modifications of GR with curvature-square terms in the action,
Kaluza-Klein-like theories, supergravity... Merging with the search for a unified
theory of all interactions, it has eventually led to string theory, a presumably finite
quantum theory of all interactions including gravity, defined in 10 dimensions,
including supersymmetry, so far difficult to reconcile with the observed world.
The canonical version of LQG was born from the discovery of loop solutions
of the Wheeler-deWitt equation, namely the formal quantization of canonical GR,
rewritten in the Ashtekar variables. The quantization of geometry was derived
within this theory and led to the spin network description of quantum geometry
(Chapters 4 and 5). The canonical theory branched into a sum over geometries
form, a` la Feynman, inspired by the functional integral Euclidean formulation
10
The name loop is proper only when e is a closed curve, or a loop, in which case the trace of Ue is
gauge invariant.
The landscape
23
Table 1.2 Main lines of development of quantum gravity

Quantum field theory perspective
General relativity perspective

The full metric is quantized
gravitational field is split
g (x) = + h (x)
and h (x) is quantized.
wave function
of the metric
[q]
The theory is non-renormalizable
R2 , Supergravity, Kaluza-Klein ... 70
Wheeler-deWitt 60
path integral
Z over geometries
R p
gR[g]
D[g] ei
Hartle-Hawking 70
Ashtekar variables 80
Loop quantization
Spinfoams 92
Unification
of interactions
geometry quantization
(Spin networks) 94
String Theory
Covariant loop theory 08
developed by Hawking and his group in the 70. This spinfoam theory (Chapter 7) merged with the canonical LQG kinematics and evolved into the current
covariant theory described in this book. The historical development of these theories is sketched in Table 1.6. For a historical reconstruction and references, see the
appendix in [Rovelli (2004)].
The rest of the book describes the covariant formulation of loop quantum gravity.
Figure 1.7
The four characters of the discussion from which quantum gravity has emerged. From
the left: Landau, Bohr, Rosenfeld and Bronstein. The photo was taken in Kharkov and
published in the newspaper Kharkovskii rabochii (The Kharkov Worker) on May 20,
1934.
24
1.7 Complements
We recall some basic SU(2) representation theory. This plays an important role in quantum
gravity. Then we compute the eigenvalues of the volume for a minimal quantum of space.
1.7.1 SU(2) representations and spinors

Definition.
SU(2) is the group of unitary 2 2 complex matrices U. They satisfy U 1 = U and det U =
1. These conditions fix the form of the matrix U as:

a b
U=
.
(1.17)
b
a
where (because of the unit determinant)
| a|2 + |b|2 = 1.
(1.18)
We write the matrix elements as U AB , with indices A, B, C, . . . taking the values 0, 1. SU(2)
has the same algebra as the rotation group SO(3) (The group SU(2) is its universal cover of
SO(3)). One can view SU(2) as a minimal building block from many points of view, including, as we shall see, quantum spacetime.
Measure.
Equation (1.18) defines a sphere of unit radius in C2 R4 . Thus, the topology of the group
is that of the three-sphere S3 . The Euclidean metric of R4 restricted to this sphere defines an
invariant measure on the group. Normalized by
Z
dU = 1 ,
(1.19)
SU(2)
this is the Haar measure, invariant under both right and left multiplication: dU = d(UV ) =
d(VU ), V SU(2). The Hilbert space L2 [SU (2)] formed by the functions (U ) on the
group square-integrable in this measure plays an important role in what follows.
Spinors.
The space of the fundamental representation of SU(2) is the space of spinors, i.e., complex
vectors z with two components,
0
z
z=
C2 .
(1.20)
z1
We shall commonly use the abstract index notation which is implicitly used by physicists
and is explicit for instance in Walds book [Wald (1984)]; that is, the notation with an index
does not indicate a component, but rather the full vector, so z A is synonymous of z for
spinors and vi is synonymous of ~v for vectors.
In Chapter 8 we study the geometrical properties of spinors in better detail, and we reinterpret spinor as spacetime objects. Here we only introduce some basic facts about them and
their role in SU(2) representation theory.
Complements
25
Representations and spin.

The vector space of the completely symmetric n-indices spinors
z A1 ...An = z( A1 ...An )
(1.21)
A10
A1
A0n
An
transforms under the action of SU(2), z A1 ...An = U . . . U z A1 ...An , and therefore defines a representation of the group. This space is denoted H j , where j = n/2 (so that
j = 0, 21 , 1, 23 , ...) and the representation it defines is called the spin-j representation. This
representation is irreducible.
0
Exercise: Show that H j has dimension 2j + 1.

Let us review some properties of these representations.
1. Consider the two antisymmetric tensors

0
1
e AB =
,
1 0
e AB =
0
1
1
0
(1.22)
These can be used for raising or lowering indices of spinors, in a way analogous to g ab
and gab for tensors but being careful about the order: using the down-left-up-right rule,
or A / A rule:
z A = e AB z B ,
z A = z B e BA ,
(1.23)
For example, we have a contraction z AA = e AB z AB = e AB z BA = z AA . Show that

1
2
3
4
B,e
AB = 2 , e
AB = 2 .
e AC eCB = A
BA e
AB e
AB
e is invariant under the action of SU(2), i.e., U AC U BD eCD = e AB .
BD U A U C = 1 .
1 = 1e
det U = e BD U 0B UD
B D
2 AC e
1
U = eUe ; that is, (U ) AB = eBD U DC eCA .
2. There are two SU(2) invariant quadratic forms defined on C2 , which should not be confused. The first is the (sesquilinear) scalar product
h z | y i = z A y A = z0 y0 + z1 y1 .
(1.24)
where the bar indicates the complex conjugate. This scalar product is what promotes C2
to a Hilbert space, and therefore is what makes the SU(2) representations unitary.
The second is the (bilinear) antisymmetric quadratic form11
(z, y) = e AB z A y B = z0 y1 z1 y0 .
The two can be related by defining the antilinear map J : C C

!
z1
A
( Jz) =
z0
(1.25)
(1.26)
so that
hz|yi = ( Jz, y).
(1.27)
All these structures are SU(2) invariant, but, as we shall see later on, ( , ) is also SL(2,C)
invariant, whereas J and h | i are not. Therefore spinors are spacetime objects, since
they carry also a representation of the Lorentz group, but as a representation of the
11
This is also indicated as [z|yi, a notation that emphasises its antisymmetry
26
Lorentz group C2 is not a unitary representation. In a precise sense, the scalar product and J depend on a choice of Lorentz frame in spacetime.
3. Most of SU(2) representation theory follows directly from the invariance of e AB . ConN
sider first a tensor product of two fundamental (j = 1/2) representations (z y) AB =
A
B
AB
z y . Show that any two-index spinor z
can be decomposed into its symmetric and
antisymmetric part
z AB = z0 e AB + z1AB ,
z0 =
1 A
z ,
2 A
z1AB = z( AB) ,
(1.28)
which are invariant under the action of SU(2). Because of the invariance of e AB , this decomposition is SU(2) invariant, scalars z0 define the trivial representation j = 0, whereas
z1AB defines the adjoint representation j = 1. Hence you have proven that the tensor
product
spin-1/2 representations is the sum of spin-0 and spin-1 representation:
N of two L
1/2 1/2 = 0 1.
4. In general, if we tensor two representations of spin j1 and j2 , we obtain a space of spinors

with (2j1 + 2j2 ) indices, symmetric in the first 2j1 and the last 2j2 indices. By symmetrizing all the indices, we obtain an invariant subspace transforming in ( j1 + j2 ) representation. Alternatively, we can contract k indices of the first group with k indices of the second group using k times the tensor e AB , and then symmetrize the remaining 2( j1 + j2 k)
indices to obtain spin-j3 representation. Show that
j1 + j2 + j3 N ,
| j1 j2 | j3 ( j1 + j2 ) .
(1.29)
These two conditions are called ClebschGordon conditions. Does this ring a bell? They are
equivalent to the fact that there exist three non-negative integers a, b, c such that
2j1 = b + c ,
2j2 = c + a ,
2j3 = a + b .
(1.30)
Exercise: Show that (1.30) implies (1.29).

This has a nice graphical interpretation, see Figure 1.8.
j2 =
c+a
2
j3 =
a+b
2
a=3
c=1
j1 =
Figure 1.8
b=2
b+c
2
Elementary recoupling.
Draw similar pictures for ( j1 , j2 , j3 ) given by i: (1/2, 1/2, 1), ii: (5/2, 5/2, 2), iii: (1, 3/2, 5/2),
iv: (5/2, 2, 5), v: (5/2, 5/2, 7/2), vi: (1, 1, 1). Find the corresponding a, b, c.
Complements
27
5. The spinor basis is not always the most convenient for the SU(2) representations. If we
diagonalize Lz in H j we obtain the well known basis | j, mi where m = j, ..., j, described in all quantum mechanics textbooks. In this basis, the representation matrices
j
Dnm (U ) are called the Wigner matrices. Mathematica gives them explicitly; they are called
WignerD[{j,m,n},, , ] and are given in terms of the Euler angles parametrization of
SU(2) (given below in (1.42)).
6. The two bilinear forms of the fundamental representation extend to all irreducible representations. Given two vectors z and y in H j , we can either take their invariant contraction
or their scalar product. In integer representations the two bilinear forms turn out to be
the same. In half-integer representations, they are different. The contraction is defined
by the projection H j H j H0 and in the z A1 ...A2j representation, it is given by
(z, y) = z A1 ...A2j y B1 ...B2j e A1 B1 ...e A2j B2j .
(1.31)
while the scalar product is given by
h z | y i = z A1 ...A2j y B1 ...B2j A1 B1 ...A2j B2j .
(1.32)
The basis that diagonalizes Lz is of course orthonormal, since Lz is self-adjoint. Therefore
h j, m | j, n i = mn ,
(1.33)
( j, m, j, n) = (1) jm m,n .
(1.34)
while a direct calculation (see for instance Landau-Lifshitzs [Landau and Lifshitz (1959)])
gives
The factor m,n is easy to understand: the singlet must have vanishing total Lz .
1.7.2 Pauli matrices

The Pauli matrices i with i = 1, 2, 3 are

0
0
1
A
,
i B =
i
1
0
i
0

1
,
0
0
1

(1.35)
Any SU(2) group element U = (U AB ) can be written in the form

U = ein
(1.36)
for some [0, 2 ] and unit vector n.
Exercise: Show that the Pauli matrices obey

i j = ij + ieijk k
(1.37)
ei e = iT = i ,
(1.38)
and
where
indicate the transpose and the complex conjugate.
Exercise: Show that

ein = cos + i n sin .
(1.39)
28
The three matrices

i
i = i
2
(1.40)
are the generators of SU(2) in its fundamental representation.

Exercise: Verify that they satisfy the relation
[i , j ] = eij k k ,
(1.41)
which defines the algebra su(2).

The Euler angle parametrisation of SU(2) is defined by
U (, , ) = e3 e2 e3 ,
(1.42)
where
[0, 2 [,
[0, [,
[0, 4 [.
(1.43)
In terms of these coordinates, the Haar measure reads

Z
dU =
1
16 2
Z 2
0
Z
0
sin d
Z 4
0
d.
(1.44)
Exercise: Verify the relations

1
1
Tr(i j ) = ij , Tr(i j k ) = eijk
2
4
1 A C
ij A C
AC
i B j D = (D B e eBD ) ,
4
BA CD = CA BD + e AD eBC
i
1h
ij Tr( Ai )Tr( Bj ) = Tr( AB) Tr( AB1 ) ,
4
Tr( A)Tr( B) = Tr( AB) + Tr( AB1 ) ,
(1.45)
(1.46)
(1.47)
(1.48)
(1.49)
for A and B SL(2, C) matrices.

The relation (1.47) is of particular importance. Notice that it can be written
graphically in the form
(1.50)
with A = 1. The reason for this writing will be more clear in Chapter 6.
Exercise: If we raise the index of the Pauli matrices with e we obtain the 2-index
spinors (i ) AB = (i ) A C eCB . Show that these are invariant tensors in the representation 1/2 1/2 1.
Complements
29
1.7.3 Eigenvalues of the volume

Problem
Equipped with this SU(2) math, compute the volume eigenvalues for a quantum
of space whose sides have minimal (non vanishing) area.
Solution
Recall (Equation (1.10)) that the volume operator V is determined by
V2 =
2
j
e Li L L k .
9 ijk 1 2 3
(1.51)
where the operators ~L a satisfy the commutation relations (1.12).

If the face of the quantum of space have minimal area, the Casimir of the corresponding representations have minimal non-vanishing value. Therefore the four
operators L a act on the fundamental representations j1 = j2 = j3 = j4 = 12 . Therefore they are proportional to the (self-adjoint) generators of SU(2), which in the
fundamental representation are Pauli matrices. That is
Lif =
i
,
2
(1.52)
The proportionality constant has the dimension of length square, is of Planck scale
and is fixed by comparing the commutation relations of the Pauli matrices with
hG
(1.12). This gives = lo2 = 8
.
c3
The Hilbert space on which these operators act is therefore H = H 1 H 1
2
2
H 1 H 1 . This is the space of objects with 4 spinor indices A, B = 0, 1, each being
2
in the 12 -representation of SU(2).
H j1 H j2 H j3 H j4 3 z ABCD
(1.53)
The operator ~L a acts on the a-th index. Therefore the volume operator acts as
0 0 0
2 3 ijk A
(V 2 z) ABCD =
e i A0 jB B0 kC C0 z A B C D .
(1.54)
9 2
Let us now implement the closure condition (1.9). Let
(1.55)
H1 1 1 1 = H1 H1 H1 H1
2 2 2 2
K 1 1 1 1 = InvSU (2) [H 1 H 1 H 1 H 1 ]
2 2 2 2
(1.56)
We have to look only for subspaces that are invariant under a common rotation
for each space H ji , namely we should look for a quantity with four spinor indices
that are invariant under rotations. What is the dimension of this space? Remember
that for SU(2) representations 21 12 = 0 1, that implies:
H 1 1 1 1 = (0 1) (0 1) = 0 1 1 (0 1 2) .
2 2 2 2
(1.57)
30
Since the trivial representation appears twice, the dimension of K 1 1 1 1 is two.

2 22 2
Therefore there must be two independent invariant tensors with four indices. These
are easy to guess, because the only invariant objects available are AB and iAB =
(i e) AB , obtained raising the indices of the Pauli matrices i A B .
i AB = i A C CB
(1.58)
Therefore two states that span K 1 1 1 1 are

22 2 2
z1ABCD = AB CD
(1.59)
(1.60)
z2ABCD
iAB iCD .
These form a (non orthogonal) basis in K 1 1 1 1 . These two states span the physical
22 2 2
SU(2)-invariant part of the Hilbert space, that gives all the shapes of our quantum
of space with a given area. To find the eigenvalues of the volume it suffices to
2
diagonalize the 2x2 matrix Vnm
2
V 2 zn = Vnm
zm .
(1.61)
Let us compute this matrix. An straightforward calculation with Pauli matrices

(do it! Eqs (1.37) and (1.38) are useful) gives
V 2 z1 =
i3
z2 ,
18
so that
V2 =
i3
18
V 2 z2 =
i3
z .
6 1
(1.62)
0
3
(1.63)
1
0
and the diagonalization gives the eigenvalues

6
V2 = .
6 3
(1.64)
The sign depends on the fact that this is the oriented volume square, which depends
on the relative orientation of the triad of normal chosen. Inserting the value =
8h G
determined above, in the last equation, we have finally the eigenvalue of
c3
the (non oriented) volume

3
1
8h G 2
V= p
.
(1.65)
c3
6 3
About 10100 quanta of volume of this size fit into a cm3 . In Chapter 7 we give a
general algorithm for computing Volume eigenvalues.
Physics without time

General relativity has modified the way we think of space and time and the way
we describe evolution theoretically. This change requires adapted tools. In this
chapter, we study these tools. They are the Hamilton function for the classical theory (not to be confused with the hamiltonian) and the transition amplitude for the
quantum theory. Unlike most of the other tools of mechanics (hamiltonian, quan
tum states at fixed time, Schrodinger
equation...) these quantities remain meaningful in quantum gravity. Let us start by introducing them in a familiar context.
2.1 Hamilton function
I feel that there will always be something missing from other methods
which we can only get by working from a hamiltonian (or maybe from
some generalization of the concept of hamiltonian).
Paul Dirac [Dirac (2001)]
Lets have a system described by a configuration variables q C where C is
the configuration space (of arbitrary dimension). We describe the evolution of this
system in the time variable t R. This means that evolution is going to be given
by q(t), namely functions R C describing the possible motions. These quantities can be interpreted operationally as follows. We have two kinds of measuring
apparatus at our disposal: a clock where we read t, and other devices for the variables q.
The physical motions q(t) that the system can follow are determined by a Lagrangian L(q, q ), where q = dq/dt, as the ones that minimize the action
S[q] =
dt L (q(t), q (t)) .
(2.1)
The action is a functional on the functions q(t). This is mechanics.

Now let us define a key object, which plays a major role in what follows. The
Hamilton function S(q, t, q0 , t0 ) (not to be confused with the hamiltonian) is a
function of four variables, q, t, q0 , t0 , defined as the value of the action on a physical
trajectory (namely a solution of the equations of motion) that starts at q at time t
and ends at q0 at time t0 .1 Do not confuse the Hamilton function with the action: the
1
31
If such trajectory is unique, the Hamilton function is a proper function. If there are many, it is mul-
32
first is a function of four variables, the second is a functional of the full trajectory,
giving a number for any function q(t). The two objects are related: the Hamilton
function is the action of a particular trajectory determined by its boundaries.
Let us denote qqt,q0 t0 (t) a physical motion
that starts at (q, t) and ends at (q0 , t0 ). That
t
q ,t
is, a function of time that solves the equations of motion and such that
0
q, t
qqt,q0 t0 (t) = q ,
(2.2)
qqt,q0 t0 (t ) = q ,
(2.3)
and assume for the moment it is unique.

The Hamilton function is defined by
S(q, t, q0 , t0 ) =
Z t0
t

dt L qqt,q0 t0 (t), q qt,q0 t0 (t) .
(2.4)
Example: free particle.

The action is
S[q] =
dt
1 2
mq .
2
(2.5)
t
q 0 , t0
The solutions of the equations of motion are

the straight motions
q0 q
,
v q = 0
t t
q, t
(2.6)
so that we immediately have the Hamilton function

0
2
Z t0
1
q q
m ( q 0 q )2
S(q, t, q0 , t0 ) =
dt m
=
.
0
2
t t
2( t 0 t )
t
(2.7)
Exercise: Show that the Hamilton function of an harmonic oscillator is

S(q, t, q0 , t0 ) = m
(q2 + q02 ) cos (t0 t) 2qq0

.
2 sin (t0 t)
(2.8)
The Hamilton function is easy to compute when we know the solution of the
equations of motion. Indeed, knowing the Hamilton function amounts to knowing
the solution of the equations of motion, as we show below.
tiple valued, if there is none, it is not defined for the corresponding values. For a cleaner treatment,
see the Complements to this Chapter.
Hamilton function
33
q0
Figure 2.1
Variation of the motion with fixed boundary time.
Properties of the Hamilton function

1. Recall that the momentum is
p(q, q ) =
L(q, q )
.
q
(2.9)
It is not difficult to show that

S(q, t, q0 , t0 )
= p(q, t, q0 , t0 ),
q
S(q, t, q0 , t0 )
= p0 (q, t, q0 , t0 ),
q0
(2.10)
where p(q, t, q0 , t0 ) = p(q, q (q, t, q0 , t0 )) and p0 (q, t, q0 , t0 ) = p(q0 , q 0 (q, t, q0 , t0 )) are

the initial and final momenta, expressed as functions of initial and final positions and times, via the actual solution of the equations of motion, determined
by these initial and final data.
To show this, vary the final point q (keeping the time fixed) in the definition
of the Hamilton function: q(t) q(t) + q(t) (See Fig.2.1). This gives

Z t0
Z t0
L
L
S =
q +
q .
(2.11)
dt L =
dt
q
q
t
t
The variation of the time derivative is the time derivative of the variation q =
d
dt q, so we can integrate by parts
S =
Z t0
t
dt
L
d L
+
q
dt q

0
L t

q +
q .
q t
t
t0
Figure 2.2
Variation of the motion with fixed boundary position.
(2.12)
34
The parentheses in the first term in (2.12) enclose the Euler-Lagrange equation and vanishes because by definition q is a solution of the equations of motions. The boundary terms give

0
L t

S =
(2.13)
q = p0 q0 p q
q t
which is precisely (2.10).

2. This result shows that the solutions of the equations of motion can be directly
obtained from the Hamilton function by taking derivatives and inverting. Indeed, inverting p(q, t, q0 , t0 ) with respect to q0 we have q0 (t0 ; q, p, t), namely the
final position as a function of the time t0 and the initial data q, p, t.
Hamilton wrote: Mr Lagranges function describes the dynamics, Mr Hamiltons function solves it. Very British.
But we are not at the real beauty yet!
3. What if we take the derivative of the Hamilton function with respect to time
rather than to q? It is easy to show that
S(q, t, q0 , t0 )
= E(q, t, q0 , t0 ),
t
S(q, t, q0 , t0 )
= E0 (q, t, q0 , t0 ).
t0
(2.14)
(Note the sign switch with respect to (2.10). The similarities with special relativity is surprising and mysterious.)
To derive (2.14), we vary S by varying the boundary time but not the boundary position (see Fig. 2.2). Lets for simplicity vary only at the final point:

Z t0

L 0 0
dt L = L|t0 t0
S =
q t = p0 q 0 L(t0 ) t0 = E0 t0 (2.15)

q t0
t
and we recognize the energy ( p0 q 0 L) = E. Thus,
~ (q0 ,t0 ) S = ( p0 , E0 ).
(2.16)
~ (q,t) S = ( p, E).
(2.17)
and similarly
Therefore we learn that the variables q and t get a sort of equal status in this
language. And this is precisely what we need for quantum gravity, where, as we
shall see, the distinction between dependent (q) and independent (t) variables
looses meaning. The Hamilton function is the natural object when we want to
treat t and q on equal footing, as we will soon be forced to.
4. Finally, the Hamilton function is a solution of the Hamilton-Jacobi equation (in
both sets of variables, that is, in (q, t) as well as in (q0 , t0 )). This follows immediately from the equations above. If Energy is expressed by the hamiltonian as
a function of position and momentum, E = H ( p, q), then, inserting (2.16) we
immediately have

S
S
+H
, q = 0,
(2.18)
t
q
Hamilton function
35
which is the Hamilton-Jacobi equation. The Hamilton-Jacobi equation has many

solutions; the Hamilton function is a peculiar solution of this equation, where
the Hamilton-Jacobi integration constants are taken to be initial position and
time.
Example: free particle

From the Hamilton function (2.7)
m ( q 0 q )2
2( t 0 t )
S(q, t, q0 , t0 ) =
we have immediately
(2.19)
S
q0 q
= m 0
= mv = p,
q
t t
m
S
= v2 = E.
t
2
Inverting the first gives the general solution of the equations of motion
p 0
( t t ).
m
q0 = q +
(2.20)
The Hamilton-Jacobi equation is

1
S
=
t
2m
and it is easy to see it is solved by (2.19).
S
q
2
(2.21)
2.1.1 Boundary terms

Care about boundary terms in the action should be taken, especially when the
action contains second derivatives or is in first order form, as is the case in gravity.
An action with second derivatives for a free particle is
S[q] =
while a first order actions is
S[q, p] =
dt
dt

m qq,
2

1 2
pq
p .
2m
(2.22)
(2.23)
If we disregard boundary terms, all these actions give the same equations of motion as for a free particle: q = 0. In fact, they differ from the conventional action
(2.5) only by total derivatives. But their values on a physical motion differ. The
Hamilton function is the value of the right action on the physical motions. The
right action is determined by which quantities we want the Hamilton function
to depend upon; that is, which quantities we are keeping fixed at the boundary in
36
the variational principle. Suppose these are q (and t). Then the variation of the motion q(t) is chosen such that q(t) = q(t0 ) = 0. It is easy then to see that S turns
out to be independent of q (t) for the action (2.5). Then the variational principle is
well defined. If we instead use (2.22), keeping q (and t) fixed at the boundary, we
have to supplement it with a boundary term
Sboundary = mqq |boundary = pq|boundary
(2.24)
for the variation to be independent of q (0).

Exercise: Try!
This is the rationale for which a boundary term is also needed in gravity when
dealing with gravitational actions. In the case of gravity the correct boundary term
was written by York and then Gibbons and Hawking [York (1972); Gibbons and
Hawking (1977)]. We will write it later on.
Notice that the action (2.22) vanishes on the physical motions. Then the Hamilton function is given entirely by the (difference between the initial and final) boundary term (2.24). The same happens in pure gravity. This does not mean that the
hamiltonian function becomes trivial, because the momenta p on the boundary
must be expressed as a function of all the boundary q0 s, and for this the equations
of motion must still be solved.
2.2 Transition amplitude

Let us now leap to the quantum theory. In the quantum theory, the object that corresponds to the Hamilton function is the transition amplitude. A quantum theory
is defined by
-
A Hilbert space H.
p corresponding to classical variables.2
Operators q,
The time variable t.
or, equivalently3 , the transition amplitude it defines.
A hamiltonian H,
Let q be a set of operators that commute, are complete in the sense of Dirac4 ,
and whose corresponding classical variables coordinatize the configuration space.
Consider the basis that diagonalizes these operators: q|qi = q|qi. The transition
amplitude is defined by
i
W (q, t, q0 , t0 ) = hq0 | e h H (t t) | qi.

2
3
4
(2.25)
We put hats over operators only when we need to stress that they are such. Whenever it is clear, we
q,
p.
drop the hat and write H, q, p for H,

By Stones theorem.
They form a maximally commuting set.
Transition amplitude
37
That is, the transition amplitude is given by the matrix elements of the evolution
operator
i
U (t) = e h Ht
(2.26)
in the |qi basis.

Notice that the transition amplitude is a function of the same variables as the
Hamilton function. The transition amplitude gives the dynamics of quantum theory. In classical physics, given something initial we compute something final. In
quantum physics, given the initial and final something, we compute the probability of the pair.

For a free particle:
2
i p
W (q, t, q0 , t0 ) = hq0 | e h 2m (t t) | qi.

R
Inserting two resolutions of the identity 1l = dp | p ih p | ,
W (q, t, q0 , t0 ) =
dp
1
2h
where we have used h p|qi =
2
i p
(2.27)
dp0 h q0 | p0 i h p0 | e h 2m (t t) | pi h p | q i
i
2
i p
dp e h p(qq ) h 2m (t t) ,
1 eipq/h .
2h
(2.28)
Solving the Gaussian integral we get

0
2
i m(q q)
W (q, t, q0 , t0 ) = A e h 2(t0 t) ,
(2.29)
q
0
2
m(q q)
m
. Recall that 2(t0 t) was the Hamilton
where the amplitude is A =
2hi (t0 t)
function of this system. Therefore
0 0
W (q, t, q0 , t0 ) e h S(q,t,q ,t ) .
(2.30)
This relation between the Hamilton function and the transition amplitude is exact
for the free particle, but in general it is still true to lowest order in h . If we can consider h small, the Hamilton function works as a phase that makes the exponential
oscillate rapidly, while the prefactor in front of the exponential varies slowly.
The rapidly varying part of the transition amplitude is given by the exponential of the Hamilton function. This is the general way in which quantum theory
encodes the classical dynamics.
2.2.1 Transition amplitude as an integral over paths

The above result is general, which can be shown as follows. Write the transition
amplitude a` la Feynman. That is, start from
W (q, t, q0 , t0 ) = hq0 | U (t0 t) | qi ,
(2.31)
38
and use the fact that the evolution operator defines a group
U (t t0 )U (t0 t00 ) = U (t t00 )
(2.32)
to write it as a product of short-time evolution operators

W (q, t, q0 , t0 ) = hq0 | U (e) . . . U (e) | qi ,
where e = (t0 t)/N.
N1
(2.33)
R At each step insert a resolution of the identity 1l =

dqn | qn ih qn | . The transition amplitudes becomes a multiple integral
0
W (q, t, q , t ) =
dqn
h q n | U ( e ) | q n 1 i .
n =1
(2.34)
This expression is exact for any N, therefore it is also true if

we take N :
W (q, t, q0 , t0 ) = lim
dqn
n =1
h q n | U ( e ) | q n 1 i .
(2.35)
Now, consider in particular the case with hamiltonian H =

p2
2m
e=
+ V (q). The evolution operator is
t0 t
p2
U (e) = e h ( 2m +V (q))e
(2.36)
namely the exponent of a sum of two operators. This can

be rewritten as the exponent of the first, times the exponent of the second, times
corrections given by commutator of the two, which are of order e2 or higher,
and may be disregarded in the large N limit, since in this limit e is small. (This
is of course far from being a rigorous step, and the resulting equalities must be
checked case by case.) For small e
2
i p
U (e) e h 2m e e h V (q)e .
(2.37)
In the |qi basis the second exponential gives just a number. The first was computed
above, in (2.28-2.29). The two together give
h q n +1 | U ( e ) | q n i e
i
h
m ( q n +1 q n )2
V ( q n )
2( t n +1 t n )2
(2.38)
multiplied by an amplitude which we absorb in an overall multiplicative factor

N . But tn+1 tn = e, so bringing everything together we obtain:
W (q, t, q0 , t0 ) = lim N
N
lim N
N
dqn e
i
h
nN=1
m ( q n +1 q n )2
V ( q n )
2e2
dqn e h S N (qn ) .
(2.39)
39
The exponent in the last equation is a discretization of the classical action. The transition amplitude can therefore be written as a multiple integral of the discretization
of the action in the limit for the discretization going to zero, namely e 0. This
limit is the definition of the functional integral
W (q, t, q0 , t0 ) =
D [q(t)] e h S[q] .
(2.40)
This nice construction was developed in Feynmans PhD thesis [Feynman (1948)].
Its interest is two-fold. First, it provides a new intuition for quantum theory as a
sum over paths. Second, in the absence of a well defined Hamiltonian operator
we can take an expression like (2.39) as a tentative ansatz for defining the quantum
theory, and study if this theory is physically interesting. This is for instance the
logic in lattice QCD, and we shall adopt a similar logic in quantum gravity.
If there is no potential term or if the potential has a simple quadratic form,
we have a Gaussian integral, which can be computed explicitly. If the rest can
be treated as a perturbation, we can compute the transition amplitude expanding
around its gaussian part. Functional integrals in fundamental physics are almost
always: either a way to keep track of the perturbation expansion (for instance in
QED) or a limit of multiple integrals (as in lattice QCD).
In a regime where h can be considered small (where classical physics is a good
approximation) we have an oscillating integral with a small parameter in front,
and this is dominated by its saddle-point expansion. The oscillations cancel everywhere except where the variation of S[q] is zero, i.e. on the classical solutions.
Assuming this is unique:
W (q, t, q0 , t0 ) =
D [q(t)] e h S[q] e h
S[qqtq0 t0 ]
0 0
e h S(q,t,q ,t ) ;
(2.41)
that is, the transition amplitudes are dominated by the exponential of the Hamilton function.5 This result is therefore general, and provides a tool for studying the
classical limit of a quantum theory.
The formal classical limit of a quantum field theory may not turn out to be particularly physically relevant, as in the case of QCD, whose interesting low-energy
phenomenology is not well described by a field theory. But general relativity, on
the contrary, works well at large distance (like electromagnetism), therefore to
have a classical regime described by general relativity is a necessary condition for
a good theory of quantum gravity. We will use the technique described above, appropriately generalised to the general covariant context, to connect the quantum
theory of gravity that we shall define, to classical general relativity.
5
This clarifies also the meaning of the cases where there is no classical solution or there is more
than one: these are simply determined by the configurations where the transition amplitude is suppressed in the classical limit, or where there is more than a single saddle point in the integral. In
other words, S(q, t, q0 , t0 ) may be ill defined as a proper function, but W (q, t, q0 , t0 ) is not.
40
2.2.2 General properties of the transition amplitude

The transition amplitude W (q, t, q0 , t0 ) is also (as a function of q0 ) the wave function
at time t0 for a state that at time t was a delta function concentrated at q. Therefore
it satisfies the Schrodinger

equation, indeed obviously, in both sets of variables:

ih W + H ih , q W = 0.
(2.42)
t
q
The Hamilton-Jacobi equation is the eikonal approximation (the equation that
gives the optical approximation to a wave equation) to the Schrodinger

equation.
In fact, in his first paper on wave mechanics [Schrodinger

(1926)], Schrodinger
introduced his equation precisely by backtracking from this approximation: he
started from the Hamilton-Jacobi formulation of classical mechanics, and interpreted it as the eikonal approximation to a wave theory. In other words, he found
the Schrodinger
equation by seeking a wave equation whose eikonal approximation would give the Hamilton-Jacobi equation. No surprise then that the phase of
the transition amplitudes be given by the Hamilton function.
There is one important difference between S(q, t, q0 , t0 ) and W (q, t, q0 , t0 ) that
must not be overlooked. The similarity of the writing can sometimes be misleading. The first depends on the configuration space variable q. The second is defined
in (2.31) where q does not necessarily indicate a classical variable: it labels the
eigenstates of the q operator. Classical variables and labels of eigenstates can be
identified only if the operator has continuum spectrum. If the spectrum is discrete, then, by definition, the q in W (q, t, q0 , t0 ) are not classical variables. They are
the discrete labels of the eigenstates of the q operator, that is, its quantum numbers.
For instance, for a particle moving on a line we can define the transition amplitude
on the momentum eigenbasis
W ( p, t, p0 , t0 ) = h p0 | U (t0 t) | pi .
(2.43)
This is simply going to be given by the Fourier transform of W (q, t, q0 , t0 ) expressing the amplitude for changing momentum. But if the particle moves on a circle,
as in the example in the first chapter, the eigenstates of the momentum are discrete
p |ni = pn |ni = nh|ni and therefore the amplitude is
W (n, t, n0 , t0 ) = hn0 | U (t0 t) | ni,
(2.44)
that is, it is not a function of the (continuous) classical momentum variable p. It

has to be so, because the amplitude determines probabilities for quantities having certain values, and some quantities can have only discrete values in quantum
theory. This observation will be important in quantum gravity, where the transition amplitude will not be a function of classical configuration-space variables, but
rather a function of the corresponding quantum numbers, which, as we shall see,
are given by spin networks.
The transition amplitude has a direct physical interpretation: it determines the
amplitude of a process. A process is characterized by its boundary quantities,
41
which are (q, t, q0 , t0 ). For the theory to make sense, there must be a way to measure, that is to assign numbers, to these quantities: they must have some operational meaning.6
The Hamilton function and the transition amplitude are tools that maintain their
meaning in a general covariant context and in quantum gravity, where many of
the other tools of classical and quantum theory are no longer available. We have
introduced them in a familiar context; now it is time to step into physics without
time.
This does not mean that we adopt an instrumentalist interpretation of quantum theory. A process
is what happens to a system S between interactions with other physical systems. The manner in
which S affects the physical systems it interact with, is described by the quantities (q, t, q0 , t0 ). This is
discussed in detail in [Rovelli (1996b)], to which we refer the interested reader for an interpretation
of quantum mechanics that make sense in the exacting context of quantum gravity.
42
2.3 General covariant form of mechanics

The quantities characterizing the initial and final state of affairs of a process are
of two kinds: the time t, and the variables q. From an instrumentalist perspective,
the tools we need to study the dynamics of the system are of two kinds: a clock
that measures t, and other instruments that measure q. We now build a formalism
where these two kind of quantities, t and q, are treated on the same footing.
By analogy, recall that a function q(t) can be expressed parametrically in terms of a couple of funct
tions

q ( ),
q(t)
(2.45)
t ( ).
t2/3
For instance, we can write the function q =

(where
t is treated as the independent variable) in terms of two
q
functions q = 2 , t = 3 (where t is treated as one of
the dependent variables). The opposite is not always true: if q = sin , t = cos ,
then q is not a proper function of t. Therefore the parametric representation is more
general than the q(t) representation of possible relations between the clock variable t and the other variables q. We will see that this wider generality is demanded
by physics.
Of course, a parametric representation is largely redundant. In the example
above, q = f ( )2 , t = f ( )3 defines the same motion q(t) for any invertible function f ( ). Therefore a parametric representation of the motions carries a large redundancy, which is to say, a large gauge invariance. This is the root of the large
gauge invariance of general relativity, namely diffeomorphism invariance.
Let us write the dynamics of a simple system in parametric form. The action
S[q] =
Z t0
t
dt L (q(t), q (t))
(2.46)
can be rewritten as a functional of two functions, q and t by changing variable

t t( )

Z 0
dq( )/d
dt( )
L q( ),
.
(2.47)
S[q, t] =
d
d
dt( )/d
The motions (q( ), t( )) that minimize this action determine motions q(t) that
minimize the original action.
Let us do this more concretely. Take for simplicity a Newtonian system, with
Lagrangian
1
L(q, q ) = mq 2 V (q).
(2.48)
2
This gives the Newton equations of motion
d
mq = q V
dt
(2.49)
General covariant form of mechanics
43
(that is, F = ma). The parametric form of this system is given by the two variables
q( ) and t( ) evolving in with Lagrangian
1 q 2
t) = m tV (q),
L(q, t, q,
2 t
(2.50)
where now the dot indicates the derivative. The equations of motion of this Lagrangian are
d q
m + t q V = 0,
d t
!
2
d
q
1
t:
V (q) = 0.
m
d
2
t
q:
(2.51)
(2.52)
Equation (2.51) is exactly the Newton equation, while (2.52) is nothing else but
energy conservation, which is a consequence of the first equation. Thus the relation
between q and t is precisely the same as the original system.
The fact that the two equations are not independent indicates that this description of the system is partly redundant, namely there is gauge invariance. The
gauge is the arbitrariness in the choice of the parameter along the motions. The
equation of motion and the action are invariant under the gauge transformations
q( ) q( 0 ( ))
and
t( ) t( 0 ( ))
(2.53)
for any differentiable invertible function 0 ( ). This means that is pure gauge:
the physics is not in the specific function q( ) and t( ), but in the relation between
q and t determined parametrically by these functions.
The hamiltonian structure of this system is important. The momenta are
2
q
1
L
V (q) ,
(2.54)
= m
pt =
2
t
t
pq =
L
q
=m ,
q
t
(2.55)
and if you try to invert these equations to express the velocities in terms of the
momenta, you see that this is not possible: the map (t, q ) ( pt , pq ) is not invertible. The reason is that the image of this map is not the full ( pt , pq ) space, but a
subspace, determined by a constraint C (t, q, pt , pq ) = 0. This is easily found from
the definition of the momenta
C = pt + Ho ( pq , q) = 0,
(2.56)
where
Ho ( pq , q) =
p2q
2m
+ V (q)
(2.57)
is easily recognized as the hamiltonian of the unparametrized Newtonian system

we started from. The constraint states that the momentum conjugate to t is (minus)
the energy.
44
What is the canonical hamiltonian H of the parametrized system defined by the

Lagrangian (2.50)? The Legendre transform from the Lagrangian gives the hamiltonian
L L
L
(2.58)
H = q
+t
q
t
and using the constraint we see immediately that
H=0
(2.59)
where the constraint is verified. More precisely, the hamiltonian is proportional to

the constraint
H C.
(2.60)
This should not be a real surprise, since the hamiltonian generates evolution
in the evolution parameter in the action, namely in , but a change in is pure
gauge, and in the hamiltonian formalism the generator of a gauge transformation
vanishes (weakly).
This does not mean in any sense that the dynamics is frozen, or other similar absurdities that one sometimes reads. The dynamics of this system is the one
described by the Newton equation above. The vanishing of the canonical hamiltonian H only means that the dynamics is expressed in this formalism by the relation
between the dependent variables q and t, rather than by the individual evolution
of these in the gauge parameter .
So, how does the hamiltonian formalism keep track of the information about
the physical evolution, if the hamiltonian vanishes? It does so by means of the
constraint
C (q, t, pq , pt ) = 0
(2.61)
as follows. For any function on phase space we can compute the equations of motion in by taking the Poisson brackets with the constraint
dA
= { A, C } ;
d
(2.62)
and we must supplement these with the constraint equation (2.61) itself and remember that the physics is not in the dependence of the variables on but in
their relative dependence when is eliminated.7 Thus C (q, t, pq , pt ) allows us to
derive all observable correlations between variables. This is why this constraint is
sometimes called the hamiltonian constraint.
It is important to emphasise that in this formulation it is not necessary to identify one of the variables as the physical time, in order to compute the observable
correlations and derive predictions for the theory. The physical phase space is interpreted as the space of the possible (solutions of the equations of) motions, rather
7
For the reader who likes more mathematical elegance, a formal symplectic treatment of this generalized form of the dynamics is described for instance in [Sundermeyer (1982)], and briefly developed
in the Complements of this chapter.
45
than the space of the initial data, and the time variable is treated on the same footing as all other variables.
Is this construction artificial? It is not. In fact, we are already used to it: recall for
instance the case of a relativistic particle. We write the action in the form
Z
q
(2.63)
S = m d x x .
The indices label four variables, but the system has only three degrees of freedom, and in fact this action is invariant under reparametrization of . The hamiltonian is zero and the constraint reads
C = p2 m2 = 0.
(2.64)
In General Relativity the action is

S[ g] =
d4 x
det g R[ g]
(2.65)
and is invariant under any reparametrization of x. The canonical hamiltonian vanishes and the information about the dynamics is coded in the constraints. This
means that the dynamics does not describe the evolution of the gravitational field
g ( x ), and other matter fields, as functions of x (this is just gauge), but rather the
relative evolution of the fields with respect to one another.
We call generally covariant, or simply covariant, this generalised formulation of mechanics. The first who understood the need of generalising mechanics
in this manner is Dirac [Dirac (1950)].
2.3.1 Hamilton function of a general covariant system

Let us get back to the parametrized system Lagrangian (2.50), keeping in mind
that this describes the same physics of a conventional Newtonian system with Lagrangian L(q, q ) = 12 mq 2 V (q). What is the Hamilton function of the Lagrangian
(2.50)? By definition,
0
S(q, t, , q , t , ) =
Z
f
i
d L(qqt,q0 t0 0 , tqt,q0 t0 0 , q qt,q0 t0 0 , tqt,q0 t0 0 )
but we can change variables in the integral and rewrite this as

!
2
Z t0
1
dq
S(q, t, , q0 , t0 , 0 ) =
dt
m
V (q) = S(q, t, q0 , t0 ).
2
dt
t
(2.66)
(2.67)
That is: (i) S(q, t, , q0 , t0 , 0 ) is independent of and 0

S
=0
(2.68)
and (ii) its value is precisely equal to the Hamilton function S(q, t, q0 , t0 ) of the
original Newtonian system! The Hamilton function of a general covariant system
does not depend on the evolution parameter, but only on the boundary values of
46

two variables q and t. The Hamilton function is well defined both for systems that
evolve in time and for systems that are generally covariant: the Hamilton function
is always the same object, it only depends on the physical quantities such as q and
t.
Notice also that the Hamilton-Jacobi equation for the Hamilton function of such
a system

S
S S
+H
, , q, t S(q, t, q0 , t0 ) = 0
(2.69)
t q
loses the first term because of (2.68), and because of (2.60) the second term reads

S S
, , q, t S(q, t, q0 , t0 ) = 0,
(2.70)
C
t q
where C is the constraint. This is the general covariant form of the Hamilton-Jacobi
equation. It treats q and t on equal footing and is determined by the constraint C.
If the constraint C has the Newtonian form (2.56), then this equation becomes
the standard non relativistic Hamilton-Jacobi equation (2.18) of the original Newtonian system, with Hamiltonian Ho ! But it can also take more general forms.
2.3.2 Partial observables

When treated on equal footing, the variables q and t are called partial observables.
Configuration space variable and time are, by definition, partial observables. Physics
is about the relative evolution of partial observables. The reason for the adjective
partial is to stress the fact that the partial observables include (for a Newtonian system) also the t variable; namely to distinguish these quantities from the
functions on the configuration or phase space, which are sometimes called observables.
The expression partial observables is important because the word observable alone generates confusion, since it takes different meanings in quantum mechanics, non-relativistic physics, general-relativistic physics, Dirac theory of constrained systems... The main source of confusion comes from mixing up two notions:
Quantities that can be predicted from the knowledge of the initial state. For
instance: the position of a particle at some later time.
Quantities that can be measured, namely for which we have measuring apparatus. For instance: the position variable q, and the time variable t.
The number of partial observables in a system is always larger than the number of
(physical) degrees of freedom, because the number of degrees of freedom is given
by the number of quantities whose evolution can be predicted by the theory, and
these, in turn are given by relations among partial observables.
The space of the partial observables is called the extended configuration space
Cext = C R. We shall denote x Cext a generic point in this space (not to be
47
confused with spacetime coordinates, sometimes denoted with the same letter).
In the examples discussed x = (q, t). Before taking these tools to the quantum
theory, we summarize the covariant formalism developed above, which leads to a
generalized definition of dynamics. This is done in the next section.
2.3.3 Classical physics without time

Let Cext be an extended configuration space and x Cext . The action of a system is
a functional of the motions x ( ) in Cext
S[ x ] =
d L( x, x )
(2.71)
invariant under reparametrizations x ( ) x ( 0 ( )). This invariance implies the

vanishing of the canonical hamiltonian and the existence of a constraint C ( x, p) =
0, where p = Lx . The Hamilton function S( x, x 0 ) is a function of two points in
Cext , defined as the value of the action on a solution of the equations of motions
bounded by the two points. The Hamilton function satisfies the constraint equation

S
C
, x = 0,
(2.72)
x
which is the covariant form of the Hamilton-Jacobi equation. Knowledge of the
Hamilton function gives the general solution of the equations of motion as follows.
The derivative
S( x, x 0 )
= p( x, x 0 )
(2.73)
p=
x
defines the initial momenta. If the initial momenta p and partial observables x
are fixed, equation (2.73) gives a relation between the partial observables x 0 . This
relation is the predictive content of the dynamical theory.
Notice that the construction does not ever require us to mention the word time
or to refer to a time variable.

In this case there are two partial observables: x = (q, t). The Hamilton function is
S( x, x 0 ) =
m ( q 0 q )2
.
2( t 0 t )
pq =
The momenta are
S( x, x 0 )
q0 q
=m 0
,
q
t t
pt =
From the first relation we derive
q0 (t0 ) = q +
S( x, x 0 )
m ( q 0 q )2
=
.
t
2( t 0 t )2
pq 0
( t t ),
m
(2.74)
(2.75)
which is the general solution of the equation of motion, here derived as a relation
between two equally treated partial observables.
48

In summary, a system invariant under reparametrizations of the Lagrangian
evolution parameter is a system where the evolution parameter is pure gauge.
The canonical hamiltonian vanishes, the momenta satisfy a constraint C ( p, x ) = 0,
and this constraint encodes the dynamical information. The Hamilton function is
independent from the Lagrangian evolution parameter and the Hamilton-Jacobi
equation it satisfies is simply (2.72). Any standard Newtonian system can be reformulated in this manner, but not every system formulated in this manner can be
expressed as a Newtonian system (an example of this is given in the Complements
of the Chapter).
2.4 Quantum physics without time

The quantum theory of a covariant system is defined by
- A Hilbert space K called the kinematical Hilbert space, where self-adjoint operators x and p x in K corresponding to classical variables x Cext and their
momenta are defined.
- A constraint operator C, whose classical limit is the constraint C ( x, p). Equivalently, we may work with the transition amplitudes W ( x, x 0 ) it defines.
Notice the absence of hamiltonian, time variable, and Schrodinger

equation.
Discrete spectrum
Assume for the moment that zero is in the discrete spectrum of C. Then the subspace of K formed by the states that satisfy the equation
C = 0
(2.76)
is a proper subspace of K, therefore it is a Hilbert space. It is called the physical state
space and denoted H. The equation (2.76) generalizes the Schrodinger

equation in
the covariant case. This can be seen in many ways. First, this equation is precisely
the Schrodinger
equation for the parametrized form of the Newtonian systems,
where, as we have seen, the constraint has the form (2.56). (Write it explicitly!)
Second, it is the wave equation whose classical limit is the Hamilton-Jacobi equation (2.72). Therefore this equation has the correct form for the Newtonian systems
and the correct classical limit. This is more than we need to take it seriously as the
definition of the quantum dynamics of the covariant quantum systems. In doing
so, we are making a genuine physical hypothesis.
Equation (2.76) is called the Wheeler-deWitt equation [DeWitt (1967)]. It was
first derived by Bryce deWitt in the context of quantum general relativity starting
from the Hamilton-Jacobi equation, interpreted as the geometrical-optic approxi
mation of a wave equation, namely precisely in the same manner as Schrodinger
found his equation starting from the Hamilton-Jacobi equation of a particle in a
Quantum physics without time
49
central potential. John Wheeler understood the importance of the equation and
took it seriously [Wheeler (1968)].
Let us see how this equation defines the transition amplitudes, which are the
actual things we need to derive predictions from the theory. There exists a map
P that sends K to H given simply by the orthogonal projection. The transition
amplitude is defined by the matrix elements of P in the basis that diagonalizes the
operators x:
W ( x, x 0 ) = h x 0 | P| x i.
(2.77)
The transition amplitude determines transition probabilities. Notice that this is a

function that treats all partial observables on the same ground. Formally, we can
rewrite the definition as
W ( x, x 0 ) = h x 0 |(C )| x i
d h x 0 |eiC | x i.
(2.78)
(the integral of an exponential is a delta function) and since C generates evolution

in the parameter , we can translate this into the Feynman language in the form
0
W ( x, x ) =
Z x0
x
D [ x ( )] e h S[ x] .
(2.79)
where the integration is over all paths that start at x and end at x 0 , for any parametrization of these. Since the action does not depend on the parametrization, the integration includes a large gauge redundancy that needs to be factored out. For a Newtonian system, gauge fixing the parameter by t = reduces this definition of
the transition function to the previous Newtonian one, equation (2.39). In the next
section we will come back to a more precise definition of the functional integral
(2.79) as a limit.
Continuum spectrum
The same construction holds if zero is in the continuous spectrum of C, but in this
context it requires more refined mathematics, analogous to those allowing us to
treat continuous-spectrum operators and their generalized eigenstates in conventional quantum theory. The need of this refined mathematics is sometimes mistakenly taken for a sign of deep issues about the nature of time or probability, but it is
not so. Let us sketch how to deal with them. A simple possibility is to pick a dense
subspace S of K whose dual defines a space S of generalized states.8 Then the
space H of the solutions of (2.76) is interpreted as the subset of S formed by the
states S such that
R (C) = 0 for any S and the map P is defined from S
to H by ( P)(0 ) = d h|eiC |0 i. The space H is still a Hilbert space, with the
scalar product h P| P0 i ( P)(0 ). The operator P is often called the projector,
by extension, even if it is a true projector operator only in the discrete-spectrum
case.
8
Giving the Gelfand triple S K S . In fact, strictly speaking the Hilbert space structure on K is
not needed.
50
As an example of continuous spectrum, consider a free Newtonian particle in

one dimension. The kinematical Hilbert space is K = L2 [ R2 , dq dt], the partialobservable operators and their momenta are the diagonal operators q and t and
and ih t
and the constraint operator is
the momenta operators ih q
C = ih
h 2 2
t 2m q2
(2.80)
so that the Wheeler-deWitt equation is precisely the Schrodingier

equation. We
can take S to be the Schwartz space and S the space of of tempered distributions
[Gelfand and Shilov (1968)]. The physical state space H is the space of solutions
in S and not in K, because they are not square-integrable in dq dt. The transition
amplitude is given by
W ( x, x 0 ) = W (q, t, q0 , t0 ) =
d hq0 , t0 |e h C |q, ti.
(2.81)
This can be computed as before by inserting resolutions of the identity

W (q, t, q0 , t0 ) =
1
2
dp
p2
dpt eip( x x) eipt (t t) e h ( pt + 2m ) .
The integration gives the delta function ( pt +
p2
2m )
(2.82)
and then a Gaussian integrap2
tion gives back the result (2.29). The operator P is simply given by ( pt + 2m ) in
Fourier transform. Notice again that we have obtained the physically-interpreted
transition function without ever referring to a physical hamiltonian, or selecting
one of the partial observables as the time parameter.
Interpretation
The equation
W (q, t, q0 , t0 ) = hq0 , t0 | P|q, ti
(2.83)
can be interpreted as follows. The unphysical state |q, ti in K is a delta-function

concentrated on the spacetime point (q, t). It represents the system being in the
configuration q at time t. For a particle, this is the event of a particle being at the
spacetime point x = (q, t). The operator P projects this kinematical state down to
a solution of the Schrodinger

equation, namely to a wave function in spacetime
which is a solution of the Schrodinger

equation and is concentrated on the space
point q at time t. The contraction with the state hq0 , t0 | gives the value of this wave
function at the point (q0 , t0 ); namely the amplitude for a particle that was at q at t
to be at q0 at t0 . This is the physical overlap of the kinematical state representing the
event particle at (q, t) and the kinematical state representing the event particle
at (q0 , t0 ). All this is just a covariant language for describing the conventional
physics of a quantum particle. The formalism treats q and t on equal footing. The
difference between the behaviour of the theory with respect to one or the other is
51
only in the fact that they enter differently in C, and therefore the solutions of the
Wheeler-deWitt equation have different properties with respect to q or t.
While the hamiltonian language (time evolution, hamiltonian, Poincare symmetry...) cannot be used in quantum gravity, the language described here remains
effective.
2.4.1 Observability in quantum gravity

So far, the language we have developed above is just a language: all the examples
considered admit a Newtonian formulation where the time variable is used as independent evolution variable. However, general relativity comes already formulated in this language, and there is no general way of selecting one of its variables
and interpreting it as the time variable. General relativity asks us to describe the
world in terms of relative evolution of partial observables rather than in terms of
evolution of degrees of freedom in time.
As an emblematic case, consider the following physical situation. You keep in
your hand a precise clock whose reading is T and throw upward a second clock,
identical to the first and synchronised to the first in the past, whose reading is T 0 .
When the second clock falls down and you grab it again in your hand, it will be
late with respect to the first. Knowing sufficient initial data, GR predicts the value
T 0 ( T ) it will display when the first clock reads T. Or, which is the same, it predicts
the value T ( T 0 ) the first will display when the second reads T 0 . Does this describe
the evolution of T as a function of the time T 0 or the evolution of T 0 as a function of
the time T? Clearly the question is silly.9 T and T 0 are partial observables, and the
theory treats them on equal footing describing their relative evolution. In General
relativity there is no preferred observable for time: time evolution is always measured with respect to some arbitrary variable. Windshield wipers slappin time.
In his great book, the Principia, Newton asked what is time? and answered
that time is not something that we observe directly. It is a convenient function
of observable quantities that we can usefully single out for playing the role of an
independent evolution parameter. This works when we disregard the dynamics of
the gravitational field, because the fixed configuration of this field itself provides
a convenient time variable. But this does not work anymore when we consider the
full quantum dynamics of spacetime. For this, we have to resort to a fully relational
view of dynamics, where the notion of a specific time variable plays no role.
This does not mean that there is no evolution in the world or that there is no
change. It means that evolution and change of the real world are too complicated
to be well represented as evolution in a single variable.
We now have to extend this technique to field theory, and in particular to a
general context. This is an important issue, and we ask the special attention of the
reader.
9
By the way, the clock in free fall during the period they are separated, namely the one moving
straight, is the one launched up.
52
We call process what happens to a system between an initial and a final interaction. Dynamics has been presented in the previous sections in terms of finite
portions of the trajectory of a system, expressed in terms of relations between the
values of physical variables at the boundaries of a process. These relations can be
coded in the Hamilton function S. In the classical theory, dynamics establishes relations between initial and final coordinates and momenta. In quantum mechanics,
trajectories between two interactions cannot be deduced from the interaction outcomes. The transition amplitudes W determine probabilities of alternative sets of
boundaries values.10
2.4.2 Boundary formalism

Taking the above idea to field theory requires us to move the boundary formalism,
championed and developed by Robert Oeckl [Oeckl (2003, 2008)]. In field theory,
we can still consider a finite portion of the trajectory of a system, but now finite
means finite in time as well as in space. Thus, we focus on a compact region spacetime M.
The transition amplitude is a function of the field values on initial and final
spacelike surfaces, but also on the sides of M: eventual timelike surfaces that
bound the box. In other words, the transition amplitude W is a function of the
values of the field on the entire boundary of the spacetime region M. Formally, W
can be expressed as the Feynman path integral of the field in M, with fixed values
on the boundary = M. The quantum state of the field on the entire boundary is
an element of a boundary Hilbert space H. The transition amplitude W is a linear
functional hW | on this space.
In the non relativistic case, the boundary Hilbert space can be identified with
the tensor product of the initial and final Hilbert spaces
H = H0 Ht ,
(2.84)
Wt ( ) := h|eiHt |i
(2.85)
and
For a field theory on a fixed spacetime, W depends on the shape and geometry of
, for instance, on the time elapsed between its initial and final sides, precisely as
in the last equation it depends on t.
But no longer so in gravity.
10
A common language for describing processes in quantum theory is in terms of preparation and
measurement. This anthropomorphic language is misleading, since it appears to involve human
intervention. The boundary of a process can be any physical interaction of the system with another
generic physical system. Quantum mechanics describes the manner physical systems affect one
another in the course of these interactions [Rovelli (1996b)]. It computes the probabilities for the
different possible effects of such interactions. The theory is characterized, and in fact its structure
is largely determined, by the fact that this description is consistent with arbitrary displacements of
what we decide to consider the boundary between processes.
53
In gravity, a transition amplitude h W | i depends on the state of the gravitational field (as well as any other field which is present) on the boundary and
this is all. Formally, this will be given by the Feynman path integral in the internal
region, at fixed boundary values of the gravitational (and other) fields on . How
do we know then the shape, namely the geometry, of ?
Here comes the magic of quantum gravity: the answer is that the shape, the size
and the geometry, of are already determined by !
In fact, the gravitational field on the boundary is precisely the quantity that
specifies the shape of ! It includes any relevant metric information that can be
gathered on the surface itself! Therefore we expect that h W | i is a function of
and nothing else.
This of course is nothing else but the field analog of the phenomenon observed
in the two previous sections for the parametrized systems, and in particular equation (2.68): the temporal information is stored and mixed among the dynamical variables, instead of being singled out and separated from other variables, as
in unparametrized Newtonian mechanics. In the general relativistic context, this
holds for temporal as well as for spacial locations: W will not be a function of space
and time variables, but simply a function of the gravitational field on the boundary (up to diffeomorphisms of ), which includes the entire relevant geometrical
information on the boundary. This determines any information that can be gathered by clocks or meters
on the boundary, because clocks and meters measure the
Rp
gravitation field (T =
g x x d).
Therefore, in quantum gravity dynamics is captured by a transition amplitude
W that is a function of the (quantum) state of the field on a surface . Intuitively, W
is the sum over geometries on a finite bulk region bounded by .11 The explicit
11
A common prejudice is that in quantum gravity we can only rely on observables at infinity, as one
often does dealing with scattering in particle theory. A source for this misleading prejudice is the
difficulty of defining bulk observables in a generally covariant theory. But this can be resolved: we
measure and describe the relativistic dynamics of our Solar System, in spite of the fact that we are
immersed in it. A second source for the prejudice is the consideration that local observables require
infinite precision and this can only be achieved with infinitely long or infinitely extended measurements: in a region of size L it does not make sense to discuss time evolution with a time resolution
better than t L2Planck /L [Arkani-Hamed et al. (2007)]. The argument, however, assumes continuous background spacetime, which is exactly what is not present in quantum gravity. Time resolution
is limited by uncertainty relations, but this is consistent with standard uncertainly relations for the
gravitational field at the boundary. What is fuzzy is the expected value of the gravitational field,
therefore physical localization of the measurement, not the possibility itself of making a measurement somewhere else than infinity. The point is related to holography (see again [Arkani-Hamed
et al. (2007)]): the Bekenstein bound limits the number of states an apparatus with given area can
resolve, therefore an apparatus localized in spacetime can only distinguish a finite number of states
and cannot resolve arbitrary small distances. This again is correct, but saved by the physical granularity of spacetime. All these arguments show that in the presence of gravity there are no local
observables in the sense of local quantum field theory: localized in arbitrary small regions. They
are all versions of Bronsteins original argument on the fact that space and time are ill-defined in
quantum gravity. The solution is not to take refuge at infinity. It is to accept observables that do not
resolve space and time more finely than Planck scale. It is conventional quantum field theory that
needs to be upgraded, in order for us to deal with observables that are local in a more general sense.
54
construction of W is the main objective in this book. It will be given explicitly in

Chapter 7.
2.4.3 Relational quanta, relational space

The interpretation of quantum mechanics developed in [Rovelli (1996b)], which
can be used in the context of quantum gravity, emphasises the relational aspect
of quantum mechanics. The theory yields probability amplitudes for processes,
where a process is what happens between interactions. Thus quantum theory describes the universe in terms of the way systems affect one another. States are
descriptions of ways a system can affect another system. Quantum mechanics is
therefore based on relations between systems, where the relation is instantiated by
a physical interaction.
The structure of general relativity is also relational, because the localization of
dynamical objects is not given with respect to a fixed background structure; instead, things are only localized with respect to one another, where thing are
all dynamical objects, included the gravitational field. The relevant relation that
builds the spacetime structure is of course contiguity: the fact of being next to
one another in spacetime. We can view a general relativistic theory as a dynamical patchwork of spacetime regions adjacent to one another at their boundaries.
Now, a fundamental discovery and universal ingredient of XX century physics is
locality. Interactions are local. That is, interactions require spacetime contiguity. But
the contrary is also true: the only way to ascertain that two objects are contiguous
is to have them interacting. Therefore locality reveals the existence a structural
analogy between the relations on which quantum mechanics is based and those
on which spacetime is woven.
Quantum gravity makes this connection explicit: a process is not in a spacetime
region: a process is a spacetime region. A state is not somewhere in space: it is the
description of the way two processes interact, or two spacetime regions passing
information to one another. Vice-versa, a spacetime region is a process: because it
is actually a Feynman sum of everything that can happen between its boundaries.
Quantum Mechanics
General relativity
Process
State
Spacetime region
Locality
Boundary, space region
As noticed, a remarkable aspect of quantum theory is that the boundary between processes can be moved at will. Final total amplitudes are not affected by
displacing the boundary between observed system and observing system.
Complements
55
The same is true for spacetime: boundaries are arbitrarily drawn in spacetime.
The physical theory is therefore a description of how arbitrary partitions of nature
affect one another. Because of locality and because of gravity, these partitions are
at the same time subdivided subsystems (in the sense of quantum theory) and partitions of spacetime. A spacetime region is a process, a state is what happens at its
boundary.
These abstract considerations will become concrete with the construction of the
theory, and its applications in the final Chapter.
2.5 Complements
2.5.1 Example of timeless system
Problem
Study a system that can be formulated in the covariant language of this chapter but not in
the Newtonian language. The system is given by the extended configuration space Cext =
R2 with coordinates a and b, and the lagrangian
b ) =
L( a, b, a,
(2E a2 b2 )( a 2 + b 2 ).
(2.86)
where E is a constant. Solve the equations of motion (Hint: the motion that minimize this
action are the geodesics of the Riemannian metric ds2 = (2E a2 b2 )(da2 + db2 ). Gauge
fix the norm of the velocity to one, and separate variables) and show that this is formally
similar to two harmonic oscillators with total energy equal to E. Show that the motions in
the extended configuration space are ellipses and therefore it is impossible to deparametrize
this system and cast it in Newtonian form. Show that the hamiltonian vanishes and the
dynamics is determined by the hamiltonian constraint
C=

1 2
p a + a2 + p2b + b2 E = 0.
2
(2.87)
Define K and the relevant operators and show that zero is in its discrete spectrum provided
that .... . (Hint: just the harmonic oscillators math!) Show that the physical phase space H
is finite dimensional. Compute its dimension (what determines it?) And write an integral
expression for the transition amplitudes W ( a, b, a0 , b0 ). Discuss the physical interpretation
of this system. What would time be here?
Solution.
b ) is clearly invariant under reparametrization of . ThereThe action S[ a, b] = d L( a, b, a,

fore the canonical hamiltonian vanishes. The momenta are
56
Figure 2.3
The motion in the space ( a, b). Which of the two variables is the time variable?
L
(2E a2 b2 ) a
=
,
a
L
(2E a2 b2 )b
L
=
pb =
L
b
pa =
(2.88)
(2.89)
and it is immediate to see that they satisfy (2.87).

