Anda di halaman 1dari 95

QUANTUM DYNAMICS OF LOOP QUANTUM

GRAVITY
A Thesis
Submitted to the Graduate Faculty of the
Louisiana State University and
Agricultural and Mechanical College
in partial fulllment of the
requirements for the degree of
Master of Science
in
The Department of Physics and Astronomy
by
Muxin Han
B.S., Beijing Normal University, Beijing, 2005
May 2007
Acknowledgements
First of all, I am grateful to Dr. Jorge Pullin for his advise and many corrections of this thesis,
and to Dr. Jonathan Dowling for all his kind help in these two years. I also would like to thank
Dr. Hwang Lee for his kind support and being a member of my committee.
I would like to thank all the people who have discussed issues with me concerning the
subject in the thesis. They are: Dr. Abhay Ashtekar, Dr. Lai-Him Chan, Dr. Weiming Huang,
Dr. Jerzy Lewandowski, Dr. Yongge Ma, Dr. Andrzej Okolow, Dr. Jorge Pullin, Dr. Carlo
Rovelli, Dr. Thomas Thiemann, Dr. Dmitry Uskov, Dr. Robert M. Wald, and Dr. Hongbao
Zhang.
This work is supported by the assistantship of LSU, the Horace Hearne Institute for Theo-
retical Physics at LSU, and funding from Advanced Research and Development Activity.
ii
Table of Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Motivation of Quantum Gravity . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Purpose of Loop Quantum Gravity . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Classical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Lagrangian Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Hamiltonian Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Foundations of Loop Quantum Gravity . . . . . . . . . . . . . . . . . . . . . . . 11
3.1 General Programme for Algebraic Quantization . . . . . . . . . . . . . . . . . 11
3.2 Quantum Conguration Space . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Cylindrical Functions on Quantum Conguration Space . . . . . . . . . . . . . 20
3.4 Loop Quantum Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.5 Spin-network Decomposition of Kinematical Hilbert Space . . . . . . . . . . . 25
3.6 Quantum Riemannian Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 31
4 Implementation of Quantum Constraints . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Solutions of Quantum Gaussian Constraint . . . . . . . . . . . . . . . . . . . . 34
4.2 Solutions of Quantum Dieomorphism Constraint . . . . . . . . . . . . . . . . 35
4.3 Hamiltonian Constraint Operator . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.4 Master Constraint Programme . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5 Quantum Matter Field on a Quantum Background . . . . . . . . . . . . . . . . . 51
5.1 Polymer-like Representation of a Scalar Field . . . . . . . . . . . . . . . . . . 51
5.2 Dieomorphism Invariant Hamiltonian of a Scalar Field . . . . . . . . . . . . 54
5.3 Hamiltonian Constraint Equation for the Coupled System . . . . . . . . . . . . 61
5.4 Master Constraint for the Coupled System . . . . . . . . . . . . . . . . . . . . 63
6 The Semiclassical Limit of Quantum Dynamics . . . . . . . . . . . . . . . . . . . 68
6.1 The Construction of Coherent States . . . . . . . . . . . . . . . . . . . . . . . 68
6.2 Algebraic Quantum Gravity Approach . . . . . . . . . . . . . . . . . . . . . . 74
7 Conclusion and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
iii
Abstract
In the last 20 years, loop quantum gravity, a background independent approach to unify general
relativity and quantum mechanics, has been widely investigated. The aim of loop quantum
gravity is to construct a mathematically rigorous, background independent, nonperturbative
quantum theory for the Lorentzian gravitational eld on a four-dimensional manifold. In this
approach, the principles of quantum mechanics are combined with those of general relativity
naturally. Such a combination provides us a picture of quantum Riemannian geometry, which
is discrete at a fundamental scale. In the investigation of quantum dynamics, the classical
expressions of constraints are quantized as operators. The quantum evolution is contained in
the solutions of the quantum constraint equations. On the other hand, the semi-classical analysis
has to be carried out in order to test the semiclassical limit of the quantum dynamics.
In this thesis, the structure of the dynamical theory in loop quantum gravity is presented ped-
agogically. The outline is as follows: rst we review the classical formalism of general relativity
as a dynamical theory of connections. Then the kinematical Ashtekar-Isham-Lewandowski rep-
resentation is introduced as a foundation of loop quantum gravity. We discuss the construction
of a Hamiltonian constraint operator and the master constraint programme, for both the cases of
pure gravity and matter eld coupling. Finally, some strategies are discussed concerning testing
the semiclassical limit of the quantum dynamics.
iv
1 Introduction
1.1 Motivation of Quantum Gravity
The current view of physics is that there exist four fundamental interactions: strong interaction,
weak interaction, electromagnetic interaction and gravitational interaction. The description for
the former three kinds of forces is quantized in the well-known standard model. The interactions
are transmitted via the exchange of particles. However, the last kind of interaction, gravitational
interaction, is described by Einsteins theory of general relativity, which is a classical theory
which describes the gravitational eld as a smooth metric tensor eld on a manifold, i.e., a 4-
dimensional spacetime geometry. There is no and hence no quantum structure of spacetime.
Thus there is a fundamental inconsistency in our current description of the whole physical
world. Physicists widely accept the assumption that our world is quantized at fundamental level.
So all interactions should be brought into the framework of quantum mechanics fundamentally.
As a result, the gravitational eld should also have quantum structure.
Throughout the last century, our understanding of nature has considerably improved from
macroscale to microscale, including the phenomena at molecule, atom, sub-atom, and elemen-
tary particle scale. The standard model of particle physics agrees with all present experimental
tests in laboratory (see e.g. [158]). However, because unimaginably large amount of energy
would be needed, no experimental tests exist for processes that happen near the Planck scale