The evolution in is generated by C. Since this is just the sum of two Harmonic oscillator
hamiltonians, the evolution in is given by
a( ) = A sin( + a )
b( ) = B sin( + b ).
(2.90)
(2.91)
A2 + B2 = 2E.
(2.92)
The constraint itself gives
Therefore the motions are closed curves in the extended configuration space ( a, b). They are
ellipses. The constant b can be set to zero by redefining . The space of the solution is
therefore two-dimensional, parametrized by A and = a . The relation between a and b,
which is the physics predicted by the system is
arcsin
b
a
arcsin
= .
A
E A2
(2.93)
Since the motions are closed this is a system that does not admit a conventional Newtonian
description. In other words, in a conventional Newtonian there is always one variable, t,
going from to +, while here all partial observables are bounded.
The kinematical Hilbert space is K = L2 [ R2 , da db]. The constraint is the sum of two Harmonic oscillator hamiltonians minus E. In the energy basis, C is diagonal and the Wheeler
de Witt equation reads

1
1
(2.94)
C | n a , n b i = ( n a + ) + ( n b + ) E | n a , n b i.
2
2
There are solutions only if E = N + 1 is an integer. The solutions are the linear combinations
of the states |n a , nb i with n a + nb = N. Therefore H is the proper subspace of K formed by
the states of the form
N
|i =
cn |n, N ni,
n =1
(2.95)
Complements
57
and the project on this subspace is

N
|n, N nihn, N n|.
P=
(2.96)
n =1
The transition amplitude is

W ( a, b, a0 , b0 ) =
ha0 , b0 |n, N nihn, N n|a, bi.
(2.97)
n =1
This can be rewritten in the form

W ( a, b, a0 , b0 ) =
ha0 |nihn|aihb0 | N nih N n|bi,
(2.98)
n =1
and the explicit form of the transition amplitude is therefore

W ( a, b, a0 , b0 ) =
n (a0 )n (a)N n (b0 ) N n (b)
(2.99)
n =1
where n ( x ) are the eigenstates of the harmonic oscillator hamiltonian.12
2.5.2 Symplectic structure and Hamilton function

Here we give a more precise mathematical picture of the formal structure of generalised
Hamiltonian mechanics, following Robert Littlejohn [Littlejohn (2013)]. If you like math this
is very pretty. Our starting point is a covariant system with extended configuration space
C with dynamics defined by a constraint C (here single, for simplicity). We call T C the
extended phase space. This space has a naturally symplectic structure = d, because it
is a cotangent space.13 If x are coordinates on C , namely partial observables, then natural
coordinates on the extended phase space are ( x, p x ). For a Newtonian system with n degrees
of freedom, these include configurations variables and time x = (q, t), as well as momenta
and energy p x = ( p, E); the dimension of the extended phase space is 2(n + 1) and C =
pt + H (q, p, t). The constraint C defines the constraint surface by C = 0 in T C ; the twoform is degenerate (presymplectic) when restricted to this surface and its orbits are the
motions. In this language the Hamilton equations read
| ( X ) = 0.
(2.100)
Pp = 0 P0 p.
(2.101)
X is the Hamiltonian vector field of C, or the vector field that generate the motions. The
space of these motions is the physical phase space ph . We call the natural projection
from to ph which sends each point to the orbit to which it belongs.
Now consider the extended boundary phase space D = T C T C . We call P and P0 , respectively, the projection of this space onto its cartesian components. This carries the symplectic form b = 0 = db = d 0 d. For a Newtonian system with n degrees of
freedom, these include the initial and final configurations variables and times ( x, x 0 ), and
the initial and final momenta and energy ( p x , p0x ); and the dimension of D is 2(2(n + 1)).
The cartesian product of the constraint surface with itself defines a subspace of D, with
co-dimension 2, where the pull-back of the symplectic form has two degenerate directions.
In this subspace, there is a surface L formed by the points p D such that
12
13
For more details on this system and its interpretation, see [Colosi and Rovelli (2003)].
It is defined by (v) := z( f v); here z T C , v Tz ( T C), and f is the natural projection T C C ,
see for instance [Arnold (1989)]. In local coordinates, z = ( x a , p a ), = p a dx a and = dp a dx a .
58
(p, p0 )
z0
zo
z
(x, x0 )
Figure 2.4
The Hamilton function S( x, x 0 ) is the line integral of b from a reference point zo to

z. If there is another point z0 above ( x, x 0 ), then S( x, x 0 ) is multivalued.
The surface L was introduced by Robert Littlejohn. For a Newtonian system we can use C
to solve for pt as a function of the other variables and we can coordinatize with (q, p, t),
L is formed by the sets (q, p, t, q0 , p0 , t0 ) such that the initial (q, p) at time t happen to evolve
to (q0 , p0 ) at time t0 . It can be coordinatised by (q, p, t, t0 ). The map R L defined
by (q, p, t, t0 ) (q, p, t, q0 , p0 , t0 ) is unique and gives the full solution of the equations of
motion, since (q0 , p0 ) are precisely the evolved data from (q, p, t) at time t0 .
The two-form b = db vanishes when restricted to L; therefore the line integral of b on
L depends only on the initial and final points of the line. Fixing an initial point zo (lying on
the diagonal), this line integral defines a function on L:
S(z) :=
:zo z
(2.102)
where L. This is a clean definition of the Hamilton function. The function S(z) does not
depend on the path, it is single valued and defined everywhere on L.
The surface L has the same dimension as C C . Consider the natural projection f :
( x, p, x 0 , p0 ) ( x, x 0 ) from D to C C . This projection maps L to C C ; the two spaces
have the same dimension, but the map is not injective nor surjective in general, therefore
the function S defined on L is sent to a function on C C which is not defined everywhere
and may have has branches. This function on C C is what is usually called the Hamilton
function, that is
S( x, x 0 ) S( x, p, x 0 , p0 ),
if ( x, p, x 0 , p0 ) L.
(2.103)
For a given ( x, x 0 ), if there is no point ( x, p, x 0 , p0 ) L then S( x, x 0 ) has no value; if there is

more than one point ( x, p, x 0 , p0 ) L then S( x, x 0 ) has more than one branch. On the other
hand, S( x, p, x 0 , p0 ) is well defined, uniquely and everywhere, on L.
The interest of this construction is that it extends to general covariant field theory. In
this case the space D = T Cb can be taken to be the cotangent space of the space Cb of
the fields on a surface bounding a region R of spacetime. A point in D represents then
a configuration of the field and its momentum on the boundary surface. The surface L is
defined as the set of these configurations which bound a solution of the field equations in
R. The line integral of the restriction of the boundary phase space one-form b to L defines
the Hamilton function. We will see in the next chapter how this applies to general relativity.
Gravity
Lets come to what we know about gravity. General relativity, the most beautiful
of the physical theories1 , has received an impressive amount of empirical corroboration in the last decades. The theory deserves to be taken seriously: what it
tells us about the world is likely to be important, and to represent a main key to
go ahead. Here we give the basics formulation of general relativity in the form
it will be used in the following chapters (tetrads, Holst action, linear simplicity
constraint...).2
3.1 Einsteins formulation

In Einsteins original formulation [Einstein (1915)], the gravitational field is a symmetric tensor field g ( x ), which can be interpreted as the pseudo-Riemannian
metric of spacetime. The dynamics is coded in the Einstein equations, which follow from the Einstein-Hilbert action
S[ g] =
1
16G
d4 x
g ( R 2),
(3.1)
where g is the determinant of the metric g ( x ), R is the Ricci scalar, and is

the cosmological constant. Greek indices , ... = 0, 1, 2, 3 are spacetime tangent
indices. The Einstein equations, in the absence of matter, are
1
R Rg + g = 0 .
2
(3.2)
The theory is defined by two constants that have both been measured. Their measured values are approximately G 6 108 cm3 g1 s1 and 1052 m2 .
The cosmological constant is sometimes presented as something more mysterious than it actually is: it is just a constant of nature, like several others. In particular, it is often wrongly confused with vacuum energy. This is like confusing the
1
2
59
[Landau and Lifshitz (1951)] Sec. 82.

For the mathematically inclined reader who is willing a more precise treatment of the geometry
used, here are some references. A good and simple introductory book which we recommend is
[Baez and Muniain (1994)]. A standard reference for mathematical physics and in particular geometry is [Choquet-Bruhat and Dewitt-Morette (2000, 2004)]. Other good texts are [Frankel (2003)] and
[Chandia and Zanelli (1997)]. If you like math, go for the classic [Kobayashi and Nomizu (1969,
1996)].
Gravity
60
charge of the electron with its radiative corrections.3 The presence of the cosmological constant implies that there is a dimensionless constant in the (quantum)
theory. Its measured value is
h G
10120 ,
(3.3)
c3
which happens to be a very small number. The cosmological constant is close to
the scale of the Hubble radius RUniverse 1/2 , therefore this number is more or
less given by the ratio between the largest and the smallest things we are aware of
in the universe (namely the cosmological scale and the Planck scale). This is
R2
c3
Universe
10120 .
h G
L2Planck
(3.4)
This is a very large number, or so it looks from our parochial human perspective;
perhaps for the gods it is still a small number. This dimensionless number sits
inside quantum gravity.4
In the following we drop wherever it does not contribute conceptually. We
come back to it in Chapter 6, where we shall see that the dimensionless number
above plays an important role in the complete theory. Also, wherever convenient
we use units where h = 8G = c = 1.
3.2 Tetrads and fermions

The formulation of the gravitational field as a pseudo-Riemannian metric cannot
be fundamentally correct, because it does not allow coupling to fermions, and
fermions exist in the world. For this, we need the tetrad formulation.
The dynamics of a fermion in flat space is governed by the Dirac equation
i I I m = 0 .
(3.5)
In order to couple this equation to gravity, we need a different description of the

gravitational field. Instead of g ( x ), we must use the tetrads eI ( x ) (an introduction
3
Vacuum energy can be computed using quantum field theory on curved spacetime, and in this
context it can be renormalized to its measured value [Hollands and Wald (2008)]. The argument
often heard that the cosmological constant should naturally be at the cut-off scale shows that we
do not yet have good control of quantum field theory at the Planck scale, not that the cosmological
constant itself is mysterious [Bianchi and Rovelli (2010a,b)].
Much has been made about the lack of naturalness in the existence of a large number in fundamental physics. We do not think this is necessarily good thinking. The heliocentric model proposed
by Aristarchus of Samos in antiquity was discarded because it required that the distance to the stars
was a number too large to be realistic, otherwise the Earth movement would have determined a
stellar parallax, which was not observed. This naturalness argument is in Ptolemy, and is wrong.
Similar naturalness arguments were used until the late XIX century against the atomic hypothesis, because the Avogadro number was too large to be realistic. When we learn more about the
universe, we see farther away, and we learn that what looks natural to us may just be so because
of our limited experience.
Tetrads and fermions
61
to tetrads and other mathematical tools can be found in [Baez (1994a)]. Other references are [Wald (1984)], [Misner et al. (1973)], [Hawking and Ellis (1973)]). Here
I = 0, 1, 2, 3 are internal flat Minkowski indices. The relation with the metric is
J
g ( x ) = eI ( x )e ( x ) I J ,
(3.6)
where I J is the Minkowski metric, which we use (with its inverse) to lower and
raise the flat Minkowsli internal indices I, J = 0, ..., 3. Geometrically, eI ( x ) is a map
from the tangent space at x to Minkowski space. It directly captures Einsteins
central intuition that spacetime is locally like Minkowski space. The metric g is
the pull back of the Minkowski metric to the tangent space: this is the meaning
of the last equation. Fermions can then be defined locally in Minkowski space as
usual.
The gravitational action can be written replacing the metric with its expression
in terms of the tetrad: S[e] = S[ g[e]].
The tetrad formalism satisfies an additional local Lorentz SO(3, 1) gauge invariance under the transformations
J
eI ( x ) I J ( x )e ( x )
(3.7)
where I J ( x ) is a Lorentz matrix.5 The metric is not affected by this transformation

because the Lorentz matrices transform the Minkowski metric I J into itself. Hence
the action is invariant.
This local Lorentz gauge invariance has a ge
ometrical interpretation: the inverse e I ( x ) of
the matrix eI ( x ) defines four vectors fields (one
for each I) at each point x. These are orthonormal:
e I ( x )eJ ( x ) g = I J .
(3.8)
Given a Riemannian manifold, the four

tetrad fields are such that everywhere they provide an orthonormal frame; vice
versa, if one has just the four vectors at each point, these define a Riemannian
manifold by defining the frame that they determine to be orthonormal.
Related to this local gauge invariance, we can introduce a connection, as in YangMills theories. The Lorentz connection is an object in the Lie algebra of the Lorentz
5
What is gauge invariance? If we consider a system in isolation, we can interpret gauge invariance
just as a mathematical redundancy in its description. However, many gauge invariant physical systems, such as gravity and electromagnetism, do not couple to other systems via a gauge invariant
quantity. Rather, the interaction couples a non-gauge-invariant quantity of the system A with a nongauge-invariant quantity of the coupled-to system B. The coupled system is still gauge invariant,
but the total gauge invariance does not factor in the product of the gauge invariance of the two
component-systems. Thus, it is not correct to regard non-gauge-invariant quantities of a system as
purely mathematical artifacts; they are a handle for possible interactions of the system. In this sense,
they represent physically meaningful quantities. See [Rovelli (2013b)] for a detailed discussion of
this point.
Gravity
62
group
IJ
JI
= .
(3.9)
This connection defines a covariant derivative D on any object that transforms

under a finite-dimensional representation of the Lorentz group. In particular, this
allows us to define a general covariant Dirac equation, which couples the fermion
field to the tetrads and the connection:
i I e I D m = 0 .
(3.10)
Let us introduce a form notation6 by writing

e I = eI dx
IJ
I J = dx .
and
(3.11)
Tetrad and connection define a geometry in the sense of Cartan, which is more
general than Riemann geometry because it includes also torsion. The torsion twoform is defined by
T I = de I + I J e J
(3.12)
(this is called the first Cartan equation) and the curvature two-form is defined by
F I J = d I J + I K K J
(3.13)
(the second Cartan equation). Given a tetrad, the condition that the torsion vanishes
de I + I J e J = 0
(3.14)
can be shown to have a unique solution [e] for the connection. This solution
is called the (torsionless) spin connection, or the Levi-Civita connection. Therefore
if the connection is torsionless, it is uniquely determined by the tetrad and Cartan geometry reduces to Riemanniann geometry (plus the extra gauge invariance
given by the tetrads). The curvature of the torsionless spin-connection is directly
related to the Riemann curvature tensor by
J
R = e I e F I J ,
(3.15)
or equivalently
J
F I J = eI e R dx dx .
(3.16)
Using this, it is a simple exercise to rewrite the Einstein-Hilbert action in the

form
Z
1
S[e] =
e I e J F KL I JKL
(3.17)
2
where F is the curvature 3.13 of the torsionless spin connection determined by

the tetrad field and I JKL is the completely antisymmetric four-index object, normalized by e0123 = 1 = e0123 . This action yields the Einstein equations (see the
6
For an introduction to differential forms, see for instance [Arnold (1989); Flanders (1963); ChoquetBruhat and Dewitt-Morette (2000, 2004)].
Tetrads and fermions
63
Complements to this chapter). We denote the Hodge dual in Minkowski space by

a star, that is: FI?J ? FI J := 12 e I JKL F KL , or TI? T JKL e I JKL and we can write (3.17)
as
Z
S[e] =
e e F? ,
(3.18)
here and subsequently we will frequently suppress contracted indices. The quantity
I J = eI eJ
(3.19)
is called the Plebanski two-form. It is a two-form with antisymmetric Minkowski

indices, and can therefore be interpreted as a two-form with values in the Lorentz
algebra.
Finally, the action of the fermion field interacting with gravity which gives the
curved Dirac equation (3.10) is
S f [, e] =
I D e J eK e L e I JKL
where D = D dx .
D
(e e e)? ,
(3.20)
3.2.1 An important sign

We wrote above that the tetrad formalism is equivalent to the metric one. This
is not exactly true and this is of importance for the quantum theory. There is a
difference between the theory defined by the Einstein-Hilbert action
SEH [ g] =
1
2
and that defined by the tetrad action

ST [ e ] =
Z p
Z
det g R d4 x,
e I e J FI?J .
(3.21)
(3.22)
These two actions are not equivalent. This can be seen by performing an internal
time-reversal operation
(i)
Te0 := e0 ,
(i)
Tei := ei ,
i = 1, 2, 3.
(3.23)
Under this transformation, SEH is clearly invariant as the metric g = e I e I is not

affected by this transformation, while ST flips sign, ST [(i) Te] = ST [e]. The difference becomes manifest by writing both actions in tensor notation and in terms of
tetrads:
Z
1
SEH [e] =
|det e| R[e] d4 x,
(3.24)
2
Z
1
ST [ e ] =
(det e) R[e] d4 x.
(3.25)
2
They differ by the sign factor
s sgn(det e).
(3.26)
Gravity
64
Hence the gravitational field defined by the triad (i) Te has a different action from
the gravitational field e, in spite of the fact that they determine the same Riemannian metric. The difference between the two gravitational fields e and (i) Te has no
effect on the matter that couples to the metric, but recall that fermions couple to
the tetrad. In fact, the dynamics of the fermion, determined by the Dirac equation
(3.10) is sensitive to the sign s: the phase of the fermion evolves in the opposite
direction in a region where s has opposite sign. Consider for instance the tetrad
field
e0 = N (t)dx0 ,
ei = dxi ,
(3.27)
where the Lapse function N (t) is continuous but becomes negative in a finite
interval in the time t. It is easy to see that this tetrad defines a flat metric. But in
the negative Lapse region the time-derivative term of the Dirac equation flips sign,
and therefore the phase of a Dirac particle rotates in the opposite direction to that
in the positive Lapse region. Since interference experiments can in principle detect
a relative phase shift, the sign s is in principle observable [Christodoulou et al.
(2012)].
In non-relativistic physics, the action (and the Hamilton function) always changes
sign for a time reversed trajectory. The Einstein-Hilbert action does not. Therefore,
in this respect the tetrad action is the natural one. All this has not much effect on
the classical theory, but it is important in the quantum theory: in defining a path
integral for the gravitational field, it is not the same whether we integrate over
metrics or over tetrads, because the integration over tetrads includes integrations
over configurations with s < 0, which contribute to the Feynman integral with a
term which is of the form
i
e h SEH [ g] ,
(3.28)
in addition to the term

i
e+ h SEH [ g] .
(3.29)
The two contributions are akin to the forward propagating and the backward
propagating path in a path integral for a relativistic particle. In this sense, one
could call a negative s region an antispacetime: a spacetime region that contributes to the action with a negative sign, as if it was time reversed. We will see
these two terms appearing over and over in the quantum theory.
3.2.2 First-order formulation

Consider a different action, which is now a function of a tetrad and a Lorentz
connection considered as independent fields
S[e, ] =
e e F [ ]? .
(3.30)
This is a simple polynomial action, sometimes denoted the Palatini action (or, better,
the tetrad-Palatini action). As first noticed by Palatini in a similar context, this ac-
Holst action and Barbero-Immirzi coupling constant
65
tion yields general relativity: the variation of the connection gives precisely equation (3.14), which states that the torsion vanishes and therefore on-shell the connection is the Levi-Civita connection. Then the variation of the tetrad gives the
Einstein equations:

: = [e],
(3.31)
e : Einstein equations.
This is called a first order formulation, and is the analog of the free-particle action
(2.23); while the formulations where the tetrad or the metric are the only variables
are called second order formulations.7
3.3 Holst action and Barbero-Immirzi coupling

constant
The Palatini action is nicely polynomial, like the action of Yang-Mills theory. Furthermore, it admits a purely geometric formulation in terms of differential forms.
One can ask which other terms with all these symmetries can be added to this action. In fact, up to boundary
terms Rthere is essentially only one other term that can
R
be added, which is e e F := e I e J F I J . Let s add it to the action with a
coupling constant 1/:
S[e, ] =
e e F? +
eeF.
(3.35)
The equations of motion of this action are again the same as those of GR: the second term has no effect on the equations of motion. What happens is that the variation with respect to the connection again gives the torsionless condition, and when
7
The first-order and second-order formulations are equivalent for pure gravity, but not so with minimally coupled fermions. The difference between the second-order action
S[, e] =
e e F [ (e)]? +
D
[ [e]] (e e e)? ,
(3.32)
D
[ ] (e e e)? ,
(3.33)
and the first-order one
S[, e, ] =
e e F [ ]? +
can be shown to be a four-fermion interaction term. This is because in the presence of fermions the
variation of does not give the vanishing of the torsion, but rather a torsion term proportional
to the fermion current. The spin connection that solves this equation has a term depending on the
fermion, which couples back to the fermions via the covariant derivative. Thus, schematically
S[, e] = S[, e, ] +
(3.34)
The minimally coupled theories are different, but in either theory we can add or subtract a fourfermion interaction, making them equivalent.
Gravity
66
this is used, the second terms becomes

Z
e I e J FKL =
R d4 x = 0,
(3.36)
which vanishes because of the symmetry properties of the Riemann tensor (its
totally antisymmetric part always vanishes). The action (3.35) can be written in
the compact form

1
S[e, ] = e e F + F

Z
1
ee ?+
F
Z
1
= (? e e + e e) F
(3.37)
and is known as the Holst action; the second term is called the Holst term8 . The
coupling constant is called the Barbero-Immirzi constant. This is the action with
which we do 4-dimensional quantum gravity, as it is the most generic one: it is
polynomial and has all the relevant symmetries. The Holst term has no effect on
classical physics, but plays a substantial role in the quantum theory. The term in
parenthesis

1
1
B = (?e e + e e) = ? +
(3.38)
plays an important role in what follows. On a t = 0 boundary, B is the derivative

of the action with respect to /t (because the quadratic part of the action is
B d), therefore B is the momentum conjugate to the connection. More precisely,
1
in front of the full action, and we
if we reinstate the dimensionful constant 8G
go to a time gauge where the restriction of ?e e on the boundary vanishes, the
momentum is the two-form on the boundary taking values in the SL(2, C) algebra
=
1
B.
8G
(3.39)
Notice that in QCD there is a similar term, which has no effect on the equations
of motion but plays a role in the quantum theory as well:
SQCD =
F F + QCD
F F.
(3.40)
The constant QCD plays a role in quantum theory, for instance in the theory of
instantons. The same happens with the Barbero-Immirzi constant .
8
Holst was the second to discover it [Hojman et al. (1980); Holst (1996)]: it is normal in science to call
something after the second discoverer.
67
Holst action and Barbero-Immirzi coupling constant
3.3.1 Linear simplicity constraint

Consider a spacelike boundary surface in the tetrad-connection formalism. The
tetrad is a map from the tangent space to the spacetime manifold to Minkowski
space. In particular, at each point of it maps the tangent space at into a 3d
linear spacelike subspace of the Minkowski space. The subgroup of the Lorentz
group that leaves this subspace invariant is the SO(3) rotation subgroup, and its
existence breaks the local SL(2, C) invariance down to SO(3) at the boundary. That
is, the boundary allows us to pick up a preferred Lorentz frame.
In coordinates, we can view this by writing the unit length vector n I normal to
all vectors tangents to , that is
J
n I e I JKL e eK eL
x x x
,
1 2 3
(3.41)
where {i }, with i = 1, 2, 3, are the coordinates of the point and x ( )

is the imbedding of the boundary into spacetime. We can use n I to gauge-fix
SO(3,1) down to the SO(3) subgroup that preserves it. That is, we can orient the
local Lorentz frame in such a way that the boundary is locally a fixed-time surface,
so that n I = (1, 0, 0, 0).
The pull-back to of the momentum two-form B I J , defined in (3.38), can be
decomposed into its electric K I = n J B I J and magnetic L I = n J (? B) I J parts, in
the same manner in which the electromagnetic tensor F I J can be decomposed into
electric and magnetic parts once a Lorentz frame is chosen. Since B is antisymmetric, L I and K I do not have components normal to , that is n I K I = n I L I = 0
and are hence three-dimensional vectors in the space normal to n, which we can
~ and ~L. In the gauge n I = (1, 0, 0, 0) these are simply
denote K
K i = Bi0 ,
Li =
1 i jk
e B ,
2 jk
(3.42)
~ = {K i } and ~L = { Li }. These are relations analogous to the ones

where we write K
that relate electric and magnetic fields to the Maxwell tensor.
Now, a simple result that has major importance for the quantum theory follows
easily: from the definition (3.38) of B we have
n I B I J = n I (?e e +
1
1
e e) I J = n I (e I J KL eK e L + e I e J ),
(3.43)
but on the boundary we have n I e I | = 0 by the very definition of n, therefore

n I B I J = n I (?e e) I J .
(3.44)
1
1
1
n I ( B? ) I J = n I (( e I e J )? ) I J = n I (?e e) I J = n I B I J .
(3.45)
For the same reason,
Gravity
68
By definition of K I and L I , the last equation gives:
~ = ~L.
K
(3.46)
In words, the magnetic and electric part of the momentum two-form B are proportional to one another, and the proportionality constant is the Barbero-Immirzi
constant. This relation is called the linear simplicity constraint and is one of the
most important equations in covariant loop quantum gravity, as we shall see in
Chapter 7.
3.3.2 Boundary term

Consider gravity defined in a compact region R of spacetime, with the topology
of a 4d ball. The boundary of R is a 3d space with the topology of a three-sphere,
which we call . In Section 2.1.1 we have seen that in order to have a well defined variational principle and to have a well defined Hamilton function, boundary terms may be required. These are known for general relativity [Gibbons and
Hawking (1977); York (1972)]. In the metric formulation, the boundary term is
SEH boundary =
k ab q ab
q d3 ,
(3.47)
where k ab is the extrinsic curvature of the boundary (its explicit form is given below in (3.55)), q ab is the three-metric induced on the boundary, q its determinant,
and are coordinates on the boundary. In the case of pure gravity without cosmological constant, since the Ricci scalar vanishes on the solutions of the Einstein
equations, the bulk action vanishes and the Hamilton function is just given by the
boundary term9
SEH [q] =
k ab [q] q ab
q d3 .
(3.48)
This is a very non-trivial functional to compute, because the extrinsic curvature

k ab [q] is determined by the bulk solution singled out by the boundary intrinsic
geometry. Therefore k ab [q] is going to be nonlocal. Knowing the general dependence of k ab from q is equivalent to knowing the general solution of the Einstein
equations.
9
Note that here the Hamilton function is a functional of the boundary metric. Therefore one cannot
rely on the difference of notation between S( ) and S[ ] to distinguish it from the action, as in finite
dimensional systems. Which is which, however, would be clear from the context: the action is a local
functional of the bulk 4-metric, the Hamilton function is a non-local functional of the boundary 3metric.
69
Hamiltonian general relativity
3.4 Hamiltonian general relativity

The task of finding a hamiltonian formulation for general relativity was first addressed by Dirac [Dirac (1958)], with the aim of quantizing the theory. The task
turned out to be difficult for two reasons. The first is that the theory has constraints. It is for this reason that Dirac developed his theory of constrained hamiltonian systems, which we briefly review below. The second is due to the horrendous intricacy of the algebra of the canonical analysis of the theory, when using
the metric variable. Dirac succeeded also in this, but the result was cumbersome.
A few years later, the task was greatly simplified by Arnowit, Deser and Misner,
who found a change of variables that simplifies all this, and has a clean geometric
interpretation. Let us start from this.
3.4.1 ADM variables

The change of variables introduced by Arnowit, Deser and Misner [Arnowitt et al.
(1962)] is the following. From g , define the three fields
q ab = gab ,
(3.49)
Na = gao ,
p
N = goo ,
(3.50)
(3.51)
where a, b = 1, 2, 3. Notice that in the last equation N is defined by the time-time

component of g with upper indices. N and Na are called Lapse and Shift
functions, q ab is called the three-metric. This change of variables turns out to be
extremely smart, for a number of reasons. The main one is that when writing the
action in terms of these variables, one discovers that the Lagrangian does not depend on the time derivatives N and N a of Lapse and Shift, and this immediately
simplifies the canonical analysis.
The second is the fact that these variables have a nice intrinsic geometrical interpretation. To see this, consider the hyper-surfaces t = constant. Assume that
these surfaces foliate spacetime. Then q ab is clearly the three dimensional metric induced on this surfaces by the spacetime metric. We use this metric to raise
and lower 3d indices a, b = 1, 2, 3. Next, consider the metric normal n to this
surface. It is not hard to see that the point y = x + Nn dt lies on the surface
t + dt. That is, the Lapse function is the proper-time elapsed between a point on
the t surface and a point on the t + dt surface, along the normal to the surface, or,
equivalently for a static observer on the surface. In a sense, the Lapse determines
the rate at which physical time elapses in the coordinates chosen. Finally, the Shift
~ = { N a = q ab Nb } gives the separation (on the t + dt surface) between the two
N
points y and (t + dt, ~x ). That is, the Shift function measures the shift of the spatial
coordinates from one constant surface to the next, with respect to the coordinates
that observers not moving on the initial surface would carry with them.
Gravity
70
The Lapse can equivalently be defined by

N 2 det q = det g
(3.52)
goo = N 2 + Na N a ,
(3.53)
or
so that the line element in these variables reads
ds2 = ( N 2 Na N a )dt2 + 2Na dx a dt + q ab dx a dx b .
(3.54)
The extrinsic curvature of the t = constant surfaces is given by

1
(q D(a Nb) )
(3.55)
2N ab
where the dot indicates the derivative with respect to t and Da is the covariant
derivative of the three-metric. Using this, the action in terms of these variables
reads
Z
Z

~ q] = dt d3 x qN k ab k ab k2 + R[q]
(3.56)
S[ N, N,
k ab =
where k = k a a and q
det q. That is, in terms of the main variables, the
Lagrangian density reads
ac bd
q( g g g ab gcd )(q ab D(a Nb) )(q cd D(c Nd) )
~ q] =
L[ N, N,
+ qNR[q].
4N
(3.57)
In this form the hamiltonian analysis is easy. The canonical momenta of Lapse
and Shift vanish because N and N a do not appear in the action. The canonical
momentum of the three metric is
p
L
= qG abcd k cd = det q (k ab kq ab ),
(3.58)
ab =
q ab
and the action written in hamiltonian form reads

Z
Z

~ q, ] = dt d3 x ab q ab NC (, q) 2N a Ca (, q) ,
S[ N, N,
where
C = Gabcd ab cd
q R[q]
(3.59)
(3.60)
is called the scalar constraint, or hamiltonian constraint, and

Ca = Da ab
(3.61)
is called the vector, or diffeomorphism constraint. Here

1
Gabcd = ( gac gbd + gad gbc gab gcd ),
2 q
(3.62)
which is called the DeWitt super metric. These must vanish because of the variation of Lapse and Shift. The hamiltonian theory is therefore entirely given by
the vanishing of the momenta associated to Lapse and Shift, the vanishing of the
71
hamiltonian and diffeomorphism constraints, and by a hamiltonian which vanishes when the constraints are satisfied.
Notice that this leads us immediately in the structure of the dynamics described
in the previous Chapter. In particular, the constraints (3.60) and (3.61) realise the
constraint (2.61) for general relativity, and the hamitonian vanishes for the reasons
explained in Chapter 2.
The formalism presented in Section 2.5.2 takes a particularly nice form for general relativity: The space Cb is the space of the intrinsic geometries q ab ( x ) of a surface with the topology of a 3 sphere. D is the space of the intrinsic and extrinsic
geometries [q ab ( x ), k ab ( x )] of this surface. L is the set of boundary (extrinsic and
extrinsic) geometries that bound a Ricci flat ball. The Hamilton function on L is the
restriction to L of the expression
S[q, k] =
and its pull back to Cb
S[q] =
k ab q ab
k ab [q]q ab
qd3 x,
qd3 x,
(3.63)
(3.64)
is the usual Hamilton function, which codes the solutions of the theory.
Notice that in the case of gravity the region R and its boundary have no metric
geometry until the field is specified, because the metric is determined by the gravitational field. Therefore there is a limit in the space of the boundary fields where
the geometry of R shrinks to a point. This is the natural reference point zo .
3.4.2 What does this mean? Dynamics

The result above is clarified by comparing it with the hamiltonian structure of two
well-known theories, which we review here briefly.
Electromagnetism
Maxwell theory can be written in terms of the potential A . The Maxwell field is
F = [ A] and the Maxwell equations are invariant under the gauge transformation A A + . The action is
S[ A] =
1
4
d4 x F F
(3.65)
and it is immediate to see that it does not depend on A 0 . Therefore A0 has the
same role in electromagnetism as the Lapse and Shift function. Notice that a gauge
transformation allows us to choose A0 as we want. For instance, we can choose it
to be A0 = 0. Analogously, Lapse and Shift can be chosen arbitrarily. They simply
determine the position on the coordinates of the next t = constant surface, after
~ = 0, called the
the first is chosen. The simplest choice is the choice N = 1, N
Gravity
72
time gauge, where the line element has the form ds2 = dt2 + q ab dx a dx b . The
momentum of A0 vanishes as the momenta of Lapse and Shift and the momentum
~ is
of A
L
a =
= F oa
(3.66)
A a
namely the electric field ~E. The action in hamiltonian terms then reads
Z
Z

~ ~E] = dt d3 x E a A a ( E2 + B2 ) + Ao C (~E)
S[ A0 , A,
(3.67)
where
C (~E) = a E a
(3.68)
is called the Gauss constraint. This is the analog of the constraints of gravity. Thus
the hamiltonian theory is determined by the vanishing of the momenta associated
to A0 , by the vanishing of the Gauss constraints, and by a non-vanishing hamiltonian.
The physical interpretation of the Gauss constraint is related to the residual
gauge symmetry left by the choice A0 = 0. In fact, we can still gauge transform
A a A a + a with a time independent function . This means that there is a
residual gauge freedom in the initial data. It is a general fact of constrained systems that any first-class constraint10 is related to a gauge freedom. In fact, it is
easy to see
R that the Gauss constraint generates the gauge transformation. Defining
C [] = d3 x ( x )C ( x ), we have
{ A, C []} = d = A
(3.69)
that is: the Poisson bracket of the variable with the constraint gives its infinitesimal
gauge transformation. Analogously, the Scalar and Vector constraints of gravity
generate gauge transformations, which corresponds, respectively, to deformations
of the t = constant surface, and to change of spatial coordinates on this surface.
There remains one feature of the hamiltonian formulation of gravity, which is
not reflected by the the Maxwell analogy: the vanishing of the hamiltonian. To
illustrate this, lets turn to the second example.
Special relativistic dynamics

The action of a single relativistic particle can be written in the form
Z
q
S = m d x x .
The momenta are
p =
10
x
L
=m
x
| x |
(3.70)
(3.71)
In Dirac terminology, a constraint is first class if its Poisson brackets with all other constraints and
with the hamiltonian vanishes when the constraints are satisfied.
73

and clearly satisfy the constraint
C = p2 m2 = 0.
(3.72)
This constraint generates the gauge transformations of the theory, which are the
reparametrizations in . To see the analogy with gravity, notice that the same dynamics can be obtained from the action

Z
x x
1
S=
d
+ Nm2 .
(3.73)
2
N
Notice that this has precisely the same structure of (3.57). Notice that the time
derivative of N does not appear. The momenta are now p = x /N and the action
in hamiltonian form is
Z

1
d p x + NC ( p) .
(3.74)
S=
2
That is, the hamiltonian vanishes when the constraint is satisfied. This clarifies
that the vanishing of the gravitational hamiltonian is simply the consequence of
the fact that the evolution parameter given by the coordinate time is not a physical
quantity, but only a free parameter. The vector constraint reduces the six degrees of
freedom in q ab to three, and the hamiltonian constraint determines the dynamics
in the form of a relation between these three variables. Therefore the theory has
two degrees of freedom per space point.
3.4.3 Ashtekar connection and triads

Instead of introducing tetrads on spacetime, it is also possible to introduce triads
on each t = constant surface. These are defined by
j
q ab ( x ) = eia ( x )eb ( x )ij ,
(3.75)
like their spacetime counterparts. Here i, j = 1, 2, 3 are flat indices. The introduction of this variable adds a local SO(3) invariance to the theory, with the geometrical interpretations of spacial rotations on fixed-time surfaces. We can also define
the triad version k ia of the extrinsic curvature by
k ia eib k ab .
(3.76)
We can consider these variables as a 9+9 dimensional canonical conjugate pair,

j
j
namely pose {k ia , eb } i ba . Notice that this increases by three the number of
variables. However, by definition the antisymmetric part of k ia eib must vanish (because it is k ab ) therefore if we want to recover the 6+6 dimensional (q ab , k ab ) phase
space we must impose the constraint
Gc = ecab k ia eib = 0.
(3.77)
It is immediate to see that the Poisson brackets of this constraint generate precisely
the local SO(3) gauge rotations. In 1986, Abhay Ashtekar realized that a curious
Gravity
74
connection introduced shortly earlier by Amitaba Sen [Sen (1982)] gives rise to
an extremely useful canonical transformation [Ashtekar (1986)].11 Consider the
connection
Aia = ia [e] + kia
(3.78)
where [e]ia is the torsionless spin connection of the triad (the unique solution to
the 3d first Cartan equation dei + ei jk j ek = 0), and is an arbitrary parameter,
and the Ashtekar electric field
Eia ( x ) =
1
j
e e abc eb eck ,
2 ijk
(3.79)
namely the inverse of the triad multiplied by its determinant. Ashtekar realized
that the connection satisfies the Poisson brackets
j
{ Aia ( x ), Ab (y)} = 0
(3.80)
(which is highly non trivial, since the connection depends on non-commuting

quantities) and
{ Aia ( x ), Ebj (y)} = ab ji 3 ( x, y).
(3.81)
Therefore Aia and Eia are canonically conjugate variables.

If = i, the connection A can be shown to be the pull back to the t = constant of
the self-dual component of the spacetime spin connection and is called the complex Ashtekar connection (or simply the Ashtekar connection). With this choice
the scalar and vector constraints turn out to have an impressively simple form:
i aj bk
C = eijk Fab
E E ,
(3.82)
i bi
Ca = Fab
E ,
(3.83)
where F is the curvature of the Ashtekar connection

j
i
Fab
= a Aib b Aia + ei jk A a Akb ,
(3.84)
and the constraint that generates the additional SO(3) local rotations is the same
as in Yang-Mills theory:
Gi = Da E ai .
(3.85)
This formulation of general relativity has long been the basis of loop quantum
gravity. It still plays a central role in the canonical theory and a less evident but
more subtle role in the covariant theory.
In more recent years, the interest has shifted to the case with real , largely
thanks to the work by Fernando Barbero [Barbero (1995)]. The connection with real
appears naturally in the hamiltonian analysis of the Holst action, with = .
But it is important to notice that (the coupling constant in the action) and (the
constant entering in the definition of the connection) do not necessarily need to be
11
Ashtekar-Barbero variables are derived in [Ashtekar and Lewandowski (2004)] with emphasis on
their covariant four dimensional character.
75
taken equal.12 With real = , the hamiltonian constraints turns out to be more
complicated. With cosmological constant, it reads

ij k
a b
2
i j
ijk c
H = Ei Ej ek Fab 2 1 + k [ a k b] +
e e Ek .
(3.86)
3 abc
The field Eia has a direct geometric interpretation as the area element. A simple calculation, indeed, shows that the area of a two surface S in a t = constant
hypersurface is
Z
q
(3.87)
AS = d2 Eia n a Eib nb .
S
To show this, choose coordinates where the surface is given by x3 = 0. Then

q
Z
Z
q
AS = d2 det(2) q = d2 q11 q22 q212
S
S
Z
Z
Z
q
q
q
(3.88)
= d2 det q q33 = d2 Ei3 Ei3 = d2 Eia n a Eib nb .
S
By introducing the two-form
Ei =
1
e E ai dx b dx c ,
2 abc
this can be written in the form

AS =
| E |.
(3.89)
(3.90)
Notice that in the limit in which the surface is small, the quantity ~ES defined by
ESi
Ei
(3.91)
is a vector normal to the surface, whose length is the area of the surface.13 In terms
of the triad, this reads
Z
1
ESi = ei jk e j ek ,
(3.92)
2
S
a formula that we have already seen in (1.11) in the first Chapter. Remember?
Now, the momentum (3.39) is conjugate to the connection and therefore it is
the canonical generator of Lorentz transformation. In particular, the generator of a
boost in, say, the z direction is
Kz =
12
13
1
Kz .
8G
(3.93)
For instance an elegant and useful formulation of the hamiltonian dynamics is given by Wieland in
[Wieland (2012)] keeping real, but using = i, namely using the complex Ashtekar connection as
main variable.
In electromagnetism, the quantity conjugate to the connection is the electric field. Here it is Ei . Hence
the length of the gravitational electric field is the area.
Gravity
76
Using the simplicity constraint (3.46), this gives

1
Lz .
(3.94)
8G
But we have just seen that Lz is the area element A of a surface normal to the z
direction. Therefore the Area element A of a surface is related to the generator K
of a boost in the direction normal to the surface by
Kz =
A
.
(3.95)
8G
This important simple equation encloses, in a sense, the full dynamics of general
relativity, as we shall later see.
K=
3.5 Euclidean general relativity in three spacetime

dimensions
In the following chapters we use the euclidean version of general relativity in
three spacetime dimensions (3d) as a toy model to introduce techniques and ideas
used in the physically relevant case of the four-dimensional lorentzian theory (4d).
Here, let us briefly define this theory.
In 3d the Riemann tensor is fully determined by the Ricci tensor, therefore the
vacuum Einstein equations Ricci = 0 imply Riemann = 0, namely that spacetime
is flat. Therefore the theory has no local degrees of freedom: spacetime is just flat
everywhere. This does not mean that the theory is completely trivial, for two reasons. First, if space has nontrivial topology it can be locally flat but have a global
dynamics (change size in time, for instance). Second, even on a trivial topology,
where there is a single physical state, this single state still determines nontrivial
relations between boundary partial observables. Let us see how this happens. In
the metric formalism, the action (including constants) is
S[ g] =
1
16G
d3 x
g R +
1
8G
d2 k ab q ab q
(3.96)
where the second term lives on the boundary = M of a compact (bulk)

spacetime region M and is needed for the reasons examined in Section 2.1.1. The
Hamilton function of the theory is a function of the restriction of the metric to the
boundary, which is given by the two dimensional metric q induced by g on . The
bulk term in the action does not contribute to the Hamilton function, because the
Ricci scalar vanishes on the solutions of the equation of motion, namely on flat
space, leaving
Z
1
d2 k ab [q]q ab q .
(3.97)
S[q] =
8G
This of course only depends on the intrinsic metric of , and not on the coordinates in which this is expressed, because of general covariance. The dependence
77
Euclidean general relativity in three spacetime dimensions

of the extrinsic curvature k ab on q ab , namely on the intrinsic metric is what codes
the dynamics. This is determined by embedding the 2d Riemannian space defined
by q into a flat 3d space. The dependence of k ab on q ab is obviously highly nonlocal. The non-triviality of the theory is therefore in the (global!) dependence of the
extrinsic curvature on the intrinsic metric.
The solution of the dynamics is expressed by giving the boundary extrinsic curvature k ab ( ) as function of the boundary partial observable q ab ( ). This is analogous to giving the solution of the equations of motion of a newtonian system by
giving p as a function of (q, t, q0 , t0 ) which in turn, as we have seen, is equivalent
to giving the dependence of q0 (t0 ) from the initial data (q, p, t).
In the triad-connection formalism, the gravitational field is described by a triad
field ei = eia dx a and an SO(3) connection i j = ai j dx a , where indices a, b... = 1, 2, 3
are spacetime indices and i, j = 1, 2, 3 are internal indices raised and lowered with
the Kronecker delta ij . In 3d, the bulk action is
1
S[e, ] =
16G
eijk ei F jk [ ],
(3.98)
where the curvature is Fi j = d i j + i k k j . The equations of motion are the

vanishing of torsion and curvature.
Exercise: Derive these equations of motion from the action.
The 4d local Lorentz invariance of the tetrad formalism is here replaced by a
local SO(3) gauge invariance under the transformations
j
eia ( x ) 7 Ri j ( x )ea ( x ) ,
R SO(3) .
(3.99)
This is 3d euclidean general relativity.

A word about notation. For the connection, and all other antisymmetric twoindex tensors, we use also the single-index notation
i =
1 i
e jk .
2 jk
(3.100)
The index i labels in fact a basis in the so(3)=su(2) Lie algebra. It is convenient also
to write the connection as the su(2) generator in the fundamental representation
of su(2), by using the Pauli matrices i = (iA B ) basis. That is, to use the notation
= i i
(3.101)
where = 2i i are the generators of su(2).

We consider this theory on a compact region R of spacetime, with trivial topology (a ball), bounded by a 2d boundary with the topology of a sphere which we
call . The pull back of the connection on the boundary is an SO(3) connection
on . The pull back of the triad is a one-form e on with values in the su(2)
algebra. It is easy to see from the action that these two variables are canonically
conjugate to one another: in analogy with the pq term of particle mechanics, the
Gravity
78
derivative term 0 normal to a boundary surface in the action has the structure
1
nc e abc eia 0 bi , where n a is the normal one form to the boundary.14 Therefore,
8G
using eab nc ecab the Poisson brackets between and e is
j
{eia (), b (0 )} = 8G ij eab 2 (, 0 )
(3.102)
where , 0 .
3.6 Complements
3.6.1 Working with general covariant field theory
Problem
Relate the Einstein-Hilbert action to the tetrad action. How does the sign difference come
about?
Solution
Let us introduce a bit of useful notation:
The gauge exterior covariant derivative D acts on a p-form with Lorentz indices as
D = d,
D I = d I + I J J ,
D I J = d I J + I K K J + J K IK .
(3.103)
It has the Leibnitz property. In this way we can write 1st Cartans equation as
De = T ,
(3.104)
where T is the torsion 2-form. Notice that (cf. (3.104)), R = d + 6= D (because

is not a tensor; the factor 2 is missing, cf. second relation in (3.103)). However,
R = D .
(3.105)
(Note that although the connection is not a tensor, is.)

For an object V I JKL with four Lorentz indices the following trace notation is useful
Tr(V ) e I JKL V I JKL .
14
x
Defined by n = eabc
1
x b
2
dx c , where (1 , 2 ) are coordinates on the boundary.
(3.106)
Complements
79
If A I J and B I J are two antisymmetric tensors, in terms of the notation used in the chapter (A B A I J B I J ), this reads
Tr( A B) = Tr( B A) = 2A B .
(3.107)
This trace has the cyclic property, as a normal matrix trace, for example Tr( a b B) =
Tr(b B a) = Tr( B a b), but also Tr( a b B) = Tr(b a B) .
1 For integration of a function on a manifold we can write
e0 e1 e2 e3 = e0 e1 e2 e3 dx dx dx dx
0
= e e e e
0
dx dx dx dx
(3.108)
3
= det(e) dx dx dx2 dx3 = s| det(e)| d4 x = s
g d4 x = s dV ,
s is the sign of the determinant of e and dV is the invariant volume on the manifold. The
appearance of s gives the sign difference between the metric and tetrad formalism. Let
us drop it in what follow, assuming s = +1. This implies that
e I e J eK e L = e I JKL |e| d4 x .
(3.109)
We have the following identities for e,

J K L
J L K
L J K
K L J
K J L
L K J
e I JKL e I MNO = M N
O + M
N O + M
N O M
N O M
N O M N
O ,

K L
K L
e I JKL e I J MN = 2 M
N N
M ,
L
e I JKL e I JKN = 6 N
,
I JKL
(3.110)
e I JKL = 24 .
We compute
Tr(e e e e) = e I JKL e I e J eK e L = e I JKL e I JKL |e|d4 x = 24|e|d4 x ,
(3.111)
where we have used (3.109) and the last of the relations (3.110). This is important when
we want to add to the action the cosmological term. Similarly, we get
Tr(e e F ) = e I JKL e I e J F KL =
1
e
F KL MN e I e J e M e N
2 I JKL
1
e
e I J MN F KL MN |e|d4 x
2 I JKL

= KM LN LM KN F KL MN |e|d4 x = 2|e| Rd4 x ,
(3.112)
where we have used (3.16) in the last step. Hence the gravitational action (3.1) can be
written as
Z
Z

1
1
S[ g[e]] =
Tr e e F =
e e F? .
(3.113)
32G
16G
Problem
Show that treating e and as independent variables in the tetrad action gives the same
physics.
Gravity
80
Solution
Varying (3.113), we get
32G S =
=2

Z

Z

Tr e e F ( ) + Tr e e F ( ) + Tr e e F ( )
Tr(e F e) +
Tr(e e D ) ,
(3.114)
where we have used the cyclic property and Eq. (3.105). Using the identity
Z
d [Tr(e e )] =
D [Tr(e e )] = 2
Tr( De e ) +
Tr(e e D ) ,
(3.115)
Eq. (3.104), and the fact that the term on the l.h.s. is a total derivative, we can write the
previous equation as
S =
1
16G
Tr(e F e)
The vanishing of the first term implies

Tr( T e ) .
e I JKL e I F JK = 0 .
(3.116)
(3.117)
To show that this is equivalent to Einsteins equations, expand the curvature 2-form F JK in
the basis, F JK = 12 R JK MO e M eO . Hence we have
e I JKL R JK MO e I e M eO = 0
e I JKL e I MOP R JK MO = 0 ,
(3.118)
where we have taken the Hodge dual. Applying (3.110), we find that this is equivalent to
1
G P L = RicP L LP R = 0 ,
2
(3.119)
which are the vacuum Einstein equations.