p
(G/c
3
)
1/2
10
33
cm and t
p
(G/c
5
)
1/2
10
43
s, which are viewed as the most
fundamental scales. The Planck scale arises naturally in attempts to formulate a quantum the-
ory of gravity, since
p
and t
p
are unique combinations of speed of light c, Planck constant ,
and gravitational constant G, which have the dimensions of length and time respectively. The
dimensional arguments suggest that at Planck scale the smooth structure of spacetime should
break down, and therefore the well-known quantum eld theory is invalid since it depends on
a xed smooth background spacetime. Hence we believe that physicists should go beyond the
successful standard model to explore the new physics near Planck scale, which is, perhaps, a
quantum eld theory without a background spacetime, and this quantum eld theory should
include the quantum theory of gravity. Moreover, current theoretical physics is thirsting for a
quantum theory of gravity to solve at least the following fundamental diculties.
Classical Gravity - Quantum Matter Inconsistency
The equation relating matter and the gravitational eld is the famous Einstein eld equa-
tion:
R

[g]
1
2
R[g]g

= T

[g], (1)
where the left hand side of the equation concerns spacetime geometry which has classical
smooth structure, while the right hand side concerns also matter eld which is funda-
mentally quantum mechanical in standard model. In quantum eld theory the energy-
momentum tensor of matter eld should be an operator-valued tensor

T

. One possible
way to keep classical geometry consistent with quantum matter is to replace T