Notice that we have three possibilities, which are called x-order formalism:
2nd-order formalism. Assumes = (e), i.e., T = 0. Then (3.114) gives the Einstein equations.
1st-order formalism. Treats and e as independent. The variation with respect to e gives
the Einstein equations, whereas the variation with respect to implies T e = 0
T = 0. That is, in the absence of matter the torsion T vanishes as a consequence of the
variational principle. This then establishes the relation = (e). 1.5-order formalism.
Treats e and as independent. However, we know that variation with respect to just
establishes the relation = (e) whatever this is. To obtain the equation of motion, it
is therefore enough to vary S with respect to e only, i.e.,
EOM =
S[e, ]
= 0.
e
(3.120)
Problem
Show that the Holst term does not modify the equations of motion.
Solution
Consider the Holst term
S [e, ]
Tr(e e F ) =
eeF =
e I e J FI J ,
(3.121)
Complements
81
Varying with respect to e gives

e S
Tr(e F e) ,
(3.122)
which implies that

e I JKL e J ( F )KL = 0 .
(3.123)
That is, we get the following equations of motion:
0 = eKLMN eKLI J e I FMN = 2(IM JN IN JM )e I FMN = 4e I FI J .
(3.124)
In the absence of torsion the last expression is precisely the Bianchi identity and vanishes.
This proves that the Holst action gives classical Einsteins equations and hence it is equally
good action for GR.
3.6.2 Problems
1 Derive the Dirac equation from the Dirac action on a curved spacetime.
2 Develop the formalism discussed in Section 2.5.2 for the tetrad-connection formulation
of general relativity.
Classical discretization
To define a quantum theory for the gravitational field, we apply the general structure devised in Chapter 2 to the theory defined in Chapter 3. For this, we need to
study how to discretize general relativity. In this chapter, we review two classic
discretizations: lattice Yang-Mills theory and Regge calculus, and then we introduce the discretisation of general relativity which we use in the following.
Lattice Yang-Mills theory is the basis for defining QCD in the strong coupling
regime, and the best tool for computing physical quantities such as hadron masses,
and comparing them with measurements [Durr et al. (2008)]. The theory is the result of some beautiful intuitions by Kenneth Wilson [Wilson (1974)], important
also for quantum gravity. But there is a crucial difference between QCD and general relativity, as we discuss in detail below. The difference is illustrated by Tullio
Regges natural discretization of general relativity, called Regge calculus [Regge
(1961)], also important for the following. The discretisation of general relativity
that we use in quantum gravity is introduced at the end of this chapter; it is a
mixture of lattice Yang-Mills theory and Regge calculus.
A discretisation is an approximation: a truncation in the number of degrees of
freedom where we disregard those likely to be irrelevant for a given problem. In
quantum field theory truncations play also a constructive role. For describing the
weak field regime we utilise Fock-space methods and perturbation theory. Fock
space is constructed by defining the N-particle state spaces. The theory is formally
given by the N limit. The same is true on the lattice. The theory is formally
defined in the limit where the number N of lattice sites goes to infinity. The same is
true in quantum gravity: the theory can be defined via a discretization/truncation.
4.1 Lattice QCD

We consider for simplicity an SU(2) Yang-Mills theory in 4 dimensions. The field
variable in the continuous theory is an SU(2) connection Ai ( x ), where i is an index
in the Lie algebra of SU(2). We also write this field as a one-form with value in the
Lie algebra
A( x ) = Ai ( x ) i dx
(4.1)
where i are the su(2) generators, namely the Pauli matrices multiplied by i/2. The
central idea of Wilson, on which loop quantum gravity also relies, is that we must
82
83
Lattice QCD
view the algebra as the tangent space to the group, and A as the log of a group
variable.
For this, fix a cubic lattice with N vertices
connected by E edges in spacetime. This of
course breaks the rotation and Lorentz invariance of the theory, which will only be recovered in a suitable limit. Let a be the length of
the lattice edges, which is determined by the
(here fixed) spacetime flat metric. Associate a
group element Ue SU(2) to each (oriented)
edge e of the lattice (and the inverse Ue1 of Ue
to the same edge with opposite orientation). Then the set of group elements Ue , for
all the edges of the lattice, provide a natural discretization of the continuous field
A. Wilsons idea is that the quantum theory is better defined starting from these
group variables than from the algebra variables. Physical quantities must then be
studied in the limit where N and a 0. The different manner in which this
limit enters gravity and Yang-Mills theory will determine the structural difference
between the two theories.
The formal relation between Ue and A is given by the equation
Ue = P e
(4.2)
that is: the group element can be seen as the path ordered exponential of the connection along the edge. In quantum gravity parlance the path ordered exponential
is commonly called holonomy, a term used with slightly different meaning in
mathematics. The definition of this object is recalled in the Complements to this
Chapter.
Expanding in the length a of the edge, this gives, to first order
Ue = 1l + aA (se ) e ,
(4.3)
where e is the unit tangent to the edge and se is the initial point of e (source).
The holonomy is invariant under all local gauge transformations A A + D
except those at the boundary points of the edge. Therefore, when truncating the
theory to the group variables, the local SU(2) gauge symmetry is reduced to a
symmetry under SU(2) rotations at the vertices of the lattice only. The gauge group
of the lattice theory is therefore SU (2)V , where V is the number of vertices. The
group variables transform as
1
Ue se Ue
te ,
v SU (2)
(4.4)
under such a gauge transformation. Here se and te are the initial and final vertices
of the edge e (source and target).
The ordered product of four group elements around a plaquette f, namely an
elementary square in the lattice
Uf = Ue1 Ue2 Ue3 Ue4
(4.5)
84
is a discrete version of the curvature. Its trace is gauge invariant. Wilson has shown
that the discrete action
S = Tr Uf + c.c.
(4.6)
approximates the continuous action in the limit in which a is small. The coupling
constant is a number that depends on a and goes to zero when a goes to zero.
Consider now the boundary of the lattice. Let l be the boundary edges or links.
If we fix the value of these and integrate the exponential of the action over the bulk
groups elements
W (U` ) =
dUe e h S[U ]
(4.7)
we obtain the transition amplitude of the truncated theory. This integral fully defines the theory. The analytic continuation of this in imaginary time can be computed numerically using Monte Carlo methods, since the integrand becomes positive definite. To get the continuous transition amplitudes, one must study the
0, N continuum limit.
4.1.1 Hamiltonian lattice theory

Of particular interest for us is the hamiltonian formulation of this theory, which
was early developed by Koguth and Susskind [Kogut and Susskind (1975)]. The
hamiltonian formulation lives on a boundary of the lattice, which we assume to be
spacelike. The hamiltonian coordinates are the group elements U` on the boundary
edges. We call the boundary edges links `, and the boundary vertices nodes n,
for reasons that will become clear later on. These group elements correspond to the
space components of the connection on the boundary, namely (in the time gauge)
they code the magnetic field. In the lagrangian language its time component, or
electric field, is coded by the edges normal to the boundary, while in the hamiltonian language it is coded in the momentum conjugate to the boundary group
elements.
The canonical configuration space is therefore SU(2) L which is a group, and the
reduced gauge is given by the gauge transformations at the nodes on the boundary. Here L is the number of links. The corresponding phase space is the cotangent
space T SU(2) L . There is one conjugate momentum Li` in the algebra of SU(2) for
each link `, and it is identified with the electric field. A cotangent space T Q carries
a natural symplectic structure (see footnote in Section 2.5.2). The symplectic structure of a space cotangent to a group is well studied. The corresponding Poisson
brackets are
{U` , U`0 } = 0 ,
{U`0 , Li` , }
j
{ Li` , L`0 }
(4.8)
i
= ``0 U` ,
= ``0 e
ij
k
k L`
(4.9)
(4.10)
(no summation over `). The Hilbert space of the discrete theory can therefore be
85
Discretization of covariant systems

represented by states (U` ), functions on the configuration space. The space of
these states carries a natural scalar product, which is invariant under the gauge
transformations on the boundary: the one defined by the SU(2) Haar measure
(, ) =
SU (2)
dU` (U` )(U` ).
(4.11)
The boundary gauge transformations act at the nodes of the boundary and transform the states as follows
(U` ) (s` U` t` ),
n SU (2).
(4.12)
The states invariant under this transformation form the Hilbert space of the gauge
invariant states which has the structure L2 [SU (2) L /SU (2) N ], where L is the number of links in the boundary and N the numbers of nodes on the boundary.
The operator corresponding to the group elements themselves is diagonal in this
basis. The operator that corresponds to the quantity conjugate to U` is the natural
derivative operator on functions of SU(2), namely the left invariant vector field ~L` .
In the classical theory, the conjugate momentum to the connection is the electric
field, namely the time-space component of the field strength. Therefore ~L` can
be identified with this quantity. (More precisely, with the flux of the electric field
across an elementary surface dual to link `.) These operators realize the Poisson
algebra above.
All these structures will reappear in loop quantum gravity.
4.2 Discretization of covariant systems

Before moving to gravity, let us study what happens to the classical and the quantum theory when we discretize a covariant system (in the sense of Chapter 2).
In section 2.2.1 we have seen that a quantum theory can be defined by a functional integral, and this can be defined as the limit of a multiple integral, namely
the limit of a discretization of the theory. See eq. (2.39)
W (q, t, q0 , t0 ) = lim
dqn e h S N (qn ) ,
(4.13)
where the discretized action at level N is
S N (qn ) =
m ( q n +1 q n )2
eV (qn ).
2e
n =1
(4.14)
Since e = (t0 t)/N, the parameter N appears in this expression twice: in the
number of steps and in the size of the time step. The second can be in part traded
by a change of integration variables, but not completely. Redefining qn qn / e,
86
we can replace S N (qn ) by
S N (qn ) =
m ( q n +1 q n )2
eV ( eqn ).
2
n =1
For concreteness, for instance, consider the simple case where V (q) =
namely a harmonic oscillator. Then
S N (qn ) =
m ( q n +1 q n )2
2 2
q ,
2
2 n
n =1
(4.15)
1 2 2
2 q ,
(4.16)
where = e. In this manner e has formally disappeared from the discrete action,
but in reality it is hidden into . To take the N limit we have to do two
things: send the upper limit of the sum to as well as sending to its critical
value c = 0.
This is well known to anybody who deals with discretized systems: the continuum limit is recovered by increasing the number of lattice sites and also scaling
down the lattice spacing to zero, or, equivalently, sending appropriate constants
(here ) to a critical limit value (here zero). This seems to be a universal feature of
discretization and it is often presented as such.
But it is not.
Something remarkable happen by (appropriately) discretising covariant systems.
t
Figure 4.1
Discretization of the parameter time.

Consider a discretization of a parametric system, like the one defined by the
action described in Section 2.3

Z
m q 2
S = d
tV ( q ) ,
(4.17)
2 t
which is physically equivalent to the system above. Discretize this action in the
natural manner. This gives
( q n +1 q n )2
N
m
(t
tn )
e2
S= e
n +1
V (qn ) .
(4.18)
( t n +1 t n )
2
e
n =1
e
Regge calculus
87
A look at this is sufficient to show that

N
S=
n =1
m ( q n +1 q n )2
( t n +1 t n ) V ( q n ).
2 ( t n +1 t n )
(4.19)
The parameter e cancels!

At first this looks magic: the discretization of the Lagrangian does not depend
on the lattice spacing! The number N appears only in the upper limit of the sum.
But a moment of reflection can convince the dubious reader that it has to be so. The
original action did not depend on the parametrization, so its discretized version
wont either.
Notice that the path integral (equation (2.79)) is now both on the qn s and the
tn s. In a sense, from the Newtonian perspective, we are integrating over the positions of the time steps of the discretization.
As we have seen in Chapter 2, the equations of motion of the continuous theory
are not independent. It is easy to check that the equations of motion of the discretized theory lose this dependence: they become independent. The equation for
the variable t, which gives energy conservation in the continuum, fixes the values
of the tn in such a way that the energy remains conserved also in the discretization.
It is only in the N limit that reparametrization invariance is restored.
Restoring of invariances in the limit in which a discretization is removed is a
general feature of discretizations. For instance, the discretization of a rotationally
invariant system typically looses rotational invariance, and this is recovered in the
limit.
The reparametrization invariant systems are characterised by the fact that there
is no parameter to rescale in the discrete action in the limit: the only place where
N appears is in the number of steps.
This peculiar behavior of the discretization of reparametrization invariant systems can be found in the good discretizations of GR. While the Wilson action
used in lattice QCD depends on a parameter that has to be scaled to its critical
value in order to gain the continuous limit, there is no similar parameter in Regge
calculus, illustrated below, which is a natural discretization of GR. It is only the
number of steps N (or cells) that has to be increased, to get a better approximation
of the theory, without any other parameter to tune.
This is a profound difference between the discretization of theories on a metric
space and (good) discretizations of general covariant theories. It is at the root of
the peculiarity of the covariant form of quantum gravity.
4.3 Regge calculus

Tullio Regge introduced a very elegant discretization of general relativity, which
is called Regge calculus. It can be defined as follows. A d-simplex is a generalization of a triangle or a tetrahedron to arbitrary dimensions: formally, the convex
88
hull of its d + 1 vertices. These vertices are connected by d(d + 1)/2 line segments1
whose lengths Ls fully specify the shape of the simplexthat is to say its metric geometry. For example a simplex of dimension d = 3 (a tetrahedron) has d + 1 = 4
vertices connected by 6 line segments whose lengths L1 , ..., L6 fully determine its
shape.
A Regge space ( M, Ls ) in d dimensions is a d-dimensional metric space
obtained by gluing d-simplices along
matching boundary (d 1)-simplices.
For instance in 2d we can obtain a surface by gluing triangles, bounded by segments, which meet at points.
In 3d we chop space into tetrahedra,
bounded by triangles, in turn bounded
by segments, which meet at points.
3D spacetime
In 4d we chop spacetime into 4simplices, bounded by tetrahedra, in turn
bounded by triangles, in turn bounded
by segments, which meet at points.
These structures are called triangulations (see Table 4.1). We assume always
that the triangulation is oriented. This means that we conventionally assign a direction to each segment, a preferred side to each triangle and a preferred cyclic
ordering to the vertices of each tetrahedron, and so on; this is just to simplify the
construction and the notation.
A moment of reflection shows that gluing flat d-simplices can generate curvature on the (d 2) simplices (sometimes called hinges). For instance, in d = 2
dimensions, we can glue four equilateral triangles as in the boundary of a tetrahedron, and there is clearly curvature on the vertices of the tetrahedron. In d = 3
dimensions, we can glue several tetrahedra all around a common segment, and
obtain a manifold flat everywhere except at this segment. In four dimensions, curvature is on the triangles. The metric of the resulting space is uniquely determined
by the length Ls of all its segments s. This follows immediately from the fact that
the metric of any d-simplex is uniquely determined by the length of its sides.
Table 4.1 Triangulations

The simplex carrying curvature, or hinge, is underlined.
2d
3d
4d
4-simplex
tetrahedron
tetrahedron
triangle
triangle
triangle
segment
segment
segment
point
point
point
Now, it is easy to see that a Riemannian manifold ( M, g) can be approximated

1
The segments are sometimes called edges in the Regge-calculus literature. Here we prefer to use
segment, in order to avoid confusion with the edges of the dual complex, introduced below.
Regge calculus
89
arbitrarily well by a Regge manifold. This means that for any ( M, g) and any e,
we can find an ( M, Ls ) (with sufficiently many simplices) such that for any two
points x and y in M the difference between the Riemaniann distance and the Regge
distance is less than e. This is the basis for the approximation of curved surfaces by
triangulations used for instance in industry. Regges idea is to approximate general
relativity by a theory of Regge manifolds.
For this, we need a notion of curvature for a Regge manifold, which converges
appropriately to Riemanns curvature. Regges notion of curvature is very beautiful. Consider first the case d = 2, which is simple. Consider a point P of the
triangulation. Around it there are a certain number of triangles tr. Let the angles
at P of these triangles be tr .
These angles can be immediately computed from the lengths, using the
b
c
elementary-geometry formula
c2 a2 b2
cos(tr ) =
,
2ab
tr
(4.20)
tr
a
which gives the angle opposite to c of a triangle with sides a, b and c.

Now, if the angles around P sum up to 2,
clearly the manifold is flat at P. If not, there is
curvature. Therefore the Regge curvature at P
can be defined as the angle
P
P ( Ls ) = 2 t ( Ls ),
(4.21)
called the deficit angle at P. In this formula,

the sum is over the triangles around P and t
is the angle at P of the triangle t.
The same logic can be used in higher
dimensions. In d dimensions, the Regge
curvature is still given by a single deficit
angle, but now P is replaced by a (d 2)
simplex (a segment in 3d and a triangle
in 4d, see Table 4.1), the sum is over the
(d 1) simplices around it (triangles in
3d, tetrahedra in 4d) and the angles become the dihedral angles of the flat dsimplices, which can be computed from
the sides using formulas of elementary
geometry. For instance, the dihedral angle on the ab side of a tetrahedron with
vertices ( a, b, c, d) is
cos ab =
cos acd cos abc cos abd

sin abc sin abd
(4.22)
90
where abc is the angle at the vertex a of the triangle of vertices a, b, c.
The geometrical interpretation of the deficit
angle is simple. If we parallel transport a vector in a loop around a d 2 simplex, the vector gets back rotated by the deficit angle (loops
wrap points in 2d, segments in 3, surfaces in
4d). Using this, Regge defines the Regge action
of a Regge manifold ( M, Ls ) to be
S M ( Ls ) =
Ah ( Ls )h ( Ls ),
(4.23)
where the sum is over the (d 2) simplices of the triangulation, or hinges (points
P in 2d, segments in 3d, triangles in 4d) and Ah is the (d 2)-volume of h (length
in 3d, area in 4d). The remarkable result obtained by Regge is then that the Regge
action (4.23) converges to the Einstein Hilbert action S[ g] when the Regge manifold
( M, Ls ) converges to the Riemann manifold ( M, g). This tells us that the Regge
theory is a good discretization of general relativity.
The Regge equations of motion are obtained by varying the action with respect
to the length. This gives two terms: the first from the variation of the A0 s and
the second from the variation of the angles. The second term vanishes. This is a
general feature of gravity theories: it is the analog of the fact that the variation
of the Riemann tensor in the Einstein Hilbert action vanishes. Thus, the Regge
equations are
Ah
( Ls ) = 0,
Ls h
(4.24)
where the sum is over the hinges adjacent to the segment s, and there is one equation per segment. We can identify h ( Ls ) with a measure of the discrete Riemann
curvature and the left hand side of the equation of motion as a measure of the
discrete Ricci tensor.
In three spacetime dimensions the hinges are the same things as the segments
and therefore the Regge equations reduce to h ( Ls ) = 0, that is: flatness, as in the
continuum case.
There is a peculiarity of the Regge discretization that plays a role in what follows. The Regge curvature is concentrated at the hinges. If we parallel transport a
vector around a hinge in a Regge manifold, the vector gets back to its original position rotated: this is the standard manifestation of curvature. However, a moment
of reflection will convince the reader that the only rotation that can result is around
an axis parallel to the hinge. That is, the rotation never rotates the hinge itself. This
is why the Regge curvature is captured by a single number, the deficit angle: because it is restricted to rotations around the hinge itself. Therefore the Riemannian
curvature of a Regge manifold is a curvature concentrated on the plane normal to
the hinge and generating a rotation on this same plane. If the plane normal to the
91
Discretization of general relativity on a two-complex

hinge is the ( x1 , x2 ) plane, for instance, the only non vanishing component of the
Riemann tensor would be R12 12 .
A curvature with these algebraic restrictions is very special. But since the physical curvature in a region is obtained by averaging the Regge curvature over the
region, and since the special feature of the Regge curvature is not a linear property,
the average curvature is not going to inherit this special property. Thus a Regge
geometry can still approximate a generic Riemannian space. But it is important to
remember that it does so with a rather restricted kind of curvature.
The action can also be rewritten as a sum over the d-simplices v of the triangulation. From (4.23) and the definition (4.21) of deficit angle, we have in fact
S M ( Ls ) = 2 Ah ( Ls ) Sv ( Ls ),
(4.25)
where the action of a d-simplex is

Sv ( L s ) =
A h ( L s ) h ( L s ).
(4.26)
This becomes particularly useful if the quantities Ah are integers (as will happen
later), in which case e2i h Ah = 1 or half integers, in which case this term gives
only a sign factor in a path integral.
If we fix the triangulation , we obtain only a finite approximation to Riemaniann manifolds and to general relativity. This defines a truncation of general relativity. It is the analog of replacing a wave equation with a finite difference equation,
or QCD with lattice QCD.
Observe the difference between the lattice discretization of Yang Mills theory
and the Regge discretization of general relativity: in the first (Section 4.1), the continuum limit is obtained by taking the number of vertices N to infinite as well as the
coupling constant (or, equivalently, the lattice spacing a) to zero. Not so for the
Regge discretization. Here the continuum limit is obtained by refining the triangulation, with no parameter to be taken to zero. This is a consequence of the covariant
character of the theory as explained in Section 4.2. The difference between lattice
QCD and Regge calculus is therefore deeply rooted in the general covariance of
general relativity. The physical size (length) of the segments of the Regge triangulation are the dynamical variables themselves, and not a fixed external quantity
that we can scale in a specific limit. The continuum limit of a proper discretization
of a general-covariant theory does not require any parameter, besides the number
of lattice steps, to be taken to a limit.
4.4 Discretization of general relativity on a

two-complex
The Regge discretization is not very good for the quantum theory, for two reasons.
First, it is based on the metric variables. The existence of fermions indicates that
92
we need tetrads at the fundamental level. More importantly, the segments of a
Regge triangulation are subjected to triangular inequalities: a segment connecting
two points P and Q cannot be longer than the sum of two segments connecting
P and R and Q and R. Therefore the configuration space of the theory, which is
the set of segments lengths that satisfies these inequalities, is a complicated space
with boundaries, and this makes the search of a quantum theory with this classical
limit far more complicated.
The tetrad-connection formulation of general relativity offers an alternative, and
leads to a discretization of the theory which is closer to that of Yang Mills theory, while at the same time retaining the specific feature of a discrete covariant
theory. This discretisation of general relativity is better described in terms of twocomplexes, which bridge between a Yang-Mills lattice and a Regge triangulation.
We introduce here this discretization of general relativity, which we use extensively in the following.
In this Chapter we give the basic construction in three spacetime dimensions,
and we restrict to the euclidean theory for simplicity, namely the theory presented
in Section 3.5. The discretization of the physical four-dimensional Lorentzian theory (which is a straightforward generalisation) is given later in Chapter 7.
The key notion we need is the dual of a triangulation, illustrated in the picture on the right,
where the tetrahedron belongs to the original triangulation, while the grey faces meeting
along edges, in turn meeting at points, form
the dual. More precisely, let be a 3d triangulation. The dual of the triangulation
is defined as follows. It is obtained by placing
a vertex within each tetrahedron, joining the
vertices of two adjacent tetrahedra by an edge
dual to the triangle that separates the two
tetrahedra, and associating a face to each segment of the triangulation, bounded
by the edges that surround the segment. These objects inherit an orientation from
their dual.
The set of vertices, edges and faces, with their boundary relations, is called a
two-complex. A two-complex can be visualized by a set of faces meeting at edges,
which in turn meet at vertices. What defines the two-complex is the combinatorial
structure of the boundary relations between these elements.
Boundary
The discretization of the bulk of a spacetime region R induces a discretization
of the boundary = R of this region. The boundary is discretized by the
boundary triangles of , separated by the boundary segments of . The end points
of the edges dual to these triangles are called nodes, or boundary vertices, and the
boundaries of the faces dual to these segments are called links, or boundary edges,
93
boundary graph
face
face
edge
vertex
edg
e
face
node
Figure 4.2
link
link
Triangulation and two-complex terminology and notation in 3d.

and together they form the graph of the boundary. The boundary graph is at the
same time the boundary of the two-complex and the dual of the boundary of the
triangulation:
= ( ) = () .
(4.27)
This terminology is important. It is summarised in Figure 4.2 and in the Tables

1.2 and 1.3. Study this carefully because we will use this terminology and notation
extensively.
We now use this construction to define a discretization of euclidean general relativity in three euclidean dimensions, namely the theory defined in Section 3.5,
where, recall, the gravitational field is described by a triad field ei = eia dx a and a
SO(3) connection i j = ai j dx a , where indices a, b... = 1, 2, 3 are spacetime indices
and i, j = 1, 2, 3 are internal indices raised and lowered with the Kronecker delta
ij . To do so, we introduce discrete variables on the two-complex.
Table 4.2 Bulk terminology and notation

Bulk triangulation
tetrahedron ()
triangle (t)
segment (s)
point (p)
Two-complex
vertex (v)
edge (e)
face (f)
Table 4.3 Boundary terminology and notation

Boundary triangulation
Boundary graph
triangle (t)
segment (s)
node, (n) (boundary vertex)

link, (`) (boundary edge)
94
Figure 4.3
Group elements are on the edges (red). Algebra elements on the segments (blue), or,
equivalently, in their dual faces
We discretize the connection, as in Yang Mills theory, by using it to assign an
SU(2) group element Ue to each edge e of the two-complex. We discretize the triad
by associating a vector Lis of R3 to each segment s of the original triangulation.
Ue
e
Lis
(4.28)
.
(4.29)
The formal relation between the continuous and the discrete variables can be
taken to be the following
Ue = P exp
SU (2) ,
(4.30)
e i R3 .
(4.31)
e Z
Lis =
Ue is the holonomy of the connection along the edge, namely the matrix of the
parallel transport generated by the connection along the edge, in the fundamental representation of SU(2). This association of a group variable to an edge is the
same as in lattice gauge theory. Lis is the line integral of the one-form ei along the
segment. The action (3.98) can be approximated in terms of these objects.
The definition (4.31) is a bit imprecise, because of the gauge. Under a gauge
transformation (3.99) the group elements U defined in (4.30) transform well,
namely as
1
Ue 7 Rse Ue R
te
(4.32)
r
t`
where se and te are the initial (source) and final (tartarget
get) vertices of the edge e (recall that all the discrete
structures are oriented) e and Rv SO(3). Therefore in
`
the discrete theory the continuous local SO(3) invariance s`
r source
is just reduced to rotations at the vertices. Not so for the
i
2
algebra variables L defined in (4.31). To correct this , assume that this definition is taken in a gauge where the connection is constant along
the segment itself (at the possible price of being distributional at the boundaries
2
There is a more precise mathematical way of doing this, which is to give a gauge equivalent definition of the quantities Li by parallel transporting the triad to the same point; see [Thiemann (2007)].
95

of the segment), as well as along the first half of each edge (edges are oriented).
In this way, also the Lis variables are invariant under all gauge transformations
except those at the vertices; and they transform covariantly, in fact, in the adjoint
representation, under rotation Rv at the vertex v.
The discretization approximates well the continuum theory when the curvature
is small at the scale of the triangulation and the segments are straight lines. Then
notice that the norm of the R3 element Lis associated to the segment s is the length
Ls of the segment
L2s = |~Ls |2 .
(4.33)
Since each face f of the two-complex corresponds to a segment s = sf of the triangulation, we can view Lis as associated to the face: Lfi = Lisf . Also, it is convenient
to view this as an element of the su(2) algebra. That is, define
Lf = Lfi i .
(4.34)
Therefore the variables of the discretized theory are:
An SU(2) group element Ue for each edge e of the two-complex.

An su(2) algebra element Lf for each face f of the two-complex.
In the 4d Lorentzian theory we will have these same variables, associated to edges
and faces.
Action
Given a face f bounded by the edges e1 , ..., en , we can
multiply the group elements Ue around it and obtain
a single group element Uf associated to the face itself.
...
f
Uf = Ue1 ...Uen .
U e2
(4.35)
U en
e1
This is the holonomy of the connection going around
the segment sf dual to the face f. If Uf is different from
the identity Uf 6= 1l, then there is curvature, and we can associate this curvature to
the segment, as in Regge calculus.
Then the discretized action reads
S=
1
Tr( Lf Uf )
8G
f
(4.36)
where Lf su(2) and Uf SU(2). On the boundary, we must close the perimeter of
the faces in order to write the quantity Uf for the faces that end on the boundary,
and write the action. Therefore there must also be group quantities U` associated to
the links of the boundary. In other words, the links of the boundaries, which are the
boundary edges, are on the same footing as the bulk edges, and are assigned group
variables as well. Using the relation (1.39) it is easy to verify that the variation of
96
3D
Figure 4.4
2D
A triangulated region of 3d spacetime and its 2d triangulated boundary.
Lf in the action (4.36) gives Uf = 1l. That is, flatness, which is equivalent to the
continuous Einstein equations in 3d.
The discretization of general relativity described in this section is similar, but
not exactly equivalent, to the Regge discretization. One difference is that it can be
coupled to fermions. A second difference is given by the sign discussed in Section
3.2.1.
Boundary variables
Let us study more in detail the boundary variables of the discrete theory, which
will play an important role in the quantum theory. These are two kinds: the group
elements U` of the boundary edges, namely the links. And the algebra elements Ls
of the boundary segment s. Notice that there is precisely one boundary segment s
per each link `, and the two cross. We can therefore rename Ls as L` , whenever `
is the link crossing the boundary segment s. Therefore the boundary variables are
formed by one couple
( L` , U` ) su(2) SU(2)
per each link ` of the graph. Since
the algebra is the cotangent space
of the group, we can write su(2)
SU(2) T SU(2). Thus the classical boundary phase space of the
discretized theory is T SU(2) L ,
where L is the number of links
of the graph. This is precisely the
same boundary phase space as for
an SU(2) Yang Mills theory on a
lattice. The Poisson brackets are
the same as in the previous section:
(4.37)
97
{U` , U`0 } = 0 ,
{U`0 , Li` } = (8G ) ``0 U` i ,

j
{ Li` , L`0 } = (8G ) ``0 eij k Lk` ,
(4.38)
namely the Poisson brackets (4.9) of an SU(2) Yang Mills theory on the lattice.3 The
dimensional constant 8G is determined by the action, because the momentum
conjugate to the connection (which has dimensionless Poisson brackets with the
connection) is not the triad, but the triad divided by 8G.
We shall discuss the 4d version of this construction in Chapter 7. For the 3d
theory, we have now all the ingredients to move to the quantum theory, which we
do in the next Chapter.
With the gauge invariant definition of Li` mentioned in the previous footnote, these Poisson brackets
can be derived from the continuous ones (3.102).
98
4.5 Complements
4.5.1 Holonomy
Given a connection A in a group G on a manifold, and a path in the manifold, the path
ordered exponential
U = P e
(4.39)
is defined as value for s = 1 of the solution of the differential equation

d
U (s) = a (s) A a ((s))
ds
(4.40)
with initial condition U (0) = 1l. Here s [0, 1] is an arbitrary parameter along the path
and a = d a /ds is the tangent of at s.
This is also the limit of large N of
UN =
(1l + a (xn ) Aa (xn ) )

n
(4.41)
where xn are N ordered points along , at distance .

Geometrically, U G is the group transformation obtained by following the path and
rotating as dictated by the connection A. In other words, it is the rotation that undergoes a
vector which is covariantly constant along .
The reason for the notation with P (path ordered) is that if we expand the exponential in
the right hand side of (4.39) in a power series in A and order each multiple integral by the
order along , we obtain the left hand side. That is, we can also write
U = 1l +
where
P
A=2
Z 1
0
ds
A+
Z s
0
1
P
2
A + ...
(4.42)
dt b (s) Ab ((s)) a (t) A a ((t))
(4.43)
(A does not commute with itself) and so on. The group element U is often called holonomy in the quantum gravity literature, although the relation with other uses of holonomy is only partial. If A is the Levi-Civita connection, its holonomy gives the standard
parallel transport of vectors in general relativity. If A is the self-dual Ashtekar connection,
its holonomy gives the parallel transport of a left handed neutrino along .
The key property of the holonomy is that it transforms nicely under gauge transformations: it is insensitive to gauge transformations localized in the bulk of and transforms
homogeneously for gauge transformations on the end points of . That is, if we gaugetransforms the connection by a gauge transformation with parameter ( x ), the holonomy
transforms as
U ((0)) U 1 ((1)).
(4.44)
Finally, important properties of the holonomy are its its variations under arbitrary variation of the connection or . A straightforward calculation gives for the first
U =
Aia ( x )
ds 3 ( x, (s)) a (s) U1 i U2
(4.45)
where 1 and 2 are the two portions in which x cuts . And for the second
U = U1 Fab ((s)) b (s)U2 .

a (s)
(4.46)
Complements
99
In particular, an infinitesimal change of parametrization of s f (s) where f (s) vanishes at

the end points gives
Z
ds f (s) a (s)
U =
a (s)
ds f (s) U1 Fab ((s)) a (s) b (s)U2
(4.47)
which vanishes because of the contraction between the antisymmetric Fab with the symmetric a b .
4.5.2 Problems
1. Equations of motion: Compute the equations of motion of the discretisation of the covariant formulation of a Newtonian system. Is energy conserved?
2. Important, do this well: Consider a tetrahedron. Place a point P in its interior and connect P with the four vertices of the tetrahedron. Let the resulting ten segments determine
a Regge space. Carefully describe the corresponding triangulation and its dual , listing all vertices, edges and faces, distinguishing those that are similar or are of different
kind. Write the boundary graph. What is the relation between the boundary graph and
the original tetrahedron? Consider then the group elements associated to all the edges
of . Find a good notation to label these. How many conditions must these satisfy in
order for the connection to be flat? (careful: there is a bubble...)
3. Gluing two complexes: Consider two spacetime region connected by a common portion
of their boundary. Let both of them be triangulated, matching at the boundary. Describe
how the dual triangulations match at the boundary. What happens to nodes and links?
Build a concrete example, giving all the vertices, edges, faces, nodes and links explicitly.
4. Holonomy: Show that the various definitions of the holonomy given above are equivalent. Derive (4.45), (4.46) and (4.47) from the formal definition of the holonomy.
5. Curvature: (difficult) What is the relation between the Regge curvature defined by the
deficit angle h and the curvature defined by Uf ? How are these two quantities related?
Part II
3D THEORY
3d Euclidean theory
In order to find a quantum theory for the gravitational field, we can apply the general structure devised in Chapter 2 to the theory defined in Chapter 3, discretized
as explained in Chapter 4. Since the resulting theory includes a number of technical complications, in this Chapter we first complete a simpler exercise: apply
this same strategy to euclidean general relativity in three spacetime dimensions,
recalled in Section 3.5. This exercise allows us to introduce a number of ideas and
techniques that will then be used in the physical case, which is the lorentzian theory in four spacetime dimensions.
There are major differences between general relativity in 3d theory and 4d. The
foremost is that the 3d theory does not have local degrees of freedom: it is therefore
infinitely simpler than the 4d theory. This is reflected in properties of the quantum
theory that do not hold in the 4d theory. In spite of these differences, the exercise
with the 3d theory is illuminating and provides insight into the working of quantum gravity, because in spite of its great simplicity the 3d theory is a generally
covariant quantum theory of geometry.
5.1 Quantization strategy

To define the quantum theory we need two ingredients:
A boundary Hilbert space that describes the quantum states of the boundary
geometry.
The transition amplitude for these boundary states. In the small h limit the transition amplitude must reproduce the exponential of the Hamilton function.
To construct the Hilbert space and the transition amplitude, we proceed in steps.
First we discretize the classical theory. Then we study the quantum theory that
corresponds to the discretized theory. Finally we discuss the continuum limit. We
use the discretization of the theory both to construct its Hilbert space, following
the ideas sketched in Chapter 1, and to define the transition amplitudes, following
the strategy sketched in Chapter 2.
Recall from the previous chapter that 3d general relativity can be truncated on
a two-complex by associating an SU(2) group element Ue to each edge e and an
su(2) algebra element Lf to each face f of the dual triangulation.
103
104
3d Euclidean theory
Ue ,
e Lf .
(5.1)
(5.2)
The first is interpreted as the holonomy of gravitational connection along the edge,
the second as the line integral of the triad along the segment dual to the face f. The
action is given in (4.36), namely
S=
1
Tr( Lf Uf ) ,
8G
f
(5.3)
where Uf is the product of the Ue of the edges around f. Gauge transformations

act as
1
Ue se Ue
te .
(5.4)
The two-complex is bounded by a graph where the boundary variables are a

group element and an algebra element
(U` , L` )
(5.5)
per each link ` of the graph . These coordinatize the phase space of the discrete
theory, which is the cotangent space to SU(2) L .
This is the discrete theory. Let us move to the quantum theory.
5.2 Quantum kinematics: Hilbert space

The quantization of a phase space which is the cotangent space to a group is standard. We are seeking operators U` and L` realizing the quantum version of (4.38),
and in particular
[U` , Li`0 ] = i (8h G ) ``0 U` i .
(5.6)
on a Hilbert space. This is a basis requirement for the theory to be constructed to

have the correct classical limit. Consider the group elements U` to be the coordinates and the algebra elements L` to be momenta. Let the Hilbert space be the
space of square integrable functions of the coordinates:
h
i
H = L2 SU(2) L .
(5.7)
The index reminds that this comes from a discretization that induces the graph
on the boundary. States are wave functions (U` ) of L group elements U` . The
scalar product compatible with the SU(2) structure is given by the group-invariant
measure, which is the Haar measure dU (see Complements to Chapter 1)
h|i =
SU(2) L
dU` (U` )(U` ) .
(5.8)
Quantum kinematics: Hilbert space
105
The (self-adjoint) natural SU(2)-covariant derivative operators on the functions

(U` ) is the left invariant vector field. This is defined as

d
ti
i
(5.9)
( J )(U ) i (Ue )
dt
t =0
To get a triad operator satisfying (5.6) it sufficient to scale this with the appropriate
dimensionfull factor:
~L` (8h G ) ~J`
(5.10)
where ~J` is the left invariant vector field acting on the argument U` . This realizes
the commutation relations (5.6).

Summarizing: we have the Hilbert space H = L2 SU(2) L , where L is the
number of links of , and operators associated with the discrete variables U` and
L` : U` is a multiplicative operator and the triad is ~L` = 8h G ~J` .
This is enough to start doing physics!
5.2.1 Length quantization

Recall that the length Ls of a segment is given by the norm Ls = |~Lf | where f is the
face dual to the segment. Therefore length of a boundary segment s is given by the
operator L` = |~L` | where ` is the link crossing s. Let us study the spectral properties of this operator. Since ~J is a generator of SU(2), its square |~J |2 is the SU(2)
Casimir, namely the same mathematical quantity as the total angular-momentum
operator; its eigenvalues are j( j + 1) for half integer js. Therefore the spectrum of
the length of a segment ` is1
q
L j` = 8h G j` ( j` + 1)
(5.11)
for half-integers j` . This is the first physical consequence of the 3d theory: length is
quantized.
Do not confuse the classical discreteness introduced by the triangulation with
the quantum discreteness determined by the fact that the length operators have
discrete spectrum. The first is classical, and is just a truncation of the degrees of
freedom. It is the analog of expanding an electromagnetic field in a box in modes,
and truncating the theory to a finite number of modes. The second is quantum
mechanical. It is the analog of the fact that the energy of each mode of the electromagnetic field is quantized: namely the field is made up by quanta.
In particular, the quantum discreteness that we find is not related to the discretization introduced in other tentative approaches to quantum gravity, such as
quantum Regge calculus or dynamical triangulations. In that context, spacetime
is discretized by means of a triangulation of fixed size, but this size is considered
unphysical and scaled down to zero in the continuum limit. Not so here. Here the
physical minimal (non-vanishing) size of a triangle is determined by h and G.
1
In 2d, h G has the dimensions of a length. This follows easily from the form of the action.
3d Euclidean theory
106
The quantum discreteness we find is a consequence of the compactness of SU(2):

the length is conjugate to the group variables and the corresponding operator is a
Laplace operator on a compact space, SU(2) and therefore has discrete spectrum.
It is this discreteness that will reappear also in 4 space-time dimensions and is the
core of the physical granularity of space.
Since length operators associated to a different (boundary) links commute (because each acts on different U` s), we can diagonalize all of them together. Therefore the j` s are good quantum numbers for labeling the kinematical boundary
states of the theory. Are there other quantum numbers? It is time to introduce the
basic structure of loop quantum gravity: the spin network basis. This is done in
the following section.
5.2.2 Spin networks

Gauge
We are not done with the definition of the boundary Hilbert space. The theory
must be invariant under the SU(2) gauge transformations (5.4). We have not yet
taken this local invariance into account. (In Maxwells theory, if we fail taking into
account gauge invariance and the Gauss law, we define three photon polarizations
instead of two.) The gauge invariant states must satisfy
1
(U` ) = (s` U`
t` )
where n SU(2) .
(5.12)
Equivalently,
~ n = 0
C
(5.13)
~ n is the generator of SU(2) transfor every node n of the boundary graph, where C
formations at the node n:
~ n = ~L` + ~L` + ~L` = 0;
C
2
3
1
(5.14)
`1 , `2 and `3 are the three links emerging from the node n. This relation is called
the closure constraint or gauge constraint or Gauss constraint.
This equation has an important direct geometrical in`1
terpretation: consider a triangle of the boundary, bounded
`2
by the three segments s1 , s2 , s3 . These are dual to the three
n
links `1 , `2 , `3 that meet at the node n at the center of the triangle. We have seen that ~L` can be interpreted as the vector
`3
describing the side of the triangle. Then (5.14) is precisely
the relation indicating that the triangle closes!
The subspace of H where (5.13) is verified is a proper subspace (because SU(2)
is compact), which we call K . We write it as
h
i
K = L2 SU(2) L /SU(2) N
(5.15)
Here L is the number of links and N the number of nodes nodes of and the
107
subscript indicates that the pattern of the SU(2) N gauge transformations (5.4)
on the SU(2) L variables is dictated by the structure of the graph. Let us study the
structure of this space.
The length operators L` are gauge invariant (~L` transforms as a vector). Therefore they are well defined on the gauge invariant Hilbert space K . Remarkably,
they form a complete commuting set, as we show below. That is, a basis of K is
given by the normalized eigenvectors of these operators, which we indicate | j` i.
An element of this basis is therefore determined by assigning a spin j` to each link
l of the graph. A graph with a spin assigned to each link is called a spin network.
We say that the links of the graph of a spin network are colored by the spin. The
spin network states | j` i form a basis of K . This is called a spin-network basis and
spans the quantum states of the geometry. Let us show this in detail.
Spin-network basis
Let us show that the quantum numbers j` are sufficient to label a basis of K . We
do so in this paragraph in a rather abstract manner. The following paragraph will
give a more concrete explicit construction of these states.
j
The Peter-Weyl theorem states that the matrix elements Dmn (U ) of the Wigner
matrices (see Complements of Chapter 1), seen as functions of SU(2), are orthogonal with respect to the scalar product defined by the Haar measure. Specifically:
Z
j0
dU Dm0 n0 (U ) Dmn (U ) =
1 jj0
mm0 nn0 .
dj
(5.16)
where d j = 2j + 1 is the dimension of the representation j. Therefore the Wigner

matrices form an orthogonal basis (up to the factor d1 this is an orthonormal basis),
j
in the Hilbert space H = L2 [SU(2), dU ], where dU is the invariant measure.

In other words, the Hilbert space L2 [SU(2)] can be decomposed into a sum of finite dimensional subspaces of
spin j, spanned by the basis states formed by the matrix

r
`
elements of the Wigner matrices D j (U ). The dimension of
Hj
r
H
j
j
these spaces is (2j + 1)2 which is the number of entries of
j
the matrix D (U ). This matrix is a map from the Hilbert
space H j of the spin-j representation to itself. Any such
map can be viewed as an element of H j H j . Therefore
the Peter-Weyl theorem states that
L2 [SU(2)] = j (H j H j )
(5.17)
This is nothing else than another way of saying that there is a basis labelled by
spins and magnetic numbers whose range is determined by the spins.
In our case we have many such spaces, therefore the boundary state space has
the structure

L2 [SU(2) L ] = ` j (H j H j ) = j` ` (H j H j )
(5.18)
3d Euclidean theory
108
The two Hilbert spaces associated to a link naturally belong to the two ends of the
link, because each transforms according to the gauge transformations at one end.
Let us therefore group together the Hilbert spaces next to
the same node, which transform all together under a gauge

j1 j2 H
j2
transformation. We obtain
H j1 r
n
L2 [SU(2) L ] = j` n (H j H j0 H j00 )
(5.19)
j3 H j3
where j, j0 , j00 are the spins coming out from the node n.
Next, we want the space of gauge invariants states. For
this we should restrict to the invariant part of any set of spaces transforming at
the same node. We obtain

L2 [SU(2) L /SU(2) N ] = j` n InvSU(2) H j1 H j2 H j3 .
(5.20)

But InvSU(2) H j1 H j2 H j3 does not exists unless the sum of three spins is integer and the three spins satisfy2
| j1 j2 | < j3 < j1 + j2 .
(5.21)
This relation is called the triangular inequality or the Mandelstam inequality. This
is a condition for the existence of the state. If it satisfied, the invariant space is
one-dimensional:

(5.22)
InvSU(2) H j1 H j2 H j3 = C
Therefore
L2 [SU(2) L /SU(2) N ] = j` C .
(5.23)
where the sum is restricted to the j` that satisfy the triangular relations. Which
is the same as saying that there is a basis | j` i labelled by j` satisfying triangular
inequalities.
Recall that the spins are the lengths of the
triangle surrounding the node: the triangular
j2
identities state that the lengths L1 , L2 , L3 of the
j3 | j1 j2 |
side of the triangle satisfy the triangular inj
3 j1 + j2
j3
equalities
j1
| L1 L2 | < L3 < L1 + L2 .
(5.24)
These are precisely the relations satisfied by a geometrical triangle. Geometry
emerges nicely from quantum geometry.
2
Recall that in the quantum theory of the composition of the angular momenta we write the tensor
j +j
product of two representations as a sum over representations in the form H j1 H j2 = j1=| j 2 j | H j .
1 2

j +j
j+ j
Therefore H j1 H j2 H j3 = j1=| j 2 j | H j H j3 = k=|3j j | Hk . This shows that the invariant
1
space, namely the spin zero representation can enter only once in the tensor product of three representations, and it does if the triangular inequalities are satisfied.
109
A generic quantum state in loop quantum gravity is a superposition of spinnetwork states
|i = C j` | j` i .
(5.25)
j`
Summarizing, the spin network states | j` i:
are an eigenbasis of all length operators,

span the gauge invariant Hilbert space, and
have a simple geometric interpretation: they just say how long are the boundary
links.
Harmonic analysis on the group.

Let us now write the states | j` i in the (U` ) representation, namely compute the
spin-network wave functions
j` (U` ) = h U` | j` i .
(5.26)
In other words, we solve explicitly the eigenvalue equation for the length operators L`
L` j` (U` ) = L j` j` (U` ).
(5.27)
in the group representation.

The Peter-Weyl theorem provides a notion of Fourier transform on the group.
A state (U ) can be written as a liner combination of states in the Wigner-matrix
basis:
(U ) =
C jmn Dmn (U ).
(5.28)
jmn
In our case the states are functions of L group elements (U` ), therefore we can
write them as
(U` ) =
j1 . . . jL
m1 . . . m L
n1 . . . n L
C j1 ...jL m1 ...m L n1 ...n L Dm1 1 n1 (U`1 ) . . . DmLL n L (U` L )
(5.29)
A state invariant under SU(2) at the nodes must be invariant if we act with a
transformation n at the node n. This acts on the three group elements of the three
links that meet at the node. Since the Wigner matrices are representation matrices,
the gauge transformation acts on the three corresponding indices. Therefore for
the state to be invariant C j1 ...jL m1 ...m L n1 ...n L must be invariant when acted upon by
a group transformation on the three indices corresponding to the same node. (Up
to normalization) there exists only one invariant object with three indices in three
SU(2) representations: it is called the Wigner 3j-symbol and written as

j1
j2
j3
m1 m2 m3
.
(5.30)
=
m1 m2 m3
3d Euclidean theory
110
See [Haggard (2011)] for a good introduction. That is, any invariant state in the
triple tensor product of the three representations is proportional to the normalized
state

j1
j2
j3
|i i =
| j1 , m1 i | j2 , m2 i | j3 , m3 i.
(5.31)
m1 m2 m3
m1 m2 m3
The 3j-symbols are the symmetric form of the Clebsh-Gordon coefficients; they are
described for instance in Chapter 14 of Landau [Landau and Lifshitz (1959)]; they
are explicitly given by Mathematica (ThreeJSymbol[{ j1 , m1 }, { j2 , m2 }, { j3 , m3 }]).
In the quantum gravity literature the Wigner 3j-symbols are usually called trivalent intertwiners.
Therefore the gauge invariant states must have the form
(U` ) =
j1 ...jL
1 m2 m3
C j1 ...jL m
. . . NL2
1
m L 1 m L
Dm1 1 n1 (U`1 ) . . . DmLL n L (U` L )
(5.32)
where all indices are contracted between the intertwiner and the Wigner matrices
D. How these indices are contracted is dictated by the structure of the graph .
There is one matrix D for each link ` and a 3j-symbol for each node n. The indices
of a node are contracted with the indices of the adjacent links.
Therefore a generic gauge invariant state is a linear combination
(U` ) =
j`
(5.33)
C j` j` (U` )
of the orthogonal states

m m2 m3
j` (U` ) = 1 1
. . . NL2
m L 1 m L
Dm1 1 n1 (U`1 ) . . . DmLL n L (U` L )
(5.34)
labelled by a spin associated to each link (do not confuse the indices n the nodes
n). These are the spin-network wave functions. We write them also in the simpler
form
j` (U` ) = h U` | j` i =
O
n
D j` (U` ).
(5.35)
These are the 3d spin network states in the group representation.
5.3 Quantum dynamics: Transition amplitudes

The transition amplitude is a function of the boundary states. We can consider a
compact region of spacetime bounded by a boundary formed by two disconnected
components, taken as past and the future boundaries. But we can also consider a
general boundary state [Oeckl (2003, 2008)], possibly connected, and define the
transition amplitude associated to any state on this boundary. This defines an amplitude for a process characterised by the given state on the boundary. That is, an
amplitude associated to a given set of boundary values of partial observables.
Quantum dynamics: Transition amplitudes
111
Fix a triangulation of spacetime. The transition amplitude is a function of the

states defined on = () , the boundary graph. Like the Hamilton function, the
transition amplitude is a function either of the coordinates or the momenta. We
write the two forms as
W (U` ) = hW |U` i
(5.36)
W ( j` ) = hW | j` i.
(5.37)
and
The subscript indicates that this is the amplitude computed on the discretization
. Notice that in the second case, where the momenta L` have discrete spectra,
W is a function of the quantum numbers, not the classical variables. This has been
discussed in detail in Section 2.2.2. The relation between the two expressions of the
transition amplitude is easy to find, because we know how to express the states | j` i
in the (generalized) basis |U` i and viceversa: the transition matrix is given by the
spin network states (5.34).
To compute the transition amplitude W of the theory discretised on the twocomplex dual to a triangulation , we use the Feynman path integral. The amplitude is given by the integral over all classical configurations weighted by the
exponential of the classical action. That is
W (U` ) = N
dUe
dLf e 8hG f Tr[Uf Lf ]
(5.38)
N is a normalization factor, where we will absorb various constant contributions.

The integral over the momenta is easy to perform, since it is an integral of an
exponential which gives a delta function. Therefore we obtain
W (U` ) = N
dUe (Uf )
(5.39)
where the delta function is over SU(2). To compute this integral, we expand the
delta function over the group in representations, using3
(U ) =
d j Tr D ( j) (U ).
(5.40)
Recall that d j is the dimension of the representation d j = 2j + 1. A group element

U SU(2) determines a rotation in SO(3): the quantity Tr D ( j) (U ) is the character
of the representation and depends on the angle but not on the rotation axis.
Using this, we can write
3
This is the SU(2) analog of the well known expansion of the delta over the circle U(1): () =
1
in , where n labels the irreducible representations of U(1), which are all one dimensional
2 n e
and given by the exponential. Similar formulas hold for any compact group.
3d Euclidean theory
112
Figure 5.1
Tetrahedra, vertices, edges, faces, and links, and their relations.
W (U` ) = N
=N
dUe
d j Tr D (Uf )
f
d jf
jf
!Z
dUe
Tr
f
D jf (U1f ) . . . D jf (Unf )
(5.41)

(5.42)
In the expression above we have made use of the fact that the representation of a
product is the product of representation, i.e. D (UV ) = D (U ) D (V ) and Uf is the
product of the group elements associated to the links 1f, ..., nf around the face f.
Consider one edge e and the corresponding
integral dUe . Observe that each edge e bounds
precisely three faces (because the edge is dual
j1
to a triangle which is bounded by three segj2
ments). Therefore each dUe integral is of the
j3
form
Z
j`
j`
j`
dU Dm11 n1 (U ) Dm22 n2 (U ) Dm33 n3 (U ) . (5.43)
This integral is easy to perform. Since the Haar measure is invariant on both sides,
the result must be an invariant in both set of indices, and, as we have seen, there is
only one such object. The normalization can be computed by considering contractions. This gives
Z
dU Dm1 1 n1 (U ) Dm2 2 n2 (U ) Dm3 3 n3 (U ) = m1 m2 m3 n1 n2 n3

j1
j2
j3
j1 j2 j3
=
.
m1 m2 m3
n1 n2 n3
(5.44)
Therefore the result of the integrals in (5.41) is a bunch of 3j-symbols contracted

among themselves. Let us see what is the precise pattern of the contractions. Each
edge produces two 3j-symbols, which we can view as located at the two ends of
the edge, because they have indices that are contracted at that end. The contractions happens therefore at each vertex. At each vertex, there are four edges, and
113
therefore four 3j-symbols, contracted among themselves. The contraction must be

an SU(2) invariant because the entire expression is invariant and therefore must
be the one obtained by the invariant contraction of magnetic indices, given by
the matrices computed for instance in Chapter 14 of Landau [Landau and Lifshitz
(1959)], or in [Haggard (2011)]

p
j j 0
gmn = 2j + 1
= m,n (1) jm .
(5.45)
m n 0
(The m,n is easily understood as conservation of the third component of the angular momentum. The sign originates from tensoring the invariant object e AB .)
The invariant contraction of four 3j-symbols gives an object called Wigner 6jsymbol (or Racah W-coefficient), indicated as follows:

j1
j4
j2
j5
j3
j6
gm a n a
m a ,n a a=1
j2
n2
j4
n4

j2
j3
m2 m3

j6
j3 j5
m6
n3 n5
j1
m1
j1
n1
j6
n6
j4
m4

j5
m5

(5.46)
or, more explicitly

j1 j2 j3
j1
j2
j3
j1
j4
j5
JM
= (1)
m1 m2 m3
j4 j5 j6
m1 m4 m5
ma

j2
j4
j6
j3
j5
j6
(5.47)
m2 m4 m6
m3 m5 m6
where M = 6a=1 m a and J = 6a=1 ja . A good detailed introduction to this algebra, with all conventions carefully fixed is [Aquilanti et al. (2010)]. The Wigner
6j-symbols are largely used in nuclear physics, molecular physics, quantum information, etc. See Chapter 14 of [Landau and Lifshitz (1959)]. They are given in
Mathematica as SixJSymbol[{ j1 , j2 , j3 }, { j4 , j5 , j6 }].
Notice that the path of contractions reproduces the structure of a tetrahedron ,
where the 3j-symbols are the four vertices of the tetrahedron and the spins are on
the six sides. The three spins in the upper row (j1 , j2 , j3 ) form a triangle; the three
spins in the lower row (j4 , j5 , j6 ) join at the vertex opposite to this triangle; and each
column is formed by spins that do not meet.
Do not confuse this tetrahedron with the tetrahedron v of the triangulation
dual to the vertex v we are considering. and v are dual to one another: the
vertices of are in the center of the faces of v and the sides of cross the sides
of v . Rather, notice, since this is important, that the tetrahedron is the boundary
graph of the triangulation formed by the single tetrahedron v . See Figure 5.2.
After integrating over all internal-edge group variables, the group variables of
the boundary edges remain. We can integrate these as well contracting with a
boundary spin network state, obtaining (carefully keeping trace of the signs [Barrett and Naish-Guzman (2009)])
114
3d Euclidean theory
Figure 5.2
The tetrahedron associated to the 6j-symbol is the boundary graph of a tetrahedron

v of the triangulation.
Figure 5.3
A vertex of the two complex and its six adjacent faces.

W ( j` ) = N
(1) jf d jf (1) Jv {6j} .

jf
(5.48)
where the sum is over the association of a spin to each face, respecting the triangular inequalities at all edges, Jv = 6a=1 ja and ja are the spins of the faces adjacent to
the vertex v. (A vertex of the two-complex is adjacent to six faces, see Figure 5.3.)
We have indicated that the normalization factor may depend on the triangulation.
This is the transition amplitude in the spin representation.
The expression (5.48) has first been written by Ponzano and Regge in the sixties [Ponzano and Regge (1968)]. Remarkably, Ponzano and Regge did not know
that the spectrum of the length is discrete, but simply guessed that the length
had to be discrete and determined by spins. The physical meaning of their ansatz
became only clear with loop quantum gravity. The connection between loop quantum gravity the Ponzano Regge model and was pointed out in [Rovelli (1993b)].
5.3.1 Properties of the amplitude

The following basic properties characterise the amplitude.
1. Superposition principle. This is the basic principle of quantum mechanics. The
amplitudes to be given by sum of elementary amplitudes (in standard quantum
115

theory, this sum is expressed by Feynmans sum over paths .)
hW | i = W ( )
(5.49)
2. Locality. This a fundamental discovery of XIX and XX century physics: interactions are local in spacetime. Therefore the elementary amplitudes can be seen
as products of amplitudes associated to spacetime points (in standard quantum field theory, this product is expressed as the exponential of an integral over
spacetime points).
W ( ) Wv .
(5.50)
v
3. Local euclidean invariance. In the physical 4d theory, this criterium will translate into the local Lorentz invariance of general relativity. Here, it fixes the SU(2)
invariance of the amplitude. We show in 6.1 at the beginning of the next Chapter that the vertex amplitude, namely the 6j-symbol can be written simply as
the projection on the SU(2) invariant part of the state on the boundary graph of
the vertex, that is
Wv = ( PSU(2) v )(1l).
(5.51)
where is a boundary state, PSU(2) is the projector on its locally invariant SU(2)
component and (1l) indicates the evaluation of the spin network state on the
identity U` = 1l.
These fundamental properties shall guide us to write the amplitudes in 4d dimensions.
5.3.2 Ponzano-Regge model

The expression (5.48) seems a bit magic. It is a very simple expression written in
terms of basic objects in the representation theory of SU(2). How can it possibly
have anything to do with general relativity?
The connection with the classical theory must be in a classical limit. In quantum mechanics, a classical limit is given by large quantum numbers, where the
quantum discreteness becomes irrelevant. Let us therefore study this expression
for large js.
In their original paper Ponzano and Regge have proven the following remarkable result. Consider a flat geometrical tetrahedron. Let its geometry be determined by the six lengths of its sides: L1 , ..., L6 . Associate six spins j1 , ..., j6 to these
lengths so that L a = ja + 12 . Let V be the volume of this tetrahedron, and S be the
Regge action of this tetrahedron. Then Ponzano and Regge have given evidence
that in the large spin limit

1

{6j}
cos S +
(5.52)
4
j
12V
This was then proven rigorously only in 1998 by Roberts [Roberts (1999)]. The
consequences of this asymptotic relation are important. The cosine is the sum of
3d Euclidean theory
116
p
L
Figure 5.4
If surface L folds over configuration space C , the relative phase of the saddle point
approximation of the path integral between two extrema includes a 4 , from a
nontrivial Maslov index.
two terms and the 4 factor can be taken out of the cosine therefore the formula
above can be written as
{6j}
1
1
eiS +
eiS ,
2 12iV
2 12iV
(5.53)
If we regard only large spins, we can disregard quantum discreteness and the sum
over spins is approximated by an integral over lengths in a Regge geometry. The
integrand is given by exponentials of the action. This is a discretization of a path
integral over geometries of the exponential of the Einstein-Hilbert action. Therefore (5.48) is precisely a concrete implementation of the path-integral sum over
geometries
Z
D [ g] e h
gR
(5.54)
formal definition of quantum gravity.

The fact that the sum has two terms with opposite sign of the action follows from
the discussion in Section 3.2.1: since we are quantizing the triad theory and not
the metric theory (recall that this is necessary because of the existence of fermions)
there are two distinct triad configurations for each metric configuration, at every
vertex. They have opposite sign and they give rise to the two terms above. The
4 term, which gives rise to an extra phase difference between the two terms is
also well understood in saddle point approximations: this phase is called Maslov
index and always appear when there is a flip of sign in the momentum between
two saddle points, namely a folding of the surface L of Littlejohn, discussed in
Section 2.5.2, over configuration space. This is discussed in detail in [Littlejohn
(1992)].
There is one last point to be discussed, to define the theory: the continuum limit.
What happens if we refine the discretization. Let us discuss this in two steps. First,
let us refine the triangulation by keeping the boundary graph fixed. Ponzano and
117

Regge proved in their paper that choosing the normalization factor to be
N = w p
(5.55)
where w is a number and p is the number of points in the triangulation, the transition amplitude does not change under a change of triangulation, up to possible
divergent terms that may appear in the refinement. Formally, divergent terms can
be reabsorbed into w as follows. First, correct the theory with a cut off, by limiting
the sum over each spin to a maximum value . Next, take w of the form
w=
wo
(5.56)
Then it is possible to prove that the resulting expression does not depend on the
triangulation in the limit . That is, the amplitude is defined by
W = lim
w p
0
(1) jf d jf (1) J {6j} .
jf =0
(5.57)
Second, what happens if we refine the boundary graph? Also in this case the
refinement has no physical effect. This can be seen as follows. Consider the transition amplitude for a boundary formed by two disconnected equal graphs . The
transition amplitude defines a map from the Hilbert space H to itself. This map
is a projector that projects out from H the physical Hilbert space. It is possible to
prove that the resulting physical Hilbert space does not depend on . Both these
properties are strictly connected to the fact that the corresponding classical theory
has no local degrees of freedom and should not be expected in 4d. Therefore we
will not insist on this here, since they are of limited interest for the physical theory
of quantum gravity.
When Regge and Ponzano found this model for quantum gravity in 3d, there
was great excitement in Princeton, where Regge was at the time. The asymptotic
result (5.52) is truly amazing: general relativity is coded in simple SU(2) representation theory combinatorics. The action of GR is just hidden in a Wigner 6j-symbol!
The idea of trying to repeat the construction in 4d was put forward immediately.
But the project turned out to be more difficult than expected. The 4d version of the
model had to wait almost half a century. It will be discussed in Chapter 7.
118
3d Euclidean theory
5.4 Complements
5.4.1 Elementary harmonic analysis
We collect here a few basic facts about the space L2 [SU (2)], which plays a major role in
quantum gravity. For a mathematically complete presentation of harmonic analysis on compact groups, see for instance [Folland (1995)].
It is good to keep in mind as reference a simple and well known example of functions on
a compact group: the space L2 [U (1)] of the square integrable functions ( ) of an angular
variable. This is for instance the space of the quantum states of a particle on a circle. The
scalar product is
(, ) =
1
2
Z 2
0
d ( )( ).
(5.58)
The space of these functions carries a representation of the group U (1) which acts simply
as ( ) ( + ), for U (1). A basis of functions orthogonal with respect to this scalar
product is given by the functions n ( ) = ein with integer n. This is the (discrete) Fourier
theorem. Notice that each of this function defines a unitary irreducible (one-dimensional)
representation of the abelian group U (1), because n ( + ) = n ( )n (). Therefore the
Fourier theorem states that L2 [U (1)] decomposes into the orthogonal sum of irreducible
representations of U (1). This can be written is the form
L2 [U (1)] = n Cn
(5.59)
where Cn is the complex plane, viewed as a representation of U (1) under the exponential
map n ( ). Also, note that the delta distribution on the circle can be expanded in terms of
this basis simply as
( ) =
1
2
ein .
(5.60)
The main features of this well known example are reproduced by the space L2 [SU (2)] of
functions (U ) over the group SU (2), square integrable with respect to the Haar measure.
The scalar product is
(, ) =
SU (2)
dU (U )(U ).
(5.61)
The space of these functions carries a representation of the group SU (2) which acts simply as (U ) (1 U ), for SU (2). A basis of functions orthogonal with respect to
j
this scalar product is given by the functions jnm (U ) = Dnm (U ), where the Dnm (U ) are
the Wigner matrices. This is the Peter-Weyl theorem. Each of this function is a matrix element of a unitary irreducible representation of the non-abelian group SU (2). Therefore the
Peter-Weyl theorem states that L2 [SU (2)] decomposes in the orthogonal sum of irreducible
representations of SU (2). This can be written is the form
L2 [SU (2)] = j Vj
(5.62)
where the sum is over the half integers j, which label the irreducible representations of
SU (2). The space Vj is the space where a D j matrix lives; it has dimension (2j + 1)2 and
transforms under the spin-j representation on either index. It can therefore be written as
H j H j where H j is the spin-j representation of SU (2). Thus
L2 [SU (2)] = j (H j H j ).
(5.63)
Complements
119
...
f
e
hvf
gve
Figure 5.5
e0
gev0
v0
Ue
Splitting the group elements U` on the edges in two.

Finally, the delta distribution on the group can be expanded in terms of this basis simply as
(U ) =
(2j + 1) Tr[ D j (U )].
(5.64)
5.4.2 Alternative form of the transition amplitude

Here we derive an alternative form of the amplitude, which will turn out to be useful for
generalizing the theory to 4d. Let us start from
Z=
(Ue ...Ue
dUe
).
(5.65)
where we consider the partition function of a triangulation without boundaries for simplic1 is a
ity. Now introduce two variables per each link l, namely Ue = gve gev0 where gev = gve
variable associated to each couple vertex-edge. Then we can write
Z=
dgve
( gve gev0 gv0 e0 ge0 v00 ...).
(5.66)
Now imagine to regroup the gev variables differently, by using hvf = gev gve0 , where e
and e0 are the two edges coming out from the vertex v and bounding the face f. Then clearly
we can rewrite the amplitude as
Z=
dhvf dgve
( gve gev0 gv0 e0 ge0 v0 ...) ( ge0 v gve hvf ).
(5.67)
(hvf hv0 f ...) ( ge0 v gve hvf ).
(5.68)
vf
Or equivalently
Z=
dhvf dgve
vf
This can be reorganized as a transition amplitude where a delta function glues the group
element around each face
Z=
dhvf
(hf ) Av (hvf )
f
(5.69)
3d Euclidean theory
120
where we have defined hf as the product of internal faces
hf =
hvf .
(5.70)
vf
If a face f is bounded by a link ` of the boundary:
hf =
hvf
vf
h` .
(5.71)
The vertex amplitude is then

Av (hvf ) =
( ge0 v gve hvf ).
dgve
(5.72)
The SU(2) integrals in this vertex amplitude are n = 4: one group element per each of the
n (here four) edges coming out of the vertex. This is a bit redundant, because a moment of
reflection shows that if we perform n 1 of these integrals the result is invariant under the
last integration variable. Therefore we can drop one of the integrations without affecting
the result. We write this by putting a prime on the measure
Z
SU(2)n
0
dgve
SU(2)n1
dgve1 ....dgven1 .
(5.73)
This is just cosmetic in 3d, because the volume of SU(2) is just unity, but it will be very
useful in 4d, where we will have to deal with a non compact group with infinite volume.
Expanding the delta function in representations gives
Av (hvf ) =
jf
0
dgve
(2jf + 1)Trj [ ge0 v gve hvf ].

f
(5.74)
Notice that the vertex amplitude is a function of one SU(2) variable per each face around
the vertex. Imagine to draw a small sphere surrounding the vertex: the intersection between
this sphere and the two-complex is a graph v . The vertex amplitude is a function of the
states in
Hv = L2 [SU(2)6 /SU(2)4 ]v
(5.75)
v is the complete graph with four nodes, the boundary graph of the vertex.
In conclusion, and going back in an obvious way to the transition amplitudes, these are
given by
W ( h` ) =
dhvf
0
dgve
(hf ) Av (hvf )
(5.76)
(2jf + 1)Trj [ ge0 v gve hvf ].
(5.77)
where the vertex amplitude is

Av (hvf ) = N
jf
This form of the amplitude will turn out to be the easiest to generalize to 4d.
Complements
121
Figure 5.6
The vertex-graph v
5.4.3 Poisson brackets

Problem
Compute the Poisson brackets (4.9) of L and U from their definition and the continuos
Poisson bracket (3.102).
Solution
From the definitions, we have
{U` , Li`0 } = { Pe
s`0
ei }
(5.78)
This is non vanishing only if ` intersects s`0 , that is, if ` = `0 . In this case (dropping the link
subscript for clarity)
{U, Li } =
s`0
ds { Pe
dtb dldt
, eia
d` a
}.
ds
(5.79)
Let us compute this using the Poisson brackets (3.102).
{U, Li } =
s`0
ds
= 8G
dsà Z
d`b
j
dt
U { (l (t))j , eia (s` (s))}U`
ds `
dt `+ b
s`0
ds
dsà Z
d`b
dt
e 2 (`(s), `(t))U`+ ij j U` .
ds `
dt ab
(5.80)
Here U` are the holonomies of the two halves of the edge cut by the segment. Since this in
invariant under reparametrization and choice of coordinates, we can use coordinates where
the link `, the segment s` and the normal n are orthogonal, and we have immediately the
the integral gives precisely unity. Therefore
{U` , Li`0 } = 8G``0 U`+ i U`
(5.81)
{U` , Li`0 } = 8G``0 U` i
(5.82)
This is a bit disturbing because the right hand side is not written in terms of the original
variables. But recall that the definition of Li was in a particular gauge, where U = 1. Using
this,
122
3d Euclidean theory
5.4.4 Perimeter of the Universe

Problem
Consider euclidean GR in three spacetime dimensions and assume spacetime has the topology of a sphere S 2 (space) times R (time). Approximate this theory by means of a
discretization, where space (namely the two sphere) is discretized by two triangles joined
by their boundaries. Call equator of the sphere the (common) perimeter of the (two) triangles. The graph dual to this triangulation is the the graph with two nodes joined by
three links. Therefore the Hilbert space is the space of functions (U, V, W ), where U, V
and W are SU(2) group elements (with a given orientation of the links). Now, suppose the
quantum state happens to be: (U, V, W ) = Tr (UV ). How long is the equator? What is its
expectation value and what is its quantum spread?
Hint: The quantum spread is zero because this state is an eigenstate of the Perimeter
P = C1 + C2 + C3 , where Ci are the three Casimirs. The mean value is the eigenvalue,
which is the sum of the three spins (one happens to be zero) times 8h G...
Bubbles and cosmological constant

If the reader is in a hurry to get to the physical 4d theory, this Chapter can be
skipped at first reading, without compromising understanding of what follows.
In this Chapter we analyse the quantum theory of 3d euclidean general relativity constructed in Chapter 5 focusing on the divergences that appear in the amplitudes and the way to correct them.
These divergences cannot be identified with the standard ultraviolet (UV) divergences of perturbative quantum general relativity, because UV divergences originate from arbitrary small spacetime regions, while here the theory is cut off at
the Planck scale. Instead, the divergences come from large spacetime regions, and
are, in a sense, infrared (IR), although the distinction is tricky in gravity, where a
large portion of spacetime can be enclosed within a small boundary. Do these IR
divergences spoil the theory?
A remarkable result, which we briefly describe in this Chapter is that these divergences disappear in the theory with a positive cosmological constant. The cosmological constant happens to function in the theory as a natural IR cut off, next
to the UV cut off provided by the Planck length.
This is beautiful, because two open issues in fundamental physics: the presence
of a small cosmological constant, and the existence of quantum field theoretical divergences, resolve one another: the cosmological constant yields to and is required
for the finiteness of the theory.
6.1 Vertex amplitude as gauge-invariant identity

Let us start the analysis of the divergences by better studying the amplitude of
a single vertex. Consider a triangulation formed by a single tetrahedron . The
boundary graph has again the shape of a tetrahedron as in figure 5.2. Therefore
the amplitude is a function of the variables of the links on this graph. Label with
a, b = 1, 2, 3, 4 the nodes of the graph (the faces of the tetrahedron) and denote
1
Uab = Uba
the boundary group elements, which live on the links. Calling and
1
Uab = Uba the boundary group elements, the transition amplitude is a function
W (Uab ).
To write it, we must construct the the two-complex. We put a vertex inside the
tetrahedron, connect it by four edges to the four boundary nodes, and draw the
six faces that connect the vertex to the six boundary links. See Figure 6.1. Using
123
124
Figure 6.1
The spinfoam of a single tetrahedron.

(5.39) (dropping the normalization for simplicity) this is given by
W (Uab ) =
dUa (Ua Uab Ub1 ).
(6.1)
ab
The integrals are easily performed. One of them turns out to be redundant, and
we obtain
W (Uab ) = (U12 U23 U31 )(U13 U34 U41 )(U23 U34 U42 ).
(6.2)
Notice that each sequence of Uab inside the deltas corresponds to an independent closed loop in the boundary graph. The interpretation of this amplitude is
therefore straightforward: the amplitude forces the connection to be flat on the
boundary. Notice that what is flat is the 3d connection, the one that generates 3d
rotations, not the 2d rotations. That is, we can imagine that there is a spacetime reference frame in each face, and these are parallel transported along the boundary
in such a way that any closed loop gives unity. In other words, W (Uab ) is just the
gauge invariant version of ab (Uab ). And in fact, this it what the formula (6.1)
says: it only makes the product of delta functions gauge invariant. The reader may
wonder why the 4th independent loop U12 U24 U41 is missing. But it is immediate
to see that if the connection is flat on the three other loops it is also flat on this loop.
In fact, it is easy to see that
hW | i =
dUab W (Uab ) (Uab ) =
dUa (Ua Ub1 ).
(6.3)
That is, W projects on the flat connections, averaged over the gauge orbits.
Let us now look at the same amplitude in the spin representation. This is given
by

j1 j2 j3
W ( jab ) =
.
(6.4)
j4 j5 j6
If all is consistent, this should be just the transform of the above, where the integral
kernel is given by the spin network states. That is
W ( jab ) =
dUab jab (Uab )W (Uab ).
(6.5)
Bubbles and spikes
125
Figure 6.2
4 .
Let us check this. Inserting the definitions,
W ( jab ) =
dUab
dUa (Ua Uab Ub1 ) a i a D jab (Uab ).
(6.6)
ab
ab
Performing the integral gives

W ( jab ) =
dUa a i a D jab (Ua ) D jab (Ub1 ).

ab
(6.7)
ab
But now the representations matrix elements act precisely on the invariant tensor,
and since these are invariant, they do not do anything, so that what remains is the
contraction of the four invariant tensors,
W ( jab ) = Tr [ a i a ]
(6.8)
namely the Wigner 6j-symbol (6.4).

This shows that the 6j-symbol is nothing else than a Fourier transform of the
gauge invariant delta functions on flat connections, in the Hilbert space associated
to the tetrahedral graph. This can be written in the notation
W ( jab ) = jab (1l).
(6.9)
or, using the the projector PSU(2) on the SU(2) invariant part of a function, the
vertex amplitude Wv can be written as
h v | Wv i = ( PSU(2) v )(1l).
(6.10)
where v is a state in the boundary of the vertex. This is the notation that we have
anticipated in the previous Chapter.
6.2 Bubbles and spikes

Consider again the single tetrahedron , but now add a point P inside it, and join
P to the four point of the original tetrahedron. Consider the six triangles deter-
126
Figure 6.3
4 . The bubble is the inside tetrahedron spanned by the four internal vertices.
mined by P and one of the six sides of the original tetrahedron. These triangles
split the original tetrahedron into four smaller tetrahedra 1 , ..., 4 . This defines a
triangulation 4 formed by four connected tetrahedra. (This step from one tetrahedron to four tetrahedra is called the 1-4 Pachner move.) Let us compute the
amplitude W4 . The boundary of this triangulation is the same as the boundary of
a single tetrahedron. Therefore the boundary variables can still taken to be the six
variables Uab as in the previous Section.
To compute the amplitude, we need to understand the two-complex 4 . This
has four vertices, each joined to the other vertices and to one boundary node. It
has two kinds of internal faces: six boundary faces, and four internal faces, each
bounded by there vertices. A moment of reflection shows that these four internal
faces surround P. Together they have the topology of a sphere. They form an example of a bubble. A bubble is a collection of internal faces that form a surface
with no boundaries. Bubbles are important. See Figure 6.3.
The amplitude is then easy to write. We call Uab the boundary group elements,
Ua and Vab the internal ones, and we have
W4 (Uab ) =
dUa dVab (Uab Ub1 Vba Ua )

ab
(Vab Vbc Vca ).
(6.11)
abc
Integrating, we have easily

W4 (Uab ) = (U12 U23 U31 )(U13 U34 U41 )(U23 U34 U42 )(U12 U24 U41 ).
(6.12)
At first sight, this may seem quite the same thing as the amplitude of a single
tetrahedron that we have computed above:
W1 (Uab ) = (U12 U23 U31 )(U13 U34 U41 )(U23 U34 U42 ) ;
(6.13)
but there is crucial difference: the last delta function. As already noticed, this is
Bubbles and spikes
127
redundant. To see this in detail, let us integrate the amplitude against an arbitrary
boundary function (Uab ). We obtain
hW4 |i =
dUab W4 (Uab )(Uab ) = (1)
dUa (Ua Ub1 )
(6.14)
That is
W4 = (1)W1 .
(6.15)
and (1) is of course an infinite factor.

The appearance of the divergence is a manifestation of the standard quantum
field theory divergences. It is strictly connected to the existence of the bubble. To
see that this is the case, reconsider the same calculation in the spin representation.
In this case,
W4 ( jab ) =
d jab {6j}.
jab ab
(6.16)
In general, in a sum like this the range of summation of the jab is restricted by the
triangular identities. Since the boundary faces have finite spins, the only possibility for an internal face to have a large spin is to be adjacent, at each edge, to at
least one other face with a large spin. In other words, a set of faces with arbitrary
large spins cannot have boundaries. Therefore to have a sum which is not up to a
maximum spin by the triangular identities the only possibility is to have a set of
faces that form a surface without boundaries in the two complex. That is, a bubble.
All this is very similar to the ultraviolet divergences in the Feynman expansion
of a normal quantum field theory, where divergences are associated to loops, because the momentum is conserved at the vertices. Here, divergences are associated
to bubbles, because angular momentum is conserved on the edges. A Feynman
loop is a closed set of lines where arbitrary high momentum can circulate. A spinfoam divergence is a closed set of faces, that can have arbitrarily high spin.
Notice however that in spite of the formal similarity there is an important difference in the physical interpretation of the two kinds of divergences. The Feynman
divergences regards what happens at very small scale. On the contrary, the spinfoam divergences concern large spins, namely large geometries. Therefore they are
not ultraviolet divergences, they are infrared. Let us see this more in detail.
Spikes
Consider a term contributing to the divergence of W4 ( jab ). This is a configuration
of the spins on the two complex where very large spins sit on the internal faces.
What is the corresponding geometry? The spins of the internal faces are the lengths
of the four bones that connect P to the boundary. Therefore the geometries creating
the divergence are geometries where these length are very large.
To get an intuition about these, consider the analogous situation in 2d instead
than in 3d. Imagine to triangulate a plane having small curvature. But imagine
128
Figure 6.4
A spike in a 2d triangulation. What matters is the intrinsic geometry of the 2d surface.
that somewhere on this plane there is a spike: an extremely thin and tall mountain. Choose a triangulation where a point P is on the top of this mountain and
is surrounded by a triangle that surrounds the bottom of the mountain. This is a
situation where the base triangle is small, but the distance of P from this triangle
is large. It is important to emphasize that all this concerns the intrinsic geometry of the plane, and not its extrinsic geometry, which is irrelevant in this context.
Therefore P is still inside the triangle, even if its distance from the boundary of the
triangle is arbitrarily large. There is no sense for it to be outside the triangle,
since what is relevant here is only the geometry of the 2d surface itself. The possibility of having a small triangle including a point at very large distance from its
boundary is just a normal feature of Riemannian space. It is because the space can
be strongly curved.
This example provides the right intuition for the spikes, and therefore the divergences of spinfoams. These are geometries which are strongly curved, and where
there is a region with a small boundary but a large internal volume.1
The astute reader may raise an objection at this point. The classical configurations of 3d gravity have
vanishing curvature. Presumably, in the large quantum number regime, configurations that are very
far from solutions of the equations of motion are suppressed, as one always expects in quantum theory. Why then the weight of the spikes are not suppressed, if these have large curvature? The answer
is subtle: it is because of the second term in the semiclassical expansion (5.53) of the vertex amplitude. If only the first term was present, the divergences would probably not be present, because the
large curvature would produce a rapidly oscillating phase in the integral, that might suppress the
spike contribution. But the second term contributes with the opposite sign to the action. Since there
are four vertices (four tetrahedral), there are four couples of such terms in the amplitude. A detailed
analysis [Christodoulou et al. (2013)] shows that divergences are associated to the mixed terms,
where the positive contribution of one tetrahedron cancel the negative contribution of another.
Turaev-Viro amplitude
129
6.3 Turaev-Viro amplitude

How to deal with the divergences? In the 3d theory, this is easy: just compute
the amplitude using a two-complex without bubbles. This is possible in general,
and therefore the theory is well defined as it is. Since there are no local degrees of
freedom, in general choosing a more complicated triangulation does not improve
the approximation.
But in 4d we will not have such a luxury, because the theory will have local
degrees of freedom, we will have to refine the triangulation, to capture all degrees
of freedom of the theory, and the bubble amplitudes will describe genuine large
radiative corrections. With this in mind, let us see whether there is a way to make
sense of bubble divergences also in 3d.
In fact, there is, and was discovered by the mathematicians Turaev and Viro in
the early 90s [Turaev and Viro (1992)]. Turaev and Viro found that it is possible
to modify the theory and obtain a version without divergences. Remarkably, the
modified theory has a physical interpretation, which we discuss in the next Section.
i
The modified theory depends on a parameter q = e r for an integer r (so that
qr = 1) and formally converges to the Ponzano-Reggae model when r , that
is q 1. The amplitudes of the Turaev-Viro theory are defined by modification of
(5.48), given by
p
Wq ( j` ) = wq
(1) jf dq ( jf ) (1) Jv {6j}q .

jf
(6.17)
The quantities dq ( j) and {6j}q appearing in this formula are the quantum dimension and the quantum 6j-symbol that appear in the representation theory of the
quantum group SU(2)q . These are defined and discussed for instance in [Carter
et al. (1995)] or [Kauffman and Lins (1994)], and we do not discuss them here,
besides a few remarks below. The normalization factor is
wq =
( q q 1 )2
.
2r
(6.18)
The reason we have not labelled the amplitude with the triangulation is that Turaev and Viro have proven that the amplitude is in fact rigorously independent
from the triangulation. It only depends on the global topology of the manifold on
which it is defined. Most important, it is finite.
For large r, all the quantities entering in the amplitude converge to the corresponding SU(2) quantities:
130
j1
j4
j2
j5
j3
j6
q
q
dj
j1
j4
j2
j5
j3
j6
+ O ( r 2 ) ,
(6.19)
= (2j + 1) + O(r 2 ) ,

2 2
w q = 3 1 + O ( r 2 ) .
r
(6.20)
(6.21)
In this sense the amplitude converges to the Ponzano Regge one for large r. But
the sum over representations is finite, because the q deformed quantities vanish
for large j. In particular, the quantum dimension is given by
2j+1 q2j1
q
2j q
2j sin r (2j + 1)
d j = (1)
= (1)
(6.22)
sin r
q q 1
and is plotted below.
q
dj
15
10
Figure 6.5
10
The quantum dimension as a function of the representation, for different values of q

approaching q = 1.
Notice from the plot that there is maximum value of j, which increases as q 1.
Therefore only spins up to a maximum value, which is
jmax
r2
2
(6.23)
enter this theory. The finiteness of the j makes the amplitude manifestly finite.
The theory predicts a maximum length for each individual segment. This works
as an infrared cut off that suppresses the infrared spike divergences. A completely
explicit expression for the quantum 6j-symbol {6j}q is given in the appendix of
[Noui and Roche (2003)].
The triangular inequalities are also modified in this setting: the q-deformed 6jsymbols vanish unless qdeformed versions of the triangular inequalities hold.
These do not correspond to the inequalities satisfies by a triangle in flat space, but
instead to the inequalities satisfied by a triangle on a sphere of finite radius.
The physical interpretation of this mathematics is discussed in the next section.
Turaev-Viro amplitude
131
6.3.1 Cosmological constant

Ponzano and Regge have proven that the asymptotic of the {6j} symbol for large
spins gives the Regge action of 3d general relativity. This indicates that the classical
limit of the theory is indeed general relativity. What does the asymptotic of the qdeformed symbol {6j}q give?
Remarkably, it was shown in [Mizoguchi and Tada (1992)] that it gives the Regge
action for the theory with cosmological constant , where the cosmological constant
is determined by the deformation parameter by
q = ei
or
=
h G
2
.
(rh G )2
(6.24)
(6.25)
Therefore the Turaev-Viro amplitude defines a quantum theory whose classical

limit is simply related to quantum general relativity with a positive cosmological
constant. The presence of the cosmological constant is related to the finiteness of
the theory. Since a few years, of course, we have learned from cosmological observations that in Nature the cosmological constant is not zero and is positive.
The quantum theory of gravity appears to be consistent only taking this fact into
account.
In Planck units the measured cosmological constant it is a very small number.
This means that the deformation parameter r is very large, or q 1, the maximum
spin is very large in the realistic (we are still in the 3d euclidean context) theory,
and the relevant quantum group is very close to SU(2).
Notice that the cosmological can be only be positive in this theory, and this is
in accord with the observed real cosmological constant.2 A positive cosmological
constant corresponds to a space of constant curvature, and therefore suggest that
space be closed: the value of can be put in correspondence with the size of space,
which has to be finite. This is dual to the fact of not having an infinite resolution
for small things (minimal length). The Planck length provides a physical cutoff in
the small, the cosmological constant provides an infrared cutoff in the large.
There is a simple physical way of understanding the connection between the
presence of a cosmological constant and a cut off in angular momentum. Let us
do this in 3+1 dimensions, where h G has the dimension of a length square and the
relation between the cosmological constant and q is
q = eihG .
(6.26)
Consider an observer in a world where there is a minimal size ` determined by the

Planck scale and a cosmological constant that determined a large distance scale L.
In a maximally symmetric geometry, the observer in this world cannot see more far
2
Quite the opposite happens in string theory, where the theory seems to like the unphysical negative
cosmological constants, as in the standard AdS/CFT scenario.
132
lP
min
Figure 6.6
horizon
fuzzy 2-sphere
The minimal angle in a universe with minimal size and maximal distance.
than the scale L = 1/(2). In the euclidean context, this is because the universe
is a sphere; in the lorentzian context, because of the cosmological horizon. Now
consider the angle under which such an observer sees an object. To make this
angle smaller, we should either move the object farther away, or make it smaller.
But there is a limitation in the smallness of the size of the object, given by `, as
well as a limitation in the distance, given by L. Therefore the minimal angle under
which an object is seen is
`
(6.27)
minimal .
L
Now suppose the observer is observing around him. What he sees can be decomposed in spherical harmonics in a sphere surrounding him. To capture smaller
objects you need higher spherical harmonics. Since the number of dimensions of
the spin j spherical harmonics is 2j + 1 and the full solid angle is 4, the spherical
harmonics of spin j capture solid angles 2 4/(2j + 1) which together with the
previous equation gives
jmax =
L2
2
=
.
2
2
h G
`
min
(6.28)
Using the definition qr = 1 and (6.23) we get

2
1 = qr = q2jmax +1 = q2jmax +1 q hG .
(6.29)
which is solved precisely by (6.26). In words, the maximum spin is the one for
which the maximal angular resolution of the corresponding spherical harmonics
is needed to capture the smallest object at the largest distance.
The geometry described by the Turaev-Viro theory can be visualized as follows.
Consider a three-sphere, with the radius given by r. Fix the north pole as the point
where the observer is, and foliate the three-sphere by two-spheres corresponding
to parallels. Consider a finite subset of these, for the distances from the North
Pole given by the values of j. Each of these is characterized by states on a spin-j
representation, with a basis | j, mi.
Complements
133
This is a neat, complete, finite theory of quantum gravity, but in 3d, where, in
fact, not much interesting happens. In the next section we start to address the
problem of describing the real world.
6.4 Complements
6.4.1 Few notes on SU(2)q
The quantum group SU(2)q is not a group and there is nothing quantum about it.
It is is not a quantum mechanical deformation of the group SU(2). It is a one-parameter
deformation of the algebra of the representations of SU(2). The deformation parameter is
not to be identified with the Planck constant. If anything, with the cosmological constant.
Consider the standard representations spaces H j of SU(2). Each space carries a representation of the su(2) algebra given by operators ~L( j) ; furthermore these spaces are related
among them by the fact that standard SU(2) representation theory defines a maps from
H j1 H j2 to H j3 . This map projects on the subspace of H j1 H j2 that transforms in the H j3
representation. The quantum group SU(2)q is defined by a deformation of the operators
~L( j) and of the map from H j to H j H j , such that some general algebraic relations be2
3
1
tween these objects continue to hold. The deformation is characterized by the fact that the
map from H j1 H j2 to H j3 differs from the map from H j2 H j1 to H j3 . In this sense, it is
non-commutative.
The first step where this is realized is in the map from H1/2 H1/2 to H0 . For the standard group SU(2) this map is given by
For SU(2)q , instead,
(, ) 7 e AB A B .
(6.30)
where
(, ) 7 e AB A B .
q
e AB =
where
0
A 1
A
0
(6.31)

(6.32)
A2 = q
(6.33)
is the deformation parameter that characterizes the quantum group. Notice that the map
from H1/2 H1/2 to H0 is not anymore symmetric.
The tensor eq satisfies the equality
q
B
C
CA D
= A1 BA D
+ A eq AC eBD ,
(6.34)
which can be written graphically in the form
(6.35)
Notice that this is a deformation of the relations (1.47) and (1.50), where we had A = 1. See
details in the Haggard thesis [Haggard (2011)].
Similarly, for SU(2)q the map from 2j copies of H1/2 , that is H1/2 ... H1/2 , 2j times to
H j is simply obtained by projecting in the fully symmetric part of the tensor product. This
134
can be obtained by summing over all permutations of the factors. Let P be the operator
that realizes the permutation of the factors. Then for SU(2)q the map from H1/2 ...
H1/2 , 2j times to H j is obtained by summing over all permutations with a weight. This is
given by the projector
Pj =
A3cr() P
(6.36)
where
c 1 =
A4cr()
(6.37)
where cr () is the (minimal) number of crossings in the permutation. These relations are
sufficient to compute the quantum dimension, defined as the trace of Pj . This is
j = (1)2j
A4j+2 A4j2
sin((2j + 1)h)
= (1)2j
(1)2j [2j + 1]
sin h
A 2 A 2
(6.38)
where q = eh . Importantly, the triangular inequalities turn out to be modified and supplemented by the conditions
2j1 , 2j2 , 2j3 j1 + j2 + j3 r 2
(6.39)
where q = e r , and these are precisely the triangular inequalities of a triangle on a sphere
with a radius determined by r. The geometry of q deformed spin networks is therefore
consistent with the geometry of constant curvature space, with the curvature determined
by the cosmological constant.
Problem
Should we expect an effect of the cosmological constant on the area eigenvalues?
Hint: Recall that these are given by the eigenvalues of the Casimir, which in turn is defined by the SU(2) generators. If SU(2) is deformed to SU(2)q , so are its generators and so
is the Casimir...
Part III
THE REAL WORLD
The real world:

4D Lorentzian theory
Remember what the dormouse said...
This is the main chapter of the book, where the physical theory of quantum gravity
is defined. The real world is, as far as all current empirical evidence indicates, 4dimensional and Lorentzian. We start from the Holst action of general relativity,
described in Section 3.3. Recall this reads

Z
1
S[e, ] = e e ? +
F
(7.1)
where the variables are the tetrad field e and a Lorentz connection . On the
boundary, the momentum conjugate to is the sl (2, C)-algebra valued two-form

1
1
1
=
B=
(e e) + (e e) .
(7.2)
8G
8G
where the electric and magnetic parts of B satisfy the linear simplicity constraint
(3.46), that is
~ = ~L.
K
(7.3)
The theory is invariant under local SL(2, C) transformations. This is a formulation of standard classical general relativity, a theory with very strong empirical
support, and we are seeking a quantum theory with this classical limit.
7.1 Classical discretization

We look for the quantum theory following the track of the 3d theory discussed in
Chapter 3.5. For this, we need two ingredients: the generalisation to 4d of the discretisation on a two-complex discussed in Section 4.4; and a way to keep track of
(7.3) in the quantum theory. The first is straightforward; the second is the technical
core of the theory. Lets start with the discretisation.
Fix a triangulation of a compact region M of spacetime.1 That is, chop M into
4-simplices. A 4-simplex can be thought as the convex region of R4 delimited by
5 points. A 4-simplex is bounded by five tetrahedra. As before, consider the dual
1
137
Later we will generalise the construction to structures more general than a triangulation.
The real world: 4D Lorentzian theory
138
boundary graph
face
face
edge
vertex
edg
e
face
node
Figure 7.1
link
link
Triangulation and two-complex terminology in 4d

of the triangulation, and focus on its vertices, edges and faces. These objects and
their relations, including for the boundary, are listed in the Table below.
Table 7.1 Bulk terminology and notation

Bulk triangulation
4-simplex (v)
tetrahedron ()
triangle (t)
segment (s)
point (p)
Two-complex
vertex (v)
edge (e)
face (f)
Table 7.2 Boundary terminology and notation

Boundary triangulation
Boundary graph
tetrahedron ()
triangle (t)
node (boundary vertex) (n)

link (boundary edge) (`)
Notice that the two-complex objects are exactly the same as those in 3d. They will play
similar roles. Only, they refer to higher dimensional objects in the triangulation. In particular,
a vertex is still dual to a chunk of spacetime,
but now this chunk is a 4-simplex, because we
are in 4d. In turn, edges are dual to tetrahedra.
See the picture here on the left.
Similarly, on the boundary we still have a
graph . Its nodes are still chunks of space,
but now these are tetrahedra, because space is 3d. The links of the boundary graph
4d
139
Figure 7.2
A vertex is a point inside the 4-simplex. It has five edges, corresponding to the five
tetrahedra bounding the 4-simplex. A triangle is dual to a face that wraps around
it, because in 4d, we can go around a triangle in the ( x, y) plane, moving in the
(t, z) plane.
connect adjacent tetrahedra and therefore are dual to triangles of the boundary triangulation. A main role is played in the theory by the faces f of the two-complex
(grey in the picture above). These are dual to triangles of the triangulation (a triangle in the ( x, y) plane is dual to a face in the (z, t) plane). A face which touches the
boundary is dual to a boundary triangle, and therefore corresponds to a boundary
link `. Geometrically, this link is the intersection of the face with the boundary.
Therefore a boundary link ` is a boundary edge, but is also associated to a face
f that touches the boundary. The diligent reader is advised to digest this well, in
order not to get lost in the following.
Variables
We discretize the connection, as in 3d, by associating a SL(2, C) group element
Ue to each edge e of the two-complex. We discretize the tetrad by associating an
element of the sl (2, C) algebra to each triangle of the triangulation. Since triangles
are dual to faces, the algebra vectors are associated to faces f, as in 3d. That is
Ue
e Bf .
(7.4)
(7.5)
The formal relation between the continuum and the discrete variables can be
taken to be the following
Ue = P e
Bf =
tf
SL(2, C) ,
B sl (2, C) .
(7.6)
(7.7)
The first integral is the holonomy of the connection along the edge, namely the
matrix of the parallel transport generated by the connection along the edge, which
we take in the fundamental representation of SL(2, C). The second is the surface
140
integral of the two-form B across the triangle tf dual to the face f. Notice that in
3d we had a one-form e to integrate along a segment dual to the faces, while now
we have a two-form B to integrate along a triangle dual to the faces. But in either
case, what we obtain is algebra element per face. The action can be approximated
in terms of these objects.2
Thus, the variables of the discretized theory are:
a group element Ue for each (internal or boundary) edge e of the two-complex;

an algebra element Bf for each face f of the two-complex;
as in 3d. As in 3d, we call U` the group elements associated to the boundary edges
`, namely the links of the boundary graph , and we call B` the algebra elements
of a face bounded by the link `.
The variable B` has a remarkable geometric interpretation. Consider a triangle
lying on the boundary. It is convenient to choose the tetrad field in the time gauge
where eo = dt, ei = eia dx a . The pull back of (e e) to the boundary vanishes and
we have
Z
1 i
e jk
e j ek .
(7.8)
Lfi =
2
tf
In the approximation in which the metric is constant on the triangle, we see that
the norm of this quantity is proportional to the area of the triangle
| Lf | =
1
A tf
(7.9)
This can be seen by choosing a gauge and coordinates where eia = ai . The vector ~Lf
itself is normal to the triangle, and has a length proportional to its area. Therefore
it is the same quantity (up to the factor 1 ) as the quantity ~ES defined in (3.91). We
now have all the ingredients to move to the quantum theory.
7.2 Quantum states of gravity

Let us start from the Hilbert space. Since the variables Ue and Lf are associated
to edges and faces respectively, the boundary variables are respectively associated
to the boundary edges, namely the links ` and the boundary of the faces, which
2
As in 3d, (7.7) can be made more precise to better deal with the gauge. Under a gauge transformation
1
the group elements U defined in (7.6) transform well, namely as Ue 7 se Ue
te where se and
te are the initial (source) and final (target) vertices of the (oriented) edge e. Therefore in the
discrete theory the continuous local SL(2, C) invariance is reduced to Lorentz transformations at
the vertices. Not so for the algebra variables B defined in (7.7), unless a suitable gauge is chosen.
To correct this, assume that this definition is taken in a gauge where the connection is constant on
the triangle itself, as well as on the adjacent tetrahedron and along the first half of each (oriented)
edge. In this way, also the Bi variables are invariant under all gauge transformation except those at
the vertices; and they transform covariantly, in fact, in the adjoint representation, under a Lorentz
transformation v at the vertex v.
141
Quantum states of gravity

are also links. Therefore, precisely as in the case of the 3d theory, the boundary
variables are B` sl (2, C) and U` SL(2, C) on each link ` of the boundary graph
. These become operators in the quantum theory. The natural quantization leads
to states (U` ), functions on SL(2, C) L , where the operator B` sl (2, C) is realized
as the generator of SL(2, C) transformations.
However, the components of these generators must satisfy the linear simplic~ = ~L, at least in the classical limit. Therefore some
ity condition (7.3), namely K
limitation of the states is needed, for this condition to hold. To realise this, we
introduce the Y map.
7.2.1 Y map
Functions on SL(2, C) can be expanded into irreducible unitary representations of
SL(2, C). The unitary representations of SL(2, C) should not be confused with the
common Lorentz representations that are familiar in physics (such as 4-vectors, or
spinors), which are not unitary; they are presented in detailed form in [Ruhl (1970)]
and are briefly recalled in the Complements to this Chapter. The key facts we need
are the following. They are labelled by a positive real number p and non negative
half-integer k. The Hilbert space V ( p,k) of the ( p, k ) representation decomposes into
irreducibles of the subgroup SU(2) SL(2, C) as follows
V ( p,k) =
M
j=k
Hj
(7.10)
where H j is a 2j + 1 dimensional space that carries the spin j irreducible representation of SU(2). In the ( p, k) representation we can therefore choose a basis of
states | p, k; j, mi, with j = k, k + 1, ... and m = j, ..., j. The quantum numbers ( p, k )
are related to the two Casimir of SL(2, C) by
~ |2 |~L|2 = p2 k2 + 1
|K
~ ~L = pk.
K
(7.11)
(7.12)
While j and m are the quantum numbers of |~L|2 and Lz respectively.

Now, we want that in the classical limit, namely in the limit of large quantum
numbers,
~ = ~L.
K
(7.13)
If so, the Casimir must satisfy
~ |2 |~L|2 = (2 1)|~L|2 ,
|K
~ ~L = |~L|2 .
K
(7.14)
(7.15)
Passing to quantum numbers, this gives

p2 k 2 + 1 = ( 2 1) j ( j + 1),
pk = j( j + 1).
(7.16)
(7.17)
142
For large quantum numbers this gives

p2 k 2 = ( 2 1) j2 ,
(7.18)
pk = j ,
(7.19)
which is solved by the two fundamental equations

p = k,
(7.20)
k = j.
(7.21)
The first of these two relations is a restriction on the set of the unitary representations. The second picks out a subspace (the lowest spin subspace in the sum (7.10))
within each representation. The states that satisfy these relations have the form.
(7.22)
| p, k; j, mi = |j, j; j, mi
Notice that these states are in one to one correspondence with the states in the
representations of SU(2). We can thus introduce a map Y as
V ( p=j,k= j)
Y : H j
(7.23)
| j; mi 7 |j, j; j, mi,
(7.24)
and all the vectors in the image of this map satisfy the simplicity constraints, in
the weak sense. That is
~ ~L | Y i = 0.
hY | K
(7.25)
in the large j limit.3 Thus we assume that the states of quantum gravity are constructed from the states |j, j; j, mi alone.
These states are in one-to-one correspondence with the SU(2) states | j; mi: the
map is given indeed by Y . This map extends immediately to a map from functions
over SU(2) to functions over SL(2, C). Explicitly, this is given by
Y :
L2 [SU(2)]
(h) =
jmn
( j)
c jmn Dmn (h)
F [SL(2, C)]
( g) =
jmn
(7.26)
(j,j)
c jmn D jm jn ( g).
(7.27)
And therefore we have a map from SU(2) spin networks to SL(2, C) spin networks.
We will see soon that the map Y is the core ingredient of the quantum gravity dynamics. It depends on the Einstein Hilbert action and codes the way SU(2)
states transform under SL(2, C) transformations in the theory. This, in turn, codes
the dynamical evolution of the quantum states of space.
The physical states of quantum gravity are thus SU(2) spin networks, or, equivalently, their image under Y . Notice that this space carries a scalar product which
3
This can be shown as follows. The only non vanishing matrix element can be
hj, j; j, m | K i | j, j; j, m0 i, and these must be proportional to the intertwiner between the representations j, j and 1. Similarly for hj, j; j, m | Li | j, j; j, m0 i. Therefore the two are proportional.
The proportionality factor can be determined by reconstructing the Casimirs from the matrix
elements.
143
is well defined: the one determined by the SU(2) Haar measure. The fact that
the scalar product is SU(2) and not SL(2, C) invariant (it is only covariant under
SL(2, C) ) reflects the fact that the scalar product is associated to a boundary, and
this picks up a Lorentz frame.
Notice that the fact that the Hilbert space of quantum gravity is formed by SU(2)
spin networks is consistent with the canonical analysis of the theory of Section
3.57. There, it was shown that the boundary degrees of freedom are given by the
3d metric. This can be expressed in terms of triads and, in turn, these give rise to
the Ashtekar canonical pair ( Aia , Eia ), which form precisely the same kinematical
phase space as that of SU(2) Yang Mills theory. The corresponding quantum states
are SU(2) spin networks. All is therefore nicely consistent. Recall also that generators ~L of SU(2) transformations are 1/ times Eia , which is the area element.
Therefore, restoring physical units from (3.39), we can identify the operator
~El = 8h G~L`
(7.28)
associated to each link of the boundary graph as the normal to the corresponding
triangle in the boundary, normalized so that the area of the triangle is
A` = 8h G |~L|2 .
(7.29)
This gives immediately the eigenvalues of the area of a single triangle to be

q
(7.30)
A j = 8h G j( j + 1),
and the eigenvalues of an arbitrary surface punctured by N punctures n1 , ..., N

q
A jn = 8h G jn ( jn + 1),
(7.31)
n
where j and the jn s are non-negative half-integers.

Notice that this is precisely the result that we have obtained in the first chapter,
where we can now fix the loop quantum gravity scale
lo2 = 8h G
(7.32)
where is now the Barbero-Immirzi constant. The parameter that we left free in
Chapter 1 is determined by the action of the general relativity and identified with
the coupling constant of the Holst term.
7.2.2 Spin networks in the physical theory

In 3d, the area quantum numbers were sufficient to determine a basis of states,
without further degeneracy. The same is not true in 4d. This can be seen as follows.
Decomposing the Hilbert space as we did in 3d, we have (see (5.17))

L2 [SU(2) L ] = ` j (H j H j ) = j` ` (H j H j ).
(7.33)
and (see (5.20))

L2 [SU(2) L /SU(2) N ] = j` n InvSU(2) H j1 H j2 H j3 H j4 .
(7.34)
144
which differ from (5.20) because now each edge is bounded by four faces, not
three. The space InvSU(2) H j1 H j2 H j3 H j4 is not one-dimensional in general. In fact, linearly independent invariant tensors in this space can be constructed
as follows

j1
j2 k
k j3
j4
m1 m2 m3 m4
ik
=
gmn
(7.35)
m1 m2 m
n m3 m4
for any k which satisfies the triangular relations both with j1 , j2 and j3 , j4 , that is
for any k satisfying
max[| j1 j2 |, | j3 j4 |] k min[ j1 + j2 , j3 + j4 ].
These states are denoted |ki. Therefore we have a space
K j1 ...j4 InvSU(2) H j1 H j2 H j3 H j4
with dimension
(7.36)
(7.37)
dim[K j1 ...j4 ] = min[ j1 + j2 , j3 + j4 ] max[| j1 j2 |, | j3 j4 |] + 1,
(7.38)
which describes the residual geometrical freedom at each tetrahedron when the
area of its faces is sharp.
It follows that a generic gauge invariant state is a linear combination
(U` ) =
C j` kn j` kn (U` )
(7.39)
j` kn
of the orthogonal states

m m2 m3 m4
j` kn (U` ) = k 1
1
. . . k L 3
N
m L 2 m L 1 m L
Dm1 1 n1 (U`1 ) . . . DmLL n L (U` L )
(7.40)
which replace (5.33) and (5.34). The difference between the 3d case and the 4d
case is that the spin networks of trivalent graphs are labelled only by the spins,
while the spin networks of graphs with higher valence are labelled also by an
intertwined quantum number k associated to the each node n. These are the
spin-network wave functions. We write them also in the simpler form
j` kn (U` ) = h U` | j` , kn i =
O
n
kn
D j` (U` ).
(7.41)
These are the 4d spin network states in the group representation.

Classically, this residual geometric freedom is described by the phase space of
the tetrahedron, namely the space of the possible shapes of a tetrahedron with
fixed areas. This is a two-dimensional space. For instance, one can take two opposite dihedral angles to coordinatize this space. This space can also be seen as the
space of the quadruplets of vectors satisfying the closure relation, with given areas, up to global rotations. Once again, the counting of dimensions gives (4 3
4 3 3 = 2). An observable on this space, in particular, is the (oriented) volume
V of the tetrahedron, which we have already studied in the first Chapter, which is
given by
2
(7.42)
V 2 = eijk Ei E j Ek .
9
145
The matrix elements of the volume can be computed explicitly in the |ki basis (see
Complements), by using standard angular momentum theory. Diagonalizing the
resulting matrix, one can then compute the eigenvalues v and the eigenstates |vi of
the volume in each Hilbert space K j1 ...j4 . In this basis, we write the spin networks
states as
j` vn (U` ) = h U` | j` , vn i =
O
n
vn
D j` (U` ).
(7.43)
where the vn form a basis that diagonalises the volume operator in the intertwined
space of the node n.
In summary, the truncated state space of quantum gravity is given by the Hilbert
space
H = L2 [SU(2) L /SU(2) N ],
(7.44)
~E` = lo2~L`
(7.45)
where the triad operator (or flux operator) associated to each link is
and ~L` is generators of SU(2) transformations on each link, namely the left invariant vector field. Notice that their algebra is the one that in Chapter 1 was simply
postulated in (1.12), and the dimensional factor in (1.13) is now determined by the
general relativity action, which fixes the scale of the momentum. Furthermore, the
closure relation (1.9) is a direct consequence of the SU(2) invariance of the states at
each node.
Increasing the complexity of the graph gives a better approximation to the full
theory. A basis of states in H is provided by the spin-network states |, j` , vn i,
with a spin j` associated to each link of the graph and a volume eigenvalue vn
associated to each node of the graph. These are eigenstates of the area and the volume operators. This is of course a separable Hilbert space4 , as the explicit discrete
basis shows.
7.2.3 Quanta of space

The states in the spin network basis have a straightforward physical interpretation
because they diagonalize the area and volume operators. They are interpreted as
quantum states of the gravitational field, forming a granular physical space. This
simple intuitive picture must be sharpened in a number of ways.
1. It is important not to confuse the physical quantum discreteness from the discreteness due to the truncation of the number of degrees of freedom. This confusion is so common that it is better to emphasize it again, at the cost of a repetition. The fact that we work on a graph is a classical truncation, analogous
to expanding a field in a box in discrete modes, and truncate the theory considering only a finite numbers of modes. There is nothing related to quantum
mechanics in this. In the truncated classical theory, the area of the faces of the
4
A persistent rumour that the Hilbert of LQG is non-separable is false.
146
Figure 7.3
Chunks of space with quantized volume are associated to nodes. The area of the
surface shared between two cells is quantized as well.
triangulation can become arbitrary small, and the geometry of a discrete triangulation can approximate arbitrary well a continuum geometry.
The quantum discreteness, on the other hand, is due to the discrete spectrum
of area and volume. This is analogous to quantized energy spectrum of each
mode of, say, the electromagnetic field. The quanta of energy on a mode are
individual photons. This is the genuine quantum phenomenon. In gravity, the
area of each triangle and the volume of each tetrahedron are quantized. This is
the genuine quantum phenomenon, with a scale set by h .
Because of this, there is no limit in which the tetrahedra become infinite small.
This is the central point of loop quantum gravity. This is the physical discreteness of space unveiled by loop quantum gravity.
2. The spin network states do not fully diagonalize the 3d metric. This is because
the area and the volume are not sufficient for determining the geometry of a
tetrahedron. The geometry of the tetrahedron is determined by 6 numbers (the
6 lengths of its sides) while areas and volume are only five numbers. The situation is like in angular momentum theory: classical angular momentum is determined by three numbers (L x , Ly , L x ) but only two operators can be diagonalized simultaneously (L2 , L x ), with the consequence that the angular momentum
~L can never be sharp. Similarly, the 3d geometry can never be sharp. Therefore
the spatial metric remains always quantum-fuzzy at the Planck scale.
It follows that the geometrical picture of tetrahedra, triangles, et cetera,
must be taken only as something meaningful in some classical approximation
and not at the fundamental scale. At the Planck scale there are tetrahedra in
the same sense in which an electron is a small rotating pebble. That is, in a
very imprecise sense. There are no tetrahedra, down at the Planck scale. There
are quantum states, formed by quanta of the gravitational field, which have the
property of giving rise to something that we describe as a three dimensional
Riemannian geometry in the limit of large quantum numbers.
3. The quanta of space described by the spin network states |, j` , vn i should not
be thought of as quanta moving in space. They are not in space. They are them-
Transition amplitudes
147
selves physical space. This can be seen by comparing them to the usual description of the quanta of the electromagnetic field. These are characterized by quantum numbers | p1 , ..., pn i where the pi s are momenta, namely Fourier transforms of position variables. Position is given with respect to a space in which
the photons live. In quantum gravity, instead, the quantum numbers do not include position. They include the intrinsic physical size of the quanta themselves
(area and volume) as well as the graph that codes the adjacency relations between these quanta. Therefore the quanta of space are located with respect to
one another, and this relation is given by the combinatorial structure of . This
is the manner loop quantum gravity realizes spatial background independence
in the quantum theory. This is the quantum analog of what happens in classical
general relativity, where the gravitational field does not live in spacetime: it is
itself spacetime.
4. Finally, a warning: the quanta of space of loop quantum gravity should not be
taken too naively as actual entities, but rather as modes of interaction. What
the theory predicts is that in any interaction where the effect of something depends on its area, this effect will be that of a quantized area. Trying to think
too literarily in terms of concrete chunks forming the quanta of space can be
misleading, as always in quantum mechanics. A pendulum is not made of its
quanta. It is the way it acts in an interaction which is characterised by quantized energy. This is important especially in applications of the theory, where
a naively realistic picture, as always in quantum mechanics, can be misleading. A basis in the Hilbert space should not be mistaken for a list of things,
because bases associated to non commuting observers are equally physically
meaningful.
7.3 Transition amplitudes

Finally, we can now complete the construction of the full theory writing its transition amplitudes.
The experience in 3d gives us the general structure of these amplitudes. Let us
recall the 3d theory: this is given in (5.69) and (5.74), which we copy here:
W ( h` ) =
SU(2)
dhvf
(hf ) Av (hvf ), .
(7.46)
where hf = hvf hv0 f ... and the vertex amplitude is

Av (hvf ) =
jf
SU(2)
0
dgve
d jf Tr jf [ ge0 v gve hvf ].
(7.47)
Since in 4d the kinematical Hilbert space is essentially the same as in 3d, the first
equation can remain unchanged in 4d. In fact, it only reflects the superposition
principle of quantum mechanics, for which amplitudes are obtained by summing
148
for amplitudes of individual stories, and a locality principle, for which the amplitudes of a story is the product of individual local amplitudes associated to separate
regions of spacetime. The actual dynamics is given by the vertex amplitude, which
determines the local physics. What is the vertex amplitude in 4d?
It must be a function Av (hvf ) of SU(2) group elements living on the graph
of the node (namely, on the boundary of a spacetime four-simplex), but it must
be SL(2, C) invariant. The amplitude (7.47) is only invariant under SU(2), and
does not know anything about SL(2, C). In order to obtain an amplitude which
is SL(2, C) invariant we must replace the SU(2) integrals in (7.47) with SL(2, C)
integrals, and somehow map the SU(2) group elements into SL(2, C) ones.
We do have the tool for this, which is the map Y . The map can be used to map
a function of SU(2) variables into a function of SL(2, C) variables. These can then
be projected on SL(2, C) invariant states and evaluated

Av () = PSL(2,C) Y (1l)
(7.48)
Let us write this explicitly. In the group representation:
Av (hvf ) =
jf
SL(2,C)
0
dgve
(2jf + 1)Trjf [Y ge0 v gve Y hvf ].
(7.49)
where the meaning of the Trace notation is

Tr j [Y gY h] = Tr[Y D (j,j) ( g)Y D ( j) (h)] =
mn
(j,j)
( j)
D jm,jn ( g) Dnm (h).
(7.50)
The vertex amplitude is a function of one SU(2) variable per each face around
the vertex. Imagine to draw a small sphere surrounding the vertex: the intersection
between this sphere and the two-complex is a graph v . The vertex amplitude is a
function of the states in
Hv = L2 [SU(2)10 /SU(2)5 ]v
(7.51)
The graph v is the complete graph with five nodes (all the nodes connected); see
Figure 7.5.
The prime in the SL(2, C) integrations, introduced in (5.73), indicates that we do
not integrate in five group variables, but only in four of them. The result of the
integration turns out to be independent on the fifth. While this was just cosmetic
in 3d, here this procedure is required, because the integration on the last SL(2, C),
which is non compact, would give a divergence if we did not take this precaution.
As it is defined, instead, the vertex amplitude can be proven to be finite [Engle and
Pereira (2009); Kaminski (2010)].
This completes the definition of the covariant formulation of the theory of loop
quantum gravity on a given two-complex.
The dynamics realizes the three criteria that we stated in Section 5.3, which we
repeat here, adapted to the physical 4d case:
1. Superposition principle. Basic principle of quantum mechanics. Amplitudes
149
Figure 7.4
The graph v , the boundary graph of the vertex, or pentagram, was called `
(Hugieia) by the Pythagoreans. They saw mathematical perfection in it. Hugieia was a
goddess of health and the expression Hugieia was used as a greeting.
are sums of elementary amplitudes (in standard quantum theory, this is Feynmans sum over paths .)
hW | i = W ( )
(7.52)
2. Locality. Fundamental discovery of XIX and XX century physics: interactions

are local in spacetime. Elementary amplitudes are products of amplitudes associated to spacetime points (in standard quantum field theory, exponential of an
integral over spacetime points).
W ()
Wv .
(7.53)
3. Lorentz invariance.
Wv = ( PSL(2,C) Y v )(1l).
(7.54)
where is a boundary state, PSL(2,C) is the projector on its locally SL(2, C)invariant component and (1l) indicates the evaluation of the spin network state
on the identity.
Figure 7.5
The vertex graph in 4d.
150
As we shall see later:

1. The theory is related to general relativity in the classical limit.
2. It is ultraviolet finite.
3. With a suitable quantum-deformation which adds the cosmological constant, it
is also infrared finite.
4. It leads to the n-point functions of perturbative general relativity, to the Friedmann equation in cosmology, and the (finite) Bekenstein-Hawking black-hole
entropy .
7.3.1 Continuum limit

The equations above define a theory with a state space defined on a given graph
, and amplitudes determined by a given two-complex C . This is a theory with
a finite number of degrees of freedom. This definition obviously correspond to a
truncation of classical general relativity, which is a theory with an infinite number
of degrees of freedom. The full theory is approximated by choosing increasingly
refined C and = C .
This procedure is the common one in quantum field theory. Concrete physical
calculations are mostly performed on finite lattices in lattice QCD, say with N vertices, and at a finite order in perturbation theory in QED. A finite order in perturbation theory means that all Feynman graphs considered have at most a number
of particles N, where N is finite, and therefore are de facto fully defined on a finite
sum of n-particle Hilbert spaces, describing arbitrary high momenta, but a finite
number of degrees of freedom. (The fact that arbitrary high momenta enter a single Feynman graph, and therefore the graph involves an infinite number of modes
does not contradict the fact that relevant excitations of these modes are those that
can be described by a finite number of particles.) In both cases, one formally considers the N limit, but concretely one always computes at finite N, with some
arguments that the rest be smaller.
Quantum gravity is defined in the same manner: one considers states and amplitudes of the truncated theory, and chooses a refinement, namely and C , sufficient
for the desired precision.
Let us first consider the state space. The crucial observation in this regard is the
following. Let 0 be a subgraph of , namely a graph formed by a subset of the
nodes and links of . Then it is immediate to see that there is a subspace H0 H
which is precisely isomorphic to the loop-gravity Hilbert space of the graph 0 .
Indeed, this is formed by all states (U` ) H which are independent from the
group elements U` associated to the links ` that are in but not in 0 . Equivalently,
H0 is the linear span of the spin network states characterized by j` = 0 for any `
that is in but not in 0 .
Exercise: Show that the two definitions above are equivalent.
151
Therefore if we define the theory on we have at our disposal a subset of states

that captures the theory defined on the smaller graph 0 . In this sense the step from
0 to is a refinement of the theory: the refined theory includes the same states,
plus others.
One can define a Hilbert space H that contains all H s by using projective-limit
techniques [Ashtekar and Lewandowski (1995)]. The family of all spaces H for all
forms indeed a projective family. Since this will not be of much concrete use, we
do not go into these mathematical niceties here. We remark, however, that there is a
stringent similarity between this structure and Fock space. Fock space can be constructed as follows. First, define the one particle Hilbert space H1 , the n-particle
Hilbert space as the symmetric tensor product Hn = H1 S ... S H1 of n copies
of H1 and the space with N or less particles H N = nN=1 Hn . The space H N 1 is
then a subspace of the space H N and the Fock space is the formal limit N .
The analogy with loop gravity is between the spaces H where has N nodes and
H N . The resulting state space is separable, as the Fock space is separable. Keep
in mind that a physicist at CERN who computes scattering amplitudes virtually
never cares about Fock space and works entirely within H N for some finite N.
Next, consider the transition amplitudes. Again, we can formally define the
transition amplitude W (h` ) as a limit of WC (h` ) where C = , when C is refined. The notion of limit is well defined5 because the set of the two complexes is
a partially ordered set with upper limit6 . Again, it is not clear whether this notion
of limit is actually useful or even the relevant one. As far as physics is concerned,
the theory is useful to the extent a subclass of graphs and two-complexes suffice
to capture the relevant physics, in the same sense in which low order Feynman
graphs, or QCD lattices with finite number of sites are sufficient to compute relevant physics.
This raises the question of the regime of validity of the approximation introduced with a given and C . The answer follows immediately from analyzing the
classical theory. The approximation is expected to be good when the discretized
theory approximates the continuum theory in the classical context. The regime of
validity of a discretization of general relativity is known. The discretization is a
good approximation when the curvature on the hinges is small. Namely when the
deficit angles, in Regge language, are small. Thus, the expansion in and C is an
expansion which is good around flat space.
The reader might immediately wonder if expanding around flat space wasnt
precisely what went wrong with the conventional non-renormalizable quantumfield-theoretical quantization of general relativity. Why are we not finding here
the same difficulties? The answer is simple, because in the quantization here the
Planck scale discreteness of space, which cuts off the infinities, is taken into account in the formalism. In the conventional Feynman perturbation expansion, it is
not.
5
6
W = limC WC means that there is a Ce such that |W WC | < e for any C > Ce .
For any C and C 0 there is a C 00 such that C 00 > C and C 00 > C 0 .
152
Figure 7.6
QED Feynman graphs and lattice QCD lattices. The two pictures converge in LQG.
7.3.2 Relation with QED and QCD

The similarities between the theory defined above and lattice QCD on a given
lattice are evident. In both cases, we have a discretization of the classical theory
where the connection is replaced by group elements, and a quantum theory defined by an integration over configurations of a configuration amplitude which is
a product of local amplitudes. The use of a triangulation instead of a square lattice
simply reflects the fact that a square lattice is unnatural in the absence or a flat
metric. The more complicated appearance of the gravity amplitude with respect
to the Wilson one is simply a reflection of the particular action of gravity.
The substantial difference, as already pointed out, is the fact that the Wilson
action depends on an external parameter (as the Yang-Mills theory depends on
an fixed external metric), while the gravity action does not (as the Einstein action
doesnt). Accordingly, the continuum limit of lattice Yang-Mills theory involves
also the tuning of a parameter () to a critical value ( = 0), while nothing of the
sort is needed for gravity.
But there are also striking similarities between the theory introduced here and
perturbative QED. The nodes of the graph can be seen as quanta of space and the
two complex can be read as a history of quanta of space, where these interact, join
and split, as in the Feynman graphs. The analogy is strongly reinforced by the
fact that the spinfoam amplitude can actually be concretely obtained as a term in
a Feynman expansion of an auxiliary field theory, called group field theory. In
this book, we shall not develop the group-field-theory presentation of quantum
gravity, for which we refer the reader for instance to Chapter 9 of [Rovelli (2004)].
The specific group field theory that gives the gravitational amplitude has been
derived (in the Euclidean context) in [Krajewski et al. (2010)]. We shall however
discuss the physical reason of this surprising convergence between the formalisms
characteristic of our two most physically successful formulations of fundamental
quantum theories: perturbative QED and lattice QCD.
Feynman graphs and the lattice QCD appears to be very different entities. But
Full theory
153
are they? A Feynman graph is a history of quanta of a field. The lattice of lattice
QCD is a collection of chunks of spacetime. But spacetime is a field: it is the gravitational field. And its chunks are indeed quanta of the gravitational field. Therefore
when the gravitational field becomes dynamical and quantized, we expect that the
lattice of lattice QCD is itself a history of gravitational quanta, namely a Feynman
graph of a quantum field theory. (This, in addition largely fixes the apparent freedom in the discretisation of the theory, which is apparent in lattice QCD.) In both
cases, an increased approximation is obtained by increasing the refinement of the
graph, or the lattice.
These vague intuitive ideas are realized in concrete in the covariant formulation
of loop quantum gravity, where, because of the general covariance of gravity, the
QED and the lattice QCD pictures of quantum field theory are beautifully merged.
7.4 Full theory