[g] by
the expectation value <

T

[g] > with respect to some quantum state of the matter on a


xed spacetime. However, in the solution of this equation the background g

has to be
changed due to the non-vanishing of <

T

[g] >. So one has to feed back the new metric


1
into the denition of the vacuum expectation value etc. The result of the iterations does
not converge in general [70]. On the other hand, some other arguments show that such a
semiclassical treatment may violate the principle of superposition in quantum mechanics
[55]. This inconsistency motivates us to quantize the background geometry to arrive at an
operator formula also on the left hand side of Eq.(1).
Singularities in General Relativity
Einsteins theory of General Relativity is considered as one of the most elegant theories
in the 20th century. Many experimental tests conrm the theory in the classical domain
[159]. However, Penrose and Hawking proved that singularities are inevitable in general
spacetimes with matter satisfying certain conditions in, by now well known, singularity
theorems (for a summary, see [91][156]). Thus general relativity as a classical theory
breaks down in certain regions of spacetime in a generic way. One naturally expects that,
in extra strong gravitational eld domains near the singularities, the gravitational theory
would probably be replaced by an unknown quantum theory of gravity.
Innities in Quantum Field Theory
It is well known that there are innity problems in quantum eld theory in Minkowski
spacetime. In curved spacetime, the problem of divergences is even more complicated,
since the renormalization process in curved spacetime is ambiguous, the expectation value
of stress tensor can be xed up to some local curvature terms, and it also depends on a
fundamental scale of spacetime. Although much progress on the renormalization have
been made [92][157], a fundamentally satisfactory theory is still far from reaching. So it
is expected that some quantum gravity theory, playing a fundamental role at Planck scale,
could provide a natural cut-o to cure the innities in quantum eld theory. The situation
of quantum eld theory on a xed spacetime looks just like that of quantum mechanics
for particles in electromagnetic eld before the establishing of quantum electrodynamics,
where the particle mechanics (actress) is quantized but the background electromagnetic
eld (stage) is classical. The history suggests that such a semi-classical situation is only
an approximation which should be replaced by a much more fundamental and satisfactory
theory.
1.2 Purpose of Loop Quantum Gravity
The research on quantum gravity is quite active. Many quantization programmes for gravity
are being carried out (for a summary see e.g. [146]). In these dierent kinds of approaches,
Among these dierent kinds of approaches, the idea of loop quantum gravity nds its roots in
researchers from the general relativity community. It follows closely the motivations of general
relativity, and hence it is a quantum theory born with background independence. Roughly
speaking, loop quantum gravity is an attempt to construct a mathematically rigorous, non-
perturbative, background independent quantum theory of four-dimensional, Lorentzian general
relativity plus all known matter in the continuum. The project of loop quantum gravity inherits
the basic idea of Einstein that gravity is fundamentally spacetime geometry. Here one believes
2
in that the theory of quantum gravity is a quantum theory of spacetime geometry with dieo-
morphism invariance (this legacy is discussed comprehensively in Rovellis book [122]). To
carry out the quantization procedure, one rst casts general relativity into the Hamiltonian for-
malism as a dieomorphism invariant Yang-Mills gauge eld theory with a compact internal
gauge group. Thus the construction of loop quantum gravity can also be applied to all back-
ground independent gauge eld theories. One can therefore claim that the theory can also be
called as a background independent quantum gauge eld theory.
All classical elds theories, other than the gravitational eld, are dened on a xed space-
time, which provides a foundation to the perturbative Fock space quantization. However general
relativity is only dened on a manifold and hence is a background independent classical eld
theory, since gravity itself is the background. So the situation for gravity is much dierent
from other elds by construction [122], namely gravity is not only the background stage, but
also the dynamical actress. Such a double character for gravity leads to many diculties in the
understanding of general relativity and its quantization, since we cannot follow the strategy in
ordinary quantum theory of matter elds. However, an amazing result in loop quantum gravity
is that the background independent programme can even help us to avoid the diculties in or-
dinary quantum eld theory. In perturbative quantum eld theory in curved spacetime, the de-
nition of some basic physical quantities, such as the expectation value of energy-momentum, is
ambiguous and it is dicult to calculate the back-reaction of quantum elds to the background
spacetime [157]. One could speculate that the diculty is related to the fact that the present
formulation of quantum eld theories is background dependent. For instance, the vacuum state
of a quantum eld is closely related to spacetime structure, which plays an essential role in the
description of quantum eld theory in curved spacetime and their renormalization procedures.
However, if the quantization programme is by construction background independent and non-
perturbative, it may be possible to solve the problems fundamentally. In loop quantum gravity
there is no assumption of a priori background metric at all and the gravitational eld and matter
elds are coupled and uctuating naturally with respect to each other on a common manifold.
In the following sections, we will review pedagogically the basic construction of a com-
pletely new, background independent quantum eld theory, which is completely dierent from
the known quantum eld theory. For completeness and accuracy, we will use detailed math-
ematical terminology. However, for simplicity, we will skip the complicated proofs of many
important statements. One may nd the missing details in the references cited. Thus our review
will not be comprehensive. We refer to Ref.[146] and [90] for a more detailed exploration,
Refs. [20] and [148] for more advanced topics. It turns out that in the framework of loop
quantum gravity all theoretical inconsistencies introduced in the previous section are likely to
be cured. More precisely, one will see that there is no UV divergence in quantum elds of
matter if they are coupled with gravity in the background independent approach. Also recent
works show that the singularities in general relativity can be smeared out in symmetry-reduced
models [43][101][48]. The crucial point is that gravity and matter are coupled and consistently
quantized non-perturbatively so that the problems of classical gravity and quantum matter in-
consistency disappear.
3
2 Classical Framework
2.1 Lagrangian Formalism
In order to canonically quantize classical gravity, a Hamiltonian analysis has to be performed
to obtain a canonical formalism of the classical theory suitable to be represented on a Hilbert
space. A well known canonical formalism of general relativity is the ADM formalism (ge-
ometrodynamics) derived from the EinsteinHilbert action[156][97], which has been problem-
atic to cast into a quantum theory rigorously. Another well-known action of general relativity
is the Palatini formalism, where the tetrad and the connection are regarded as independent dy-
namical variables. However, the dynamics of the Palatini action has similar diculties at the
time of quantization as the dynamics derived from the EinsteinHilbert action [4][87]. In 1986,
Ashtekar presented a formalism of true connection dynamics for general relativity with a rela-
tively simple Hamiltonian constraint, and thus opened the door to apply quantization techniques
from gauge eld theory [2][3][123]. However, a drawback of that formalism is that the canon-
ical variables are complex, which need the implementation of complicated reality conditions
if one is to represent real general relativity. Moreover, the quantization based on the complex
connection could not be carried out rigorously, since the internal gauge group is noncompact.
In 1995, Barbero modied the Ashtekar new variables to give a system of real canonical vari-
ables for dynamical theory of connections [34]. Then Holst constructed a generalized Palatini
action to support Barberos real connection dynamics [93]. Although there is a free parameter
(Barbero-Immirzi parameter ) in the generalized Palatini action and the Hamiltonian constraint
is more complicated than the Ashtekar one, the generalized Palatini Hamiltonian with the real
connections is widely used by loop theorists for the quantization programme. All the following
analysis is based on the generalized Palatini formalism.
Consider a 4-manifold, M, on which the basic dynamical variables in the generalized Pala-
tini framework are a tetrad e