Summarizing, for any graph we have a Hilbert space
with operators
H = L2 [SU(2) L /SU(2) N ] ,
(7.55)
~E` = lo2~L` = 8h G~L` ,
(7.56)
interpreted as the flux of the densitized inverse triad on the surfaces bounding
the space regions labelled by the nodes. The transition amplitudes between these
states can be computed at each order in a truncation determined by a two-complex
C bounded by . They are given (reinserting a normalization factor dependent on
the two-complex) by
WC (h` ) = NC
SU(2)
dhvf
( hf ) Av ( hv f ).
f
(7.57)
where the vertex amplitude is

Av (hvf ) =
jf
SL(2,C)
0
dgve
(2jf + 1) Trjf [Y ge0 v gve Y hvf ].
(7.58)
Together with the definition of the Y map, which we repeat here:

Y : | j, mi 7 |j, j; j, mi .
(7.59)
Given an observable O defined on H, we define its expectation value as
hOi =
hW |O|i
.
hW | i
(7.60)
154
g0
e0
h
Figure 7.7
The wedge.
This is the full definition of the theory, in the limit in which the cosmological constant vanishes.
For each choice of a graph we have a given truncation of the theory. Refining
the graph, we obtain a better approximation. At given graph, for each choice of
two-complex C we have a given truncation of the transition amplitudes. Refining
the two-complex, we have a better approximation. The difference with the 3d theory is therefore substantial: in 3d the topological invariance of the theory, namely
the absence of local degrees of freedom implies that a refinement of the triangulation has no effect on the quantum theory; while in 4d a more refined triangulation
can capture more of the infinite number of local degrees of freedom. These refinements determine what is called the continuum limit of the theory. A priori,
there is no parameter to tune in approaching this limit, for the reasons explained
in Chapter 2.
This is the theory that we analyze in detail in the rest of the book. The construction of this theory is the result of a long process, to which a large number of people
have participated. A very condensed historical account with the main references
is in the Complements of the last Chapter.
7.4.1 Face-amplitude, wedge-amplitude and the kernel P

The amplitude above can be equivalently rewritten as a product of face amplitudes instead of a product of vertex amplitudes. For this, it suffices to expand the
delta function in representations and perform the dhvf integrals. Then all jf in the
different amplitudes that pertain to the same face are constrained to be equal and
we obtain
WC (hf ) = NC
jf
SL(2,C)
0
dgve
Trjf [Y ge0 v gve Y Y gev0 gv0 e00 Y ...]
(7.61)
where the trace runs around the face, with the proper orientations. The trace is the
face amplitude.
Alternatively, the wedge amplitude is defined by
W ( g, h) =
(2j + 1) Trj [Y gY h].
(7.62)
It is associated to a wedge, as in Figure 7.7. If we Fourier transform this to the basis

formed by the Wigner matrix elements,
W ( g, j, m, m0 ) = h j, m0 |Y gY | j, mi.
(7.63)
Full theory
155
and take g = eiKz , this gives the amplitude for a transition between the state | j, mi
and the state | j, m0 i as observed by an accelerated observer undergoing a boost of
Lorentzian angle .
Wm,m0 ( ) = h j, m0 |Y eiK Y | j, m0 i.
(7.64)
This matrix element, like several other quantities related to SL(2, C) representations can be found explicitly in [Ruhl (1970)], in terms of an hypergeometric function, and reads
Wm,m0 ( ) = mm0 em e (1+i)( j+1) 2 F1 ((1 + i)( j + 1), m + j + 1, 2j + 2; 1 e2 ).
(7.65)
The full amplitude can then be written in terms of the wedge amplitude in the
simple wedge form
WC (h` ) = NC
0
dhvf dgve
(hf ) W ( ge0 v gve , hvf ).

w
(7.66)
Finally it is often convenient to rewrite the vertex amplitude (7.58) in the form
Av ( h ` ) =
SL(2,C)
dgn0
1
, h ` ],
P[ gs(`) gt(`)
(7.67)
where the `, n notation refers to the nodes and the links of the vertex graph, and
the kernel is
P[ g, h] =
d j Trj [Y gY h].
(7.68)
Explicitly:
P[h, g] =
(j,j)
( j)
(2j + 1) Djm jn ( g) Dm n (h) .
(7.69)
This kernel is a key quantity that appear repeatedly in the calculations. It expresses
the link between SU(2) and SL(2, C) .
7.4.2 Cosmological constant and IR finiteness

In Section 6.3 we have seen that in three dimensions there is a version of the transition amplitudes, given by the Turaev-Viro model, which is finite, and whose
classical limit gives general relativity with a positive cosmological constant. The
same happens in four dimension. The quantum deformation of the SL(2, C) group
needed for defining the theory has been derived by Buffenoir and Roche [Buffenoir
and Roche (1999)] and Noui and Roche [Noui and Roche (2003)], and the theory
has been constructed by Fairbairn and Meusburger [Fairbairn and Meusburger
(2012)] and by Han [Han (2011b)], who has also studied the classical limit of the
156

theory [Han (2011b); Ding and Han (2011)]. In this text we do not enter in the definition of the theory with cosmological constant, because it would require more
detail on quantum groups. We refer to the original papers above.
We only state the two key results:
1 The transition amplitudes of the q-deformed theory are finite.
2 The classical limit of the vertex amplitude is related to the Regge action with
cosmological constant.
The theory with cosmological constant depends on two dimension-full parameters: the Planck length lo and the cosmological constant . One of these can be
taken as the unit of length, so ultimately quantum gravity depends on a single
dimensionless constant, which is a very small number (with respect our limited
human imagination):
h G 10120 .
(7.70)
Since the cosmological constant is of the order of the cosmological scale, this number is of the order of the ratio between the smaller and the largest observable scales
around us.
7.4.3 Variants
There is a number of possible alternatives in the definition of the theory, and aspects that we have left open, which we mention here because they might play a
role in the future.
We have not specified how the normalization factor NC depends on C . We discuss this below, when dealing with the continuous limit. Recall that a nontrivial NC was needed to define the finite Turaev-Viro theory in 3d.
We have defined the theory with p = j. But the alternative
p = ( j + 1)
(7.71)
first suggested by Alexandrov [Alexandrov (2010)] is of substantial interest

and may be preferable; with this choice the linear simplicity condition holds
weakly exactly, and not just for large j [Ding and Rovelli (2010b)]. See the last
section in the Complements of this Chapter.
We have defined the theory on two-complexes and graphs dual to a triangulations. However, more general complexes can be used, and these might be
relevant. First, we can take a generalized triangulation, obtained by gluing
4-simplices, but dropping the requirement that two 4-simplices be connected
by at most one tetrahedron. This is a generalization that we definitely take in
the following.
More generally, we can take an arbitrary two-complex, not necessarily dual
157
Full theory
to a triangulation. In particular, where vertices and edges have arbitrary valence. It is easy to see that the formulas above extent immediately to this
case.This generalization of the theory has been pointed out and developed
by the Polish school [Kaminski et al. (2010a)]. The theory is then defined on
generic two-complexes. In particular, the nodes of the graphs need not be
four-valent. The geometrical interpretation of such nodes continues to hold,
thanks to a theorem, proven by Minkowski, with states that a set of n vectors
satisfying n`=1 ~E` = 0 uniquely defines a polyhedron with n faces, where the
vectors are normal to these faces and have a length proportional to the area of
the faces. The corresponding geometry and in particular the volume has been
explored in particular in [Bianchi et al. (2011b); Dona and Speziale (2010)] and
all the rest go trough. One of the advantages of this version of the theory is
that it becomes closer to the canonical formulation of loop quantum gravity,
where nodes have arbitrary valence.
In the Euclidean version of the theory, the Polish school has also explored the
possibility of modifying the face amplitude term, given by the delta function
in (7.57), it is not yet clear if a corresponding Lorentzian version of this exists
[Bahr et al. (2011)].
The amplitude associated to a given two-complex can be obtained as a Feynman amplitude in the perturbative expansion of an associated quantum field
theory with fields defined on a group [Krajewski et al. (2010)]. For a general
introduction to group field theories (GFT), see [Krajewski (2011)] and references therein. Full GFT amplitudes are given by sums over two-complexes,
and not by refining the two complex, as here. In spite of the apparent difference, the two definitions agree under some conditions (in particular, on the
normalization constants NC ), as shown in [Rovelli and Smerlak (2012)]. It is
not known, at present, if these conditions are satisfied for the physical quantum gravity theory.
An alternative definition of the Y map has been defined in [Baratin et al. (2010)],
starting from group field theory. In the same paper a derivation of the formalism much closer to a simple path integral formulation of the discretized
theory has been developed.
Although it does not appear to be a viable physical theory, we mention here also
the Barrett-Crane theory [Barrett and Crane (2000)], because it has played a
major role in the development of the current theory. The current theory was
in fact found as a modification of the Barrett-Crane theory, in order to address
some shortcomings it had. The Barrett-Crane model can be seen as an appropriate limit of the theory defined in this chapter. We refer to [Perez
(2012)] for a simple introduction of this model.
158
In the next chapter we discuss coherent states techniques. These techniques

can also be used to directly define the theory directly in coherent-state basis. One of the (independent) paths in which the theory presented here was
first found, in the Euclidean context, is using these techniques [Freidel and
Krasnov (2008)]. In the Euclidean context, this approach gives a slightly different theory if > 1, not so in the Lorentzian context, where the direct use
of these techniques is not entirely clear yet. Again, see [Perez (2012)] for a
detailed discussion.
The theory defined by the equations above is of course not written in stone.
Perhaps it needs to be adjusted or modified, in order to describe Nature. That is,
in order to be fully consistent and to agree with experiments. At present, however,
it seems to yield a possibly consistent and complete theory of non-perturbative
quantum gravity, finite and with a classical limit consistent with what we know
about the world.7
7.5 Complements
7.5.1 Summary of the theory
Kinematics

Hilbert space: H = L2 SU(2) L /SU(2) N , where the action of the quotient is given by the
combinatoric structure of the graph .
Operators: Triad operator ~E` = 8Gh~L` where ~L` is the left-invariant vector field (the
derivative in the algebra), acting on the group elements on the links `.
On each node we have the operators G``0 = ~E` ~E`0 . The norm A` = G`` = |~E` |
is the area of the faces of the tetrahedron punctured by the link `.
Volume associated to a node: Vn2 = 92 ~E1 (~E2 ~E3 ); does not depend on the triplet
used because of the gauge invariance at the node.
A` and Vn form a complete set of commuting observables.
States |, j` , vn i spin-network basis. The quantum geometry is discrete and fuzzy in the small
and gives a Riemannian geometry in the large.
Dynamics
Supergravity might still yield a finite theory, but at present we do not know how to extract from
it the nonperturbative information needed to describe Planck scale physics in the four-dimensional
world without exact supersymmetry that we experience. String theory is likely to be finite, but
its connection with the world we experience, and therefore its predictive power, is even frailer.
Theoreticians working on these theories have long predicted low-energy supersymmetry, which so
far hasnt shown-up.
Complements
159
Transition amplitude: Depends on a 2-complex C

WC (h` ) = NC
SU(2)
dhf
hfv Av (hfv )
f
vf
Vertex amplitude: Calling h` = hvf the variables on the links of the vertex graph, and n the
nodes of the vertex graph
Av ( h ` ) =
SL(2,C)
dgn 0
( j)
(j,j)
d j Dmn (h` ) D jm jn ( gn gn0 1 )
The integration is over one gn for each node (edge of v), except one. The 4product
is over 10 faces f , and D ( j) and D ( p k) are matrix elements of the SU(2) and SL(2, C)
representations. These are connected by the simplicity map.
Simplicity map:
Y :
H j H p,k
| j; mi 7 |j, j; j, mi .
with
p = j,
k=j
(7.72)
160
7.5.2 Computing with spin networks

A good and detailed introduction to spin network theory and techniques is [Haggard (2011)],
to which we refer the reader. Here we give a few basics. Recall that the dimension of the
invariant part of the tensor product of three SU(2) irreducible representations is
(
h

i 1 if the triangular inequalities are satisfied
dim Inv H j1 H j2 H j3
(7.73)
0 otherwise.
recall that if the triangular inequalities are satisfied, there exist one invariant state
|i =
m1 ,m2 ,m3
m1
m2 m3
| j1 , m1 i | j2 , m2 i | j3 , m3 i.
Its components are proportional to the Wigner 3j symbol

j1
j2
j3
m1 m2 m3 =
.
m1 m2 m3
(7.74)
(7.75)
To get invariant objects, we can contract magnetic indices by means of the Wigner matrix,
which we use to raise and lower magnetic indices.
gmm0 = (1) jm m,m0 .
(7.76)
m1
(7.77)
The Wigner 3-j symbol is normalized, that is:

m2 m3
m1
m2 m3
= 1.
A diagrammatic notation is very convenient to keep track of the pattern of index contraction. For this, we write
m1
m1
m2 m3
b
ed
j1
j2
m2
Then equation (7.77) reads
(7.78)
j3
m3
j1
j2
= 1= 1.
j3
(7.79)
And in this notation, the Wigner 6j symbol reads
j#
j6
5
edb
j1
j3
j2
"!
j1
j4
j2
j5
j3
j6
j4
With these tools, we can find the matrix elements of the generator of the rotation group
Li in any representation j. These are tensors with two indices in the j representation and
one index in the adjoint. Therefore they must be proportional to the Wigner 3-j symbols

h j, m0 | Li | j, mi = Li 0 m = c i m0 m ,
(7.80)
m
Complements
161
or
Li
m0 m
=cim
(7.81)
Where we have changed basis from the magnetic m = 1, 0 1 basis to the real i = 1, 2, 3 =
x, y, z basis in the adjoint representation of SU(2). The proportionality constant c can be
found by recalling that the Casimir is
~L2 = j( j + 1)1l,
(7.82)
therefore, the trace of the Casimir is j( j + 1) times the dimension of the representation,
which is 2j + 1. Thus
j( j + 1)(2j + 1) = Tr~L2 = ik Lim m Lkm0 m = c2 mm i mm0 i = c2 .
0
This fixes the normalization. Thus we conclude

L
imm0
j( j + 1)(2j + 1)
(7.83)
(7.84)
b
ed
1
m0
Intertwiners
Let us now study the Hilbert space Hn associated to a tetrahedron. This is the invariant part
of the tensor product of four SU(2) irreducibles

Hn = K j1 ...j4 = Inv H j1 H j2 H j3 H j4 .
(7.85)
Tensors in this space are called intertwiners.
They have the form
|i =
m1 ...m4
m1 m2 m3 m4 | j1 , m1 i . . . | j4 , m4 i
(7.86)
and are invariant under the diagonal action of
SU(2). That is
D ( j1 ) m1 n1 (U )...D ( j) m1 n1 (U ) n1 n2 n3 n4 = m1 m2 m3 m4 .
If U = ei~~ and we take || << 1 we can expand
D (U )|i = (1 + i~ ( L j1 + L j2 + L j3 + L j4 ))|i = |i
(7.87)
(7.88)
where each L j acts on a different factor space. That is, intertwiners satisfy the closure constraint.
4
~L` |i = 0 .
(7.89)
`=1
This equation says that the sum of the normals to the triangles of a tetrahedron, normalized
to their areas, vanishes.
It is not difficult to find a basis of invariant tensors, starting from the trivalent intertwines.
Indeed

0

m1 m2 m
m m3 m4
1 m2 m3 m4
0
2k
+
1
g
(7.90)
m
=
mm
k
j1
j2
k
k
j3
j4
is clearly invariant. Equation (7.90) can be expressed diagrammatically as
162
m1
m4
de
ed
k
m2
m1
2k + 1
m3
m4
defed
k
m2
(7.91)
m3
with a virtual link with spin k in the node.

Exercise: Given the graph 1 with two nodes connected by three links oriented from the first
to the second node, and the graph 2 formed by four nodes all connected among themselves.
Write the corresponding spin network states j1 j2 j3 (Ul ) and j1 j2 j3 j4 j5 j6 (Ul ).
Exercise: Consider now the first graph and the state 1 = Tr (UVWV ), where Ul =
(U, V, W ). Is it a physical state? What about 2 = Tr (UV 1 WV 1 )? If so, expand it in
the spin network basis.
Exercise: Consider second graph and the following states:
h
i
3 = Tr ( X 1 VY )7 ,
(7.92)
Which of them belong to physical Hilbert space, with what identification between the group
elements and the oriented links?
1 = Tr(UWV 1 ) ,
2 = Tr(VYZW ) ,
3 = Tr( X 1 UWY ) ,

Exercise: Consider the state |1 , 12 , 1, 12 . Write this state as a sum of products of loops,
where a loop is the trace of a product of a sequence of Us along a closed cycle on the graph.
Any spin network state can be written in this way. This is the historical origin of the denomination loop quantum gravity for the theory. [Hint: You may want to read a discussion
on p. 237 of [Rovelli (2004)].]
Observables
Here we introduce the observables in Hn . An observable O commutes with the constraint
(7.89) C = 4`=1 ~L` |i = 0 for the generator of the rotations. That is [O, C] = 0. Two of these
observables are (~L1 + ~L2 )2 and ~L2` .
[ ~L2` , C ] = 0
[ (~L1 + ~L2 )2 , C ] = 0
[ (~L1 + ~L2 )2 , ~L2 ] = 0
(7.93)
(7.94)
(7.95)
They are invariant under rotations. They can be diagonalized simultaneously. They form a
maximal set of commuting observables
~L2 |i = j` ( j` + 1)|i
`
(7.96)
(~L1 + ~L2 )2 measures k

(~L1 + ~L2 )2 | k i = k(k + 1)| k i
(~L3 + ~L4 )2 | k i = k(k + 1)| k i
where
where
| j1 j2 | k j1 + j2
| j3 j4 | k j3 + j4
(7.97)
(7.98)
The range of k is dictated by the triangular inequalities. Thus dim Hn = k max k min + 1
The states | k i form an orthogonal basis because they are eigenstates of an Hermitian
operator.
Complements
163
Problem
Verify explicitly that the states |ik i are orthogonal and normalized.
Solution
The proportionality constant in
(7.99)
can be determined by closing both sides of the equation (Wigner-Eckart theorem). The left
hand side gives unit, so that = 1/(2k + 1). Then
j1
j2
h k | k0 i =
2k + 1 2k0 + 1
k0 =
2k + 1 2k0 + 1
j3
j4
1
0.
2k + 1 kk
(7.100)
7.5.3 Spectrum of the volume

Recall that the volume operator reads
q
3
2
Vn =
(8Gh) 2 |~L1 (~L2 ~L3 )|
3
(7.101)
We have four vectors ~Li but any (oriented) triplet gives the same result because of the constraint
~L1 + ~L2 + ~L3 + ~L4 = 0.
(7.102)
The volume operator has eigenstates | v i
Vn | v i = v| v i
(7.103)
In order to compute | v i we start by computing
~L1 (~L2 ~L3 ) | q i = q| q i

Then we will have
Vn | q i =
3
2
(8Gh) 2
3
We need the matrix elements of this operator
|q|| q i.
Qk k = h k | ~L1 (~L2 ~L3 ) | k0 i

0
(7.104)
(7.105)
(7.106)
these form a d d matrix with eigenvectors h k | q i . We can compute the matrix elements
and diagonalize.
164
j1
j2
k0
k0
j3
j4
j1
j2
Figure 7.8
k0
1
j3
j4
A grasping is a connection made by a line carrying spin 1, using (7.84).

Let us start by observing that
i h
i
i h
1h~
j
j
( L1 + ~L2 )2 , ~L1 ~L3 = ~L1 ~L2 , ~L1 ~L3 = L1i , L1j L2i L3 = ieijk L1i L2 L3k = i~L1 (~L2 ~L3 ).
2
(7.107)
Using this,
0
Qk k = h k | ~L1 (~L2 ~L3 ) | k0 i
i
= h k | [(~L1 + ~L2 )2 , ~L1 ~L3 ] | k0 i
2
i
= h k | (~L1 + ~L2 )2~L1 ~L3 | k0 i h k | ~L1 ~L3 (~L1 + ~L3 ) | k0 i
2
i
= (k(k + 1) k0 (k0 + 1))h k | ~L1 ~L3 | k0 i.
2
(7.108)
(7.109)
(7.110)
(7.111)
The last term can be computed diagrammatically (Figure 7.8) using

j1
j2
j1
j2
k
1
j3
j4
=
k0 = k + 1
k
k
1
j3
j4
k+1k1
k+1k1
j1
k
1
j2
j1
k1

j3
k

j3
k1
(7.112)
1
j4
That is, cutting the graph by inserting a resolution of the identity in terms of trivalent
intertwines. The eigenvalues are complex conjugate (see k + 1 k 1 above). We need to
compute just one so that we get automatically the other. This gives
q
q
h k | ~L1 ~L3 | k0 i = 2k + 1 2k0 + 1 j1 ( j1 + 1)(2j1 + 1) j3 ( j3 + 1)(2j3 + 1).

(7.113)
times the two tetrahedral graphs in Figure 7.8.
Notice that the triangular inequalities in these graphs force k and k0 to different precisely
by one unit. A little algebra gives finally the the only non vanishing matrix elements are
1
ak = Qk k1 = (k(k + 1) (k 1)k) 2k + 1 2k 1
2

q
q
j1
j1 ( j1 + 1)(2j1 + 1) j3 ( j3 + 1)(2j3 + 1)
k
1
j2
j1
k1

j3
k
1
j4
(7.114)

j3
k1
Complements
165
and their transposes, and the matrix has the form
0
+ a1
0
a1
0
+ a2
k0
Qk = 0
a2
0
..
.
...
..
(7.115)
Matrices of this form have several peculiar proprieties, in particular they have a non-degenerate
spectrum. The diagonalization of these matrices, however, is not straightforward, and must
be done numerically.
Since the spectrum is not degenerate, we can use the eigenbasis of the volume and use
the eigenvalues of ~L1 (~L2 ~L3 ) as good labels for the basis. Notice that in the matrix the
eigenvalues have opposite sign, but when we take the modulus in the volume we are going
to miss this sign: at each node we need to specify the corresponding volume and orientation.
There are special values of j for which the {6j} have a simple analytic form. They can be
found on Landaus book or with Mathematica.

2( a + 21 , b + 12 , c)
a 1
a
= (1)(a+b+c) p
(7.116)
c b c1
a( a + 1)(2a + 1) 2c + 1 2c 1 c
p
where ( a, b, c) = 14 ( a + b + c)( a + b c)( a b + c)( a + b + c) is Herons formula for
the area of a triangle of sides ( a, b, c). Using this,
2( j1 + 12 , j2 + 12 , k ) 2( j3 + 21 , j4 + 12 , k)
1

ak = 2k
2
2k + 1 2k 1 k
2k + 1 2k 1 k
4
1
1
1
1
=
( j1 + , j2 + , k)( j3 + , j4 + , k)
2
2
2
2
2
4k 1
(7.117)
(7.118)
Example
Take the case with all spin equal j1 = j2 = j3 = j4 = j Then the above formulas simplify to
ak =
In particular:
If j =
1
2
1 k2 (k2 (2j 1)2 )
.
4
4k2 1
(7.119)
, then a1 = 43 , the possible virtual spins are k = 0, 1, the dimension of the node
Hilbert space is d = 2, and the matrix is

0
3
0 +i
.
Qk k =
i 0
4
This has eigenvalues q =
3
4
and q+ = +
3
4
and eigenvectors
1
|q i = (|k = 0i i |k = 1i) .
2
(7.120)
Therefore the full (oriented) volume operator has these eigenvectors with eigenvalues (the sign codes the orientation)
s
3
2
3
(8Gh) 2
(7.121)
V =
3
4
166
If j = 1 , then a1 = 2 , a2 = 5 . The possible virtual spins are k = 0, 1, 2, the dimen3

3
sion of the node Hilbert space is d = 3, and the matrix is
0
2i
0
1
k0
2i
0
5i
(7.122)
Qk =
3
0
5i
0
The eigenvalues are qo = 0 and q = 3 and the eigenvectors |qo i = 25 + |2i

and |q i = 2 |0i 3i |1i + |2i. The eigenvalues of the volume are
5
v0 = 0,
vI =
3 1
2
(8Gh) 2 3 4 .
3
(7.123)
Problem
Recall the volume formula
j
V 2 = ceijk Lij1 L j2 Lkj3 1l j4
(7.124)
where L ji are in ji representation of SU(2) and similarly 1l j4 is the identity matrix in j4 representation, and c is some constant.
1. Derive the value of c. [Hint: Take a tetrahedron having three sides equal to the three
orthogonal basis vectors in R3 .]
2. Prove that the volume does not depend on which three (out of four) links we choose.
2 , this gives the same
In other words, if we denote the volume operator in (7.124) as V123
2
2
result as if we had chosen V124 or V234 . [Hint. Recall that on physical states we have
j
( L11 + L22 + L33 + L44 ) = 0 .]

3. Calculate the volume eigenvalues for K11 1 1 and K1111 .
2 2
7.5.4 Unitary representation of the Lorentz group and the Y

map
We review here the unitary representation of the Lorentz group. More precisely, the representations of SL(2, C), the double cover of SO(3,1), the component of the Lorentz group
connected to the identity. The Lie algebra of SL(2, C) is
[ J I J , J KL ] = IK J JL + IL J JK + JK J IL JL J IK .
(7.125)
We fix an SU(2) subgroup of SL(2, C) (as groups of matrices, SU(2) is simply the subgroup
of the unitary matrices, of course.) In the abstract groups, this depends on the choice of a
Lorentz frame, namely a timelike vector t I = (1, 0, 0, 0). SU(2) is the little group that preserves t I . The generators J I J of SL(2, C) then split into rotations and boosts.
1 I
e
J JK t L = (0, Li )
2 JKL
K I = J I J t J = (0, K i )
LI =
(7.126)
(7.127)
with the Lie algebra
[ Li , L j ] = eij k Lk ,
[ Li , K j ] = eij k K k ,
[K i , K j ] = eij k Lk .
(7.128)
Complements
167
Finite dimensional representations

The unitary representations of SL(2, C), should not be confused with he well-known finite
dimensional representations of the Lorentz group, which are ubiquitous in fundamental
theoretical physics. These are for instance:
1. The fundamental, of Weyl, or spinor representation. This is defined on C2 . Recall we
write spinors in C2 as z = z A = (z0 , z1 ). The generators of the rotations are the Li =
i = i 2i matrices, precisely as the fundamental SU(2) representation. The generators
of boosts are the non Hermitian operators K i = 2i .
2. The vector representation. Is the fundamental of SO(3,1), formed by the Minkowski vectors: X 0 I = I J X J , with I, J = 0, 1, 2, 3.
3. The adjoint representation. This is given by tensors with two antisymmetric Minkowski
indices, like the electromagnetic field F I J . If a specific frame is fixed, or a specific time
direction n I , then these tensors split into a magnetic and an electric part, as we have
seen for B. In particular, the generators of the algebra, which are of course in the adjoint
representation, split into boosts and rotations.
4. Etcetera.
These representations are all non-unitary. There is no scalar product defined on them. (The
Minkowski scalar product X I X I = ( X 0 )2 | ~
X |2 is not positive definite.) Let us therefore
move to the unitary representations.
Infinite dimensional representations

The representations that are of interest for us are the unitary irreducible representations in
the principal series. These are described in detail in the book [Ruhl (1970)]. SL(2, C) has
rank 2 and therefore two independent Casimirs. These can be taken to be
1
~ 2 ~L2 ,
J JIJ = K
2 IJ
1
~ ~L.
C2 = e I JKL J I J J KL = K
8
(7.129)
C1 =
(7.130)
Accordingly, the unitary representations are labelled by two quantum numbers: p R

and k 12 N. The representation space is V ( p,k) and is infinite dimensional: dim V ( p,k) =
. On this space act the Hermitian generators J I J = J J I . The value of the Casimirs in
these representations is given above in (7.12). V ( p,k) is a reducible representation of SU(2).
It admits the decomposition (7.10) into an infinite number of (different) SU(2)-invariant
blocks For instance, if U SU(2),
( p, 12 )
(U ) =
..
(7.131)
where there first 2 2 matrix is D ( 2 ) and so on. The generators of rotations preserve this
decomposition (the blocks are invariant under SU(2) transformation) while the boost generator K i takes out of these blocks. It sends Hi in H j1 H j H j+1 .
The orthonormal basis | p, k; j, mi of V ( p,k) , called the canonical basis, is obtained diagonalizing the operators C1 , C2 , ~L2 and Lz . The generators in this basis are ([Gelfand et al.
168
(1963)], page 187-189)

L3 | j, mi =m| j, mi,
q
L+ | j, mi = ( j + m + 1)( j m)| j, m + 1i,
q
L | j, mi = ( j + m)( j m + 1)| j, m 1i,
q
q
K3 | j, mi = ( j) j2 m2 | j 1, mi ( j) m| j, mi + ( j+1) ( j + 1)2 m2 | j + 1, mi,
q
q
K + | j, mi = ( j) ( j m)( j m 1)| j 1, m + 1i ( j) ( j m)( j + m + 1)| j, m + 1i
(7.132)
q
( j+1) ( j + m + 1)( j + m + 2)| j + 1, m + 1i,
q
q
K | j, mi =( j) ( j + m)( j + m 1)| j 1, m 1i ( j) ( j + m)( j m + 1)| j, m 1i
q
+ ( j+1) ( j m + 1)( j m + 2)| j + 1, m 1i,
where
and
L = L1 iL2 ,
K = K1 iK2
s
kp
i ( j2 k2 )( j2 + p2 )
.
,
( j) =
( j) =
j
j ( j + 1)
4j2 1
(7.133)
Problem
Show that if we define the map Y by
p = ( j + 1),
(7.134)
k=j
(instead than p = j, k = j), then the simplicity conditions are weakly realized exactly and
not just in the large j limit.
Solution
p
Recall that
| jmi =
j( j + 1)(2j + 1). Let us compute these matrix
m0 where c =
elements on the image of Y: h p, j; jm0 | Li | p, j; j, mi is invariant in the space Inv( j j 1).
There is only such an object that is the {3j} and the same computation that fixes the coefficient in the SU(2) representations holds here. Thus h p, j; j, m0 | Li | p, j; j, mi = c im m0 .
To do the same for K i , we need care because K i goes out of the SU(2) representation
spaces. But what we actually need is to compute this on the image of the map Y
h jm | L I
c im
h p, j; j, m0 | K i | p, j; j, mi = im m
(7.135)
i m0
~ Tr 1l j = (2j + 1) pj = cimm0 imm0 = c
Lim m0 Km
= ~L K
(7.136)
(2j + 1) pj
p
=
=
.
c
j( j + 1)(2j + 1)
j+1
(7.137)
On the left and on the right we have the same spin, therefore we are still inside a block of
(7.131), that is in fact the one with the lowest weight. We have
because the intertwiner is normalized. The ratio of the two proportionality factors gives
169
Complements
If p = ( j + 1), this c = and the ratio does not depend on j. Thus, if we consider the map
~ | i =
Y defined by p = ( j + 1) and we consider the states |i, |i Imm Y then h | K
~L | i, namely
h
|
c
~ | i = h | ~L | i .
h | K
(7.138)
~ and ~L
On the Imm Y we have k = j and p = ( j + 1), therefore the matrix elements of K
satisfy the relation
~ = ~L
K
(7.139)
Because of this intriguing result, the theory may be better defined with p = ( j + 1) instead than p = j [Alexandrov (2010); Ding and Rovelli (2010a)]. The question is open.
Classical limit
The theory defined in the previous chapter does not resemble much general relativity. Where are the Einstein equations? Where is Riemannian geometry? Where
is curvature? And so on. In this chapter, we show how the classical limit emerges
from the quantum theory defined in Chapter 7.
If we start from the Schrodinger

equation of a non relativistic particle, we can
recover the dynamics of a classical particle by studying the evolution of wave
packets. A wave packet is a state where both position and momentum are not
too quantum-spread. In the limit where the actions in play are much larger than
the Planck constant, wave packets behave like classical particles. A useful class of
wave-packet states is given by the coherent states, which are a family of states
labelled by classical variables (position and momenta) that minimize the spread of
both.
Coherent states are the basic tool for studying the classical limit also in quantum gravity. They connect the quantum theory defined in the previous chapter
with classical general relativity. In this chapter we study the coherent states in the
Hilbert space of the theory, and their use in proving the large distance behavior of
the vertex amplitude and connecting it to the Einsteins equations.
8.1 Intrinsic coherent states

Consider a state in the spin network basis |, j` , vn i. Can we associate a 3d classical geometry to it? At first sight, one is tempted to answer as follows: take a
triangulated 3d manifold obtained by gluing flat tetrahedra where triangles have
areas j` and tetrahedra have volume vn . But a moment of reflection shows that this
strategy is weak: the geometry of a tetrahedron is only partially specified by giving areas and volume. The shape of a tetrahedron is specified by six numbers, not
five. Therefore a state |, j` , vn i is in fact highly non classical, in the sense that some
variables of the intrinsic geometry it describes are maximally quantum-spread.
The same problem can be seen from a different perspective. Fix the four areas
of a tetrahedron. The resulting quantum state space is the Hilbert space Hn associated to a single node of the graph. A basis in this space is given by the states |ik i,
defined in (7.35). We have seen that these are eigenstates of ~L1 ~L2 . They diagonalize the dihedral angle 12 between the faces 1 and 2. On the other hand, the angle,
say 13 between the faces 1 and 3, namely the observable ~L1 ~L3 is spread.
Given a classical tetrahedron, can we find a quantum state in H such that all
170
Intrinsic coherent states
171
the dihedral angles are minimally spread around the classical values? The solution
to this problem is represented by the intrinsic coherent states, which we describe
below.
8.1.1 Tetrahedron geometry and SU(2) coherent states

Lets reconsider the geometry of a classical
tetrahedron sketched in the first Chapter. A
tetrahedron in flat space can be determined by
giving three vectors, ~ea with a = 1, 2, 3, representing three of its sides emanating from a
vertex P. If we choose a non-orthogonal coordinate system where the axes are along these
vectors and the vectors determine the unit of
coordinate length, then eia is the triad and
h ab = ~ea ~eb .
~e1
~e3
~e2
(8.1)
is the metric in these coordinates. The three vectors

~E a = 1 e abc ~eb ~ec
(8.2)
2
are normal to the three triangles adjacent to P and their length is the area of these
faces. The products
~E a ~Eb = (det h) h ab
(8.3)
define the matrix h ab which is the inverse of the metric h ab = ~ea ~eb . The volume of
the tetrahedron is
q
1
1
V = ~e1 (~e2 ~e3 ) =
23 |~E1 (~E2 ~E3 )|.
(8.4)
3!
3!
We extend the range of the index a to 1, 2, 3, 4, and denote all the four normals, normalised to the area, as ~E a . It is easy to show that these satisfy the closure condition
4
~Ea =
a =1
Aa ~na = 0.
(8.5)
f =1
(Physical intuition: a vector quantity normal to a face and proportional to the area
is the force due to the pressure. If we put the tetrahedron in a gas and increase the
pressure, the tetrahedron does not move, namely the sum of all the forces gives
zero. This is what indicates the closure equation (8.5).)
The dihedral angle between two triangles is given by
~E1 ~E2 = A1 A2 ~n1 ~n2 = A1 A2 cos 12 .
(8.6)
Classical limit
172
Now we move to the quantum theory. Here, the quantities ~E a are quantized as
~Ea = 8Gh~L a
(8.7)
in terms of the four operators ~L a , which are the (hermitian) generators of the rotation group (recall (1.12)!):
j
[ Lia , Lb ] = iab eij k Lka .
(8.8)
The commutator of two angles is

j
[~E1 ~E2 , ~E1 ~E3 ] = (8Gh)4 [ L1i , L1j ] L2i L3
= (8Gh)
j
ieijk L1i L2 L3k
= (8Gh)~E1 (~E2 ~E3 )
(8.9)
(8.10)
(8.11)
From this commutation relation, the Heisenberg relation follows:

1
(8.12)
(~E1 ~E2 ) (~E1 ~E3 ) 8Gh|h~E1 (~E2 ~E3 )i|
2
p
where h Ai = h | A | i and A = h | A2 | i (h | A | i)2 . Now we want to look
for states whose dispersion is small compared with their expectation value: semiclassical states where
(~Ea ~Eb )
1
a, b.
(8.13)
|~Ea ||~Eb |
SU(2) coherent states

Consider a single rotating particle. How to write a state for which the dispersion
of its angular momentum is minimized? If j is the quantum number of its total
angular momentum, a basis of states is
| j, mi H j
(8.14)
[ L x , Ly ] = iLz
(8.15)
Since
we have the Heisenberg relations

L x Ly
1
| h Lz i |
2
(8.16)
Every state satisfies this inequality. Can we saturate it? A state that saturates (8.15),
namely for which L x Ly = 12 | h Lz i | is given by | j, ji. This can be shown as follows. | j, ji is an eigenstates of Lz
Lz | j, ji = j| j, ji
(8.17)
173
~
L
Figure 8.1
The angular size of the cone that gives the spread goes as 1/
j.
so that Lz = 0. Then
h L x i = h Ly i = 0
1
1
j
1
h L x i 2 = h L2x + L2y i = h L2 L2z i = ( j( j + 1) j2 ) = .
2
2
2
2
Therefore we conclude that
L x = Ly =
j
.
2
(8.18)
(8.19)
(8.20)
Therefore the state | j, ji saturates the uncertainty relations

L x Ly =
j
2
1
j
| h Lz i | = .
2
2
(8.21)
In the large j limit we have

q
1
L
2
q x = p
= p
j 0
j ( j + 1)
2( j + 1)
h~L2 i
(8.22)
Therefore this state becomes sharp for large j.

The geometrical picture corresponding to this calculation is transparent: the
state | j, ji represents a spherical harmonic maximally concentrated on the North
pole of the sphere, and the ratio between the spread and the radius decreases with
the spin.
Thus we have found a good candidate for a coherent state: | j, ji. What about the
other states of the family? Well, these are easily obtained rotating the state | j, ji
into an arbitrary direction ~n. Let us introduce Euler angles , to label rotations,
are as in Figure 8.2. Then let zi = (0, 0, 1) and define the matrix R SO(3) of the
form R = eiLz eiLy , by R~n ij z j = ni . With this, define
| j, ~ni = D~n ( R)| j, ji.
(8.23)
The states | j, ~ni form a family of states, labelled by the continuous parameter ~n,
which saturate the uncertainty relations for the angles.
Classical limit
174
~n
Figure 8.2
~n
For a generic direction ~n = (n x , ny , nz )

Some of their properties are the following.
~L ~n| j, ~ni = ( D ( R) Lz D 1 ( R))| j, ~ni

= ( D ( R) Lz D
(8.24)
(8.25)
( R)) D ( R)| j, ji = D ( R) Lz | j, ji
= j D ( R) | j, ji = j| j, ~ni.
(8.26)
Therefore
h j, ~n | ~L | j, ~ni = j ~n
and
~)=
(~L m
~ )2
1 (~nm
(8.27)
r
j
.
2
The expansion of these states in terms of Lz eigenstates is

j
where m (~n) = h j, m | D ( R) | j, ji = D ( j) ( R)m .
| j, ~ni = m (~n)| j, mi
m
(8.28)
The most important property of the coherent states is that they provide a resolution of the identity. That is
1l j =
2j + 1
4
S2
d2~n | j, ~nih j, ~n|.
(8.29)
The left hand side is the identity in H j . The integral is over all normalized vectors,
therefore over a two sphere, with the standard R3 measure restricted to the unit
sphere.
Exercise: Prove this. (Hint: use the definition of the coherent states in terms of
Wigner matrices, note the invariance under U(1) and promote the S2 integral to an
SU(2) integral, use the Peter-Weyl theorem.)
Finally, observe that by taking tensor products of coherent states, we obtain coherent states. This follows from the properties of mean values and variance under
tensor product.
175
8.1.2 Livine-Speziale coherent intertwiners

Armed with these tools, it is now easy to introduce coherent tetrahedra states. A classical tetrahedron is defined by
the four areas A a and the four (normalized) normals ~n a , up
to rotations. These satisfy 4a=1 ja~n a = 0. Therefore let us
consider the coherent state
| j1 , ~n1 i | j2 , ~n2 i | j3 , ~n3 i | j4 , ~n4 i
(8.30)
in H1 . . . H4 and project it down to its invariant part in

the projection
P : H1 . . . H4 Inv(H1 . . . H4 )
(8.31)
|| ja , ~n a i P(| j1 , ~n1 i | j2 , ~n2 i | j3 , ~n3 i | j4 , ~n4 i)
(8.32)
The resulting state
is the element of H that describes the semiclassical tetrahedron. The projection

can be explicitly implemented by integrating over SO(3)
|| ja , ~n a i =
SO(3)
dR(| j1 , R~n1 i | j2 , R~n2 i | j3 , R~n3 i | j4 , R~n4 i).
(8.33)
Or, more properly, over the SU(2) action on the states
|| ja , ~n a i =
SU(2)
dh ( D j1 (h)| j1 , ~n1 i D j2 (h)| j2 , ~n2 i D j3 (h)| j3 , ~n3 i D j4 (h)| j4 , ~n4 i).
(8.34)
These are called the Livine-Speziale coherent intertwines, and are essential tools
in analysing the theory.
It can be shown that if we expand this state in any intertwiner basis
|| ja , ~n a i = k (~n a )| k i
(8.35)
we find that for large j the coefficients k (~n a ) = m1 m2 m3 m4 m1 (~n1 ) . . . m4 (~n4 )

1
( k k o )2
have the form k (~n a ) e 2 2 eik : they are concentrated around a single value
k which determines the value of the corresponding dihedral angle, and have a
phase such that when changing basis to a different intertwined basis, we still obtain a state concentrated around a value.
Some properties of these states are the following. For large j,
h(n` ) | Ea Eb | (n` )i ja jb~n a~nb
and
(~E A ~Eb )
1
| Ea || Eb |
(8.36)
Finally, by combining coherent intertwiners at each node, we can define a coherent state in H which is a wave packet peaked on a classical triangulated geometry:
j` ,~ns
,~nt` (U` )
= ` D ( j` ) (U` ) n n (~n` ).
(8.37)
Classical limit
176
Thick wedges
Thin wedge
Triangle
Figure 8.3
Tetrahedron
Thin and thick wedges between spacelike tetrahedra in Minkowski space.
8.1.3 Thin and thick wedges and time oriented tetrahedra

If we consider a compact region in a pseudo-Riemannian manifold bounded by
a spacelike surface , the outgoing normal to will be partially future oriented
and partially past oriented. In a triangulation of , some of the tetrahedra will be
future oriented and some past oriented. In such a Lorentzian context, a coherent
intertwiner can be obtained from the Euclidean one (8.34) by time reversing the
normals of past-pointing tetrahedra [Bianchi and Ding (2012)]. Time reversal is a
well defined operator in the representations we use and its effect on a coherent
states is given by
T | j, ~ni = (1) j | j, ~ni,
(8.38)
and thus on the (past-pointing) coherent intertwiner by

T || ja , ~n a i = (1)b>a jab || ja , ~n a i.
(8.39)
A link of the graph between two nodes represents a triangle joining two tetrahedra. In spacetime, this triangle is the edge of a wedge, which can then be of two
types: a thick wedge if the incident tetrahedra have the same time orientation, a thin
wedge otherwise [Barrett et al. (2010)].
When dealing with 3d geometries embedded in spacetime, is convenient to introduce a quantity ` to denote the Lorentzian geometry of the link ` in the following way:
` =
0, thick wedge
, thin wedge
(8.40)
We shall use this in the last Chapter.

Before using these states to discuss the classical limit of quantum gravity, let us
pause to introduce some facts about spinors, since these coherent states simplify
considerably when written in spinor language.
Spinors and their magic
177
8.2 Spinors and their magic

The fundamental representation of SL(2, C) is defined on the same vector space as
the fundamental representation of SU(2), namely H 1 = C2 , the space of couples of
2
complex number. The elements of this space are called spinors and are fundamental
and ubiquitous objects. Something magic about spinors is how they encode three
dimensional and four dimensional physical objects, rotations and Lorentz transformations. Michael Atiyah, one of the great living mathematicians has written
about spinors:
Nobody fully understands spinors. Their algebra is clear, but their geometrical significance remains
mysterious. In a sense, they describe the square root of geometry. To fully understand the square
root of -1 has taken centuries, the same could be true for spinors.1
We have already mentioned spinors repeatedly in the text. Here we summarise

and expand on their properties and illustrate their use to describe coherent quantum states of physical space.
Recall that a spinor z C2 is a two-dimensional complex vector
0
z
z=
.
z0 , z1 C
(8.41)
z1
There is a large and confusing variety of notations used to denote spinors. Depending on the context, we use the notations
0
z
= z A = | z i.
(8.42)
z=
z1
The space of these objects carries the fundamental representation of SL(2, C)

0
a c
z
,
g SL(2, C)
(8.43)
gz =
b d
z1
where ac bd = 1, and of course a representation of the subgroup SU(2)

0
a b
z
hz =
,
h SU(2).
b a
z1
(8.44)
From now on g will always indicate an SL(2, C) element and h an SU(2) element.
There are two bilinear forms which are defined on this space. The first is
(w, z) = n A z B e AB = w0 z1 w1 z0 .
(8.45)
( gw, gz) = (w, z).
(8.46)
It is antisymmetric and nvariant under SL(2, C)

The second is the scalar product of H 1
2
hw|zi =
w A zB
Quoted in [Farmelo (2009)], Chapter 31.
AB = w0 z0 + w1 z1 .
(8.47)
Classical limit
178
This is invariant under SU(2)
h hw | hz i = h w | z i
(8.48)
but not under SL(2, C). (This is the reason why the fundamental representation of
SL(2, C) is not unitary.) Indeed
h gw | gz i = h w | ( g g)z i
(8.49)
If we write g as a rotation times a boost, the rotation cancels in ( g g) but not the
boost. Therefore the scalar product depends on the choice of a given frame, namely
an SU(2) subgroup of SL(2, C). It has information about the choice of a Lorentz
frame. The same information can be coded in the map J : C2 C2 defined by
!
0
z
z1
J
=
(8.50)
z1
z0
In fact, it is easy to see that
h w | z i = ( Jw, z).
(8.51)
In words: we can define the (frame dependent) scalar product in terms of the
(frame independent) bilinear form and the (frame dependent) map J. These are
the basic structures of spinor space.2
8.2.1 Spinors, vectors and bivectors

Spinors and 3-vectors
Let us now see why spinor space is the square root of geometry, in the words of
Atyah. The point is that to each spinor n C2 we can associate a three-dimensional
real vector.
~n = hn |~ | ni.
2
(8.56)
We write z A instead of the lighter notation z A to avoid confusion with the notation used for instance by Wolfgang Wieland, where z0 and z1 do not indicate the complex conjugate of z0 and z1
respectively, but rather the complex conjugate of z1 and z0 respectively. Wieland notation is
z A = z A , z A = e AB z B , z A = ( Jz) A , z A = ( Jz) A ,
(8.52)
which leads to
h z | w i = z A w A ,
(z, w) = z A w A .
(8.53)
Another notation used in the field is the one used by Freidel and Speziale, who write
|zi = z,
|z] = Jz,
(8.54)
so that
[z|wi = (z, w).

One advantage of this notation is that it renders explicit the fact that ( , ) is not symmetric.
(8.55)
179
Figure 8.4
Penrose representation of a spinor: a vector with a flag.

The Pauli matrices are hermitian and this vector is real. Explicitly, it is given by
2Re(n0 n1 )
~n = 2Im(n0 n1 ) .
(8.57)
| n0 |2 | n1 |2
With a norm
|~n|2 = h n | n i .
(8.58)
|nihn| = |~n|1l + ~n ~
(8.59)
Viceversa,
where the left hand side is the matrix n B n A . As you can see, the map from spinors
to vectors is quadratic, therefore a vector is the square of a spinor, and a spinor
is the square root of a vector. Now a real number x has two square roots: x.
Similarly, a vector has more than one spinor square root. This is simple to see
since n and ei n clearly give the same vector. Therefore a spinor has more information than a vector: in addition there is the phase . Roger Penrose pictures spinors
as arrows with a little flag: the arrow indicates the vector, the flag (which can rotate
around the arrow) indicates the phase.
The important point to absorb here is that we can associate a normalized spinor
n (with a free phase) to a normalized vector ~n.
Spinors and 4-vectors

But spinors also have a four-dimensional interpretation in Minkowsky space. For
this, it is sufficient to construct the quadruplet of matrices I = (1l,~ ) and extend
the definition (8.56) to
n I = h n | I | n i.
(8.60)
remarkably, this transforms as a vector under the action of SL(2, C) on the spinor
(try). The vector (8.60) is null, because it follows from (8.58) that
n0 = |~n|,
(8.61)
and therefore n I n I = (n0 )2 |~n|2 = 0. Therefore a spinor determines a vector in

space and null vector in spacetime.
Classical limit
180
Spinors and bivectors

But suppose now that we have now a spinor in H 1 (as opposed to a spinor in
2
C2 ). That is, we have a spinor, but also a scalar product defined in C2 , or, equivalently, the map J. Then a spinor defines a spacelike bivector in Minkowski space.
This is quite obvious, since a three-vector and a Lorentz frame define a timelike
two-plane in Minkowski space, and its dual defines a spacelike plane. Equivalently, in 3d a vector determines a plane, and this plane can be seen as a plane
in Minkowski space if space is a slice of Minkoski space. For instance, the vector
(0, 0, 1) in the frame where the time direction is (0, 0, 0, 1) determines the spacelike
plane (0, x, y, 0). Lets see how this works explicitly with spinors. The bivector is
defined by
n I J = e I J KL hn | [K | ni h Jn | L] | Jni.
(8.62)
Indeed, recall that J acts as parity. Then, for instance, the spinor n = (1, 0) determines the 3-vector ~n = (0, 0, 1), the null 4-vector hn | K | ni = (1, 0, 0, 1), the parity reversed null four vector h Jn | L | Jni = (1, 0, 0, 1) and therefore the bivector
[I J]
n I J = 2 3 , that determines the plane 23.

Therefore a spinor determines a spacelike bivector. This relation plays an important role in the theory. On the boundary, spinors are associated to (oriented) links,
and represent the vector normal to a triangle. In the bulk, they are associated to
faces, and represent the bivector parallel to the triangles of the bulk triangulation.
All this is realized by the coherent states, where spinors play a crucial role.
8.2.2 Coherent states and spinors

The spinor n = (1, 0) is the eigenvector of Lz with eigenvalue 12 and unit norm.
Therefore it is the state | j = 21 , m = 12 i, which is a coherent state. Since all other
coherent states in the j = 12 representation are obtained rotating | 12 , 12 i and since
rotation preserves the norm of the spinor, it follows that all normalised spinors n
describe coherent states in the fundamental representation, that is
|ni = |1/2, ~ni.
(8.63)
where n and ~n are related by (8.56). Indeed, from (8.56) it follows that
~
1
hn | ~L | ni = hn | | ni = ~n = j ~n.
2
2
(8.64)
Normalized spinors are coherent states for the normalized 3-vectors they define.
This is quite remarkable, and even more remarkably is that it extends to any representation. Indeed, the tensor product of coherent states is a coherent state, as can
be checked from the definitions. Consider the state
| j, ni = n ... n
(8.65)
181
2j times. This is in the spin-j representation, and is precisely the coherent state
| j, ~ni that satisfies
h j, n | ~L | j, ni = j~n.
(8.66)
and is minimally spread.
8.2.3 Representations of SU(2) and SL(2,C) on functions of

spinors and Y map
In the Complements of the first Chapter we have constructed the irreducible representations of SU(2) by tensoring spinor space. The spin-j representation space H j
can be realised by symmetric tensors y A1 ...A2j with 2j spinor indices. There is another realisation of the spin spaces H j which is particularly useful. It is defined on
a space of functions f (z) of spinors. More precisely, the finite dimensional vector
space H j can be realised as the space of the totally symmetric polynomial functions f (z) of degree 2j. This is indeed quite immediate, starting from the y A1 ...A2j
states: the corresponding polynomial function of z is simply
f (z) = y A1 ...A2j z A1 ... z A2j
(8.67)
and satisfies the homogeneity condition

f (z) = 2j f (z).
(8.68)
Clearly SU(2) acts on these functions via its transpose action
(U f )(z) = f (U T z).
(8.69)
Let us see how coherent states look in this representation. For spin 1/2, clearly
a coherent state is represented very simply by the linear function
f n (z) n A z A h z | n i
(8.70)
up to a normalization. If we take the symmetrized tensor product of this state with

itself 2j times, we obtain the coherent state in the j representation in the surprisingly simple form
( j)
f n (z) h z | n i2j .
The normalization can be computed, giving the normalized states
r
2j + 1
( j)
h z | n i2j .
f n (z) =
(8.71)
(8.72)
This is a very straightforward representation of the coherent state with angular

momentum j~n.
This concrete realization of SU(2) representation spaces turns out to be very useful to relate SU(2) representations with SL(2, C) unitary representations, because
a similar representation exists for SL(2, C). In the previous Chapter, indeed, we
have constructed the SL(2, C) representation space V ( p,k) in the canonical basis
Classical limit
182
| p, j; jm0 i. This is conceptually simple, but not easy to work with. A realisation of
the representation space of V ( p,k) which is more convenient for computing is in
terms of functions of spinors f (z) where z C2 . The representation ( p, k) is defined on the space of the homogenous functions of spinors that have the property
f (z) = 1+ip+k
1+ipk
f ( z ).
(8.73)
The representation is given by the transpose action

g f ( z ) = f ( g T z ).
(8.74)
The translation between the canonical basis and the spinor basis is computed explicitly in [Barrett et al. (2010)], where it is found to be
r
2j + 1
j
j
f m (z) = h z | p, k; j, m i =
h z | z i ip1 j Dmk ( g(z))
(8.75)
whith
g(z) =
z0
z1
z1
z0
(8.76)
In these representations (both for SU(2) and SL(2, C)) the scalar product between two functions is given by an integral in spinor space. The invariant integral
in spinor space is defined by
h f |gi =
f g
(8.77)
with
=
i 0 1
(z dz z1 dz0 ) (z0 dz1 z1 dz0 ).
2
(8.78)
This can be expressed more easily exploring the homogeneity properties of the
functions. Indeed, by homogeneity these are uniquely determined by their value
on z1 = 1. Thus we can define

z
F (z) = f
(8.79)
1
which has the same information as f (z). Then the scalar product has the simple
form
h f |gi =
d2 z G ( z ) F ( z )
(8.80)
where d2 z = dz dz = dx dy if z = x + iy is the standard integration measure on

the complex plane. In exchange, the group action becomes a bit more complicated
(see [Gelfand et al. (1963)], pg 247):

az + b
k 1+ip
k1+ip
g f (z) = ( az + b)
( a z + b)
f
,
(8.81)
cz + d
Classical limit of the vertex amplitude
183
where
g=
a
c
b
d
(8.82)
These spinor representations are particularly convenient because the Y map

takes a particularly simple form in this language. Since the embedding of Hj in
V ( p,k) is given by
f (z) h z | z i1+ipk f (z),
(8.83)
Y f (z) = h z | z i1+(i1) j f (z).
(8.84)
we have simply
This is the map at the basis of loop gravity in this representation!