I
and an so(1, 3)-valued connection
I J

(not necessarily torsion-


free), where the capital Latin indices I, J, ... refer to the internal S O(1, 3) group and the Greek
indices , , ... denote spacetime indices. A tensor with both spacetime indices and internal
indices is named as a generalized tensor. The internal space is equipped with a Minkowskian
metric
I J
(of signature , +, +, +) xed once for all, such that the spacetime metric reads:
g

=
I J
e
I

e
J

.
The generalized Palatini action in which we are interested is given by [20]:
S
p
[e

K
,
I J

] =
1
2
_
M
d
4
x(e)e

I
e

J
(
I J

+
1
2

I J
KL

KL

), (2)
where e is the square root of the determinant of the metric g

,
I J
KL
is the internal Levi-Civita
symbol, is the real Barbero-Immirzi parameter, and the so(1, 3)-valued curvature 2-form
I J

of the connection
I J

reads:

I J

:= 21
[

I J
]
=

I J

I J

+
IK


J
K
,
here 1

denote the so(1, 3) generalized covariant derivative with respect to


I J

acting on both
spacetime and internal indices. Note that the generalized Palatini action returns to the Palatini
4
action when
1

= 0 and gives the (anti)self-dual Ashtekar formalism when one sets


1

= i.
Moreover, besides spacetime dieomorphism transformations, the action is also invariant under
internal S O(1, 3) rotations:
(e, ) (e

) = (b
1
e, b
1
b + b
1
db),
for any S O(1, 3) valued function b on M. The gravitational eld equations are obtained by
varying this action with respect to e

I
and
I J

. We rst study the variation with respect to the


connection
I J

. One has

I J

= (d
I J
)

+
IK


J
K
+
IK


J
K
= 21
[

I J
]
by the denition of covariant generalized derivative 1

. Note that
I J

is a Lorentz covariant
generalized tensor eld since it is the dierence between two Lorentz connections [107][104].
One thus obtains
S
p
=
1
2
_
M
d
4
x(e)e

I
e

J
(
I J

+
1
2

I J
KL

KL

)
=
1

_
M
(
I J

+
1
2

I J
KL

KL

)1

[(e)e

I
e

J
],
where we have used the fact that 1

for all vector density


of weight +1 and
neglected the surface term. Then it gives the equation of motion:
1