This allows us to write immediately the action of the Y map on coherent states:
from (8.84) we obtain
r
2j + 1
hz | Y | j, ~ni =
h z | z i 1+(i1) j h z | n i2j .
(8.85)
Or, for later convenience

p
hz | Y | j, ~ni =
2j + 1
e j [(i1) ln h z | z i +2 ln h z | n i] .
hz|zi
(8.86)
Notice how straightforward is the form of the Y map when acting on spinors
states. This expression gives the form of the coherent state | j, ~ni after it is mapped
by Y into the appropriate SL(2, C) representation.
8.3 Classical limit of the vertex amplitude

We now use the tools developed above to rewrite the vertex amplitude using
spinors. Recall the vertex amplitude is given by (7.58), which we copy here:
Av (hvf ) =
jf
SL(2,C)
(2jf + 1) Trjf [Y ge0 v gve Y hvf ].
0
dgve
(8.87)
It is convenient to drop the subscript v, label the edges emerging from the vertex
with labels a, b = 1...5 and the faces adjacent to the vertices as ( ab), where a and b
are the two edges bounding the face at the vertex. Then the amplitude reads
Av (h ab ) =
jab
SL(2,C)
dg0a
(2jab + 1) Trjab [Y ga 1 gb Y hab ].
This is the quantity we want to study.
ab
(8.88)
Classical limit
184
8.3.1 Transition amplitude in terms of coherent states

The trace in the last equation can be written inserting two resolutions of the identity in terms of coherent states
Tr j [Y gg0 Y h] =
S2
~ h j, m
~ | Y gg0 Y | j, ~ni h j, ~n | h | j, m
~ i.
d~n, dm
(8.89)
Let us focus on the first of these matrix elements, which is the crucial one. Writing
it in the spinor basis, it gives
~ | gg0 | Y j, ~ni =
hY j, m
C2
~ | gz i h g0 z | Y j, ~n i
d h Y j, m
Using the representation (8.85) of coherent states in the spinor basis, and introducing the (common but a bit confusing) notation
Z0 = g0 z
Z = gz,
(8.90)
this gives
~ | gg0 | Y j, ~ni =
hY j, m
(2j + 1)
C2
where
0
d
e j S(n,m,Z,Z ) (8.91)
0
0
hZ|Zi hZ |Z i
hZ|Zi
h Z | m i2 h Z 0 | n i2
+ i ln
.
0
0
hZ|Zi hZ |Z i
h Z0 | Z0 i
S(n, m, Z, Z0 ) = ln
(8.92)
Let us now insert this in the vertex amplitude (8.88). We need unit vectors ~n ab
(where ~n ab and ~nba indicate different objects). The amplitude is a functional on the
space Hv of the states living on the boundary of the vertex graph v , namely on
the boundary of the 4-simplex dual to the vertex. Let us choose a coherent state in
Hv . In particular, pick a quadruplet of normalized vectors ~n ab for each node a of
v . These define a state | jab , ~n ab i. A moment of reflection shows that the amplitude
of this state is
Av ( jab , ~n ab ) h Av | jab , ~n ab i =
SL(2,C)
dg0a
(2jab + 1) h jab , nab | Y ga 1 gb Y | jba , nba i.

f
(8.93)
To obtain a coherent tetrahedron state there is no need to integrate over SU(2),
since the integration over SL(2, C) already does the job. Using the result above,
this gives
Av ( jab , ~n ab ) = ( jab )
SL(2,C)
(2j +1)2
dg0a
C2
d ab
eab
|Z ab | |Zba |
jab S(n ab ,nba ,Z ab ,Zba )
(8.94)
ab
with ( jab ) = ab
and Z ab = ga z ab and Zba = gb z ab . Notice that jab = jba
and z ab = zba but ~n ab and Z ab are in general different from ~nba and Zba .
The last equation gives the form of the vertex amplitude in terms of coherent
states. This is the starting point for studying the classical limit of the dynamics,
which is what we do in the next Section.
185
1.0
1.0
0.8
0.5
0.6
-2
0.4
Figure 8.5
-0.5
0.2
-2
-1
-1
-1.0
e j f ( x) for large j in the real and the imaginary case.
Saddle point
In any quantum theory, the classical limit is a limit of large quantum numbers. In
general, this is the regime where actions are large with respect to h and therefore
quantum effects can be disregarded. To study the classical limit of the dynamics,
we have therefore to study what happens to the transition amplitude in the limit
of large quantum numbers. Let us see in particular what happens when j is large.
If the spins jab are large, the integral (8.94) can be evaluated using the saddle
point approximation [Barrett et al. (2010)]. The saddle point approximation for a
one-dimensional integral is
s

Z
1
2
f ( xo )
j f (x)
g( xo ) e
1+o
.
(8.95)
dx g( x ) e
=
j
j f ( xo )
R
where xo is the saddle point, namely a point where the first derivative of f ( x )
vanishes (here assumed to be unique for simplicity). The generalization of this
formula to integrals in d dimension is
Z
Rd
dx d g( x ) e j f ( x) =
2
j
d
(det H2 f ) 2 g( xo ) e f (xo )
1+o

1
.
j
(8.96)
where H2 f is the Hessian (the matrix formed by the second derivatives) of f at

the saddle point, which is now a point where the gradient of f vanishes. For this
to hold, the real part of f must be negative. Intuitively, what happens is rather
simple: if f is real, a large j gives a narrow gaussian around the maximum of f . If
f is imaginary, when j is large, the exponential oscillates very fast, and the integral
is cancelled out by the fast oscillations. The points where this does not happen
are the ones where the derivative of f vanishes, because in these points the fast
variation of the exponential in x are suppressed. The two cases are illustrated in
figure 8.5.
The imaginary case is of particular interest because it is what gives the standard
classical limit of quantum mechanics from the Feynman path-integral formulation:
the classical trajectories are given by an extremum of the action because in the
limit where h is small quantum interference cancels out the amplitude except for
the saddle point of the path integral. It is good to keep this general fact in mind to
follow what happens here.
Classical limit
186
We do not give here the full details of the calculation to find the saddle point of
the integral (8.94). These can be found in the classic paper [Barrett et al. (2010)]. See
also [Magliaro and Perini (2011b); Han and Zhang (2013); Ding and Han (2011)]
for recent analyses. We only sketch the main steps. We have to find the stationary
points of the action (8.92). Let us start from the real part. This is given by
Re[S] =
log
ab
|h Z ab | n ab i|2 |h Zba | nba i|2

.
h Z ab | Z ab i h Zba | Zba i
(8.97)
This is always smaller than unit, because the scalar product | h Z ab | n ab i |2 is necessarily smaller than the norm h Z ab | Z ab i as coherent states are normalized. The
maximum is therefore when log vanishes. An obvious solution is when
n ab =
Z ab
,
|Z ab |
nba =
Z ab
,
|Z ab |
nba = eiab
Zba
,
|Zba |
(8.98)
but this is not the only solution, because there might be a phase, which cancels in
the products. The general solution is instead
n ab = eiab
Recalling the meaning of Z, this gives

1
g
a n ab =
Zba iab 1
e ga nba .
Za b
Zba
.
|Zba |
(8.99)
(8.100)
Let us now look at the extrema of the action under variations of the spinor variables z ab . The explicit calculation gives an equation similar, but not identical to the
last one:
Z
ga n ab = ba eiab ga nba .
(8.101)
Za b
Finally, we now look at the extrema of the action under variations of the group elements ga and the integration variables z ab . The first is relatively easy. The group
variables enter Z ab . A first order variation of the group element gives the action of
the algebra element. Therefore the saddle point equations for the group elements
give the vanishing of the action of an infinitesimal SL(2, C) transformation on the
action. Such an action can be decomposed in boosts and the rotations, but by construction the action of the boosts is proportional to the action of the rotations in
the relevant representations. Therefore the needed invariance is only for rotations.
This can be moved from the Z to the normals n by change of variables, and the
result is
jab |nab i = 0
(8.102)
This is beautiful and surprising: the saddle point equations for the group integral
are precisely the closure conditions for the normal at each of the boundary nodes
of the vertex graph. A priori, we have chosen an arbitrary set of normals, but
the dynamics suppresses all possible sets of n ab unless these satisfy the closure
187
Figure 8.6
Two tetrahedra in the boundary of a vertex are connected by a face.

constraint at each node Namely, unless they define a proper tetrahedron a at
each node a of the vertex graph.
Now, because of equation (8.102), the spins and the normals on the vertex graph
define four tetrahedra. These are three-dimensional objects, and can be thought
lying in a common three-dimensional surface of Minkowski space left invariant
by the SU(2) subgroup of SL(2, C). A vector in defines a surface in to which it
is normal. A Lorentz transformation can then act on this surface and move it to an
arbitrary (spacelike) surface in Minkowski space. In terms of spinors, this action
is simply given by the action of an SL(2, C) elements on the spinor. This makes
these equations immediately transparent: they imply the existence of five Lorentz
transformations ga that rotate the five tetrahedra a in Minkowski space, in such a
way that the b face of the tetrahedron a is parallel to the a face of the tetrahedron
b . That is, the group elements ga rotate the five tetredra from a common spacelike
surface into place to form the boundary of a five-simplex.
Therefore at the critical point the variables describe precisely a 4-simplex in
Minkowski space. The value of the action at the saddle point is then
S = i jab ab
(8.103)
ab
plus a term that determines only a sign in the amplitude. What is ab ? It is the
difference between the Lorentz transformations to the opposite side of each side of
a triangle in the 4-simplex. Therefore it is dihedral angle between two tetrahedral
But jab (in units 8Gh = 1) is the area of the face of the boundary faces of the
4-simplex. It follows that S on the critical point is the Regge action of the four
simplex having the boundary geometry determined by the 10 areas jab .
More precisely, since it is possible to construct to four simplices with opposite
orientation from the same boundary data, the vertex amplitude, in the large j limit,
is proportional to
A( jab , n ab ) ceiSRegge ( jab ) + c0 eiSRegge ( jab ) .
(8.104)
188
Classical limit
The constants c and c0 are not equal, due to the presence of the same Maslov index
that appears in 3d. Here we have assumed that the boundary tetrahedra define a
non-degenerate Lorentzian 4-simplex. For the other possible case and a detailed
derivation and discussion, including the details neglected here, see [Barrett et al.
(2010)].
Summarising, if the state on the boundary of a vertex represents a geometry
which can be the boundary geometry of a 4-simplex, the amplitude of the vertex
is the Regge amplitude associated to the corresponding 4-simplex. Importantly, if
the normals are not the ones determined by the geometry of the 4-simplex, the
amplitude is suppressed.
The Regge action of a 4-simplex is the Hamilton function of the 4-simplex. The
full amplitude over a two-complex is then obtaining by summing over spins. The
spin configurations such that no normals exist for which the tetrahedra match are
suppressed. The remaining ones are those for which there are lengths at the boundaries of the four-simplices, that determine the corresponding areas a the faces. The
sum over these is equivalent to a sum over truncated geometries, where the weight
is determined by the Regge action. The result gives, in the classical limit, the Regge
Hamilton function of the boundary data.
The geometrical interpretation of all the quantities appearing in the integral is
then transparent. The SL(2, C) group elements gve are holonomies of the spinconnection that transport from each vertex v to the tetrahedron e that separates
two vertices. The spins jf are the areas of the corresponding triangles, and so on.
Finally, notice that local Lorentz invariance at each vertex is implemented without destroying the discreetness of the three dimensional geometry.
In conclusion, the vertex amplitude of the 4d theory approximates the Regge
action of a 4-simplex. This result is the 4d version of the Ponzano-Regge theorem,
which was long searched, and is the cornerstone of the theory.
8.3.2 Classical limit versus continuum limit

In the previous Section we have discussed the large-j limit, which is a classical limit.
In the previous Chapter we have discussed the limit in which the two-complex C is
refined, which is a continuous limit. The two limits are conceptually and practically
very different and should not be confused. Unfortunately, confusion between the
two limits abounds in the literature. So, let us try to get some clarity here. The
relation between the two limits is summarised in Table 8.3.2.
The transition amplitudes on a fixed two-complex that we have defined sit in
the top left corner of the Table. They yield an increasingly better approximation
by refining the two complex. As we refine the two complex, we move towards
the bottom of the Table, namely towards the (ideal) exact transition amplitudes of
quantum gravity.
If we look at increasingly large spin values of the boundary state at fixed triangulation, we move towards the right in the table. At each fixed two-complex C ,
dual to a triangulation , the classical (large j) limit of the transition amplitudes
189
Regge
Hamilton function
S ( l e )
Loop-gravity
transitionamplitudes
WC (h` )
C
Continuum limit
Table 8.1 Relation between continuum limit and classical limit of the transition
amplitudes
Exact quantum gravity

transition amplitudes
W ( h` )
General relativity
Hamilton function
S(q)
Classical limit
is related to the Hamilton function of the Regge truncation of general relativity

defined on that triangulation .
If we refine the triangulation, the Regge discretization is known to converge
to classical general relativity. Therefore as we (correspondingly) refine the twocomplex, the transition amplitudes define a quantum theory with more degrees of
freedom, whose classical limit is closer and closer to continuum general relativity.
This is the way amplitudes are defined in covariant loop gravity.
Recall again that also in QED and QCD, concrete calculations are generally performed on the truncated theory. This theory provides an approximation which is
good in certain regimes.
The regimes where the classical limit is good in quantum gravity are those involving scales L which are much larger than the Planck scale.
L LPlanck .
(8.105)
The regimes where the truncation is good are suggested by the Regge approximation. This is good when the deficit angles are small. This happens when the scale
of the discretization is small with respect to curvature scale Lcurvature .
L Lcurvature .
(8.106)
Therefore a triangulation with few cells, and, correspondingly, a two-complex

with few vertices, provide a approximation in the regimes (determined by the
boundary data) where the size of the cells considered is small with respect to the
curvature scale (of the classical solution of the Einsteins equation determined by
the given boundary data).
Refining the triangulation leads to including shorter length-scale degrees of free-
Classical limit
190
Figure 8.7
1
2
1090
1
2
1
2
Planck-scale geometry
1090r
1090
Large geometry
A small space and a large space

dom. But the physical scale of a spinfoam configuration is not given by the graph
or the two complex. It is given by the size of its geometrical quantities, which is
determined by the spins (and intertwiners). The same triangulation can represent
both a small and a large size of spacetime. A large chunk of nearly flat spacetime can be well approximated by a coarse triangulation, while a small chunk of
spacetime where the curvature is very high requires a finer triangulation. In other
words, triangulations are like all engineers discretizations: they do not need to be
uselessly fine: they need to be just as fine as to to capture the relevant curvature.3
On the same Hilbert space H determined by the same graph, there are states
representing small and large chunks of space
A detailed analysis of the continuous limit for large trianguations, taking into
account the need of working at a scale intermediate between the Planck scale and
the curvature scale for recovering the classical theory has been developed in particular by Muxin Han in [Han and Zhang (2013); Han (2013)]. Hans analysis,
which we do not report here, studies the expansion that leads from the full loop
quantum gravity amplitudes to the effective action as a two-parameter expansion:
1/ L2Planck /L2 measures quantum corrections, while L2 /L2curvature measures the high energy corrections. Interestingly, Han uncover a relation between
and these two parameters, of the form
1
| | 1 2 ,
(8.107)
illustrated in Figure 8.8.
Figure 8.8
The Einstein sector of loop quantum gravity.

The red region in Figure 8.8 is where the expansion is good.4 It is called the
3
4
Misunderstanding of this point has generated substantial confusion in the literature.

These results clarify the source of the apparent flatness notices by same authors [Bonzom (2009);
Extrinsic coherent states
191
Einstein sector of loop gravity. It defines a subspace of spinfoam configurations,

where configurations are interpreted as Lorentzian geometries satisfying
LPlanck L Lcurvature .
(8.108)
When the fluctuations of the spinfoam variables are in this sector, the perturbative
expression of the spinfoam amplitude is then given by
W
e LPlank
( jf ,gve ,zvf )
d4 x gR[ g ](1+o ( L2 /L2curvature )+o ( L Planck /L))
(8.109)
where g is the Lorentzian metric approximated by the spin foam data ( jf , gve , zvf ).
The leading contributions come from the configurations ( jf , gve , zvf ) which give
g satisfying the Einstein equations.
8.4 Extrinsic coherent states

There is one additional topic we have to cover before concluding this chapter. The
astute reader may wonder why the intrinsic coherent states introduced in Section
8.23 where called intrinsic. The reason is that they are coherent states for the operators ~L` that represent the intrinsic geometry of space, but they are not coherent
states for the operators representing the extrinsic geometry. The quantum spread
of the intrinsic geometry of space is minimalized by these states, but not its conjugate momentum, namely its time variation. Therefore these are not truly semiclassical states in the sense in which a common wave packet is. In fact, these states
are eigenstates of the area variables, and therefore the variable conjugate to the
area is completely quantum-spread. The geometrical interpretation of the variable
conjugate to the area of a triangle between two tetrahedra is the four-dimensional
dihedral angle between the 4d normals of the two tetrahedra. This is clearly the
discrete version of the extrinsic curvature, which is the variable conjugate to the
3-metric in the ADM variables. The states that are both coherent in the intrinsic
and extrinsic geometry are the extrinsic coherent states.
There are different manners to construct the extrinsic coherent states [Bahr and
Thiemann (2009); Freidel et al. (2009); Freidel and Speziale (2010b); Bianchi et al.
(2010a)]. Here we present a simple one, which makes their meaning transparent
[Bianchi et al. (2010a,c,d)]. To see what goes on, let us start from a simple example. Consider a one-dimensional quantum system with position q and momentum
p. A wave packet peaked on the phase space point (q, p) is (in the Schrodinger
representation and disregarding normalizations)
h x | q, p i q,p ( x ) = e
( x q )2
+ hi px
22
(8.110)
Hellmann and Kaminski (2013)] in the spinfoam amplitudes: a naive large scale limit cannot be
taken, because it leads outside the correct regime of validity of the low energy approximation: the
naive j limit on a fixed triangulation yields a flat geometry.
Classical limit
192
This state has spread x in position and p h / in momentum, since its

Fourier transform is (up to a normalization)
h k | q, p i e
(k p/h)2
+iqk
2/2
(8.111)
For later convenience, observe that this state can be written in the form (always
disregarding normalisation)
q,p ( x ) = e
( xz2)
z = qi
2
p.
h
(8.112)
where z is the complex variable

(8.113)
That is, a wave packet peaked on the phase space point (q, p) can be written as a
Gaussian function with a complex position.
We need an analog of this state in H , peaked both on group variables and
on their conjugate. Let us first consider L2 [SU(2)]. A state in this space which is
completely peaked on group variables is clearly given by a delta function on the
group. The (generalized) state
(U ) = (U )
(8.114)
is sharp on the origin U = 1l, while the state

(U ) = (Uh1 )
(8.115)
is sharp on the group element h SU(2). These states are of course completely
spread in the conjugate variable, as it is clear by Fourier transforming on the
group:
(U ) =
d j Tr j [U ].
(8.116)
But we can spread this out around the origin by adding a Gaussian factor in
momentum space. To see how this can be done, recall that a Gaussian can be obtained by acting on the delta function with the operator exp{t2 }, where 2 is
the Laplacian and t = 2 . Doing the same on the group and using the Laplacian
~L2 gives the state
1l,0 (U ) =
d j etj( j+1) Tr j [U ],
which is peaked on U = 1l with spread

Similarly,
h,0 (U ) =
(8.117)
t as well as on j = 0 with spread 1/ t.
d j etj( j+1) Tr j [Uh1 ],
(8.118)
is peaked on U = h and j = 0 with the spread t and 1/ t, respectively. How

then do we obtain a state peaked on j 6= 0? In the case of the particle, this is
193
done by adding the oscillating factor exp{ipx/h}. But we have also seen that this
can be obtained complexifying the argument. Let us do the same on the group. A
complexification of SU(2) is given by SL(2, C). Let us therefore consider the state
H (U ) =
d j et j( j+1) TrD( j) (UH 1 ).
(8.119)
labeled by a variable H SL(2, C) such that

H=e
it E2
lo
(8.120)
where h SU(2) and E = ~E i ~2 is in the algebra su(2). Notice that the Wigner
matrix is here evaluated on the SL(2, C) element H: by this we mean its analytic
continuation on the complex plane. This is a good guess for a wave packed peaked
both in the group variable and on its conjugate variable. In fact, an explicit calculation gives
hH | ~E | H i ~
= E,
(8.121)
h H | H i
with the same spread as before. The spread is U` t and E` lo2 1/t.
If we study a phenomenon at a scale E L2 lo2 j lo2 , then
hH | U | H i
=h,
h H | H i
E
1
.
E
tj
(8.122)
For this to be small, we need
t
1
.
j
(8.123)
Therefore we must have at the same time
1
t 1.
(8.124)
j
Notice then that t determine an intermediate scale L2wp = lo2 t such that
lo Lwp L.
(8.125)
The scale Lwp is a scale intermediate between the scale of the phenomenon L and
the Planck scale lo . It gives the size of the wave packet. For being in a semiclassical
regime, this must be much smaller that the scale considered but still much larger
than the Planck scale.
It is then easy to generalized these states to spin network states, by making them
invariant under SU(2) at the nodes. An extrinsic coherent state on a graph is then
labelled by an SL(2, C) variable H` associated to each link ` of the graph, and is
defined by
H` (U` ) =
SU(2)
1
dhn d j` et j` ( j` +1) Tr D ( j` ) (U` hs` H`1 h
t ` ).
j`
(8.126)
Classical limit
194
where, for each `, the nodes s` and t` are the source and the target. It is convenient
to introduce the integral kernel
K (U, H ) =
d j et j( j+1) Tr D(l) (UH 1 ).
(8.127)
called the heat kernel on the group, because it is the analytic continuation on
SL(2, C) of the solution of the diffusion equation for the Laplacian on the group,
and write
H` (U` ) =
SU(2)
1
dhn K (U` , ht` H` h
s ` ).
(8.128)
The kernel intertwines between SU(2) and SL(2, C), but sould not be confused
with the Y map, to which it is unrelated.
The states H` are the extrinsic coherent states of loop quantum gravity. They
are peaked both on the intrinsic and the extrinsic curvature of a 3d triangulation.
Because of the SU(2) integrations in their definition, what matters is not the specific direction of each normal, but the SU(2) invariant quantities, which are the
lengths of the normals, namely the areas of the faces, and the angles between any
two normals of a node. The relation between the data H` and the geometry can be
clarified as follows.
A Lorentz transformation H can always be written as the product of a rotation
R, a boost e p
3
2
in the z direction and another rotation R0 . In turn, any rotation R

3
can be written as the product of R~n and a rotation eiq 2 around the z axis, where
R~n SU(2) is the rotation matrix that rotates a unit vector pointing in the (0, 0, 1)
direction into the unit vector ~n, around the axis normal to both. Using this we can
always write
H` = R~ns(`) eiz`
3
2
1
R~
n
(8.129)
t(`)
for a complex variable

z` = k ` + i
2t
A .
lo2 `
(8.130)
The geometrical interpretation of H` is then given by the four quantities ( A` , k ` , ~ns , ~nt ).
In fact, it is not difficult to show [Bianchi et al. (2010a)] that the state H` can be
rewritten in the form
H` =
0 2
d j` et( j` j` )
j`
eik` j` ,~ns
,~nt`
(8.131)
where j`0 = A` /lo2 and j` ,~ns ,~nt are the intrinsic coherent states (8.23) defined in
`
`
Section 8.1.2 in this Chapter. The two vectors ~ns and ~nt represent the normals to
the triangle `, in the two tetrahedra bounded by this triangle, A` is the area of this
triangle, because the spin, which determines the area is peaked on j`0 = A` /lo2 , and
k ` determine the variable conjugate to the area, which is related to the extrinsic geometry at the triangle, namely the angle between the four-dimensional normals to
195
the two tetrahedra.5 This geometrical interpretation allows us to build discretized

spaces with intrinsic and extrinsic geometry, from the H` .
If we assume the metric to be flat within each tetrahedron, then the metric
of the triangulation can be discontinuous at each triangle, because the area and
the normal of the triangle matches from one side to the other, but not necessarily so the length of the triangle sides. These discretized geometries are called
twisted geometries and have been studied in depth by Speziale and Freidel
[Freidel and Speziale (2010b)]. See also [Freidel and Speziale (2010a); Rovelli and
Speziale (2010); Magliaro and Perini (2012); Haggard et al. (2013)].
These states are the starting point of a full semiclassical analysis of the dynamics,
analog to the analysis of standard quantum mechanics in terms of wave packets.
We will use them repeatedly in the physical applications in the last chapters.
More precisely, k ` is determined by the holonomy of the Ashtekar connection along the link, which
depends both on the dihedral angle and the component of the 3d connection normal to the triangle
[Rovelli and Speziale (2010)].
Matter
The universe is not made by the gravitational field alone. There are also fermions
and Yang-Mills fields, and, so it seems, scalar fields. In this Chapter we illustrate
how fermions and Yang Mills fields can be coupled to the gravitational field, following [Bianchi et al. (2010b)] (see also [Han and Rovelli (2013)]). The resulting
theory is still ultraviolet-finite, because the quantization of the geometry acts as
a physical cut off on the high-momentum Feynman integrals, also for the matter
fields. In a sense, fermions and Yang-Mills behave as if they were on a Planck size
lattice.
9.1 Fermions
The dynamical coupling of a fermion to quantum gravity is simple in the spinfoam
formalism. Consider first 4d euclidean space, for simplicity. The Dirac action reads
SD = i
d4 x D 6 D
(9.1)
plus complex conjugate, always understood here and in what follows. Let us focus
on a chiral spinor C2 with two complex components A , A = 0, 1. Its action is
the projection of SD on one of its two helicity components. This reads
S=i
d4 x I I
(9.2)
where I = (1l,~), ~ are the Pauli matrices and denotes the hermitian conjugate
of . This can be rewritten in form notation as
S=i
I d e J eK e L e I JKL
(9.3)
where e I is a tetrad field, for the moment just euclidean.

Fix a coordinate system x = ( x I ) and discretize spacetime by chopping it into
the union of 4-cells v. Consider the dual complex, with vertices v (with coordinates
xv ) connected by (oriented) edges e. Call |v| the number of edges bounded by v.
Approximate the field ( x ) by its values v = ( xv ) at each vertex. Discretize the
derivative on each edge as
te se ( xtIe xsIe ) I
196
(9.4)
Fermions
197
where se and te are the source and the target of the edge e. This gives
ueI I
te se
è
(9.5)
where è = | xtIe xsIe | is the length of the edge e and ueI = ( xtIe xsIe )/è is the unit
vector parallel to e. The action (9.2) can then be discretized as a sum over 4-cells:
S v Sv .
v+e v
4iVv
e ueI I
,
(9.6)
Sv
|v| e
è
v
where the sum is over the edges bounded by v, Vv is the volume of the 4-cell v,
and the factor |v4| is included to take into account the fact that multiple edges overcount the derivative. The second term above cancels in subtracting the complex
conjugate. Now consider the 3-cell e dual to the edge e. Assume this is orthogonal
to the edge. Each 4-cell v can be partitioned into the union of |v| pyramids with
base e and height he = è /2. The 4-volume of these is 14 he ve where ve is the
3-volume of e . Using this,
Sv
where
i
e ve e v+e ,
2 e
v
e I ue I
(9.7)
(9.8)
is the -matrix in the direction of the edge e. This can be written as

ve e I
e I JKL e J eK e L ,
(9.9)
where e I = dx I is the tetrad one-form. Notice now that each term of the sum depends only on edge quantities. This suggest to consider writing the full discretized
action as a sum of edge terms
S = i se ve e te .
(9.10)
More in general, say that the collection of edges e, together with the data (ve , ue I )
approximate a flat metric at a scale a if [Ashtekar et al. (1992)]
Z
d4 x I I ( x ) =
ve
e
I ue I
(9.11)
for any quadruplet of one-forms I = I dx that varies slowly at the scale a. Then
(9.10) is a discretization of the Weyl action if the two-complex approximates a flat
metric. Equation (9.10) is a very simple expression that discretizes the fermion
action.1 Its simplicity recalls the simplicity of the free particle hamiltonian on a
1
Fermions on a lattice suffer from the fermion-doubling problem and the related chiral anomaly.
Because of the absence of a regular triangulation and the integration on the gravitational variables
that we introduce below, however, here fermions are essentially on a random lattice, where the
(obvious) species doubling problem does not arise.
Matter
198
graph [Rovelli and Vidotto (2010)]. It could have also been directly guessed from
(9.9) and the form (9.3) of the action.
In view of the coupling with gravity, and in order to better understand boundary
states, it is convenient to move from vertex variables to edge variables. Let xe be
the intersection between e and e and introduce edge variables e = ( xe ). In
the approximation in which we are working, where the second derivative of the
field can be neglected, e 12 (se + te ). Using this and the fact that the sum of
quadratic terms averages to zero, we can write
i
v ve e e .
4
ve
S=
(9.12)
where the sum is over all couples of adjacent vertices-edges. Then observe that
(always in this approximation) we can express the vertex fermion as an average
over the corresponding boundary edge fermions: v = |v1| e0 v e0 . This gives
S
4|v| 0
v
e,e v
e0 ve e e ,
(9.13)
Thus, we consider the discretization of the fermion action defined by associating

to each 4-cell v with boundary fields e the action
Sv = i e0 ve e e .
(9.14)
ee0
where we have assumed here for simplicity that all vertices have the same valence
and we have absorbed a constant in a redefinition of the field. This is an expression
that can be used to couple the fermion to quantum gravity.
Quantum fermion field on a 2-complex

Consider the fermion partition function of a 2-complex characterized by the quantities (ve , e ). This will be given by the Berezin integral
Z
De eiS .
(9.15)
1
de de eve e e ,
v2e
(9.16)
Z=
Choose the integration measure to be
De =
which realizes the scalar product at each edge, seen as a boundary between two
4-cells, and where we interpret the field e as an anticommuting variable, in order
to take Pauli principle into account. The volume ve in the exponent is needed for
dimensional reasons and to keep into account the fact that in the classical theory
h|0 i =
d3 x q ( x ) 0 ( x )
(9.17)
Fermions
199
contains the 3-volume factor q [Rovelli and Vidotto (2010)]. The definition of
Berezin integral is that the only non vanishing integral is
Z
dd a b c d = e ac ebd = ( ab cd ad bc ).
(9.18)
Expand the vertex action in Taylor series

eiSv = 1 +
e1 ve2 e2 e2
(9.19)
e1 e2
e1 e2 e3 e4
(e1 ve2 e2 e2 )(e3 ve4 e4 e4 ) + ...
The series stops because there can be at most 4 fermions (two and two ) per
edge. Each edge integration in (9.15) gives zero unless on the edge there is either no
fermion, or a term e e , or a term e e e e . The volume factors in the measure
cancel those in the vertex action. The result is that (9.15) becomes a sum of terms,
each being the product of traces of the form
Ac
Z=
(9.20)
{c} c
where {c} is a collection of oriented cycles c, each formed by a closed sequence of

oriented edges on the 2-complex: c = (e1 , ..., e N ), and each trace is
Ac = (1)|c| Tr [e1 ...eN ],
(9.21)
where |c| is the number of negative signs from (9.18). For every edge, there cannot
be more than two fermionic lines (Pauli principle) and if there are two lines, these
are anti-symmetrized (by (9.18)). Explicitly, each term reads
I
Ac = (1)|c| Tr [I1 ...In ] ue11 ...ueInn

and is therefore a Lorentz-invariant contraction of the normals to the
(9.22)
3-cells.2
Fermions in interaction with gravity

Let us now return to the Lorentzian framework, and come to our first key technical observation. Assume that all edges are time-like. Consider an SL(2, C) matrix
ge that rotates the unit vector u I = (1, 0, 0, 0) into the vector ueI (the phase is irrelevant, and can be fixed by requiring that ge is a pure boost)
J
u J ge I = u e I ,
(9.24)
where g is the vector representation of SL(2, C). Recall the transformation properties of the I matrices,
gI J J = g I g,
2
(9.25)
If we do not include the volume factors in (9.16), then (9.21) is replaced by

Ac = (1)|c| ve1 ...veN Tr [e1 ...eN ].
This alternative will be studied elswhere.
(9.23)
Matter
200
and observe that

ueI I = ge 0 ge = ge ge ,
(9.26)
because 0 = 1. Therefore we can write the discretized action (9.10) in the form
S = i ve se ge ge te .
(9.27)
Sv = i ve e0 ge ge e .
(9.28)
and the vertex action as

ee0
What is the geometrical interpretation of these SL(2, C) group elements?

Consider a single chiral fermion. This can be thought as a quantum excitation of
a single mode of a fermion field, or as a fermionic particle with spin 21 at one space
1
point.3 The fermion transforms in the fundamental representation H 2 C2

1
of SU (2). The vector space C2 is also the carrier space H ( 2 ,0) of the fundamental
representation of SL(2, C), and this determines the Lorentz transformation prop1
1
erties of the fermion. But H 2 is a unitary representation, while H ( 2 ,0) is not; in
other words, C2 can be equipped with a scalar product which is SU (2) invariant, but this scalar product is not SL(2, C) invariant.4 This becomes particularly
transparent if we write h|i = 0 , which shows that the scalar product is the
time component of a 4-vector I . The Dirac action exploits this dependence by
constructing a Lorentz scalar contracting the Lorentz vector s I = I d with the
covector u I giving the direction of the derivative, where I = d(u I ).
At the light of this discussion, the geometrical interpretation of the matrices
ge is clear. In the discretized theory, a Lorentz frame is determined at each 3-cell
by the 4-normal to the 3-cell, which is the direction along which the derivative
is computed. The matrices ge parallel-transport the fermion from the fixed reference frame at the center of the 4-cell to a frame at the center of a 3-cell, namely
at the boundary of the 4-cell, where the normal to the 3-cell is oriented in the
time-direction (1, 0, 0, 0). The scalar product that defines the action is taken in that
frame. The construction is Lorentz invariant, since the preferred frame of the scalar
product is determined by the normals along which the variation of the fermion is
computed.
Now, (9.28) is of particular interest for generalizing the discretization to a curved
spacetime. In fact, we can discretize a curved geometry in terms of flat 4-cells glued
along flat 3-cells, as in Regge calculus. Curvature is then confined on 2-cells. This
implies that the holonomy of the spin connection around a 2-cell can deviate from
unity. Therefore in general on a curved space there is no way of choosing a reference frame in each 4-cell which is parallel transported to itself across all 3-cells. In
other words, the parallel transport from se to e may be different from the parallel transport from te to e. Therefore the generalization of the fermion action to a
3
4
On the notion of particle in the absence of Poincare invariance, see [Colosi and Rovelli (2009)].
If we fix a basis in C2 , the scalar product is given by h|i = a b ab and the tensor ab is invariant
under SU (2) but not under SL(2, C).
Fermions
201
curved spacetime, discretized a` la Regge, can be obtained replacing (9.27) with

S = i ve se ge se gete te .
(9.29)
where gev is the holonomy of the spin connection from a coordinate patch covering
the (flat) 4-cell v to one covering the (flat) 3-cell e. Equivalently, replacing the vertex
amplitude (9.28) by
Sv = i ve e0 ge se gete e .
(9.30)
e0 e
The fermion partition function on the 2-complex representing a curved spacetime,

is therefore as before, with the only difference that the amplitude (9.22) of each
cycle is replaced by
Ac = (1)|c| Tr [ ge1 v1 ge1 v2 ...ge n vn gen v1 ].
(9.31)
where (v1 , e1 , v2 , e2 , ..., vn , en ) is the sequence of vertices and oriented edges crossed
by the cycle c. This can be written in a form more easy to read by defining
g = ( g1 ) = eg e.
(9.32)
Ac = (1)|c| Tr [ gv1 e1 ge1 v2 ...gvn en gen v1 ].
(9.33)
Notice that for a rotation g = g while for a boost g = g1 . Using this,
where the sequence of vertices and edges is in the cyclic order. The full partition
function (9.20) becomes

1
Z = (-1)|c| Tr ( gse e gete )eec ,
(9.34)
2 e c
{c} c
where eec = 1 according to whether the orientations of the edge and the cycle
match.
The action (9.27) and the amplitude (9.33) are of particular value for coupling
the fermion field to quantum gravity because they depend on the geometry only
via the two quantities ve and gve , which are precisely the quantities that appear in
the gravitational spinfoam amplitude. This we do in the next section.
Coupling to quantum gravity

Recall that the gravity transition amplitudes can be written as
WC (h` ) = NC
jf
SL(2,C )
0
dgve
(9.35)
For simplicity, consider here the partition function (no boundaries)

ZC = NC
jf
SL(2,C )
0
dgve

f
(9.36)
Matter
202
It is pretty obvious now how to couple fermions to quantum gravity: it suffice to

write the fermion partition function above on the quantum sum over states. The
SL(2, C) matrices gev in this expressions can be identified with the SL(2, C) matrices gev in (9.27) and (9.33).They have the same geometrical interpretation as parallel transport operators from the edge to the vertex; and second, more importantly,
the asymptotic analysis of the vertex amplitude in [Barrett et al. (2010)] shows that
the saddle point approximation of the integral is on the value of gev that rotates
the (arbitrary) Lorentz frame of the 4-cell into a Lorentz frame at the 3-cell where
the time direction is aligned with the normal of the 3-cell, which is precisely the
geometry described in the previous paragraph. Therefore in the limit in which we
move away from the Planck scale, these group elements take precisely the value
needed to yield the fermion action.
The obvious ansatz for the dynamics in the presence of fermions, is therefore
Z

0
dgve
Z = NC
Trjf Y ge0 v gve Y
SL(2,C)
{c} jf
(1)|c|
c
ef

Tr ( gese ge te )eec .
e c n
(9.37)
where {c} labels families of worldlines running along the edges of the foam. The
sum is over all families that do not overlap more than once. Notice that the definition of the in this expression depends on the choice of a specific SU (2) subgroup at each edge, but this dependence drops from the total expression, because
of the SL(2, C) integrations, precisely as discussed in [Rovelli and Speziale (2011)].
Therefore Lorentz invariance is implemented in the bulk.
This expression defines a quantum theory of gravity interacting with fermions.5
9.2 Yang-Mills fields

Suppose now that the fermion lives in fundamental representation of a compact
group G. Then the above theory is invariant under global G trasformations. To
make it invariant under local gauge transformations we can introduce a group
element Uve G associated to each wedge (ve), and replace (9.10) by
S = i se Use e ve e Ute e te .
(9.38)
This naturalness of the fermion dynamics in gravity was already observed early in the loop quantum gravity literature [Morales-Tecotl and Rovelli (1994, 1995)] and is surprising. A fermion, is essentially an extra face of spin 21 of a quantum of space, which is non-local over the 2-complex.
At fixed time, it can be seen as a non-local loop that disappears outside spacetime, to reappear
far away, like a Wheeler-Smolin Kerr-Newman fermion: the picture of fermions as wormholes
suggested by John Wheeler long ago [Wheeler (1962); Sorkin (1977)], and considered by Lee Smolin
[Smolin (1994)] in the context of loop gravity.
Yang-Mills fields
203
The quantum kinematics on the boundary is then evident: spinfoams carry representations of SL(2, C) and intertwiners at the nodes have a possible extra leg
representing fermions in (antisymmetric products of) the fundamental representation of SL(2, C) G.
What is the dynamics? One possibility of obtaining it is simply to keep only the
gravity and fermion terms in the action. The Yang-Mills action is then generated
by the one-loop radiative corrections to the fermion action in the Yang-Mills field,
as suggested by Zeldovich [Adler (1982)].
Z=
{c} jf
SL(2,C)
(1)|c|
c
0
dgve
dUve Tr j f
f
e f

Tr ( geseUeseUete ge te )eec .
ecn
Y ge0 v gve Y
(9.39)
This expression defines a minimally-coupled spinfoam formulation of the EinsteinWeyl-Yang-Mills system. In principle, it can be used to compute all quantum gravity amplitudes order by order. The analysis of the the theory of loop gravity with
matter is in a primitive stage and little has been done so far [Han and Rovelli
(2013)].
Part IV
PHYSICAL APPLICATIONS
10
Black holes
The last three chapters of the book introduce to the main current applications of
the theory to physical situations. Results have been obtained mainly in three directions:
thermal properties of black holes,

early cosmology,
scattering theory.
The theory is evolving rapidly in these directions and therefore these are only
shortly described in the following, limiting to basic facts and pointing to some
relevant literature. We begin with black holes, objects that have now been found
numerous in the sky, but still remain among the most fascinating and mysterious
in the universe.
10.1 Bekenstein-Hawking entropy

Stephen Hawking has made the remarkable theoretical discovery that quantum
field theory predicts that a black hole is hot [Hawking (1974, 1975)]. More precisely, a quantum field on a spacetime where a black hole forms evolves from its
vacuum state to a stationary state with a thermal distribution of outgoing radiation at temperature
TH =
h c3
.
k8GM
(10.1)
The surprising aspect of this result is not that particles are created: there is particle
creation anytime a quantum field interacts with a non-stationary potential; what
is surprising is the thermal nature of the outgoing radiation.
This result adds credibility to a series of previous physical intuitions [Bekenstein (1973)] and precise mathematical results [Bardeen et al. (1973)] indicating
that classical black holes behave in a way which is akin to thermal systems. They
rapidly evolve to an equilibrium state characterized by a quantity (the horizon
area) which cannot decrease in (classical) physical processes, just like the entropy
of a statistical system. This is not completely surprising either: entropy measures
of the amount of information about the microstates which is not captured by the
macroscopic variables describing a system with many degrees of freedom. A black
207
208
Black holes
hole is a system with a horizon that screens information, precisely as the description of a system by macroscopic parameters does. Therefore, it is reasonable to
expect similarities between the physics of the observables accessible from outside
the horizon and statistical mechanics. In a stationary context, it is then reasonable
to expect thermal properties for black holes.
The thermal properties of a macroscopic state are captured by giving its entropy as a function of macroscopic variables. In the microcanonical setting, entropy measures the volume of the region of the microscopic phase space determined by these macroscopic variables. That is, entropy measures the information
lost in the coarse graining. The entropy S that yields the temperature computed
by Hawkings quantum field theory calculation is given by a celebrated formula:
the Bekenstein-Hawking entropy:
SBH =
kc3
A.
4h G
(10.2)
where A is the area of the black horizon and the subscript BH stands either for
Black hole, or for Bekenstein-Hawking. This is a beautiful formula. It contains
the Boltzman constant k, the speed of light c, the Planck constant h and the Newton
constant G; therefore it pertains to a regime where all fundamental theories are
relevant: statistical mechanics, special relativity, quantum mechanics and general
relativity. The curious factor 4 appears a bit strange at first, but it is not so if
we remember that the coupling constant of general relativity is not G but rather
8G, which is in the denominator of the Einstein-Hilbert action and on the right
hand side of the Einstein equations. The origin of the curious four is therefore
4 = 8/2. That is, in natural units k = c = h = 8G = 1, the BekensteinHawking formula reads
SBH = 2 A;
(10.3)
the 4 is now a more presentable 2.

In standard thermodynamics, the expression of the entropy as a function of the
macroscopic variables can be computed using statistical mechanics. That is, it is
possible to compute the value of the entropy. For instance, we can compute the
volume of the phase space corresponding to given macroscopic variables. Can the
same be done for the Bekenstein-Hawking entropy? This is an open challenge for
any theory of quantum gravity.
Much fuss is sometimes been made about the fact that h enters (10.2) and the fact
that it enters in the denominator. But this is wrong. The entropy of an electromagnetic field in equilibrium in a cavity of volume V = L3 , expressed as a function of
the relevant mechanical macroscopic quantity, which is the energy E, is
s
4 4 2 L3 E3
,
(10.4)
S=k
3
15c3 h 3
which contains h downstairs as well. The presence of h depends on the fact that h
209
Local thermodynamics and Frodden-Ghosh-Perez energy

determines the size of the physical quantum granularity that underlines the statistical behavior, and this is true both for electromagnetism and for gravity. If we take
h to zero, both the entropy of a black hole and the entropy of a black body diverge.
The reason is that they are no longer bound by quantum discreteness. This yields
the ultraviolet catastrophe contemplated by the theoreticians who tried to apply
statistical mechanics to classical fields, at the end of the XIX century.
In fact, the fact that there is no ultraviolet catastrophe was precisely the evidence that let Planck and Einstein realise the discreteness of the photons [Einstein
(1905b)]. In the same manner, the fact that the entropy of a black hole is finite and h
appears in the denominator of (10.2) is evidence that there is a similar granularity
in quantum spacetime, preventing a gravitational ultraviolet catastrophe.
Let us make the analogy sharper. From (10.4) we can see what is the size of a
single cell in phase space: this is given when S/k is order unity, that is by LE/c h ,
but L is also the wavelength therefore a single cell is characterised by
E h
(10.5)
A h G.
(10.6)
where is the frequency. In other words, we can read out (or actually, Einstein
did) the size of the scale from the expression of the entropy. In the same manner,
the single of the quanta of gravity can be read out from the Bekestein-Hawking
entropy by equating it to a number of order unity, which gives a quantisation of
the area:
In agreement with the results of loop quantum gravity.
In the case of electromagnetism, we can compute the entropy directly from the
quantum theory of the electromagnetic field. Can we do the same for gravity?
10.2 Local thermodynamics and

Frodden-Ghosh-Perez energy
An important step in the right direction has been taken by Frodden, Ghosh and
Perez [Ghosh and Perez (2011)]. The mass M of a Schwartzshild black hole can
be identified with energy measured at infinity. The same energy measured at a
Schwartzshild radius R is scaled by the redshift factor
q
M
E( R) = goo ( R) M = q
.
(10.7)
1 2GM
R
Consider an observer at coordinate distance r 2GM from the horizon. The energy of the black hole for this observer is
r
M
2GM
E (r ) = q
M
.
(10.8)
r
1 2GM
2GM+r
210
Black holes
Near the horizon the coordinate distance r is related to the physical distance d by
r
2GM
1
d = grr r = q
r
= 2GMr.
(10.9)
r
1 2GM
2GM+r
Therefore the energy measured at a physical distance d from the horizon is

E(d) =
2GM2
.
d
(10.10)
Now recall that the area of the black hole horizon is A = 4 (2GM )2 . Also, a
standard calculation in classical general relativity shows that an observer staying
at a fixed distance d from the horizon maintains an acceleration a = 1/d (in its
own frame). Thus, for a static observer near the horizon, with acceleration a, the
black-hole energy is
E=
aA
.
8G
(10.11)
This beautiful formula has been confirmed by Frodden, Ghosh and Perez in a number of ways. For instance, let dE be the energy of a particle of small mass m falling
from infinity into the black hole, as measured by the observer at short distance by
the horizon. When the mass enters the black hole, the area of the hole increases by
an amount dA where
a dA
.
(10.12)
dE =
8G
The interest of this formula is that it is local on the horizon. An observer near the
horizon can interact at most with an infinitesimal patch of the horizon. If this patch
has area dA, the observer can associate an energy dE to this patch, given by the last
equation. A variation of energy dE entering or leaving the hole near the observer
is reflected in a local variation dA of the area, and dE and dA are related by (10.12).
Several alternative derivation of the Froddent-Gosh-Perez equations can be given.
For instance, for a uniformly accelerated observer in a flat region, the proper time
is just the dimensionless boost parameter for a boost centred in the horizon,
scaled by the acceleration a. (See Exercise, at the end of the chapter.) The energy
E is the generator of evolution in proper time, and for this observer K = E/a is
a boost generator. Inserting this in (3.95) we get immediately the Frodden-GoshPerex relation (10.11). This shows that the relation is not constrained to black holes,
but in fact is valid for any event horizon: an accelerated observer sees an event
horizon, and a local change of area dA of this horizon is related to an energy
density dE crossing it by (10.11), as a consequence of the Einstein equations. Still
another derivation of this relation can be obtained studying the corner term associated to an action of a finite region of spacetime [Carlip and Teitelboim (1995);
Bianchi and Wieland (2012); Smolin (2012)].
Bill Unruh has shown [Unruh (1976)] that an observer moving with acceleration
211
Kinematical derivation of the entropy

a in the vacumm state of a quantum field observes a temperature
h a
(10.13)
2 k
Combining this with the above result, we can use Clausius definition of entropy
to obtain
dE
2k a dA
dA
dS =
=
=k
.
(10.14)
T
h a 8G
4h G
The r.h.s. is precisely the Bekenstein-Hawking entropy SBH . Therefore the Unruh
temperature T and the Frodden-Ghosh-Perez energy E are related by the thermodynamical relation
T=
dE = T dSBH .
(10.15)
If we red-shift the energy and the temperature from the distance d of the horizon
all the way to infinity, this relation becomes
dM = TH dSBH .
(10.16)
where M is the ADM mass of the black hole and TH is the Hawking temperature.
Therefore the Hawking temperature is nothing else than the Unruh temperature
which an accelerated observer measures in the vicinity of the horizon, red-shifted
to infinity.
This derivation of the Bekenstein-Hawking entropy relies on the classical Einstein equations and the properties of the vacuum of a quantum field on flat spacetime. Namely on the separate use of classical general relativity on the one hand
and quantum field theory on the other. Can it be repeated in the context of the full
quantum theory of gravity, without relying on semiclassical approximations?
Two avenues have been studied in order to address this question. The first is
based on using statistical arguments: counting the number of states on the horizon. The second is based on the identification of the relevant entropy with the
entanglement entropy across the horizon. As we shall see later, the two are not alternative as they may seem at first sight, in fact, they are two sides of the same
coin. Lets first briefly sketch them separately.
10.3 Kinematical derivation of the entropy

The first possibility is to realize that an observer external to the black hole does not
have access to the degrees of freedom within the interior of the hole. Therefore she
sees the hole simply as a two dimensional surface evolving in time. The interior of
the hole does not affect the external evolution. Therefore the horizon can be treated
as an independent dynamical system interacting with the exterior.
Consider for simplicity a non-rotating and non-charged black hole. At equilibrium, this system is formed by a spherical surface with area A. In a thermodynamical context, the system will undergo statistical fluctuation. In other words,
Black holes
212
Figure 10.1
Black hole horizon punctured by the spin network links.

the horizon can shake in a thermal context, since Heisenberg uncertainty allows
it to do so, and this is the origin of its thermal behavior.
We can address the problem in the micro-canonical framework, and compute
the entropy by counting the states with given energy. The relevant energy is not
the energy measured at infinity, which includes whatever is happening in the exterior, but rather the energy measured from the vicinity of the horizon. As shown
above, this is determined by the Area of the horizon. The problem of computing
the entropy of a black hole reduces then to the problem of computing the number
of states of a 2d surface with area A.
This number diverges in the classical theory, but in loop gravity it is finite because of the discreteness of the geometry. The calculation can be done as follows.
The Gibbs ansatz states that at equilibrium and at temperature T = 1 the probability pn for a system of being in a state with energy En is proportional to the
Boltzman factor exp En . In loop quantum gravity the contribution to the area of
single link of the graph with spin j arriving crossing the horizon is given by
q
A j = 8G j( j + 1).
(10.17)
Since this link carries a spin j representation, there are in fact d j = 2j + 1 orthogonal states in which the system can be. The probability for having spin j at inverse
temperature , taking the degeneracy into account is therefore
p j ( ) (2j + 1)e E .
(10.18)
where E is the energy. Using the energy (10.11) and the explicit values of the area
(7.30) for a link of spin j, this reads
aA
p j ( ) (2j + 1)e 8G = (2j + 1)e a j( j+1) .
(10.19)
The proportionality factor
p j ( ) = Z 1 ( ) (2j + 1)e a
j ( j +1)
(10.20)
Kinematical derivation of the entropy
213
is fixed by j p j = 1, and is the partition function
Z ( ) = (2j + 1)e a j( j+1) .
(10.21)
This is the partition function describing the Gibbs state of a single link at inverse
temperature , if the system is in equilibrium with respect to the flow of the accelerated observer. The statistical entropy of the Gibbs state is by definition
S = p j log p j ,
(10.22)
and it is immediate, taking derivatives, to see that it satisfies the standard thermodynamical relation
E = TS + F
(10.23)
F = T log Z
(10.24)
where
is the Helmholtz free energy. Let as assume that the system is at the Unruh temperature = 2/a (the Unruh temperature is an independent input); then we have
immediately from the Frodden-Gosh-Perez (10.11) relation,
2 aA
+ log Z
a 8G
A
=
+ log (2j + 1)e2 j( j+1)
4G
j
S=
(10.25)
The solution for of the equation
(2j + 1)e2
j ( j +1)
=1
(10.26)
is called 0 in the quantum gravity literature, and can be easily found numerically
to be 0.274. Thus, if
= 0 0.274
(10.27)
then the Free Energy vanishes and we have that the statistically entropy is precisely the Bekenstein Hawking entropy. 1
More refined calculations of the number of quantum-geometry states with given
area have been developed extensively using canonical methods, and we refer to
1
If is not equal to 0 , we can guess a closed of the partition function to be

Z=e
A ( a 2 o )
8G
(10.28)
(see [Bianchi (2012a)]), which gives the entropy

S=
E
A
A
o
o A
+ log Z =
(1 ) =
,
T
4G
4G
4G
which is the general result in the LQC literature.
(10.29)
214
Black holes
the literature for the details [Rovelli (1996a); Ashtekar et al. (1998, 2005)]. The general result is that the entropy is finite and is proportional to the area, fully confirming the Bekenstein-Hawking result and deriving it from first principles, provided
that the Barbero-Immirzi constant satisfies is chosen as in the last equation.
An alternative path that avoids this restriction by using a grand-canonical framework where the number of punctures on the black hole is governed by a chemical
potential has been recently developed in [Frodden et al. (2011); Ghosh et al. (2013)].
We refer the reader to the current literature for the present state of this approach,
and we focus instead in the next section on an alternative approach, which takes
the dynamics more explicitly into account and is based on the covariant theory
presented in this book.
10.4 Dynamical derivation of the entropy

The core of the covariant derivation of the Bekenstein-Hawking formula is the
observation that this relates extensive quantities, and therefore is local on the horizon. This fact is made explicit by the Frodden-Ghosh-Perez relations given above.
The observer near the horizon interacts at most with a small patch of the horizon.
Therefore the problem can be studied at the level of a single elementary horizonsurface triangle, which is to say, at the level of a single link of a spin network graph
crossing (or, in the jargon of trade, puncturing) the horizon.
The second idea is to identify the entropy appearing in the Hawking-Bekenstein
formula as entanglement entropy. The idea is old, and has been defended in the
past by Jacobson, Parentani and others. It had difficulties to be accepted for two
reasons. The first is that the entanglement entropy seems at first to dependent on
the number of existing fields, and this appears to be in conflict with the universality of the Bekenstein-Hawking formula. We will see that this objection is wrong:
the entanglement entropy does not depend on the number of fields, because the
fields are not independent at the Planck scale, where gravity dominates. The second difficulty has been the reluctance to accept the idea that entanglement entropy
could give rise to thermodynamcal relations. But it does, and this is nowadays becoming more clear in various fields of physics.
On the basis of this, the idea is to recover the ingredients of the Frodden-GoshPerez derivation presented above in Section 10.2, but in the context of the full
quantum theory, using its dynamics restricted to a very small patch of spacetime,
and the properties of entangled states.
The classical local geometry at a scale much smaller than the black-hole radius
2GM is flat, with the static Frodden-Ghosh-Perez observer moving along an accel-
215
Dynamical derivation of the entropy

t
Figure 10.2
The wedge is bounded by a uniformly accelerated trajectory between P and P0 and

the two segments connecting P and P0 to the origin H on the horizon. The Wedge
amplitude is a function of the boost parameter and evolves from P to P0 .
erated trajectory with acceleration a (See Exercise at the end of the Chapter). As we
have seen, a trajectory at constant acceleration is generated by the boost generator
Kz in the direction normal to the horizon surface. If we multiply Kz by a we obtain
a transformation whose parameter along the trajectory is the proper time. Therefore H = aKz is the generator of proper time evolution, that is the hamiltonian,
for the accelerated observer. This is the operator that moves the observer along a
hyperbolic trajectory, or, equivalently, determines the change of the geometry as
perceived from the static observer at fixed distance from the horizon.
The boost generator Kz is a well defined operator in quantum gravity because
the states sit in representations of the Lorentz group. Consider a link crossing the
horizon. This is described by a state in the representation of SU(2) associated to the
corresponding spin. Say it is an eigenstate of the area with spin j. Assume j 1
since we are interested in the macroscopic limit. We are looking at regions small
with respect to 2GM but large with respect to the Planck scale. The horizon is by
assumption in the plane normal to the z direction. Therefore consider the coherent
state with this orientation, which is | j, ji. (Here we have gauge fixed the internal
frame to the external directions, for simplicity.) The area operator is
A = 8Gh| L2 |.
(10.30)
A = 8Ghj.
(10.31)
Therefore this state has area (j 1)
The evolution in spacetime of this state is governed by the map Y , which maps it
to
Y | j, ji = |j, j; j, ji
(10.32)
E = hj, j; j, j | H | j, j; j, ji = hj, j; j, j | ahKz | j, j; j, ji.
(10.33)
Its energy with respect to the accelerated observer is
Black holes
216
But since we are in a -simple representation, this is

E = hj, j; j, j | ahLz | j, j; j, ji = a h j
(10.34)
This can be rewritten in the form

E = ah
A
aA
=
,
8Gh
8G
(10.35)
Which is precisely the Frodden-Ghosh-Perez energy.

This result, obtained by Eugenio Bianchi in [Bianchi (2012a)], is very remarkable.
To appreciate, observe that one need that Einstein equations to derive the FroddenGhosh-Perez relation. Here, the same relation emerges from the quantum theory
using only the map Y and the SU(2) and SL(2,C) representation theory.
This confirms, indeed, that the information about the Einstein equations is encoded in the Y map. Notice the manner in which the Barbero-Immirzi constant
cancels.
The second main ingredient for deriving the Bekenstein-Hawking relation in the
quantum theory is the observation that |j, j; j, ji is a thermal state at temperature
T = h /2 for the flow generated by the operator Kz . This is shown by Bianchi in
the original paper [Bianchi (2012a)] by explicitly coupling a thermometer to this
state, following the accelerated trajectory, and computing the ratio between up
and down transitions among the thermometer states.
In fact, this is a general result for an accelerated observer in a Lorentz invariant
context, as first realized by Unruh. To see this is, consider the wedge amplitude
(7.62)
W ( g, h) =
(2j + 1) Tr j [Y gY h],
(10.36)
which defines the dynamics of the theory. The amplitude for a transition between
the state | j, mi and the state | j, m0 i as observed by an accelerated observer undergoing a boost of Lorentzian angle is given in (7.64)
Wj,m,m0 ( ) = h j, m0 |Y eiK Y | j, m0 i
(10.37)
together with its explicit form. The correlation function between an observable A
and an observable B respectively at the beginning and the end of this evolution is
G AB ( ) = hW ( )| AB| i,
(10.38)
where is the boundary state. What is the boundary state representing a state
which is locally Lorentz invariant? To answer this, consider a key property of the
Lorentz group: a boost rotates a spacelike vector into a spacelike vector and never
moves a vector out of the wedge. But a boost with imaginary parameter is in fact
a rotation, and can move a vector across the light cone. In particular, a boost with
imaginary parameter 2i moves a vector around the Minkowski plane, and a boost
with imaginary parameter 2i rotates the half line to the = 0 one. If we
217

take the boundary state to be given by the Lorentz dynamics outside the wedge,
we can therefore postulate
= eiK (2 ) .
This is in fact nothing else than the quantum gravity version of the old formal argument that gives an intuitive version of
the Bisognano-Wichmann theorem [Bisognano and Wichmann (1976)] which gives
the mathematics underlying the Unruh effect. The standard formal argument runs
as follows. Consider a quantum field on
Minkowski space. The vacuum state |0i can
be written as the path integral on the fields
of the lower half plane.
h |0i =
(10.39)
'+
'
D [] eiS .
(10.40)
If we split the field in its components at the left and the right of the origin, e
have
h , + |0i =
D [] eiS .
(10.41)
But his can be read as the evolution generated by a rotation from the negative to
the positive x axis, thus
h , + |0i = h |eK | + i
(10.42)
where K is the boost generator. If we now trace over to get the density matrix
representing the statistical state restricted to the positive axis, we have
= Tr [h , + |0ih0| , + i] = Tr [h |eK | + ih + |eK | i] = e2K .
(10.43)
Evolving this with with eiK to the boundary of the wedge we have the wedge
boundary state (10.39).
Bringing together the amplitude and the state, we have
G AB ( ) = Tr [eiK (2 ) BeiK A]
(10.44)
where the trace is taken in the unitary Lorentz representation. This correlation
function satisfies the KMS property, which is the mark of a thermal configuration
(see the Complements to this Chapter for a simple introduction to these notions),
with dimensionless temperature
T=
1
2
(10.45)
with respect to the dimensionless evolution parameter . Scaling the evolution
Black holes
218
parameter to the proper time s = a, where a is the acceleration of the observer

gives the Unruh temperature
TH =
h a
.
2
(10.46)
Above we have seen that the covariant dynamics gives the Frodden-Ghosh-Perez
energy. Here we have seen that it gives the Unruh temperature. The two together,
as we have seen above, give the Bekenstein-Hawking entropy, with the correct
factor. This shows that the Bekenstein-Hawking entropy SBH can be derived entirely in the quantum theory. It also shows that it can be viewed as an effect of the
quantum correlations across the horizon, because this is the source of the entropy
captured by the thermal state of the Bisognano-Wichmann theorem.
The calculation can be summarized as follows. The quantum state of spacetime
near the horizon has a form that is compatible with local Lorentz invariance. If we
restrict it to the algebra of observables of an observer that keeps herself outside
the hole, this is represented by the state
= e2K
(10.47)
where K is the generator of boosts normal to the horizon. This state has an entropy
S = Tr [ log ].
(10.48)
Now suppose this state is modified by an arbitrary process that changes it by an

amount . Then the change in entropy is
1
S = Tr [ log ] = Tr [() log ] Tr [ ].
(10.49)
The second term vanishes because must remain normalized in the change. The
first term, using (10.47) and the relation between the boost generator and the Area
derived above gives
S = Tr [()2K ] = 2Tr [()
A
A
A
] = 2
=
= SBH .
8Gh
8Gh
4Gh
(10.50)
So we find the (differential form of the) Bekenstein-Hawking relation.