[(e)e

I
e

J
] =
1
4
1

I JKL
e
K

e
L

] = 0,
where

is the spacetime Levi-Civita symbol. This equation leads to the torsion-free Cartans
rst equation:
1
[
e
I
]
= 0,
which means that the connection
I J

is the unique torsion-free Levi-Civita spin connection


compatible with the tetrad e

I
. As a result, the second term in the action (2) can be calculated
as:
(e)e

I
e

I JKL

KL
=

,
which is exactly vanishing, because of the symmetric properties of Riemann tensor. So the
generalized Palatini action returns to the Palatini action, which will certainly give the Einstein
eld equation.
2.2 Hamiltonian Formalism
To carry out the Hamiltonian analysis of action (2), suppose the spacetime M is topologically
R for some 3-dimensional compact manifold without boundary. We introduce a foliation
5
parameterized by a smooth function t and a time-evolution vector eld t

such that t

(dt)

= 1
in M, where t

can be decomposed with respect to the unit normal vector n

of as:
t

= Nn

+ N

, (3)
here N is called the lapse function and N

the shift vector [156][97]. The internal normal vector


is dened as n
I
n

I
. It is convenient to carry out a partial gauge xing, i.e., x a internal
constant vector eld n
I
with
I J
n
I
n
J
= 1. Note that the gauge xing puts no restriction on
the real dynamics
1
. Then the internal vector space V is 3+1 decomposed with a 3-dimensional
subspace W orthogonal to n
I
, which will be the internal space on . With respect to the internal
normal n
I
and spacetime normal n

, the internal and spacetime projection maps are denoted by


q
I
i
and q

a
respectively, where we use i, j, k, ... to denote the 3-dimensional internal space indices
and a, b, c, ... to denote the indices of space . Then an internal reduced metric
i j
and a reduced
spatial metric on , q
ab
, are obtained by these two projection maps. The two metrics are related
by:
q
ab
=
i j
e
i
a
e
j
b
, (4)
where the orthonormal co-triad on is dened by e
i
a
:= e
I

q
i
I
q

a
. Now the internal gauge group
S O(1, 3) is reduced to its subgroup S O(3) which leaves n
I
invariant. Finally, two Levi-Civita
symbols are obtained respectively as

i jk
:= q
I
i
q
J
j
q
K
k
n
L

LI JK
,

abc
:= q

a
q

b
q

c
t

,
where the internal Levi-Civita symbol
i jk
is an isomorphism of Lie algebra so(3). Using the
connection 1-form
I J

, one can dened two so(3)-valued 1-form on :

i
a
:=
1
2
q

a
q
i
I

I J
KL
n
J

KL

,
K
i
a
:= q
i
I
q

I J

n
J
,
where is a spin connection on and K will be related to the extrinsic curvature of on shell.
After the 3+1 decomposition and the Legendre transformation, action (2) can be expressed as
[93]:
S
p
=
_
R
dt
_

d
3
x[

P
a
i
[
t
A
i
a
1
tot
(A
i
a
,

P
b
j
,
i
, N, N
c
)], (5)
from which the symplectic structure on the classical phase space is obtained as
A
i
a
(x),

P
b
j
(y) :=
i
j

a
b

3
(x, y), (6)
where the conguration and conjugate momentum are dened respectively by:
A
i
a
:=
i
a
+ K
i
a
,

P
a
i
:=
1
2

abc

i jk
e
j
b
e
k
c
=
1

_
det qe
a
i
,
1
However, there are some arguments that such a gauge xing is a non-natural way to break the internal Lorentz
symmetry (see e.g. [131]).
6
here det q is the determinant of the 3-metric q
ab
on and hence det q = ()
3
det P. In the
denition of the conguration variable A
i
a
, we should emphasize that
i
a
is restricted to be the
unique torsion free so(3)-valued spin connection compatible with the triad e
a
i
. This conclusion
is obtained by solving a second class constraint in the Hamiltonian analysis [93]. In the Hamil-
tonian formalism, one starts with the elds (A
i
a
,

P
a
i
). Then neither the basic dynamical variables
nor their Poisson brackets depend on the Barbero-Immirzi parameter . Hence, for the case
of pure gravitational eld, the dynamical theories with dierent are related by a canonical
transformation. However, as we will see, the spectrum of geometric operators are modied by
dierent value of , and the non-perturbative calculation of black hole entropy is compatible
with Bekenstein-Hawkings formula only for a specic value of [68]. In addition, it is argued
that the Barbero-Immerzi parameter may lead to observable eects in principle when the grav-
itational eld is coupled with Fermions [112]. In the decomposed action (5), the Hamiltonian
density 1
tot
is a linear combination of constraints:
1
tot
=
i
G
i
+ N
a
C
a
+ NC,
where
i