Say for simplicity we consider a local change of state given by a change j in the
spin of a link crossing the horizon. Then we have the change in the energy
E = a h j
(10.51)
and the change in entropy

S = E/T =
a h j
= 2j.
h a/2
(10.52)
Thus up to a possible additive constant, the entropy associated to a link is a simple

function of the spin
S = 2j
(10.53)
219

and since the area of a link is A = 8Gh j, this gives
S=
2A
A
=
8Gh
4h G
(10.54)
again the Bekenstein-Hawking formula.
10.4.1 Entanglement entropy and area fluctuations

The two previous sections seem to point to different physical interpretations for
the origin of the black hole entropy. The first is the idea that this is the effect of the
quantum fluctuations of the horizon. The second points to the idea that the black
hole entropy is given by the entanglement entropy across the horizon. Is there a
contradiction?
No. The two interpretations are not in contradiction with one another; they are
two faces of the same phenomenon, the local quantum fluctuations (of anything)
near the horizon. On the one hand, these fluctuations generate geometry fluctuations, and therefore can be visualized as a quantum shaking of the horizon, or
a Planck scale thickness of the horizon, due to the fact that this cannot be resolved at a scale smaller than the Planck scale. On the other hand, we can hold a
mental picture where the horizon is fixed in some (imaginary) background coordinates, and the quantum field theoretical fluctuations have finite wavelengths that
correlate the value of any field across the horizon.
For some time the possibility that the black hole entropy be related to entanglement entropy across the horizon has been considered implausible because of an
obvious objection: the horizon fluctuations are a purely geometrical phenomenon
and therefore can justify a universal formula like the Bekenstein-Hawking formula. But the horizon entanglement is a feature of all quantum fields that are
present, and therefore the entropy should increase if the number of fields increase.
Why then the Bekenstein-Hawking formula does not depend on the number of
fields? The solution to this puzzle has recently been found by Eugenio Bianchi
[Bianchi (2012b)], who has computed the variation of the entanglement entropy
Sent and of the horizon area A in a physical process, using standard quantum field
theory and classical general relativity. He has found that these variation are finite
and given by
A
,
(10.55)
Sent =
4h G
independently from the cut off and from the number of fields. The key to the mystery is that Sent does not increase with the number of fields, because at small
scale the fields are not independent. The cut off on the fields modes goes into effect when the total energy density of the fields reaches the Planck scale. The total
energy density depends on the individual fields fluctuations plus their (negative)
potential gravitational energy. Therefore he cut off scale on the wavelength of an
individual field depends on the presence of other fields, which are fluctuating at
small scale as well. Alltogether, the result is that the entanglement entropy does
220
Black holes
not dependent on he number of fields. (On this, see also the intriguing recent results in [Ghosh et al. (2013)] and [Frodden et al. (2012); Han (2014)].)
The issue of understanding black-hole entropy directly in quantum gravity has
been largely clarified by these loop gravity results. The entropy of a black hole is
a local phenomenon due to the fact that stationary observers are accelerating. It is
related to quantum fluctuations of the horizon or, equivalently, to entanglement
across the horizon. The entropy can be computed from first principles, including the famous 1/4 Hawking factor, it is finite and the calculation can be done
for realistic black holes, such as a Schwartzschild or Kerr hole. This is beautiful
achievement of the theory which for the moment is not matched by any of the
other tentative quantum theories of gravity.
10.5 Complements
10.5.1 Accelerated observers in Minkowski and Schwarzshild
Exercise: Consider a uniformly accelerated particle in Minkowski space, following the trajectory
t = l sinh ,
x = l cosh .
(10.56)
(10.57)
show that the acceleration is constant and is a = 1/l, the proper time among the trajectory
is s = l, the distance between the particle and the origin is constant and given by l, the null
line x = t is a horizon for this particle, and no signal emitted at t = 0, x < 0 can reach the
particle. What is the Unruh temperature felt by this particle?
Exercise: Consider a uniformly accelerated particle in Schwarzshild spacetime, following

the trajectory
r = 2GM + e.
(10.58)
where e 2GM. Show that show that the acceleration is a = 1/l where l is the physical
distance of the particle from the horizon. Show that this distance is constant (do not confuse
it with e)! Compute the temperature felt by this particle using the Hawking temperature at
infinity and the Tolman law. Compare with the previous exercise.
10.5.2 Tolman law and thermal time

If we ignore general relativity, two systems are in equilibrium when they have the same
temperature. An isolated column of gas in a (Newtonian) gravitational field, for instance,
will thermalise to an equilibrium state where the pressure decreases with altitude by the
temperature is constant. This, however, is no longer true when general relativity is taken
221
Complements
into account. At equilibrium, the temperature T (measured by a standard thermometer) of

the gas will decrease with the altitude as
T
g
= 2
T
c
(10.59)
where g 9.8m/s2 is Galileo acceleration. The covariant version of this relation is

T || || = constant,
(10.60)
where || || is the norm of the Killing field along which equilibrium established. This fact,
which at first is quite surprising was first derived theoretically by Tolman and Ehrenfest
in the early thirties [Tolman (1930); Tolman and Ehrenfest (1930)], and then re-derived in a
number of ways by many authors.
The simplest way to understand it is simply to observe that gravity slows down time.
In a stationary gravitational field, stationary clocks do not remain synchronised. Therefore
there are two distinct notions of time flow, in a stationary field: the common time given
by the Killing field, along which spacetime is stationary, and the local time s, given by the
proper time along each trajectory at rest. The ratio of the two is given by the norm of the
Killing field. A clock is a devise that measures the transition probability between energy
levels, and these levels are determined by the proper time, not by . But an equilibrium
configuration is in equilibrium with respect to the flow of the Killing field, namely in , not
in s. Therefore the different regions in a global equilibrium configuration are red-shifted
with respect to the other, by the ratio between and s.
Another way if viewing the same effect is to realise that in general relativity energy
weight, and therefore accumulates at lower altitude. More precisely, when energy leaves
a lower part of the system and reaches a hight part of it, it arrives red shifted, therefore
diminished. It follows that we cannot anymore maximise entropy by assuming that the energy lost by one subsystem is gained by the other: we must include the red shift factor in
the calculation, and this gives (10.59)
The equilibrium time parameter is an example of thermal time [Rovelli (1993a); Connes
and Rovelli (1994)]. Temperature can be defined by the ratio between the thermal time and
the proper time T d/ds. The proportionality factor fixes the scale of the Killing field and
thermal time [Rovelli and Smerlak (2010)]. Fixing it to
T=
h d
k ds
(10.61)
is particularly convenient because it makes dimensionless and it identifies it with the

number of individual quantum cells jumped over by a system at equilibrium: see [Haggard
and Rovelli (2013)].
10.5.3 Algebraic quantum theory

We describe here the basic formalism of thermal states in the algebraic language, used in
the text. This is more general and more powerful than the standard Hamiltonian language
for describing thermal states.
We also give a direct physical interpretation of the KMS condition, in terms of thermometer coupling.
Consider a quantum system with Hilbert space of states H, hamiltonian H and an algebra
A of self-adjoint observables A, B.... Every state H defines a positive linear function on
the algebra by
( A ) = h | A | i
(10.62)
and so does, more in general, any density matrix

( A) = Tr [A].
(10.63)
Black holes
222
The hamiltonian defines a flow on the algebra by

t ( A) = eiHt AeiHt .
(10.64)
The algebra A with the flow t and the set of its positive state provides a language for
talking about quantum theory which is more general than the Hilbert space one (see below).
Let us use this language to describe thermally.
10.5.4 KMS and thermometers

Consider a quantum system described by an observable algebra A, with observables A, B, ...,
which is in a state : A C. Let t : A A with t R be a flow on the algebra. We say
that the state is an equilibrium state with respect to t if for all t,
( t A ) = ( A ),
(10.65)
and that its temperature is T = 1/ if

f AB (t) = f BA (t + i)
(10.66)
f AB (t) = (t A B).
(10.67)
where
An equivalent form of (10.66) is given in term of the Fourier transform fAB of f AB as

fAB ( ) = e fAB ( ).
(10.68)
The canonical example is provided by the case where A is realized by operators on a

Hilbert space H where a Hamiltonian operator H with eigenstates |ni and eigenvalues En
is defined, the flow and the state are defined in terms the evolution operator U (t) = eiHt
and the density matrix = e H = n e En |nihn|, by
t ( A) = U 1 (t) AU (t)
(10.69)
( A) = tr[A]
(10.70)
and
respectively. In this case, it is straightforward to verify that equation (10.65) and (10.66)
follow.
However, the scope of equations (10.65) and (10.66) is wider than this canonical case. For
instance, these equation permit the treatment of thermal quantum field theory, where the
Hamiltonian is ill-defined, because of the infinite energy of a thermal state in an infinite
space. It is indeed important to remark that these two equations capture immediately the
physical notions of equilibrium and temperature. For equation (10.65), this is pretty obvious: the state is in equilibrium with respect to a flow of time if the expectation value of any
observable is time independent.
The direct physical interpretation of equation (10.66) is less evident: it describes the coupling of the system with a thermometer.
To see this, consider a simple thermometer formed by a two-state system with an energy
gap e, coupled a quantum system S by the interaction term
V = g(|0ih1| + |1ih0|) A.
(10.71)
where g is a small coupling constant. The amplitude for the thermometer to jump up from
Complements
223
the initial state |0i to the final state |1i, while the system moves from an initial state |i i to a
final state | f i can be computed using Fermi golden rule to first order in g:
W+ (t) = g
=g
Z t
Z t
dt (h1| h f |) t (V ) (|0i + |i i)
dt eite h f |t ( A)|i i.
(10.72)
The probability for the thermometer to jump up is the modulus square of the amplitude,
summed over the final state. This is
P+ (t) = g2
Z t
dt1
Z t
dt2 eie(t1 t2 ) (t2 ( A )t1 ( A))
where we have used the algebraic notation ( A) = hi | A|i i. If the initial state is an equilibrium state
P+ (t) = g2
Z t
dt1
Z t
dt2 eie(t1 t2 ) f AA (t1 t2 ).
and the integrand depends only of the difference of the times. The transition probability
per unit time is then
p+ =
dP+
= g2 fAA (e)
dt
(10.73)
which shows that (10.67) is precisely the quantity giving the transition rate for a thermometer coupled to the system. It is immediate to repeat the calculation for the probability to
jump down, which gives
p = g2 f AA (e).
(10.74)
And therefore (10.68) expresses precisely the fact that the thermometer thermalizes o at
temperature , that is p+ /p = e e
These observations show that equations (10.65) and (10.66) define a generalization of
the standard quantum statistical mechanics, which fully captures thermal properties of a
quantum system. The structure defined by these equations is called a modular flow in the
mathematical literature: t is a modular flow, or a Tomita flow, for the state , and is the
basic tool for the classification of the C algebras. In the physical literature, the state is
called a KMS state (for the time flow). In order to show that a system is in equilibrium and
behave thermally in a certain state, it is sufficient to show that the state satisfies these two
equations.
Notice that in the context of Hilbert space quantum mechanics, a state can satisfy these
two equations also if it is a pure state. The standard example is provided by the vacuum
state of a Poincare invariant quantum field theory, which is KMS with respect to the modular flow defined by the boost in a given direction. Being Poincare invariant, the vacuum
is invariant under this flow, and a celebrated calculation by Unruh shows that it is KMS at
inverse temperature 2.
In this case, the physical interpretation is simply given by the fact that an observer stationary with respect this flow, namely an accelerated observer at unit acceleration, will measure a temperature 1/2. Unruhs original calculation, indeed, follows precisely the steps
above in equations (10.71-10.74).
10.5.5 General covariant statistical mechanics and quantum

gravity
The relations between gravity, quantum and statistics form a net of intertwined problems,
where much of our confusion on the fundamental aspects of nature lies. At the core of the
Black holes
224
difficulty is the fact that we lack a coherent statistical theory for the gravitational field (as we
have one for the electromagnetic field), even in the classical case. The reason is related to the
peculiar manner time appears in general relativity, which makes the standard tools of statistical mechanics useless. The effort of extending the power of statistical reasoning to general
covariant theories is, in our opinion, among the deepest and most beautiful problems open
at the core of our present understanding of the physical world. As beautiful as, and most
likely strictly related to, quantum gravity. Here we only sketch briefly a few results and
ideas in a research line in this direction under development by the authors, pointing to the
relevant literature.
Thermal time in the classical theory. If all the variables are on the same footing, it is not
clear what equilibrium means. In such a situation (generically) any statistical state
can be seen as an equilibrium state, because it is invariant under a flow in phase
space: the hamiltonian flow generated by H = ln . Since is a density on phase
space, it has the dimension of inverse action, therefore we need a constant to fix its
logarithm. Looking ahead, we do so by taking h = 1 units. The quantity H is called
the thermal hamiltonian of the state and the (dimensionless) flow parameter
is called thermal time [Rovelli (1993a)]. The thermal hamiltonian of a Gibbs state
is the standard hamiltonian scaled by the temperature and the temperature is the
ratio between the thermal time and the physical time t:
T=
h
,
k t
(10.75)
an expression we have already seen in (10.61) generalises to gravity, where t becomes the proper time s which varies, as does the temperature, from point to point.
Mean geometry. If we study the thermal fluctuations of the gravitational field, then it is
reasonable to call equilibrium a state where the these fluctuations are centred
around a mean geometry having a Killing field which can be identified with the
thermal time flow. This construction is developed in [Rovelli (2013a)].
Thermal time in quantum field theory. The idea of thermal time gets bite in quantum field
theory, where the thermal time flow is the modular flow of the state, which is
independent on the state up to unitaries [Connes and Rovelli (1994)]. The modular
flow of the vacuum state of a quantum field theory, restricted to the algebra of the
observables on the Rindler wedge x > |t|, is the flow of the boost, the thermal time
is the boost parameter and the local temperature defined by (10.75) is the Unruh
temperature. But the thermal flow makes sense also in non stationary situations,
like for instance a Friedmann cosmology, where the thermal time of the cosmic
background radiation state turns out to be the cosmological time [Rovelli (1993c)].
The zeroth principle. Remarkably, when two systems interact, equilibrium is given by the
equality of their thermal times [Haggard and Rovelli (2013)]. This has a physical
interpretation: thermal time is time in the units given by the time a quantum state
moves to a state orthogonal to itself. Therefore two interacting systems in equilibrium interact with the same number of distinct states: the net flux of information
they exchange vanishes. This notion of equilibrium, defined in terms of information rather than in term of a preferred time variable, generalises to the covariant
systems where no preferred time is available [Haggard and Rovelli (2013)]. In turn,
the combined state of the two systems itself can select a time variable among the
combination of the individual times [Chirco et al. (2013)].
Statistical states in the boundary formalism. In Section 2.4.2 we have argued that in gravity we need to associate states to the full boundary of a spacetime region. If
can be split into a past and a future component, this Hilbert space has the structure
H = H past H f uture . This space contains states that are not of the tensor product
of a state in H past and a state in H f uture . What do these represent? It is not difficult
to see that they represent statistical states [Bianchi et al. (2013)].
.
225
Figure 10.3
Complements
Lens shaped spacetime region with spacelike boundaries and corners (filled circles).
Statistical versus quantum fluctuations. In particular, if there is a corner joining the future and the past components of the boundary, as in Figure 10.3 and if a quantum
version of the equivalent principle holds, implying that the corner is locally like
a Minkowski state in the UV, then it follows that all boundary states are mixes.
Because locally in the corner the state has the form
= e2K
(10.76)
where K is the boost generator at the corner. This supports the idea that statistical
fluctuations and thermal fluctuations become indistinguishable when gravity and
horizons are in play.
These brief notes are of course far from exhausting this fascinating topic, which still seeking for clarity.
11
Cosmology
The major application of loop gravity to a realistic physical phenomenon is early
cosmology. This is also the most important application, because it appears to be
the most likely to lead to predictions that could confirm the theory. Observational
cosmology is advancing rapidly and there is hope that loop gravity could lead
to specific testable predictions in cosmological observations, in particular in the
cosmic background radiation.
As for the case of black hole thermodynamics, there are two main avenues for
applying loop gravity to cosmology. The first is based on the canonical formulation
of the theory, the second on the covariant formulation. Both are based on the same
idea as classical relativistic cosmology: focusing on the dynamics of the very large
scale degrees of the universe [Einstein (1917)]. Here we touch on the first only
briefly, and we discuss a bit more in detail the second. We begin by reviewing the
classical formulation of a simple cosmological model, concentrating on the aspects
relevant for the quantum theory.
11.1 Classical cosmology

Cosmology is not the study of the totality of the things in the universe. It is the
study of a few very large-scale degrees of freedom. It is a fact (intuited by Einstein)
that at very large scale the universe appears to be approximately homogenous and
isotropic. Therefore it admits an approximate description obtained expanding in
modes around homogenous-isotropic geometries.
The spatial curvature of the universe appears to be very small at the scale we
access. One must not conclude from this observation that the universe is spatially
flat: humanity has already made this mistake once: the Earths average curvature
is very small at our scales and we concluded that the Earth is flat. Let us not repeat
this mistake (many do). But to a good approximation a spatially flat universe approximates the patch of the Universe that we see, and since a spatially flat universe
is a bit easier to describe than a closed one, at least for some purposes, it makes
sense to use a spatially flat metric here.
Thus, consider a Friedman-Lematre metric, where the function a(t) is the scale
factor, and write it in terms of tetrads:
J
ds2 = N (t)dt2 + a(t)2 d~x2 = I J eI (t) e (t) dx dx
where, introducing eI dx = e I ,
226
(11.1)
Classical cosmology
227
e0 = N (t)dt
ei = a(t)dxi .
Inserting this metric into the action gives

Z
3
3
2
S[ a] =
dt N a a a + Smatter
16G
3
(11.2)
(11.3)
The momentum conjugate to the scale factor is
L
3Vo
=
N aa
(11.4)
a
8G
where L is the Lagrangian and Vo is the coordinate volume of the space considered.
This volume cannot be taken infinite without introducing a confusing infinite in
the formalism. So we take it finite. Normalizing the coordinates so that Vo = 1 has
the consequence that a3 is the actual volume of the universe. With this choice, the
boundary term in the gravitational action is then
pa =
3
2.
N aa
(11.5)
8G
The dynamics of the scale factor defines a one-dimensional dynamical system,
which is the ground for studying physical cosmology. Solution to this system and
perturbations around these solutions describe the large scale dynamics of the universe quite well. The recent observations by the Planck satellite [Planck Collaboration (2013)] have confirmed the credibility of this picture to a surprising degree
of accuracy.
The equation of motion obtained varying the Lapse function N is an equation
In the presence of a matter energy density
that ties a and a.
Sboundary = p a a =
( t ) = a3
Smatter
N (t)
(11.6)
the variation of the Lapse gives

2
a
8
= + G,
a
3
3
(11.7)
which is the Friedmann equation, the main equation governing the large scale
dynamics of the universe. After taking this equation into account, we can fix the
Lapse, say at N = 1. The Friedman equation gives a as a function of a at every time.
Therefore we can view the boundary term simply as a function of a. It follows that
the Hamilton function is a pure boundary term of the form
S ( ai , a f ) = S ( a f ) S ( ai )
where
(11.8)
3
2
aa
(11.9)
8G
and a depends on a via the Friedmann equation. Therefore we must expect the
transition amplitude of the quantum theory in the small h limit to factorize
S( a) =
W ( ai , a f ) e h S(ai ,a f ) = e h S(a f ) h S(ai ) = e h S(a f ) e h S(ai ) W ( a f )W ( ai )
(11.10)
Cosmology
228
where
i
.
W ( a) e h aa
(11.11)
ai .
kia = eia = a
(11.12)
This is the classical limit to which a cosmological quantum transition amplitude

must converge, if it is has to agree with standard cosmology.
Let us translate this to the Ashtekar variables defined in Section 3.4.3 . The
Ashtekar-Barbero connection Aia = ia + kia conjugate to the densitized inverse
j
triad Eia = 12 e abc eijk eb eck is the sum of the spin-connection ia of the triad and the
extrinsic curvature of the spacial slide kia multiplied by the Barbero-Immirzi constant , that we here take equal to , as usual in loop quantum cosmology. In the
flat case ia = 0 because space is flat and the extrinsic curvature is diagonal and a
multiple of the time derivative of the scale factor
Therefore the Ashtekar-Barbero variables are reduced to

Eia = a2 ia = pia ,
Aia = a ai = cai
(11.13)
which satisfy
{c, p} =
8G
.
3Vo
(11.14)
Inserting these variables in the hamiltonian constraint (3.86) gives (up to an irrelevant overall constant)
2
cp
p3 .
(11.15)
H=
H = 0 is the Friedmann equation for pure gravity. Adding the matter energy
density (assumed constant in space) we obtain
c2
8
= + G,
3
3
2 p
(11.16)
which translated back into the a(t) variable is (11.7), namely the standard Friedmann equation for a flat universe with cosmological constant. The Hamilton function factorizes as above and reads
S( p)
c( p) p
(11.17)
where c( p) is the solution of the Friedmann equation. Notice that cancels. For
instance, if there is no matter, we expect
r
3
S( p)
p2.
(11.18)
3
Thus for the classical limit to hold, the transition amplitude must have the small h
behavior
i
W ( p) e h
p2
(11.19)
229
Canonical Loop Quantum Cosmology
11.2 Canonical Loop Quantum Cosmology

A large body of work has been developed to study the canonical quantization of
the dynamical system defined by the dynamics of the scale factor, as well as the
perturbations around it. We refer the interested reader to recent reviews, such as
[Ashtekar and Singh (2011)] and [Bojowald (2011)], which follow slightly different
paths, and references therein.
The main result of this work is that the dynamics of the quantum dynamical system describing the evolution of the scale factor is well-defined also around what
appears in the classical approximation as the initial singularity. That is, the Big
Bang singularity does not appear to be a physical event, but only an artifact of the
classical approximation. In this, it is analogous to the possibility for an electron to
fall into the nucleus of the atom, which is predicted by the classical approximation but not by the quantum theory of the electron. The quantum evolution of the
universe near the Big Bang is well defined, and no singularity develops.
Interestingly, the quantum corrections can be approximated in terms of a modified classical dynamical evolution for a wave packet. This gives a modified Friedman equation. In the case = 0 and k = 0, for instance, one obtain

2
8G
a
=
1
,
(11.20)
a
3
c
where the critical density c is a constant with a value that is approximatetively
half of the Planck density (Pl 5.1 1096 kg/m3 ). In the absence of the correction
term, a cannot change sign at small scales. In the presence of the correction term, a
can vanish, and the dynamics of the scale factor bounces: semiclassical trajectories describe a collapsing universe that bounces into an expanding one. We do not
give a derivation of (11.20) here, see [Ashtekar and Singh (2011)], but a few words
on how this happens may be useful.
The idea is to study a quantum system with Friedman cosmology as classical
limit using two key inputs from the full theory.
i.) The first is the idea that the proper operators are not the local ones but the integrated ones. In particular, the relevant operator is not the connection but
rather its holonomy. This translates into cosmology in the request that the operator well defined in the quantum theory is not c but rather eic , for suitable
.
ii.) The second is that the dynamics is modified to take into account the quantum
discreteness of space. This is obtained by replacing the curvature in the hamiltonian constraint with the holonomy around a small circuit encircling the
smallest possible quantum of area.
The first input allows the definition of a suitable Hilbert space where the operators
eic are defined. The second permits the definition of the dynamics. A detailed
analysis of the resulting system and its semiclassical limit yields then the above
230
Cosmology
result. The resolution of cosmological singularities can be seen to be quite generic
[Singh (2009); Singh and Vidotto (2011)].
The Poisson brackets (11.14) become commutators in the quantum theory, and
yield Heisenberg uncertainty relations
cp >
8h G
.
3Vo
(11.21)
By taking Vo arbitrarily large, the uncertainties can be made arbitrarily small. This
is analogous to the fact that by taking an object made by sufficiently many atoms
we can have its position and velocity arbitrarily well defined. In other words, loop
quantum cosmology looks only at the large modes of the field, which are averages
in space of the gravitational field, and is blind to the local quantum fluctuation
of the gravitational field, which are dominated by shorter wavelength modes. It
follows that the theory, so far, does not describe the actual fluctuating geometry a
the bounce. The bounce is not determined by the universe being small, but by the
matter energy density reaching Planck scale, and this can happen, depending on
the matter content of the universe, at any size of the universe.
We also report here, without deriving it, the effect of quantum gravity on cosmological perturbations. Remarkably this as well can be summarized in a correction to the standard cosmological perturbation theory. The Mukanov equation that
governs the dynamics of the perturbations [Mukhanov et al. (1992)]
z
v 2 v v = 0,
z
is corrected to [Cailleteau et al. (2012a,b)]

v 1 2
2 v
c
z
v = 0.
z
(11.22)
(11.23)
An electron does not fall into the Coulomb potential because of the uncertainty
principle, which forbids concentrating it into too small a region of space without
having a large momentum that allows it to escape. In a sense, quantum mechanics
provides an effective repulsive force at short distance. The same happens for the
universe: when too much energy density is concentrated into too small a region,
the uncertainty principle prevents further collapse and acts as an effective strong
repulsive force.
The precise interpretation of this semiclassical bounce is in our opinion not
completely clear yet, because more understanding of the quantum fluctuations of
spacetime at the bounce is needed. The full physics of the gravitational field near
the bounce is not yet well understood. Is spacetime still approximately classical at
the bounce or do the full quantum fluctuations make the very notion of spacetime
ill defined around the bounce, as the notion of a classical trajectory is ill defined
for an electron falling into a Coulomb potential? In the second case, the picture of
a previous contracting phase of the universe bouncing into a later expanding
one might be naive for some purposes: spacetime itself becomes ill defined in early
cosmology.
Spinfoam Cosmology
231
Figure 11.1
Two tetrahedra, glued by all their faces, triangulate a 3-sphere.
11.3 Spinfoam Cosmology

The approach to quantum cosmology based on the covariant formulation of loop
gravity was introduced in [Bianchi et al. (2010e)] and is called spinfoam cosmology. The expansion at the basis of the covariant theory is suitable for cosmology,
because it is an expansion in the number of degrees of freedom. Since a cosmological model is the reduction of general relativity to a small number of degrees of
freedom, it can be naturally described at low order in the spinfoam expansion.
Consider a closed homogeneous and isotropic universe, with the spacial geometry of a 3-sphere S3 of radius a, evolving in time. We want to study the quantum
dynamics of the degree of freedom a(t) as well as the quantum dynamics of the
large-scale deformations of the three sphere.
In a finite lapse of time, a compact universe evolves from a 3-sphere to a 3sphere. Therefore we want the transition amplitude form a state of quantum space
with the topology of a 3-sphere to different quantum state of space with the topology of a 3-sphere. To represent one such states, we have to choose a discretisation
of a 3-sphere. Spinfoam cosmology was first studied using the simplest cellular
decomposition of S3 . This can be obtained by gluing two tetrahedra by all their
faces.1 The graph dual to this triangulation of S3 is obtained by placing a node
on each tetrahedron and connecting the two nodes accords each of the four faces.
(Figure 11.1). This graph is called 2 , or dipole graph: it is formed by 2 nodes
1
Figure 11.2
The common sphere S2 can be obtained by taking two triangles and gluing them by their sides.
Five tetrahedra triangulate a 3-sphere.
Cosmology
232
connected by 4 links. Quantum cosmology on this graph was first studied in the
hamiltonian language [Rovelli and Vidotto (2008); Battisti et al. (2010); Borja et al.
(2010, 2011)]. More general regular graphs were studied in [Vidotto (2011)]. For
instance, we can obtain a three sphere by pairwise gluing five tetrahedra, as in
the boundary of a four-simplex (Figure 11.2).2 In the following we use a generic
regular graph . The first step to describe the large scale geometry of the universe
is to write the boundary states in the Hilbert space H , representing semiclassical
configurations. Then we can take a couple of these states, which can be thought of
as in and out states, and compute the corresponding transition amplitude.
11.3.1 Homogeneus and isotropic geometry

Fix a regular graph, namely a graph where all the nodes have the same number
of links.3 As an example, we can think about 5 , the graph formed by five nodes
all connected. Pick a cellular decomposition of a (metric) three sphere formed by
equal cells and dual to the regular graph. Approximate each cell with a flat cell
having the same volume, face area and average area normal. This defines normals
~nl associated to each oriented link of the graph. By regularity, all areas and extrinsic curvatures on the graph are equal. We can construct a coherent state on the
graph, based on these geometrical data, using the coherent states developed in
Section 8.4. The coherent state is determined by an SL(2, C) element H` for each
link, where H` is determined by the two (distinct) normals associated to two cells
bounding the face, and a complex number z, which is the same for all links. The
real and imaginary parts of z are determined by the area and the extrinsic curvature on the face. We assume on the basis of (8.120), that the relation between these
and the coefficients of an isotropic geometry are given simply by
p
(11.24)
z = c + it 2
lo
(For a finer analysis, see [Magliaro et al. (2011)]). Notice that the Planck length in
the denominator is imposed by (8.120) as well as by dimensions.
Let us now seek quantum states that describe the
geometry of these triangulated spaces. For this, consider the extrinsic coherent states H` (h` ) in the
Hilbert space H introduced in Section 8.4. Recall that
these are determined by the data (cl , pl , ns(l ) , nt(l ) )
With these assumptions, a homogeneous isotropic
5
coherent state on a regular graph is described by a
single complex variable z, whose imaginary part is
proportional to the area of each regular face of the cellular decomposition (and
2
3
The common sphere S2 can be obtained by taking fours triangles and gluing them by their sides, as
in the boundary of a tetrahedron.
There are an infinite number of such graphs. For instance, two nodes can be connected by arbitrary
number of links. Examples of regular graphs with N > 2 are given by the (dual of) the Platonic
solids.
Spinfoam Cosmology
233
it can be put in correspondence with the total volume) and whose real part is
related to the extrinsic curvature [Rovelli and Speziale (2010)]. We denote H` (z)
this state, and H` (z,z0 ) = H` (z) H` (z0 ) the state on two copies of the regular
graph, obtained by tensoring an in and a out homogeneous isotropic state.
These states are peaked on an homogenous and isotropic geometry, but since they
are genuine coherent states, they include the quantum fluctuations around this
geometry, for the degrees of freedom captured by the graph.
11.3.2 Vertex expansion

For a boundary state formed by two disconnected components, the transition amplitude determines the probability to go from one state to the other. For the coherent states that have just been introduced, this reads
WC (z0 , z) = hz0 | W | zi =
dh`
dh0` z0 (h0` ) WC (h0` , h` ) z (h0` )
(11.25)
The transition amplitude can be computed by approximating it with a truncated

version of the theory, namely by choosing a two-complex C bounded by the boundary graph, that is by two 5 graphs. At the lowest order, we have have the contribution of a two-complex without vertices.
hz0 | WCo | zi
(11.26)
This vanishes if the initial and final states are sufficiently far apart, which is the
non-trivial case we are interested in. Let us therefore study the first-order term,
determined by a two-complex with a single vertex
hz0 | WC1 | zi
(11.27)
We write this transition amplitude using (7.67) and (7.68)
where
WC1 (h0` , h` ) = W (h0` )W (h` )

W ( h` ) =
SL(2,C)
dge0
`=1
P ( h ` , ge )
(11.28)
(11.29)
Cosmology
234
The integration over the SL(2, C) elements hn associated to the edges imposes
Lorentz invariance. It is over all the gn but one. Using these, and the definition
(8.126) of the coherent states, a short calculation gives
WC1 (z0 , z) = W (z)W (z0 )
(11.30)
where
W (z) =
Z
Z
dh`
dgn0
dgn0
`=1
Kt (h` , H` (z)) P( gn , h` )
(j )
(j ,j )
(2j` +1) e2tj` ( j` +1) Dmm` 0 ( H` (z)) Djm0`jm`
`=1 j`
( G` ). (11.31)
The transition amplitude factorizes, as we expected from the classical theory. Each
individual term W (z) can be interpreted as a Hartle-Hawking wave function of
the universe determined by a no-boundary initial condition [Hartle and Hawking
(1983)], namely the amplitude to go from nothing to a give state. The reason this
is equivalent to the transition amplitude from a given geometry to another is the
reflection of the fact that in the classic theory the dynamics can be expressed as a
relation between the scale factor and its momentum, which is a relation at a single
time. The probability of measuring a certain out coherent state does not depend
on the in coherent state.
Let us now evaluate W (z) in the semiclassical regime
11.3.3 Large-spin expansion

We are interested in (11.31) in the regime where the volume of the 3-sphere (which
is determined by p) is large compared to the Planck scale. By (11.24), this corresponds to the regime where the imaginary part of z is large. From (8.129), we have
3
1
D ( j) ( H` (z)) = D ( j) ( R~ns ) D ( j) (eiz 2 ) D ( j) ( R~
n n ).
(11.32)
When the imaginary part of z is large, Im z 1, the Wigner matrix

3
D ( j) (eiz 2 ) =
eizm |mihm| .
m
(11.33)
is dominated by the term m = j, therefore

3
D ( j) (eiz 2 ) eizj | jih j|,
(11.34)
where | ji is the eigenstate of L3 with maximum eigenvalue m = j in the representation j. Inserting this into (11.32) gives
1
D ( j) ( H` (z)) = D ( j) ( R~ns ) | ji eizj h j| D ( j) ( R~
n n ).
(11.35)
and recalling the definition (8.23) of the intrinsic coherent states this is
D ( j) ( H` (z)) = eizj | j, ~n` ih j, ~n` | .
(11.36)
Spinfoam Cosmology
235
Inserting this in (11.31) gives

W (z) =
dgn0
(j ,j )
(2j` +1) e2tj` ( j` +1) eizj` h j` , ~n` | Djm0`jm`
`=1 j`
( G` )| j` , ~n` i.(11.37)
Each of the sums is a gaussian sum that peaks the spin on a large value, which,
using (11.24), is
jo =
p
,
4lo2
(11.38)
the same for all `s. Therefore we can write
(2j+1) e2tj( j+1) eizj
W (z) =
!L
( jo ) .
(11.39)
where
( jo ) =
(j ,j )
dgn0 h jo , ~n` | D jm0 ojmo ( G` )| jo , ~n` i.
(11.40)
`=1
This integral can be computed in the large spin regime using saddle-point methods
and the formulas of the previous chapter. The matrix elements are simply peaked
around the origin gn = 1l, and the result of the integration is determined by the
Hessian, which is a rational function of jo . Here we are inly interested in the phase
of W (s) and therefore we can write
W (z)
2tj( j+1) izj
(2j+1) e
!L
z2
e L 8t
(11.41)
The classical limit is obtained by the rapidly oscillating phase of the amplitude, as
explained in Section 2.2. From the definition (11.24) of z, this is
W (z) e
Lcq
4lo2
(11.42)
Translated in standard variables this is

2
L aa
W (z) ei 8Gh
(11.43)
which is the expected behavior (11.11). This indicates that we can recover the classical limit from spinfoam cosmology. This confirms that the transition amplitudes
computed in spinfoam cosmology yield the classical result in the appropriate limit.
For more details and more results in spinfoam cosmology, including the inclusion of the cosmological constant, see [Bianchi et al. (2011a); Vidotto (2011); Enrique F. Borja et al. (2011)].
236
Cosmology
11.4 Maximal acceleration

Spinfoam cosmology is not yet developed to the point of allowing us to describe
the quantum physics of the near-black-hole region directly. However, it does provide an indirect element of evidence for the resolution of the classical singularity,
because it directly predicts a limit on the acceleration. The idea that quantum gravity may limit the acceleration is old [Caianiello (1981); Caianiello and Landi (1985);
Brandt (1989); Toller (2003)]. In covariant loop quantum gravity, it can be seen as a
direct consequence of the quantisation of the area and the simplicity relations that
relate area and boost [Rovelli and Vidotto (2013)]. To see this, consider a uniformly
accelerated observer in Minkowski space, with acceleration a. The wedge defined
by a portion of its trajectory is depicted in Figure (10.2). Specifically, taking the
boost angle of this wedge = 1, it is an easy calculation to see that the Lorentzian
area of the wedge is
Al =
1
2a2
(11.44)
Therefore the Area of this wedge measures the acceleration of the observer. But
in turn, the Lorentzian area of this wedge is tied to the Euclidean area Ae by the
linear simplicity constraint
Al =
1
Ae
and since the minimum non vanishing eigenvalue of Ae is 8Gh

it follows that the maximal value of the acceleration is
r
1
amax =
.
8Gh
(11.45)
p
1/2(1/2 + 1),
(11.46)
In turn, a maximal acceleration screens the classical singularity. The resolution

of classical singularities under the assumption of a maximal acceleration has been
studied using canonical methods for Rindler [Caianiello et al. (1990)], Schwarzschild
[Feoli et al. (1999)], Reissner-Nordstrom [Bozza et al. (2000)], Kerr-Newman [Bozza
et al. (2001)] and Friedman-Lematre [Cainiello et al. (1991)] metrics. Here we consider can sketch a simple argument using a homogeneous and isotropic cosmological model, with vanishing spatial curvature and pressure. The dynamics is governed by the Friedman equation (we use here R for the scale factor in order not t
confuse it with the acceleration a)
8G
R 2
=
2
3
R
(11.47)
R3 is the matter energy density. Any co-moving observer is accelerating with

respect to his neighbours in this geometry. The growing acceleration approaching
a classical singularity is bounded by the existence of a maximal acceleration a
Physical predictions?
237
R/R.
This gives a maximal value of the energy density

3
3 R 2
3 2
=
`
max
=
.
8G R2 max
8G min
h (8G )2
(11.48)
So one recovers a Planck-scale maximal energy density as in the canonical loop

theory.
11.5 Physical predictions?

The possibility of seeing quantum gravitational effects in the CMB is presently
under intense study. We do not enter here in any details, since this would take
us too far into cosmology and the subject is rapidly evolving. We only indicate a
recent result, from which the interest reader can follow the field. To study quantum
gravitational effects on the CMB one needs to study both the effects of the modified
dynamics of the scale factor and that of the quantum fluctuations of the geometry
themselves. The net effect of these is equivalent to having a dressed smooth
geometry [Ashtekar et al. (2009)]. Using this result, one can repeat the standard
cosmological derivation of the power spectrum of the CMB, but extend it to the
pre inflationary era. A recent result, for instance, is the computation of the power
spectrum in [Agullo et al. (2012)]. We give an example of the results in the Figure
11.3, take from this paper.
Figure 11.3
Ratio of predicted power spectrum with and without quantum gravity effects, as a
function of the mode [Agullo et al. (2012)].
For a detailed review, see for instance [Agullo and Corichi (2013)]. This is a field
that is rapidly growing and is of major importance, in view of the possibility of
testing the theory.
12
Scattering
In the two previous chapters we have discussed applications of loop gravity to
physically relevant, but specific situations, black holes and early cosmology. How
to extract the entire information from the theory systematically, and compare it
with the usual way of doing high energy physics?
In conventional field theory, knowledge of the n point functions
W ( x1 , . . . , x n ) = h0 | ( x n ) . . . ( x1 ) | 0i,
(12.1)
amounts to the complete knowledge of the theory, as emphasised by Arthur Wightman in the fifties [Wightman (1959)]. From these functions we can compute the
scattering amplitudes and everything else. Can we recover the value of all these
functions, from the theory we have defined in this book? This, for instance, would
allow us to compare the theory with the effective perturbative quantum theory
of general relativity, which, although non renormalizable is nevertheless usable at
low energy. More in general, it would connect the abstruse background independent formalism needed for defining quantum gravity in general, with the tools of
quantum field theory that we are used to, from flat space physics.
The answer is yes, the value of the n point functions, can be computed from
the theory we have defined in this book. This require a careful understanding of
how the information about the background around which the n-point functions
are defined is dealt with in the background-independent theory. This is done in
this chapter.
12.1 n-point functions in general covariant theories

The n-point functions (12.1) should not be confused with the transition amplitudes. Transition amplitudes, such as
W [ in , out ] = h out |eiHt | in i
(12.2)
are defined between arbitrary (fixed-time) field configurations . While n-point

functions, such as (12.1), are amplitudes for quanta (excitations) of the field over a
preferred state (the vacuum, |0i).
To illustrate the distinction between these two quantities and clarify the relation
between the two, consider a simple system with one variable q(t) that evolves in
238
239
n-point functions in general covariant theories

time. The variable q(t) plays here the role of the fields ( x ). The transition amplitude is
0
W (q, t, q0 , t0 ) = hq0 | eiH (t t) | qi;
(12.3)
W (t, t0 ) = h0 | q(t0 ) q(t) | 0i;
(12.4)
this determines the probability of going from |qi to |q0 i in a time t0 t. The 2-point
function, on the other hand, is
it is a function that depends only on t and t0 and not on q and q0 . It is the transition
amplitude between the states q(t)|0i and q(t0 )|0i, where q(t) are (Heisenberg) operators at different times. The transition amplitude carries information about the
fields, while the n-point function carries information about the quanta of the fields
on a given state. The transition amplitude is the probability to see the system in
a configuration q0 (t0 ) if we have seen it in the configuration q(t). If the system is
in the vacuum state and we create a quantum at time t, then the 2-point function
gives the probability to see a quantum at a later time t0 .
The relation between these two quantities is as follows. Write the Heisenberg
operators in terms of Schrodinger

operators in the expression for the 2-point function
0
W (t, t0 ) = h0 | q(t0 ) q(t) | 0i = h0 | e+iHt qeiHt e+iHt qeiHt | 0i
(12.5)
The vacuum state is invariant under time evolution. So we can write

0
W (t, t0 ) = h0 | q eiH (t t) q | 0i
=
=
dq
dq
dq0 h 0 | q0 i q0 W (q, t, q0 , t0 ) q h q | 0 i
(12.6)
dq0 W (q, t, q0 , t0 ) qq0 o (q) o (q0 )
(12.7)
where now q and q0 are just numbers. The quantity W (q, t, q0 , t0 ) is the transition
amplitude: it does not carry information about any particular state. It characterizes the full dynamics of the theory. On the other hand, W (t, t0 ) characterizes the
dynamics of the theory in the particular state o . The n-point functions can be
written in terms of the transition amplitude, the field operators and a state o , to
be specified.
There is another way to write a two-point function, in terms of a path integral:
W (t, t0 ) =
Dq q(t)q(t0 )eiS[q]
(12.8)
In this expression the state is implicitly fixed by the boundary conditions at infinity
in the path integral. The relation between this formula and the previous one is
interesting. The path integral is (the limit of) a multiple integral in variables q(t).
Lets partition these in five groups: the single variables q(t) and q(t0 ) for fixed
values of t and t0 , and the three classes of times in which t and t0 partition the real
line.
Scattering
240
t0
Then clearly (12.8) reduces to (12.7) where the transition amplitude is the result of
the integration in the region [t, t0 ]
0
W (q, t, q , t ) =
Dq ei
R t0
t
(12.9)
and the two states are the result of the integrations in the past and future regions
o (q) =
Dq ei
Rt
(12.10)
with appropriate boundary conditions at infinity.

Let us now promote these same considerations to field theory. Let us fix an (increasing) time ordering for the points xn , so we do not have to worry here about
T-products. The n-point function can also be written as a path integral.
W ( x1 , . . . , x n ) = h0 | ( x n ), . . . , ( x1 ) | 0i =
D ( xn )...( x1 ) eiS[]
(12.11)
For simplicity, let us first restrict to a two-point function. We can still break the
integration into the five regions as above, obtaining
W ( x, x 0 ) = h0 | ( x 0 )( x ) | 0i =
DD0 W [, t, 0 , t0 ] (~x ) 0 (~x 0 ) o [0 ]o [],
(12.12)
where now spacetime splits into
t
five regions and the transition
0
0
amplitude W [, t, 0 , t0 ] is the field
t
x
propagator in the intermediate
band, obtained by integrating on
t
x
the fields in that band. Observe
x
that in order to perform this integration we have to fix the boundary values of the fields at the initial slice, at the final one, but also at spatial infinity.
Once this is understood, it is natural to consider also a different possibility: to
split (12.11) differently. Instead of selecting a band bounded by two equal time
surfaces, let us select an arbitrary compact region R, as in the figure below.
Then we can write
Wb ( x, x 0 ) =
Db W [ b ] (~x ) (~x 0 ) b [ b ]
t
t
x0
(12.13)
t
x
where b are fields on the boundary of
x
the region, W [b ] is a functional integral that depends only on the boundary
fields, (~x ) and (~x 0 ) are field operator in two distinct points on the boundary,
241
n-point functions in general covariant theories

and b [ b ] is the state of the field on the boundary that describes the integration
outside the region.
In this manner we have found a connection with the boundary formalism of
Section (2.4.2). The n point functions can be written using the boundary formalism:
the background state determines the boundary state, the field operators sit on the
boundary.
Now lets return to quantum gravity. In covariant loop quantum gravity we
have all the ingredients needed to write the last equation. We have the transition amplitude for a field on a boundary, the field operator, and a notion of coherent states that can describe the quantum gravitational field on the boundary
of a region corresponding to a given classical configuration. By combining these
ingredients, we can reconstruct n-point functions over a background, from the
background-independent quantum gravity theory. The background enters in determining the boundary state.
The region considered not need to be a square, in fact, we can take it to be a
region bounded by space like hyper-surfaces, as below.
x0
x
The crucial observation, now, was already made in (2.4.2): in a background dependent theory, in addition to the boundary value of the field, we must also fix
the shape of the boundary regions, its geometry, its size, the time lapsed from the
beginning to the end of the process, and so on, in order to define the boundary
amplitude
W [b ] =
D ei
L[]
(12.14)
But in quantum gravity, due to general covariance, this expression does not depend on the shape or dimension of R. Assigning the state b [ ] of the boundary
field, indeed, amounts to giving the geometry of the boundary, including its shape its
size, the time lapsed from the beginning to the end of the process, and so on!
This is the true magic of quantum gravity: background dependence disappears
entirely, and the transition amplitude is only a function of the field: there is nothing
else in the universe to be considered besides the dynamical fields in the process.
The position of the n points arguments of the n-point function is well defined
with respect to the background metric around which the boundary state is peaked.
Thus, pick for instance a region R of Minkowski space, with a given metric
geometry, and approximate it with a triangulation sufficiently fine to capture the
relevant dynamical scale of the phenomenon to be studied. Determine the extrinsic
Scattering
242
and extrinsic geometry of the boundary of this region, and pick a quantum state
H of gravity picked on these data. Then the two points ~x and ~x 0 sit on nodes
of the boundary graph. The operators ~E` associated to these nodes and directions
determined by the links describe quanta over the boundary state.
This is the solution of a longstanding confusion in quantum gravity: formally,
if S[ g] is a general covariant action and D [ g] a generally covariant measure, the
quantity
W ( x1 , ..., xn ) =
D [ g] g( x1 )...g( xn ) eiS[ g]
(12.15)
is independent from the position of the points ( x1 , ..., xn ) (as long as they do not
overlap), because general covariance can move the points around leaving everything else invariant. Therefore (12.15) is a wrong definition of n-point functions
in a generally covariant theory. It has nothing to do with the quantity (12.1) computed expanding on a background. The mistake is that the nn in (12.1) are physical
distances, computed in the background metric, while the nn in (12.15) are coordinate
distances. To have a quantity depending on physical distances, we must explicitly
specify the background state on a boundary surface, and have the nn s defined by
positions in this metric.
In more formal terms, (12.13) is nontrivial because both the positions of the
points and b transform together under a diffeomorphism.
12.2 Graviton propagator

Let us focus on the 2-point function, which in perturbation theory is defined by
Wabcd ( x, y) = h0 | gab ( x ) gcd (y) | 0i
(12.16)
W ( x, y) = h0 | gab ( x ) gcd (y) | 0i n a ( x )mb ( x ) nc (y)md (y)
(12.17)
where gab ( x ) is the metric field. It is convenient to contract this with two couples
of vectors at x and y, which we shall choose appropriately in a moment
We fix a compact metric region M in the background metric, having x and y on its
boundary = M. Then has an intrinsic and an extrinsic geometry associated
~ tangent to . Passing to triads on we can write
to it. We choose vectors ~n and m
W ( x, y) = h0 | ~Ea ( x ) ~Eb ( x ) ~Ec (y) ~Ed (y) | 0i n a ( x )mb ( x ) nc (y)md (y).
(12.18)
W ( x, y) = hW | ~Ea ( x ) ~Eb ( x ) ~Ec (y) ~Ed (y) | i n a ( x )mb ( x ) nc (y)md (y).
(12.19)
and translate this expression in one based on M
where now is a state on a surface with the topology of a three sphere, peaked on
the geometry of . This state bring the background information with it.
To compute the last expression we resort to approximations. To this aim, we replace M with a triangulated space . The finer the triangulation , the best will be
243
Graviton propagator
the approximation. The triangulation induces a triangulation on the boundary
and we search the state in the spin network Hilbert space H where is the
graph dual to the boundary triangulation. We choose as the extrinsic coherent
state (8.126) determined by the (discretised) intrinsic and extrinsic geometry of .
We identify the points x and y as nodes in the graph, or thetrahedra in the boundary of the triangulation, and fix the vectors n and m to be normal to the faces of
the corresponding tetrahedra.
Then (12.19) becomes a well defined expression: the boundary state is the
state (8.126) chosen as mentioned, the operators ~Eb ( x )nb ( x ) are the triad operators
(7.56) associated to the corresponding links of the spin network. The transition
amplitude W is the amplitude (7.57), associated to the two-complex dual to the
bulk triangulation.
At the lowest nontrivial approximation we can take the triangulation formed by
a single four-simplex. Since we are expanding around flat space, this is reasonable,
as there is no background curvature. That is, we are in the right regime for this expansion to have a chance to work. Then the boundary graph is the regular graph
formed by five connected tetrahedra and W is directly given by the vertex amplitude (7.58). Thus, using equations (8.126), (7.56) and (7.58), the two point function
(12.19) becomes a completely explicit expression. Computing it is just a matter of
technical ability.
The computation for arbitrary boundary size is hard, but it simplifies in the limit
where the boundary geometry of is large with respect to Planck scale. In this case
it is possible to use the techniques developed in [Barrett et al. (2010)] replacing the
vertex amplitude (7.58) with its saddle point approximation (8.103)-(8.104).
Let us see in a bit more detail, following [Bianchi and Ding (2012)]. The expectation value of an operator is given in (7.60). The operator we are here interested in
is the (density-two inverse-) metric operator q ab ( x ) = ij Eia ( x ) Ebj ( x ). We focus on
the connected two-point correlation function G abcd ( x, y) on a semiclassical boundary state |0 i. It is defined as
G abcd ( x, y) = hq ab ( x ) qcd (y)i hq ab ( x )i hqcd (y)i .
(12.20)
Expressing it in terms of triads contracted with the normals to the tetradra faces,
this reads
c
d
c
d
abcd
i h Ena Enb i h Em
Em
i,
Em
Gnm
= h Ena Enb Em
(12.21)
The first problem is to construct the correct boundary state. This is a coherent
Scattering
244
state on the graph
5 =
(12.22)
with five nodes v a dual to tetrahedra of the 4-simplex and ten links lab , ( a < b)
dual to the corresponding meeting triangles. In constructing it, we should take
care to keep track of the orientations of the time orientation of the tetrahedra and
the Thin/Thick nature of the links discussed in Section 8.1.3. See [Bianchi and Ding
(2012)]. This allows us to define the intrinsic coherent states
|5 , jab , ~n ab i = exp (i
ab jab )|5 , jab , ~nab i.
(12.23)
a<b
that depend on a chosen temporal orientation of the boundary tetrahedra, on the

spin of the links and the normals to the faces of the tetrahedra.
A Lorentzian extrinsic coherent state peaked both on intrinsic and extrinsic
geometry can be given by a superposition of Lorentzian coherent spin network
states. Instead of using the full machinery of coherent states, it is simpler to just
choose a wave packet
|o i = jo ,o ( j)| j, ~ni ,
(12.24)
jab
with coefficients jo ,o ( j) given by a gaussian times a phase,

jo ,o ( j) = exp i oab ( jab jo ) exp
jab jo jcd jo
p
p
.
jo
jo
ab
ab,cd
(12.25)
where o labels the simplicial extrinsic curvature, which is an angle associated to
the triangle shared by the tetrahedra; the 10 10 matrix (ab)(cd) is assumed to be
complex with positive definite real part [Bianchi et al. (2009)]. The quantity jo is
an external input in the calculation and sets the scale of the background geometry
of the boundary, on which the boundary state is fixed. It therefore establishes the
distance between the two points of the the 2-point function we are computing.
We can now insert the state (12.24), the form of the vertex amplitude given in
(8.94) and (8.92) in the expectation values (7.60) for the 2-point function (12.21).
(ab)(cd)
Graviton propagator
245
This gives
abcd
Gnm
=
c Ed | j, ~
c Ed | j, ~
ni j j hW | Ena Enb | j, ~ni j j hW | Em
ni
j j hW | Ena Enb Em
m
m
.
j j hW | j, ~ni
j j hW | j, ~ni
j j hW | j, ~ni
(12.26)
c Ed | j, (~
The matrix elements hW | Ena Enb | j, (~n)i and hW | Ena Enb Em
n)i can be
m
given an explicit integral expressions following [Bianchi et al. (2009)], by introduction an insertion
As in [Bianchi et al. (2009)], one can introduce an insertion
1
a i
Qiab h jab , ~n ab ( )| Y g
nba ( )i,
a gb Y ( Eb ) | jab , ~
(12.27)
and expressing them in terms of this insertion. With some work [Bianchi and Ding
(2012)], we arrive at the expression
R 4 10
R 4 10
R 4 10
S
S
j j d g d z qnab eS j j d g d z qcd
j j d g d z qnab qcd
me
me
abcd
,
Gnm =
R 4 10
R 4 10
R 4 10
j j d g d z eS
j j d g d z eS
j j d g d z eS
(12.28)
where S is given in (8.92),
qnab Ana Abn .
(12.29)
and
Aiab jab
hi Zba , ba i
.
h Zba , ba i
(12.30)
The low-energy behaviour of the 2-point function is determined by the large-jo

asympototics of the correlation function (12.28) because jo is the parameter in the
boundary state which determine the distance between the two points in the background metric. This can be done [Bianchi et al. (2009)] rescaling the spins jab and
( jo ) ab by an integer so that jab jab and ( jo ) ab ( jo ) ab , thus the two-point
function (12.28) can be reexpressed as G (). To study large-spin limit turns then to
study large- limit, via stationary phase approximation. The stationary phase approximation of the integral has been studied in detail and we refer to the literature
[Alesci and Rovelli (2008); Bianchi et al. (2009); Bianchi and Ding (2012); Magliaro
and Perini (2012); Rovelli and Zhang (2011)], and in particular to [Bianchi and
Ding (2012)]. The result is that in the classical limit, introduced in [Bianchi et al.
(2009)], where the Barbero-Immirzi parameter is taken to zero 0, and the spin
of the boundary state is taken to infinity j , keeping the size of the quantum geometry A j finite and fixed (The finite and fixed area A corresponds to
the finite and fixed distance between the two points where the correlation functions are defined), the two-point function exactly matches the one obtained from
Lorentzian Regge calculus [Regge (1961)]. In turn, in this limit this matches that of
general relativity (12.16).
Therefore the semiclassical limit of the theory gives back usual linearised gravity, indicating that the theory is General Relativity in an appropriate limit.
246
Scattering
It is important also to go beyond first rider in the vertex expansion. The asymptotic analysis of spinfoams with an arbitrary number of vertices is studied in 0
limit in [Magliaro and Perini (2011b,a, 2013)]. Without taking 0, the large-j of
spinfoams with an arbitrary number of vertices is studied in [Conrady and Freidel
(2008)] with a closed manifold. An extensive discussion and analysis is given by
Muxin Han in [Han (2013)].
Some aspect of this result are not fully clear. Among these the peculiar limit
in which the classical result emerge. The same limit was considered in [Bojowald
(2001)] in the context of loop quantum cosmology, and its reason is probably related to the relation (8.107) illustrated in Figure 8.8. But the situation is not completely clear yet. Also, the current calculation is involved. One has the impression
that a simpler version of it would exist, especially since most of the complications
lead eventually to terms that disappear.
The quantum corrections are contained in higher orders. At the time of writing, only preliminary calculations of radiative corrections with bubbles have been
obtained [Riello (2013)]. This is an extremely important open direction of investigation in the theory. A number of questions regarding for instance the proper normalisation of the n point functions and similar, are still unclear, and much more
work is needed.
Ideally, one would expect that the low energy behaviour of the n point functions
computed in the non perturbative theory would agree with the one computed
using the non renormalizable perturbative expansion, while at higher orders, the
calculations in the non perturbative theory would amount to fixing all infinite free
parameters of the non renormalizable theory. Whether this is the case is still an
open question.
The true importance of these results, on the other hand, is in the fact that they
indicate the manner to compute background dependent quantities from the non
perturbative theory, a problem that has long been a source of great confusion in
quantum gravity.
13
Final remarks
We close with a short historical note on the development of the theory presented,
and a few considerations on the main problems that remain open.
13.1 Brief historical note

The quantum theory of gravity presented in this book is the result of a long path
to which many have participated. We collect here a few notes about this path and
a few references for the orientation of the reader. A comprehensive bibliography
would be impossible.
The basic ideas are still those of John Wheeler [Wheeler (1968)] and Bryce DeWitt [DeWitt (1967)] for what concerns the state space, and those of Charles MisnerMisner (1957), Stephen Hawking [Hawking (1980)], Jonathan Halliwell [Halliwell and Hawking (1985)] and Jim Hartle [Hartle and Hawking (1983)] for what
concerns the sum over geometries. The theory would not have developed without
Abhay Ashtekars [Ashtekar (1986)] introduction of his variables and grew out
of the loop space representation of quantum general relativity, which was conceived in the late eighties [Rovelli and Smolin (1988, 1990); Ashtekar et al. (1991)].
Graphs were introduced by Jurek Lewandowski in [Lewandowski (1994)], spin
networks and discretisation emerged from loop quantum gravity in the nineties
[Rovelli and Smolin (1995b,a); Ashtekar and Lewandowski (1997)]. The geometrical picture used here derives from [Bianchi et al. (2011b)]. The idea of using
spinfoams for quantum gravity derives from the work of Hirosi Ooguri [Ooguri
(1992)], who generalised to 4d the Boulatov model [Boulatov (1992)], which in turn
is a generalisation to 3d of the 2d matrix models. Ooguri theory in 4d is not general
relativity, but has a structure which matches remarkably well that of loop quantum
gravity and therefore it appeared natural to try to import his technique into gravity
[Rovelli (1993b)], yielding the spinfoam ideas [Reisenberger and Rovelli (1997)].
The problem of adapting the Ooguri model to gravity by suitable constraining it
was discussed long in the nineties and finally brilliantly solved by John Barrett
and Louis Crane [Barrett and Crane (1998, 2000)], who devised the first spinfoam
model for 4d quantum gravity. The model prompted the start of the group field
theory technology [De Pietri et al. (2000)], but it turned out to have problems,
which became clear when it was used to compute the graviton two-point function,
which did not turn out correct [Alesci and Rovelli (2007)]. The effort to correct it
lead to the model presented here, which was developed in a joint effort involving
247
248
Final remarks
numerous people [Engle et al. (2007, 2008a); Freidel and Krasnov (2008); Livine
and Speziale (2007)]. Its euclidean version was found independently and simultaneously by two groups. The Lorentzian version appeared in [Engle et al. (2008b)].
The extension of the theory to arbitrary complexes was rapidly developed by the
Warsaw group [Kaminski et al. (2010a,b)].
The theory was puzzling and seen with suspicion until John Barrett and his collaborators succeeded in the tour de force of computing the semiclassical approximation of the amplitude [Barrett et al. (2010)] and this turned out to match the
Regge theory. Shortly after the version with cosmological constant was introduced
[Han (2010); Fairbairn and Meusburger (2012)] building on the mathematical work
in [Noui and Roche (2003)], and the theory proved finite. The coherent states techniques have been developed independently by many people [Thiemann (2006);
Livine and Speziale (2007); Bianchi et al. (2010a); Freidel and Speziale (2010b)]. The
importance of bubbles for the study if divergences was undressood early [Perez
and Rovelli (2001)], but radiative calculations started only recently [Perini et al.
(2009); Riello (2013)]. Fermion coupling has been found in [Bianchi et al. (2010b)].
Spinfoam cosmology was introduced in [Bianchi et al. (2010e)] and extended to the
case with many nodes and many links in [Vidotto (2011)]. The application of spinfoams to black holes was developed in [Bianchi (2012a)]. The technique for computing n-point functions in a background independent context was introduced in
[Rovelli (2006)] using the boundary formalism developed by Robert Oeckl [Oeckl
(2003)]. The Lorentzian two point function has been computed in [Bianchi and
Ding (2012)]. n-point functions for higher n have been computed in [Rovelli and
Zhang (2011)]. The asymptotic of the theory has been studied recently in [Han and
Zhang (2013)] and in [Han (2013)].
This is far from being a complete bibliography of loop quantum gravity, and we
sincerely apologise to the many scientists that have contributed importantly to the
theory and we have not mentioned here or elsewhere in the book; there are many
of them.
13.2 What is missing

The main open questions in the theory are two. The first regards its consistency,
the second the possibility of making predictions. In addition, the theory still needs
to be further developed, or perhaps modified, to become complete.
Consistency
1. With the cosmological constant, the transition amplitudes are finite at all orders
and the classical limit of each converges to the truncation of classical limit of
GR on a finite discretization of spacetime; in turn, these converge to classical GR
when the discretization is refined. This gives a coherent approximation scheme.
249
What is missing
However the approximation scheme may go wrong if the quantum part of the
corrections that one obtains refining the discretization is large. These can be
called radiative corrections, since they are somewhat analog to standard QFT
radiative corrections: possibly large quantum effects effects that appear taking
the next order in an approximation. It is not sufficient for these radiative corrections to be finite, for the approximation to be viable: they must also be small.
Since the theory includes a large number: the ratio of the cosmological constant
scale over the Planck scale (or over the observation scale), these radiative corrections a priori could be large. Understanding whether these are under control
(in suitable physical regimes) is the key open problem on which the consistency
of theory hinges today.
Since large contributions to amplitudes come from large spins and these are
cut off by the cosmological constant, these potentially dangerous large numbers
can be studied as divergences in the = 0 theory. This is the path taken by Aldo
Riello in [Riello (2013)], where some bubble divergences have been computed.
The results are so far reassuring, as the cut off comes in only logarithmically in
the divergent terms, suggesting at worse a logarithm of the cosmological constant to enter the amplitudes. But a general analysis is needed, and the problem
is fully open.
It is then essential to study whether other bubble divergences appear in the
= 0 theory, besides those considered in [Riello (2013)].
2. Can a general result about these divergences be obtained?
3. A separate question is given by the contribution of very fine discretizations.
Does something non trivial happen to theory in approaching the continuum
limit? Are there phase transitions? Although this seems unlikely at present, this
is a possibility. A large amount of work is being developed for studying the
continuum limit is model theories that share similarities with quantum gravity
see for instance [Bonzom and Livine (2011)], [Dittrich et al. (2013)] and references therein. The results of these works may be of great relevance for getting
information on the continuum limit of quantum gravity.
Completing the theory

1. The matter sector of the theory has not been sufficiently developed. Can we
obtain the Dirac propagator from the vertex?
2. The q-deformed version of the theory has been sketched [Han (2011a,b); Fairbairn and Meusburger (2012)] but it is very little developed [Ding and Han
(2011)]. It deserves to be studied and better understood. In particular, can we
reobtain in this theory the results obtained in the = 0 theory?
250
Final remarks
Modifying the theory?