1
2

i
jk

jk
t
, N
a
and N are Lagrange multipliers. The three kinds of constraints in the
Hamiltonian are expressed as [20]:
G
i
= D
a

P
a
i
:=
a

P
a
i
+
k
i j
A
j
a

P
a
k
,
C
a
=

P
b
i
F
i
ab

1 +
2

K
i
a
G
i
,
C =

2
2
_
det q

P
a
i

P
b
j
[
i j
k
F
k
ab
2(1 +
2
)K
i
[a
K
j
b]
]
+ (1 +
2
)
a
(

P
a
i
_
det q
)G
i
, (7)
where the conguration variable A
i
a
performs as a so(3)-valued connection on and F
i
ab
is the
so(3)-valued curvature 2-form of A
i
a
with the well-known expression:
F
i
ab
:= 2D
[a
A
i
b]
=
a
A
i
b

b
A
i
a
+
i
jk
A
j
a
A
k
b
.
In any dynamical system with constraints, the constraint analysis is essentially important be-
cause they reect the gauge invariance of the system. From the above three kinds of constraints
of general relativity, one can know the gauge invariance of the theory. The Gauss constraint
G
i
= 0 has crucial importance in formulating the general relativity into a dynamical theory of
connections. The corresponding smeared constraint function, g() :=
_

d
3
x
i
(x)G
i
(x), gener-
ates a transformation on the phase space as:
A
i
a
(x), g() = D
a

i
(x)

P
a
i
(x), g() =
k
i j

j
(x)

P
a
k
(x),
which are just the innitesimal versions of the following gauge transformation for the so(3)-
valued connection 1-form A and internal rotation for the so(3)-valued densitized vector eld

P
respectively:
(A
a
,

P
b
) (g
1
A
a
g + g
1
(dg)
a
, g
1

P
b
g).
7
To display the meaning of the vector constraint C
a
= 0, one may consider the smeared constraint
function:
+(

N) :=
_

d
3
x(N
a

P
b
i
F
i
ab
(N
a
A
i
a
)G
i
).
It generates the innitesimal spatial dieomorphism by the vector eld N
a
on as:
A
i
a
(x), +(

N) = [

N
A
i
a
(x),

P
a
i
(x), +(

N) = [

P
a
i
(x).
The smeared scalar constraint is weakly equivalent to the following function, which is re-
expressed for quantization purpose as
S(N) :=
_

d
3
xN(x)

C(x)
=

2
2
_

d
3
xN

P
a
i

P
b
j
_
det q
[
i j
k
F
k
ab
2(1 +
2
)K
i
[a
K
j
b]
]. (8)
It generates the innitesimal time evolution o . The constraints algebra, i.e., the Poisson
brackets between these constraints, play a crucial role in the quantization programme. It can be
shown that the constraints algebra of (7) has the following form:
g(), g(

) = g([,

]),
g(), +(

N) = g([

N
),
g(), S(N) = 0,
+(

N), +(

) = +([

N,

N

]),
+(

N), S(M) = S([

N
M),
S(N), S(M) = +((N
b
M M
b
N)q
ab
)
g((N
b
M M
b
N)q
ab
A
a
))
(1 +
2
)g(
[

P
a

a
N,

P
b

b
M]
det q
), (9)
where det qq
ab
=
2

P
a
i

P
b
j

i j
. Hence the constraints algebra is closed under the Poisson
brackets, i.e., the constraints are all rst class. Note that the evolution of constraints is consistent
since the Hamiltonian H =
_

d
3
x1
tot
is a linear combination of the constraints functions. The
evolution equations of the basic canonical pair read
[
t
A
i
a
= A
i
a
, H, [
t

P
a
i
=

P
a
i
, H.
Together with the three constraint equations, they are completely equivalent to the Einstein
eld equations. Thus general relativity is cast as a dynamical theory of connections with a
compact structure group. Before nishing the discussion of this section, several issues should
be emphasized.
8
Canonical Transformation Viewpoint
The above construction can be reformulated in the language of canonical transformations,
since the phase space of connection dynamics is the same as that of triad geometrodynam-
ics. In the triad formalism the basic conjugate pair consists of densitized triad