1. The two versions of the Y map with p = j and p = ( j + 1) are equally viable
and interesting for the moment. Is there something that selects one of the two
theories?
2. The possibility of expressing the theory as a group field theory has been studied in [Ben Geloun et al. (2010)]. While transition amplitudes on individual discretizations are unaffected by this, this formulation leads to a different definition of the continuum limit, where contributions of increasingly refined discretizations are summed. This is a possible alternative definition of the theory,
certainly of interest. In this context, a possible alternative to the formulation
given here is also the theory defined by Daniele Oriti and Aristide Baratin in
[Baratin et al. (2010)].
Physical predictions
1. The most likely window towards observational support to the theory seems at
present to come from cosmology. While canonical LQC is well developed, the
covariant theory is only at its infancy [Bianchi et al. (2010e); Vidotto (2011); Rennert and Sloan (2013)] and much needs to be done to bring it to level. Computing cosmological transition amplitudes to next relevant orders in the Lorentzian
theory and for less trivial two complexes is a necessary step. The recent results in writing cosmological states obtained in [Alesci and Cianfrani (2013)]
by Alesci and Cianfrani could provide useful tools for this. Comparison with
canonical Loop Quantum Cosmology [Ashtekar (2009); Bojowald (2005)] is also
to be done.
2. The idea that propagation can be affected by quantum gravity, considered and
then partially discarded some years ago because of the proof that the theory
is Lorentz invariant, is not necessarily wrong, as Lorentz invariant corrections
to the propagator are possible. This is also a direction to explore. For this, one
needs to study higher order the corrections to the graviton propagator computed in [Bianchi et al. (2009); Bianchi and Ding (2012)]. Is there a Planck correction to the form of the classical propagator?
3. As mentioned Riello results in [Riello (2013)] indicate that radiative corrections
are proportional to the logarithm of the cosmological constant. Can this be used
to derive some physics? The scaling of G or ? The vacuum energy renormalization?
4. A recent suggestion [Rovelli and Vidotto (2014)] is that a star that collapses
gravitationally can reach a further stage of its life, where quantum-gravitational
pressure counteracts weight, as it happens in loop cosmology. This is called a
Planck star. The duration of this stage is very short in the star proper time,
yielding a bounce, but extremely long seen from the outside, because of the
huge gravitational time dilation. Loop quantum gravity indicates that the onset
of quantum-gravitational effects is governed by energy density, or by accelera-
What is missing
251
tion not by size therefore the star can be much larger than planckian in this
phase. The object emerging at the end of the Hawking evaporation of a black
hole can can then be larger than planckian by a factor (m/m P )n , where m is
the mass fallen into the hole, m P is the Planck mass, and n is positive. These objects could have astrophysical and cosmological interest: primordial black holes
ending their evaporation now could produce a detectable signal, of quantum
gravitational origin. The large factor amplifying quantum gravitational effects
is here the ratio of the Hubble time tH (the life of the primordial black hole) to
the Planck time tPl , giving a wavelength [Rovelli and Vidotto (2014)]
r
tH
L
1014 cm
(13.1)
3
tPl Planck
which is well within the window of currently detectable signals. Is this suggestion viable? Can the dynamics of a Planck star be studied with the loop theory?
There is much yet to do in quantum gravity. Waiting for some empirical support, we are not sure the track is the good one. But the sky is much more clear than
just a few years ago, in quantum gravity. So, here is the final
Exercise:
Show that the theory defined in this book is fully consistent; if it is not, correct it.
Find observable and testable consequences.
Using these, test the theory.
If the theory is confirmed, we will feel good, and feeling good is good enough ...
.
References
Adler, S L. 1982. Einstein Gravity as a Symmetry Breaking Effect in Quantum Field

Theory. Rev. Mod. Phys., 54, 729.
Agullo, Ivan, and Corichi, Alejandro. 2013. Loop Quantum Cosmology. grqc(Feb.), 52, arXiv:1302.3833.
Agullo, Ivan, Ashtekar, Abhay, and Nelson, William. 2012. A Quantum Gravity Extension of the Inflationary Scenario. Phys. Rev. Lett., 109, 251301,
arXiv:1209.1609.
Alesci, Emanuele, and Cianfrani, Francesco. 2013. Quantum-Reduced Loop Gravity: Cosmology. Phys. Rev., D87, 83521, arXiv:1301.2245.
Alesci, Emanuele, and Rovelli, Carlo. 2007. Complete LQG propagator: Difficulties
with the Barrett-Crane vertex. Physical Review D, 76(10), 104012, arXiv:0708.0883.
Alesci, Emanuele, and Rovelli, Carlo. 2008. The complete LQG propagator: II.
Asymptotic behavior of the vertex. Phys. Rev., D77, 44024, arXiv:0711.1284.
Alexandrov, Sergei. 2010. New vertices and canonical quantization. Phys. Rev.,
D82(July), 24024, arXiv:1004.2260.
Aquilanti, Vincenzo, Haggard, Hal M, Hedeman, Austin, Jeevanjee, Nadir, Littlejohn, Robert G, and Yu, Liang. 2010. Semiclassical Mechanics of the Wigner
6j-Symbol. arXiv:1009.2811.
Arkani-Hamed, Nima, Dubovsky, Sergei, Nicolis, Alberto, Trincherini, Enrico, and
Villadoro, Giovanni. 2007. A Measure of de Sitter entropy and eternal inflation.
JHEP, 0705, 55, arXiv:0704.1814.
Arnold, V I. 1989. Mathematical methods of classical mechanics. Springer Verlag.
Arnowitt, Richard L, Deser, Stanley, and Misner, Charles W. 1962. The dynamics
of general relativity. arXiv:0405109 [gr-qc].
Ashtekar, A. Lectures on nonperturbative canonical gravity.
Ashtekar, A. 1986. New Variables for Classical and Quantum Gravity. Phys. Rev.
Lett., 57, 22442247.
Ashtekar, A, and Petkov, V. 2013. Springer Handbook of Spacetime. Springer.
Ashtekar, A, Baez, J, Corichi, A, and Krasnov, Kirill. 1998. Quantum geometry and
black hole entropy. Phys. Rev. Lett., 80, 904907, arXiv:9710007 [gr-qc].
Ashtekar, Abhay. 2009. Loop Quantum Cosmology: An Overview. Gen. Rel. Grav.,
41, 707741, arXiv:0812.0177.
Ashtekar, Abhay. 2011. Introduction to Loop Quantum Gravity. PoS, QGQGS2011,
1, arXiv:1201.4598.
252
253
References
Ashtekar, Abhay, and Lewandowski, Jerzy. 1995. Projective techniques and functional integration for gauge theories. J. Math. Phys., 36, 21702191, arXiv:9411046
[gr-qc].
Ashtekar, Abhay, and Lewandowski, Jerzy. 1997. Quantum theory of geometry. I:
Area operators. Class. Quant. Grav., 14, A55A82, arXiv:9602046 [gr-qc].
Ashtekar, Abhay, and Lewandowski, Jerzy. 1998. Quantum theory of geometry. II:
Volume operators. Adv. Theor. Math. Phys., 1, 388429, arXiv:9711031 [gr-qc].
Ashtekar, Abhay, and Lewandowski, Jerzy. 2004. Background independent quantum gravity: A status report. Class. Quant. Grav., 21, R53, arXiv:0404018 [gr-qc].
Ashtekar, Abhay, and Singh, Parampreet. 2011. Loop Quantum Cosmology: A
Status Report. Class.Quant.Grav., 28, 213001, arXiv:1108.0893.
Ashtekar, Abhay, Rovelli, Carlo, and Smolin, Lee. 1991. Gravitons and loops. Phys.
Rev., D44, 17401755, arXiv:9202054 [hep-th].
Ashtekar, Abhay, Rovelli, Carlo, and Smolin, Lee. 1992. Weaving a classical geometry with quantum threads. Phys. Rev. Lett., 69(Mar.), 237240, arXiv:9203079
[hep-th].
Ashtekar, Abhay, Engle, Jonathan, and Van Den Broeck, Chris. 2005. Quantum
horizons and black hole entropy: Inclusion of distortion and rotation. Class.
Quant. Grav., 22, L27, arXiv:0412003 [gr-qc].
Ashtekar, Abhay, Kaminski, Wojciech, and Lewandowski, Jerzy. 2009. Quantum
field theory on a cosmological, quantum spacetime. Phys. Rev., D79, 64030,
arXiv:0901.0933 [gr-qc].
ATLAS Collaboration. 2012. Observation of a new particle in the search for the
Standard Model Higgs boson with the ATLAS detector at the LHC. Phys. Lett.
B, 716(1), 129.
Baez, J C. 1994a. Generalized Measures in Gauge Theory. Lett. Math. Phys., 31,
213223.
Baez, J C (ed). 1994b. Knots and Quantum Gravity. Oxford Lecture Series in Mathematics and its Applications, vol. 1. Oxford, U.K., New York, U.S.A.: Clarendon
Press; Oxford University Press.
Baez, J C, and Muniain, J P. 1994. Gauge Fields, Knots, and Gravity. Singapore; River
Edge, U.S.A.: World Scientific Press.
Baez, John C, and Munian, Javier P. 1994. Gauge fields Knots and Gravity. World
Scientific, Singapore.
Bahr, Benjamin, and Thiemann, Thomas. 2009. Gauge-invariant coherent states for
Loop Quantum Gravity II: Non-abelian gauge groups. Class. Quant. Grav., 26,
45012, arXiv:0709.4636.
Bahr, Benjamin, Hellmann, Frank, Kaminski, Wojciech, Kisielowski, Marcin, and
Lewandowski, Jerzy. 2011. Operator Spin Foam Models. Class. Quant. Grav., 28,
105003, arXiv:1010.4787.
Baratin, Aristide, Dittrich, Bianca, Oriti, Daniele, and Tambornino, Johannes.
2010.
Non-commutative flux representation for loop quantum gravity.
arXiv:1004.3450 [hep-th].
254
References
Barbero, J Fernando. 1995. Real Ashtekar variables for Lorentzian signature space
times. Phys. Rev., D51, 55075510, arXiv:9410014 [gr-qc].
Barbieri, Andrea. 1998. Quantum tetrahedra and simplicial spin networks. Nucl.
Phys., B518, 714728, arXiv:9707010 [gr-qc].
Bardeen, James M, Carter, B, and Hawking, S W. 1973. The Four laws of black hole
mechanics. Commun.Math.Phys., 31, 161170.
Barrett, John, Giesel, Kristina, Hellmann, Frank, Jonke, Larisa, Krajewski, Thomas,
and Others. 2011a. Proceedings of 3rd Quantum Geometry and Quantum Gravity School. PoS, QGQGS2011.
Barrett, John W, and Crane, Louis. 1998. Relativistic spin networks and quantum
gravity. J. Math. Phys., 39, 32963302, arXiv:9709028 [gr-qc].
Barrett, John W, and Crane, Louis. 2000. A Lorentzian signature model for quantum general relativity. Class. Quant. Grav., 17, 31013118, arXiv:9904025 [gr-qc].
Barrett, John W, and Naish-Guzman, Ileana. 2009. The Ponzano-Regge model.
Class.Quant.Grav., 26, 155014, arXiv:0803.3319.
Barrett, John W, Dowdall, Richard J, Fairbairn, Winston J, Hellmann, Frank, and
Pereira, Roberto. 2010. Lorentzian spin foam amplitudes: graphical calculus and
asymptotics. Class. Quant. Grav., 27, 165009, arXiv:0907.2440.
Barrett, John W, Dowdall, R J, Fairbairn, Winston J, Hellmann, Frank, and Pereira,
Roberto. 2011b. Asymptotic analysis of Lorentzian spin foam models. PoS,
QGQGS2011, 9.
Battisti, Marco Valerio, Marciano, Antonino, and Rovelli, Carlo. 2010. Triangulated Loop Quantum Cosmology: Bianchi IX and inhomogenous perturbations.
Phys.Rev.D, 81, 64019.
Bekenstein, J D. 1973. Black Holes and Entropy. Phys. Rev. D, 7, 23332346.
Ben Geloun, Joseph, Gurau, Razvan, and Rivasseau, Vincent. 2010. EPRL/FK
Group Field Theory. Europhys. Lett., 92, 60008, arXiv:1008.0354 [hep-th].
Bertone, Gianfranco. 2010. Particle Dark Matter. Cambridge University Press.
Bianchi, Eugenio. 2012a. Entropy of Non-Extremal Black Holes from Loop Gravity.
arXiv:1204.5122.
Bianchi, Eugenio. 2012b. Horizon entanglement entropy and universality of the
graviton coupling. arXiv:1211.0522.
Bianchi, Eugenio, and Ding, You. 2012. Lorentzian spinfoam propagator. Phys.Rev.,
D86, 104040, arXiv:1109.6538.
Bianchi, Eugenio, and Rovelli, Carlo. 2010a. Is dark energy really a mystery? Nature, 466, 321, arXiv:1003.3483.
Bianchi, Eugenio, and Rovelli, Carlo. 2010b. Why all these prejudices against a
constant? Feb., arXiv:1002.3966, http://arxiv.org/abs/1002.3966v3.
Bianchi, Eugenio, and Wieland, Wolfgang. 2012. Horizon energy as the boost
boundary term in general relativity and loop gravity. arXiv:Arxives:1205.5325.
Bianchi, Eugenio, Magliaro, Elena, and Perini, Claudio. 2009. LQG propagator
from the new spin foams. Nucl. Phys., B822, 245269, arXiv:0905.4082.
Bianchi, Eugenio, Magliaro, Elena, and Perini, Claudio. 2010a. Coherent spinnetworks. Phys. Rev., D82, 24012, arXiv:0912.4054.
255
References
Bianchi, Eugenio, Han, Muxin, Magliaro, Elena, Perini, Claudio, and Wieland,
Wolfgang. 2010b. Spinfoam fermions. arXiv:1012.4719.
Bianchi, Eugenio, Magliaro, Elena, and Perini, Claudio. 2010c. Spinfoams in the
holomorphic representation. Phys.Rev.D, 82, 124031.
Bianchi, Eugenio, Magliaro, Elena, and Perini, Claudio. 2010d. Spinfoams in the
holomorphic representation. Phys. Rev., D82, 124031, arXiv:1004.4550 [gr-qc].
Bianchi, Eugenio, Rovelli, Carlo, and Vidotto, Francesca. 2010e. Towards Spinfoam
Cosmology. Phys. Rev., D82, 84035, arXiv:1003.3483.
Bianchi, Eugenio, Krajewski, Thomas, Rovelli, Carlo, and Vidotto, Francesca.
2011a. Cosmological constant in spinfoam cosmology. Phys. Rev., D83, 104015,
arXiv:1101.4049 [gr-qc].
Bianchi, Eugenio, Donà, Pietro, and Speziale, Simone. 2011b. Polyhedra in loop
quantum gravity. Phys. Rev., D83, 44035, arXiv:1009.3402.
Bianchi, Eugenio, Haggard, Hal M., and Rovelli, Carlo. 2013. The boundary is
mixed. June, arXiv:1306.5206.
Bisognano, J J, and Wichmann, E H. 1976. On the Duality Condition for Quantum
Fields. J.Math.Phys., 17, 303321.
Bohr, N, and Rosenfeld, L. 1933. Det Kongelige Danske Videnskabernes Selskabs.
Mathematiks-fysike Meddeleser, 12, 65.
Bojowald, Martin. 2001. The semiclassical limit of loop quantum cosmology. Classical and Quantum Gravity, 18(Sept.), L109L116, arXiv:0105113 [gr-qc].
Bojowald, Martin. 2005. Loop quantum cosmology. Living Rev. Rel., 8, 11,
arXiv:0601085 [gr-qc].
Bojowald, Martin. 2011. Quantum cosmology. Lect.Notes Phys., 835, 1308.
Bonzom, Valentin. 2009. Spin foam models for quantum gravity from lattice path
integrals. Phys.Rev., D80, 64028, arXiv:0905.1501.
Bonzom, Valentin, and Livine, Etera R. 2011. Yet Another Recursion Relation for
the 6j-Symbol. arXiv:1103.3415 [gr-qc].
Borja, Enrique F, Diaz-Polo, Jacobo, Garay, Inaki,

and Livine, Etera R. 2010. Dynamics for a 2-vertex Quantum Gravity Model. Class.Quant.Grav., 27, 235010.
Borja, Enrique F, Garay, Inaki,

and Vidotto, Francesca. 2011. Learning about Quantum Gravity with a Couple of Nodes. SIGMA 8 (2012), 015, 44 pages.
Born, M, and Jordan, P. 1925. Zur Quantenmechanik. Zeitschrift fur
Physik,
34(September), 858888.
Born, M, Jordan, P, and Heisenberg, W. 1926. Zur Quantenmechanik II. Zeitschrift
fur
Physik, 35, 557615.
Boulatov, D V. 1992. A Model of three-dimensional lattice gravity. Mod. Phys. Lett.,
A7, 16291646, arXiv:9202074 [hep-th].
Bozza, V, Feoli, A, Papini, G, and Scarpetta, G. 2000. Maximal acceleration effects
in Reissner-Nordstrom space. Phys. Lett., A271, 3543, arXiv:0005124 [gr-qc].
Bozza, V, Feoli, A, Lambiase, G, Papini, G, and Scarpetta, G. 2001. Maximal acceleration effects in Kerr space. Phys. Lett., A283, 5361, arXiv:0104058 [gr-qc].
Brandt, Howard E. 1989. Maximal proper acceleration and the structure of spacetime. Found. Phys. Lett., 2, 39.
256
References
Bronstein, M P. 1936a. Kvantovanie gravitatsionnykh voln (Quantization of Gravitational Waves). Zh. Eksp. Tear. Fiz., 6, 195.
Bronstein, M P. 1936b. Quantentheorie schwacher Gravitationsfelder. Phys. Z.
Sowjetunion, 9, 140157.
Buffenoir, Eric, and Roche, Philippe. 1999. Harmonic analysis on the quantum
Lorentz group. Commun.Math.Phys., 207(3), 499555, arXiv:9710022 [q-alg].
Caianiello, E R. 1981. Is there a maximal acceleration? Lett. Nuovo Cim., 32, 65.
Caianiello, E R, and Landi, G. 1985. Maximal acceleration and Sakhrovs limiting
temperature. Lett.Nuovo Cim., 42, 70.
Caianiello, E R, Feoli, A, Gasperini, M, and Scarpetta, G. 1990. Quantum corrections to the space-time metric from geometric phase space quantization. Int. J.
Theor. Phys., 29, 131.
Cailleteau, Thomas, Mielczarek, Jakub, Barrau, Aurelien, and Grain, Julien. 2012a.
Anomaly-free scalar perturbations with holonomy corrections in loop quantum
cosmology. Class.Quant.Grav., 29, 95010, arXiv:1111.3535 [gr-qc].
Cailleteau, Thomas, Barrau, Aurelien, Grain, Julien, and Vidotto, Francesca. 2012b.
Consistency of holonomy-corrected scalar, vector and tensor perturbations in
Loop Quantum Cosmology. Phys.Rev., D86, 87301, arXiv:1206.6736.
Cainiello, E R, Gasperini, M, and Scarpetta, G. 1991. Inflation and singularity
prevention in a model for extended-object-dominated cosmology. Classical and
Quantum Gravity, 8(4), 659.
Carlip, Steven, and Teitelboim, Claudio. 1995.
The Off-shell black hole.
Class.Quant.Grav., 12, 16991704, arXiv:9312002 [gr-qc].
Carter, J Scott, Flath, Daniel E, and Saito, Masahico. 1995. The Classical and Quantum
6j-symbol. Princeton University Press.
Chandia, Osvaldo, and Zanelli, Jorge. 1997. Torsional topological invariants (and
their relevance for real life). arXiv:hep-th/9708138 [hep-th].
Chirco, Goffredo, Haggard, Hal M, and Rovelli, Carlo. 2013. Coupling and Equilibrium for General-covariant Systems.
Choquet-Bruhat, Yvonne, and Dewitt-Morette, Cecile. 2000. Analysis, Manifolds and
Physics, Part II. North Holland.
Choquet-Bruhat, Yvonne, and Dewitt-Morette, Cecile. 2004. Analysis, Manifolds and
Physics, Part I. 2 edn. North Holland.
Christodoulou, Marios, Riello, Aldo, and Rovelli, Carlo. 2012. How to detect an
anti-spacetime. Int.J.Mod.Phys., D21, 1242014, arXiv:1206.3903 [gr-qc].
Christodoulou, Marios, Langvik, Miklos, Riello, Aldo, Roken,

Christian, and Rovelli, Carlo. 2013. Divergences and Orientation in Spinfoams. Class. Quantum
Grav., 30, 055009, arXiv:1207.5156.
CMS Collaboration. 2012. Observation of a new boson at a mass of 125 GeV with
the CMS experiment at the LHC. Phys. Lett. B, 716(1), 3061.
Colosi, Daniele, and Rovelli, Carlo. 2003. A simple background-independent
Hamiltonian quantum model. Phys. Rev., D68, 104008, arXiv:0306059 [gr-qc].
Colosi, Daniele, and Rovelli, Carlo. 2009. What is a particle? Class. Quant. Grav.,
26, 25002, arXiv:0409054 [gr-qc].
257
References
Connes, Alain, and Rovelli, Carlo. 1994. Von Neumann algebra automorphisms
and time thermodynamics relation in general covariant quantum theories. Class.
Quant. Grav., 11, 28992918, arXiv:9406019 [gr-qc].
Conrady, Florian, and Freidel, Laurent. 2008. Path integral representation of spin
foam models of 4d gravity. Class. Quant. Grav., 25, 245010, arXiv:0806.4640.
De Pietri, Roberto, Freidel, Laurent, Krasnov, Kirill, and Rovelli, Carlo. 2000.
Barrett-Crane model from a Boulatov-Ooguri field theory over a homogeneous
space. Nucl. Phys., B574, 785806, arXiv:9907154 [hep-th].
DeWitt, Bryce S. 1967. Quantum Theory of Gravity. 1. The Canonical Theory. Phys.
Rev., 160, 11131148.
Ding, You, and Han, Muxin. 2011. On the Asymptotics of Quantum Group Spinfoam Model. Mar., arXiv:1103.1597.
Ding, You, and Rovelli, Carlo. 2010a. Physical boundary Hilbert space and volume
operator in the Lorentzian new spin-foam theory. Class. Quant. Grav., 27, 205003,
arXiv:1006.1294.
Ding, You, and Rovelli, Carlo. 2010b. The volume operator in covariant quantum
gravity. Class. Quant. Grav., 27, 165003, arXiv:0911.0543.
Dirac, P A M. 1950. Generalized Hamiltonian dynamics. Can. J. Math., 2, 129148.
Dirac, P A M. 1956. Electrons and the vacuum. In: Dirac Papers. Florida State
University Libraries, Tallahassee, Florida, USA.
Dirac, P A M. 1958. Generalized Hamiltonian Dynamics. Proc. R. Soc. Lond., 246,
326332.
Dirac, Paul. 2001. Lectures on Quantum Mechanics. Cambridge Monographs on
Mathematical Physics. New York : Belfer Graduate School of Science, Yeshiva
University.: Dover Books on Physics.
Dittrich, Bianca, Martin-Benito, Mercedes, and Steinhaus, Sebastian. 2013. Quantum group spin nets: refinement limit and relation to spin foams. Dec.,
arXiv:1312.0905.
Dona, Pietro, and Speziale, Simone. 2010. Introductory lectures to loop quantum
gravity.
Durr, S, Fodor, Z, Frison, J, Hoelbling, C, Hoffmann, R, and Others. 2008. Ab-Initio
Determination of Light Hadron Masses. Science, 322, 12241227, arXiv:0906.3599
[hep-lat].
Einstein, A. 1905a. On a heuristic viewpoint of the creation and modification of
light. Ann d Phys.
Einstein, Albert. 1905b. Zur Elektrodynamik bewegter Korper.

Annalen der Physik,
17(891).
Einstein, Albert. 1915. Zur Allgemeinen Relativitatstheorie. Preuss. Akad: Wiss.
berlin Sitzber., 778786.
Einstein, Albert. 1916. Grundlage der allgemeinen Relativitatstheorie. Annalen der
Physik, 49, 769822.
Einstein, Albert. 1917. Cosmological Considerations in the General Theory of Relativity. Sitzungsber. Preuss. Akad. Wiss. Berlin (Math. Phys. ), 1917, 142152.
258
References
Engle, Jonathan, and Pereira, Roberto. 2009. Regularization and finiteness of the
Lorentzian LQG vertices. Phys. Rev., D79, 84034, arXiv:0805.4696.
Engle, Jonathan, Pereira, Roberto, and Rovelli, Carlo. 2007. The loop-quantumgravity vertex-amplitude. Phys. Rev. Lett., 99, 161301, arXiv:0705.2388.
Engle, Jonathan, Pereira, Roberto, and Rovelli, Carlo. 2008a. Flipped spinfoam
vertex and loop gravity. Nucl. Phys., B798, 251290, arXiv:0708.1236.
Engle, Jonathan, Livine, Etera, Pereira, Roberto, and Rovelli, Carlo. 2008b.
LQG vertex with finite Immirzi parameter.
Nucl. Phys., B799, 136149,
arXiv:0711.0146.
Enrique F. Borja, Inaki

Garay, and Francesca Vidotto. 2011. Learning about quantum gravity with a couple of nodes. SIGMA, 7.
Fairbairn, Winston J, and Meusburger, Catherine. 2012. Quantum deformation of two four-dimensional spin foam models. J.Math.Phys., 53, 22501,
arXiv:1012.4784 [gr-qc].
Farmelo, Graham. 2009. The Strangest Man. The Hidden Life of Paul Dirac, Quantum
Genius. Faber and Faber Ltd.
Feoli, A, Lambiase, G, Papini, G, and Scarpetta, G. 1999. Schwarzschild field with
maximal acceleration corrections. Phys. Lett.Lett., A263, 147153, arXiv:9912089
[gr-qc].
Feynman, R P. 1948. Space-time approach to nonrelativistic quantum mechanics.
Rev. Mod. Phys., 20, 367387.
Flanders, Harley. 1963. Differential Forms: With Applications to the Physical Sciences.
Dover.
Folland, Gerald B. 1995. A Course in Abstract Harmonic Analysis. CRC Press.
Frankel, Theodore. 2003. The Geometry of Physics: An Introduction. 2 edn. Cambridge University Press.
Freidel, Laurent, and Krasnov, Kirill. 2008. A New Spin Foam Model for 4d Gravity. Class. Quant. Grav., 25, 125018, arXiv:0708.1595.
Freidel, Laurent, and Speziale, Simone. 2010a. From twistors to twisted geometries. Phys.Rev., D82, 84041, arXiv:1006.0199 [gr-qc].
Freidel, Laurent, and Speziale, Simone. 2010b. Twisted geometries: A geometric
parametrisation of SU(2) phase space. Phys.Rev., D82, 84040, arXiv:1001.2748.
Freidel, Laurent, Krasnov, Kirill, and Livine, Etera R. 2009. Holomorphic Factorization for a Quantum Tetrahedron. arXiv:0905.3627.
Frodden, Ernesto, Ghosh, Amit, and Perez, Alejandro. 2011. Black hole entropy in
LQG: Recent developments. AIP Conf.Proc., 1458, 100115.
Frodden, Ernesto, Geiller, Marc, Noui, Karim, and Perez, Alejandro. 2012. Black
Hole Entropy from complex Ashtekar variables. Dec., 5, arXiv:1212.4060, http:
//arxiv.org/abs/1212.4060.
Gambini, Rodolfo, and Pullin, Jorge. 2010. Introduction to loop quantum gravity.
Oxford University Press.
Gelfand, I M, Minlos, R A, and Shapiro, Z.Ya. 1963. Representations of the rotation
and Lorentz groups and their applications. Pergamon Press.
Gelfand, I.M., and Shilov, G.E. 1968. Generalized functions 1b5. Academic Press.
259
References
Ghosh, Amit, and Perez, Alejandro. 2011. Black hole entropy and isolated horizons
thermodynamics. Phys.Rev.Lett., 107, 241301, arXiv:1107.1320.
Ghosh, Amit, Noui, Karim, and Perez, Alejandro. 2013. Statistics, holography, and
black hole entropy in loop quantum gravity. Sept., arXiv:1309.4563, http://
arxiv.org/abs/1309.4563.
Gibbons, G W, and Hawking, S W. 1977. Action integrals and partition functions
in quantum gravity. Physical Review D, 15(10), 27522756.
Gorelik, Gennady, and Frenkel, Victor. 1994. Matvei Petrovich Bronstein and Soviet
Theoretical Physics in the Thirties. Birkhauser Verlag.
Haag, Rudolf. 1996. Local Quantum Physics: Fields, Particles, Algebras. Springer.
Haggard, Hal M. 2011. Asymptotic Analysis of Spin Networks with Applications to Quantum Gravity. http://bohr.physics.berkeley.edu/hal/pubs/
Thesis/.
Haggard, Hal M, and Rovelli, Carlo. 2013. Death and resurrection of the zeroth principle of thermodynamics. Journal of Modern Physics D, 22, 1342007,
arXiv:1302.0724.
Haggard, Hal M, Rovelli, Carlo, Wieland, Wolfgang, and Vidotto, Francesca. 2013.
The spin connection of twisted geometry. Phys.Rev., D87, 24038, arXiv:1211.2166.
Halliwell, J J, and Hawking, S W. 1985. The Origin of Structure in the Universe.
Phys. Rev., D31, 1777.
Han, Muxin. 2010. 4-dimensional Spin-foam Model with Quantum Lorentz
Group.
Han, Muxin. 2011a. 4-dimensional Spin-foam Model with Quantum Lorentz
Group. J. Math. Phys., 52, 72501, arXiv:1012.4216 [gr-qc].
Han, Muxin. 2011b. Cosmological Constant in LQG Vertex Amplitude. Phys.Rev.,
D84, 64010, arXiv:1105.2212.
Han, Muxin. 2013. Covariant Loop Quantum Gravity, Low Energy Perturbation
Theory, and Einstein Gravity. arXiv:1308.4063.
Han, Muxin. 2014. Black Hole Entropy in Loop Quantum Gravity, Analytic Continuation, and Dual Holography. Feb., 5, arXiv:1402.2084, http://arxiv.org/
abs/1402.2084.
Han, Muxin, and Rovelli, Carlo. 2013. Spin-foam Fermions: PCT Symmetry,
Dirac Determinant, and Correlation Functions. Class.Quant.Grav., 30, 75007,
arXiv:1101.3264.
Han, Muxin, and Zhang, Mingyi. 2013. Asymptotics of Spinfoam Amplitude
on Simplicial Manifold: Lorentzian Theory. Class.Quant.Grav., 30, 165012,
arXiv:1109.0499 [gr-qc].
Hartle, J B, and Hawking, S W. 1983. Wave Function of the Universe. Phys. Rev.,
D28, 29602975.
Hawking, S W. 1974. Black hole explosions? Nature, 248, 3031.
Hawking, S W. 1975. Particle creation by black holes. Commun. Math. Phys., 43,
199220.
Hawking, Stephen W. 1980. The path integral approach to quantum gravity.
260
References
Hawking, Stephen W, and Ellis, George F R. 1973. The large scale structure of spacetime. Cambridge Monographs on Mathematical Physics. Cambridge, U.K.: Cambridge University Press.
Heisenberg, W. 1925. Uber quantentheoretische Umdeutung kinematischer und mechanischer Beziehungen.
Hellmann, Frank, and Kaminski, Wojciech. 2013. Holonomy spin foam models:
Asymptotic geometry of the partition function. July, 63, arXiv:1307.1679.
Hojman, R, Mukku, C, and Sayed, W A. 1980. Parity violation in metric torsion
theories of gravity. Phys.Rev., D22, 19151921.
Hollands, Stefan, and Wald, Robert M. 2008. Quantum field theory in curved
spacetime, the operator product expansion, and dark energy. Gen.Rel.Grav., 40,
20512059, arXiv:0805.3419.
Holst, Soren.
1996. Barberos Hamiltonian derived from a generalized HilbertPalatini action. Phys. Rev. D, 53, 59665969, arXiv:9511026 [gr-qc].
Kaminski, Wojciech. 2010. All 3-edge-connected relativistic BC and EPRL spinnetworks are integrable. arXiv:1010.5384 [gr-qc].
Kaminski, Wojciech, Kisielowski, Marcin, and Lewandowski, Jerzy. 2010a. SpinFoams for All Loop Quantum Gravity.
Class. Quant. Grav., 27, 95006,
arXiv:0909.0939.
Kaminski, Wojciech, Kisielowski, Marcin, and Lewandowski, Jerzy. 2010b. The
EPRL intertwiners and corrected partition function. Class. Quant. Grav., 27,
165020, arXiv:0912.0540.
Kauffman, L H, and Lins, S L. 1994. TemperleyLieb Recoupling Theory and Invariants of 3-Manifolds. Annals of Mathematics Studies, vol. 134. Princeton, U.S.A.:
Princeton University Press.
Kiefer, Claus. 2007. Quantum Gravity (International Series of Monographs on Physics).
2 edn. Oxford University Press, USA.
Kobayashi, Shoshichi, and Nomizu, Katsumi. 1969. Foundations of Differential Geometry, Volume II. John Wiley & Sons.
Kobayashi, Shoshichi, and Nomizu, Katsumi. 1996. Foundations of Differential Geometry, Volume I. Wiley-Interscience.
Kogut, John B, and Susskind, Leonard. 1975. Hamiltonian Formulation of Wilsons
Lattice Gauge Theories. Phys. Rev., D11, 395.
Krajewski, Thomas. 2011.
Group field theories.
PoS, QGQGS2011, 5,
arXiv:1210.6257.
Krajewski, Thomas, Magnen, Jacques, Rivasseau, Vincent, Tanasa, Adrian, and Vitale, Patrizia. 2010. Quantum Corrections in the Group Field Theory Formulation of the EPRL/FK Models. Phys. Rev., D82, 124069, arXiv:1007.3150 [gr-qc].
Landau, L, and Peierls, R. 1931. Erweiterung des Unbestimmheitsprinzips fur die
relativistische Quantentheorie. Zeitschrift fur Physik, 69, 5669.
Landau, L D, and Lifshitz, E M. 1951. The classical theory of fields. Oxford: Pergamon.
Landau, L D, and Lifshitz, E M. 1959. Quantum mechanics. Oxford: Pergamon.
261
References
Lewandowski, J. 1994. Topological Measure and Graph-Differential Geometry on
the Quotient Space of Connections. Int. J. Mod. Phys. D, 3, 207210.
Liberati, Stefano, and Maccione, Luca. 2009. Lorentz Violation: Motivation and
new constraints. Ann. Rev. Nucl. Part. Sci., 59, 245267, arXiv:0906.0681 [astroph.HE].
Littlejohn, Robert. 2013. Lectures in Marseille.
Littlejohn, Robert G. 1992. The Van Vleck Formula, Maslov Theory, and Phase
Space Geometry. Journal of Statistical Physics, 68, 750.
Livine, Etera R, and Speziale, Simone. 2007. A new spinfoam vertex for quantum
gravity. Phys. Rev., D76, 84028, arXiv:0705.0674.
Magliaro, E., and Perini, C. 2011a. Emergence of gravity from spinfoams. EPL
(Europhysics Letters), 95(3), 30007, arXiv:1108.2258.
Magliaro, Elena, and Perini, Claudio. 2011b. Regge gravity from spinfoams.
arXiv:1105.0216.
Magliaro, Elena, and Perini, Claudio. 2012. Local spin foams. Int.J.Mod.Phys., D21,
1250090, arXiv:1010.5227 [gr-qc].
Magliaro, Elena, and Perini, Claudio. 2013. Regge gravity from spinfoams. International Journal of Modern Physics D, 22(Feb.), 1350001, arXiv:1105.0216.
Magliaro, Elena, Marciano, Antonino, and Perini, Claudio. 2011. Coherent states
for FLRW space-times in loop quantum gravity. Phys.Rev.D, 83, 44029.
Misner, Charles W. 1957. Feynman quantization of general relativity. Rev Mod
Phys, 29, 497.
Misner, Charles W, Thorne, Kip, and Wheeler, John A. 1973. Gravitation (Physics
Series). W. H. Freeman.
Mizoguchi, Shunya, and Tada, Tsukasa. 1992. Three-dimensional gravity from the
Turaev-Viro invariant. Phys.Rev.Lett., 68, 17951798, arXiv:9110057 [hep-th].
Morales-Tecotl, H A, and Rovelli, Carlo. 1995. Loop space representation of quantum fermions and gravity. Nucl. Phys., B451, 325361.
Morales-Tecotl, Hugo A, and Rovelli, Carlo. 1994. Fermions in quantum gravity.
Phys. Rev. Lett., 72, 36423645, arXiv:9401011 [gr-qc].
Mukhanov, Viatcheslav F, Feldman, H A, and Brandenberger, Robert H. 1992. Theory of cosmological perturbations. Part 1. Classical perturbations. Part 2. Quantum theory of perturbations. Part 3. Extensions. Phys.Rept., 215, 203333.
Noui, Karim, and Roche, Philippe. 2003. Cosmological deformation of Lorentzian
spin foam models. Class. Quant. Grav., 20, 31753214, arXiv:0211109 [gr-qc].
Oeckl, Robert. 2003. A general boundary formulation for quantum mechanics
and quantum gravity. Phys. Lett., B575, 318324, arXiv:0306025 [hep-th].
Oeckl, Robert. 2008. General boundary quantum field theory: Foundations and
probability interpretation. Adv. Theor. Math. Phys., 12, 319352, arXiv:0509122
[hep-th].
Ooguri, Hirosi. 1992. Topological lattice models in four-dimensions. Mod. Phys.
Lett., A7, 27992810, arXiv:9205090 [hep-th].
Perez, Alejandro. 2012. The Spin Foam Approach to Quantum Gravity. Living
Rev.Rel., 16, 3, arXiv:1205.2019.
262
References
Perez, Alejandro, and Rovelli, Carlo. 2001. A spin foam model without bubble
divergences. Nucl. Phys., B599, 255282, arXiv:0006107 [gr-qc].
Perini, Claudio, Rovelli, Carlo, and Speziale, Simone. 2009. Self-energy and vertex
radiative corrections in LQG. Phys. Lett., B682, 7884, arXiv:0810.1714.
Planck Collaboration. 2013. Planck 2013 results. I. Overview of products and scientific results. arXiv:1303.5062.
Ponzano, G, and Regge, Tullio. 1968. Semiclassical limit of Racah coefficients. In:
Bloch, F (ed), Spectroscopy and group theoretical methods in Physics. Amsterdam:
North-Holland.
Regge, T. 1961. General relativity without coordinates. Nuovo Cim., 19, 558571.
Reisenberger, Michael P, and Rovelli, Carlo. 1997. *Sum over surfaces* form of
loop quantum gravity. Phys. Rev., D56, 34903508, arXiv:9612035 [gr-qc].
Rennert, Julian, and Sloan, David. 2013. Towards Anisotropic Spinfoam Cosmology. arXiv:1304.6688 [gr-qc].
Riello, Aldo. 2013. Self-Energy of the Lorentzian EPRL-FK Spin Foam Model of
Quantum Gravity. Phys. Rev., D88, 24011, arXiv:1302.1781.
Roberts, Justin. 1999. Classical 6j-symbols and the tetrahedron. geometry and Topology, 3, 2166, arXiv:9812013 [math-ph].
Rovelli, Carlo. 1993a. Statistical mechanics of gravity and the thermodynamical
origin of time. Class. Quant. Grav., 10, 15491566.
Rovelli, Carlo. 1993b. The basis of the Ponzano-Regge-Turaev-Viro-Ooguri model
is the loop representation basis. Phys. Rev. D, 48, 27022707, http://arxiv.org/
abs/hep-th/9304164v1.
Rovelli, Carlo. 1993c. The Statistical state of the universe. Class. Quant. Grav., 10,
1567.
Rovelli, Carlo. 1996a. Black Hole Entropy from Loop Quantum Gravity. Physical
Review Letters, 77(16), 32883291, arXiv:9603063 [gr-qc].
Rovelli, Carlo. 1996b. Relational Quantum Mechanics. Int. J. Theor. Phys., 35(9),
1637, arXiv:9609002 [quant-ph].
Rovelli, Carlo. 2004. Quantum Gravity. Cambridge, U.K.: Cambridge University
Press.
Rovelli, Carlo. 2006. Graviton propagator from background-independent quantum gravity. Phys. Rev. Lett., 97, 151301, arXiv:0508124 [gr-qc].
Rovelli, Carlo. 2011. Zakopane lectures on loop gravity. PoS, QGQGS2011, 3,
arXiv:1102.3660.
Rovelli, Carlo. 2013a.
General relativistic statistical mechanics.
arXiv,
D87, 84055, http://arxiv.org/abs/1209.0065papers2://publication/doi/
10.1103/PhysRevD.87.084055.
Rovelli, Carlo. 2013b. Why Gauge? arXiv:1308.5599.
Rovelli, Carlo, and Smerlak, Matteo. 2010. Thermal time and the Tolman-Ehrenfest
effect: temperature as the speed of time.
Rovelli, Carlo, and Smerlak, Matteo. 2012. In quantum gravity, summing is refining. Class.Quant.Grav., 29, 55004, arXiv:1010.5437 [gr-qc].
263
References
Rovelli, Carlo, and Smolin, Lee. 1988. Knot Theory and Quantum Gravity. Phys.
Rev. Lett., 61, 1155.
Rovelli, Carlo, and Smolin, Lee. 1990. Loop Space Representation of Quantum
General Relativity. Nucl. Phys., B331, 80.
Rovelli, Carlo, and Smolin, Lee. 1995a. Discreteness of area and volume in quantum gravity. Nucl. Phys., B442, 593622, arXiv:9411005 [gr-qc].
Rovelli, Carlo, and Smolin, Lee. 1995b. Spin Networks and Quantum Gravity.
arXiv.org, gr-qc, arXiv:9505006 [gr-qc].
Rovelli, Carlo, and Speziale, Simone. 2010. On the geometry of loop quantum
gravity on a graph. Phys. Rev., D82, 44018, arXiv:1005.2927.
Rovelli, Carlo, and Speziale, Simone. 2011. Lorentz covariance of loop quantum
gravity. Phys. Rev., D83, 104029, arXiv:1012.1739.
Rovelli, Carlo, and Vidotto, Francesca. 2008. Stepping out of Homogeneity in Loop
Quantum Cosmology. arXiv:0805.4585.
Rovelli, Carlo, and Vidotto, Francesca. 2010. Single particle in quantum gravity
and BGS entropy of a spin network. Phys.Rev., D81, 44038, arXiv:0905.2983.
Rovelli, Carlo, and Vidotto, Francesca. 2013. Evidence for Maximal Acceleration
and Singularity Resolution in Covariant Loop Quantum Gravity. Physical Review
Letters, 111(9), 091303, arXiv:1307.3228.
Rovelli, Carlo, and Vidotto, Francesca. 2014. Planck stars. arXiv:1401.6562.
Rovelli, Carlo, and Zhang, Mingyi. 2011. Euclidean three-point function in loop
and perturbative gravity. Class. Quant. Grav., 28, 175010, arXiv:1105.0566 [gr-qc].
Ruhl, W. 1970. The Lorentz group and harmonic analysis. New York: W.A. Benjamin,
Inc.
Schrodinger,
Erwin. 1926. Quantisierung als Eigenwertproblem. Annalen der
Physik, 13, 29.
Sen, A. 1982. Gravity as a spin system. Phys. Lett. B, 119, 8991.
Shannon, C E. 1948. A Mathematical Theory of Communication. The Bell System
Technical Journal, XXVII(3), 379.
Singh, Parampreet. 2009. Are loop quantum cosmos never singular? Class. Quant.
Grav., 26, 125005, arXiv:0901.2750.
Singh, Parampreet, and Vidotto, Francesca. 2011. Exotic singularities and spatially
curved Loop Quantum Cosmology. Phys.Rev., D83, 64027, arXiv:1012.1307.
Smolin, Lee. 1994. Fermions and topology. arXiv:9404010 [gr-qc].
Smolin, Lee. 2012. General relativity as the equation of state of spin foam.
arXiv:1205.5529.
Sorkin, R D. 1977. On the Relation Between Charge and Topology. J. Phys. A: Math.
Gen., 10, 717.
Stone, Douglas. 2013. Einstein and the Quantum: The Quest of the Valiant Swabian.
Princeton University Press.
Sundermeyer, Kurt. 1982. Constrained dynamics : with applications to Yang-Mills theory, general relativity, classical spin, dual string model. Springer-Verlag.
Thiemann, T. 2007. Modern Canonical Quantum General Relativity. Cambridge, U.K.:
Cambridge University Press.
264
References
Thiemann, Thomas. 2006. Complexifier coherent states for quantum general relativity. Class. Quant. Grav., 23, 20632118, arXiv:0206037 [gr-qc].
Toller, M. 2003. Geometries of maximal acceleration. arXiv:0312016 [hep-th].
Tolman, Richard C. 1930. On the Weight of Heat and Thermal Equilibrium in
General Relativity. Phys. Rev., 35, 904924.
Tolman, Richard C, and Ehrenfest, Paul. 1930. Temperature Equilibrium in a Static
Gravitational Field. Phys. Rev., 36(12), 17911798.
Turaev, V G, and Viro, O Y. 1992. State sum invariants of 3 manifolds and quantum
6j symbols. Topology, 31, 865902.
Unruh, W G. 1976. Notes on black hole evaporation. Phys.Rev., D14, 870.
Vidotto, Francesca. 2011. Many-nodes/many-links spinfoam: the homogeneous
and isotropic case. Class. Quant Grav., 28(245005), arXiv:1107.2633.
Wald, Robert M. 1984. General relativity. Chicago, USA: Univ. Pr.
Wheeler, J A. 1962. Geometrodynamics. New York: Academic Press.
Wheeler, John A. 1968. Superspace and the nature of quantum geometrodynamics.
In: DeWitt, Bryce S, and Wheeler, John A (eds), Batelles Rencontres. Bennjamin,
New York.
Wieland, Wolfgang. 2012. Complex Ashtekar variables and reality conditions for
Holsts action. Annales Henri Poincare, 13, 425, arXiv:1012.1738 [gr-qc].
Wightman, Arthur. 1959. Quantum field theory in terms of vacuum expectation
values. Physical Review, 101(860).
Wilson, Kenneth G. 1974. Confinement of Quarks. Phys.Rev., D10, 24452459.
York, James W. 1972. Role of conformal three-geometry in the dynamics of gravitation. Physical Review Letters, 28, 1082.
Index
SL(2, C) little group, 166

Why not trying this? physics, 9
2-complex, 149
3d action, 95
3d boundary variables, 96
4-simplex, 137
action 3D, 140
Alexandrov, Sergei, 156
antispacetime, 64
area fluctuations, 219
Area of the horizon, 212
Aristarchus of Samos, 60
Aristotle, 19
Arkani-Hamed, 53
Ashtekar electric field, 74
Ashtekar variables, 22, 228
Ashtekars electric field, 14
Ashtekar, Abhay, vii, 73, 247
Ashtekar-Barbero connection, 228
Atiyah, Michael, 177
Atyah, Michael, 178
background independence, 147
Baez, John, vii
Baratin, Aristide, 250
Barbero-Immirzi constant, 143
Barrett, John, vii, 247
Bekenstein, 150
Bekenstein-Hawking entropy, 208, 211, 218, 219
Berezin integral, 198, 199
Bianchi, Eugenio, vii, 219
Big Bang, 229
Big Bang singularity, 229
Bisognano-Wichmann theorem, 217
bivector, 180
black hole, 207
black hole entropy, 150
black holes at LHC, 4
Bohr, Niels, 6, 23
Boltzmann constant, 208
boost generator, 215
Boulatov model, 247
bounce, 229
Bronstein, Matvei, 7, 23
bubbles, 125
Buffenoir, Eric, 155
Cartan equations, 62
265
Cartan geometry, 62
Cartan, Elie,
62
CERN, 3, 7, 151
classical limit, 189
Clausius, 211
closure condition, 171
closure constraint, 106
closure equation, 171
closure relation, 13
continuum limit, 150
contracting phase, 230
Copernicus, 5, 6
cosmological constant, 123
Crane, Louis, 247
critical density, 229
critical point, 18
dark energy, 4
dark matter, 4
deficit angle, 89, 90
deformation parameter, 133
Descartes, Rene, 19
DeWitt, Bryce, 247
dihedral angle, 171
Dirac action, 196
Dirac equation, 60, 62, 63
Dirac, P.A.M., 20
Dirac, Paul, 69
dual of a triangulation, 138
edges, 138
Einstein, Albert, 5, 6, 209
Einstein-Hilbert action, 59, 62
Einstein-Weyl-Yang-Mills system,, 203
Einstein. Albert, 226
Engle, John, vii
Entanglement entropy, 219
Euler angles, 173
extra dimensions, 3
faces, 138
Fairbairn, Winston, 155
Faraday, Michael, 19
fermion doubling, 197
fermions, 21, 60
Feynman, Richard, 22
first order formulation, 65
flag, 179
Fock space, 151
Index
266
four-fermion interaction, 65
Freidel, Laurent, 178, 195
Friedman-Lematre metric, 226
Friedmann equation, 150, 227, 228
Frodden-Ghosh-Perez energy, 209, 211, 216, 218
fuzziness, 10, 19
Galilei, Galileo, 5
Gambini, Rofolfo, vii
gauge invariance, 61
Gauss constraint, 106
general covariant quantum fields, 20
geometry, 195
Giesel, Kristina, vii
Giordano Bruno, 15
granular structure, 17
graph, 20
group element, 94, 139
group field theory, 152, 247
Haggard, Hal, 133
Haggard, Hall, vii
Halliwell, Jonathan, 247
Hamilton function, 31, 227, 228
hamiltonian constraint, 228
Han, Muxin, 155, 246
Harmonic analysis, 109
Hartle, Jim, 247
Hawking temperature, 211
Hawking, Stephen, 23, 207, 247
Hawkings, Steven, 150
Heisenberg relation, 172
Heisenberg uncertainty, 7, 212
Heisenberg, Werner, 6
Herons formula, 165
Higgs particle, 3
hinge, 88
Hodge dual, 63
holonomy, 21
Holst action, 137
homogeneity and isotropy, 226
horizon, 8
Hubble radius, 60
Hugieia, 149
information, 19
infrared divergences, 123
integral in spinor space, 182
Jacobson, Ted, 214
Kaluza-Klein, 22
Kepler, Johannes, 5
Kiefer, Claus, vii
KMS property, 217
Koguth, John, 84
Landau, Lev, 6, 23
Lapse function, 64, 69, 71, 227
lattice QCD, 22, 152
Levi-Civita symbol, 12
Lewandowski, Jerzy, vii

Lewandowski, Jurek, 247
LHC, 4
line of force, 21
link, 20
locality, 149
loop, 11
loops, 20, 22
Lorentz invariance, 4
Lorentzian angle, 216
Maslov index, 188
Maxwell equations, 9
Maxwell, James, 19
Meusburger, Catherine, 155
minimal length, 19
Misner, Charles, 247
modified Friedman equation, 229
Mukanov equation, 230
Muxin Han, 190
natural units, 19
naturalness, 60
Newton, Isaac, 5, 19
non-renormalizability, 22
Noui, Karim, 155
null vector, 179
Oeckl, Robert, 248
ontology, 20
Ooguri, Hirosi, 247
Oogury, 247
Oriti, Daniele, 250
orthogonal basis, 107
Pachner moves, 126
Palatini action, 64
Parentani, Renaud, 214
parity, 180
partial observables, 46
Pauli principle, 198
Penrose, Roger, 179
pentagram, 149
Perez, Alejandro, vii
Peter-Weyl theorem, 107, 118
phase transitions, 18
Planck length, 8, 18
Planck mass, 251
Planck star, 250
Planck, Max, 209
Planck-scale cut-off, 19
Plebanski two form, 63
Poincare symmetry, 5
pressure, 171
primordial black holes, 251
Ptolemy, 60
Pullin, Jorge, vii
punctures, 214
Pythagoreans, 149
Index
267
QCD, 150152, 189

QED, 150, 152, 189
quanta of area, 15
quanta of space, 10, 16
quanta of volume, 15
quantisation postulate, 10, 14
quantum correlations across the horizon, 218
quantum dimension, 129, 134
quantum geometry, 9
quantum group, 133
truncation, 145
Turaev-Viro, 129
Turaev-Viro model, 129
two complex, 92
random lattice, 197

Regge action, 187, 188
Regge equations, 90
Regge Hamilton function, 188
Regge space, 88
Regge, Tullio, 82, 87
regular graph, 232
renormalisation group, 18
Riello, Aldo, 249
Riemann curvature, 62
Riemann tensor, 62
Riemannian manifold, 6
Robert Oeckl, 52
Roche, Philippe, 155
vacuum energy, 59
vertices, 138
Sahlmann, Hanno, vii

scale factor, 226
scaling, 18
second order formulation, 65
Sen, Amitaba, 74
separability of LQG Hilbert space, 145
Shift function, 69, 71
Smolin, Lee, 202
source, 140
Speziale, Simone, 178, 195
spike, 128
spikes, 127
spin connection, 62
spin-network basis, 107
spinfoam cosmology, 231
spinors, 177
Stalin, Joseph, 7
Standard Model, 3
string theory, 22
supergravity, 22
superposition principle, 148
supersymmetric particles, 3
supersymmetry, 22
Susskind, Leonard, 84
target, 140
tetrad-Palatini action, 64
tetrads, 60
Thiemann, Thomas, vii
time gauge, 72
time-reversal, 63
torsion, 62
triangular inequality, 108, 134
ultraviolet catastrophe, 209

ultraviolet divergences, 123
unification, 4
Unruh effect, 217
Unruh temperature, 211, 218
Unruh, Bill, 210
Warsaw group, 248

wedge amplitude, 154, 216
Wheeler, John, 202
Wheeler, John , 247
Wheeler-deWitt equation, 22
Wightman, Arthur, 238
Wilson, Kenneth, 82
Wolfgang, Wieland, 178
Yang-Mills field, 202
Zakopane lecture, vii
Zeldovich, 203

Introduction LQ G

Uploaded by

Copyright:

Available Formats

Introduction LQ G

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction LQ G

Uploaded by

Copyright:

Available Formats

Covariant Loop Quantum Gravity

An elementary introduction to Quantum Gravity and Spinfoam Theory

CARLO ROVELLI AND FRANCESCA VIDOTTO

Cover: Martin Klimas, Jimi Hendrix, House Burning Down.

1 Spacetime as a quantum object

2 Physics without time

6 Bubbles and cosmological constant

Part III The real world

7 The real world: 4D Lorentzian theory

Part IV Physical applications

Spacetime as a quantum object

1.1 The problem

Spacetime as a quantum object

forces, black holes at LHC...). Anything beyond these theories is speculative. It is

Fock space of photons and electrons

Feynman diagrams expansion

computation scattering of neutral particles in Plank units

Spacetime as a quantum object

1.2 The end of space and time

And we stand on their shoulders.

The end of space and time

Spacetime as a quantum object

Conventional quantum field theorist

Post-Maldacena string theorist

Genuine quantum-gravity physicist

Pre-general-relativistic physics is conceived on spacetime. The recent developments of

1.3 Geometry quantized

Spacetime as a quantum object

+ 2k q2 has discrete spectrum with eigenvalues (E(n) = En (n) )

where = k/m. That is, energy is quantized: it comes in discrete quanta.

The magic circle: discreteness is kinematics

Spacetime as a quantum object

where ij k is the totally antisymmetric (Levi-Civita) symbol. SU(2) representation

The four vectors ~L a , normals to the faces.

They satisfy the closure relation

(Defined in this manner, a negative value of V 2 is simply the indication of a

Spacetime as a quantum object

Quantization of the geometry

[ Lia , Lb ] = iab lo2 ij k Lka ,

An image from De triplici minimo et mensura (1591) by Giordano Bruno, showing

Spacetime as a quantum object

The shape of the quanta of space and the fuzziness of the

Areas and volumes have discrete eigenvalues.

Spacetime as a quantum object

1.4 Physical consequences of the existence of the

Physical consequences of the existence of the Planck scale

1.4.2 Fuzziness: disappearance of classical space and time

Spacetime as a quantum object

1.5 Graphs, loops and quantum Faraday lines

Table 1.1 The ontology of the world in contemporary physics

General-covariant quantum fields

Graphs, loops and quantum Faraday lines

A set of adjacent quantum polyhedra and the graph they determine

Spacetime as a quantum object

1.6 The landscape

Table 1.2 Main lines of development of quantum gravity

General relativity perspective

gravitational field is split