E
a
i
=

P
a
i
and extrinsic curvature K
i
a
. The Hamiltonian and constraints read
1
tot
=
i
G

i
+ N
a
C
a
+ NC
G

i
=
k
i j
K
j
a

E
a
k
, (10)
C
a
=

E
b
j

[a
K
j
b]
, (11)
C =
1
_
det q
[
1
2
det qR +

E
[a
i

E
b]
j
K
i
a
K
j
b
], (12)
where
a
is the S O(3) generalized derivative operator compatible with triad e
a
i
and R is
the scalar curvature with respect to it. Since

E
a
i
is a vector density of weight one, we have

E
a
i
=
a

E
a
i
+
k
i j

j
a

E
a
k
= 0.
One can construct the desired Gauss law by
G
i
:=
1

E
a
i
+ G

i
,
=
a

P
a
i
+
k
i j
(
j
a
+ K
j
a
)

P
a
k
,
which is weakly zero by construction. This motivates us to dene the connection A
a
i
=

i
a
+K
i
a
. Moreover, the transformation from the pair (

E
a
i
, K
j
b
) to (

P
a
i
, A
j
b
) can be proved to
be a canonical transformation [34][146], i.e., the Poisson algebra of the basic dynamical
variables is preserved under the transformation:

E
a
i


P
a
i
=

E
a
i
/
K
j
b
A
j
b
=
j
b
+ K
j
b
,
as

P
a
i
(x), A
j
b
(y) =

E
a
i
(x), K
j
b
(y) =
a
b

j
i
(x, y),
A
i
a
(x), A
j
b
(y) = K
i
a
(x), K
j
b
(y) = 0,

P
a
i
(x),

P
b
j
(y) =

E
a
i
(x),

E
b
j
(y) = 0.
The Preparation for Quantization
The advantage of a dynamical theory of connections is that it is convenient to be quan-
tized in a background independent fashion. In the following procedure of quantization,
the quantum algebra of the elementary observables will be generated by holonomy, i.e.,
connection smeared on a curve, and electric ux, i.e., a densitized triad smeared on a 2-
surface. So no information about a background geometry is needed to dene the quantum
9
algebra. In the remainder of the thesis, in order to incorporate also spinors, we will en-
large the internal gauge group to be S U(2). This does not damage the prior constructions
because the Lie algebra of S U(2) is the same as that of S O(3). Due to the well-known
nice properties of compact Lie group S U(2), such as the Haar measure and Peter-Weyl
theorem, one can obtain the background independent representation of the quantum alge-
bra and the spin-network decomposition of the kinematic Hilbert space.
Analysis on Constraint Algebra
The classical constraint algebra (9) is an innite dimensional Poisson algebra. Unfortu-
nately, it is not a Lie algebra since the Poisson bracket between two scalar constraints has
structure function depending on dynamical variables. This causes problems when solving
the constraints quantum mechanically. On the other hand, one can see from Eq.(9) that
the algebra generated by Gauss constraints forms not only a subalgebra but also a 2-side
ideal in the full constraint algebra. Thus one can rst solve the Gauss constraints indepen-
dently. It is convenient to nd the quotient algebra with respect to the Gauss constraint
subalgebra as
+(

N), +(

) = +([

N,

N

]),
+(

N), S(M) = S([

N
M),
S(N), S(M) = +((N
b
M M
b
N)q
ab
),
which plays a crucial role in solving the constraints quantum mechanically. But the subal-
gebra generated by the dieomorphism constraints can not form an ideal. Hence the pro-
cedures of solving the dieomorphism constraints and solving Hamiltonian constraints
are entangled with each other. This leads to certain ambiguity in the construction of a
Hamiltonian constraint operator [134][149]. Fortunately, the Master Constraint Project
addresses the above two problems as a whole by introducing a new classical constraint al-
gebra [149]. The new algebra is a Lie algebra where the dieomorphism constraints form
a 2-side ideal. We will come back to this point in the discussion on quantum dynamics of
loop quantum gravity.
10
3 Foundations of Loop Quantum Gravity
In this chapter, we will begin to quantize the above classical dynamics of connections as a
background independent quantum eld theory. The main purpose of the chapter is to construct
a suitable kinematical Hilbert space 1
kin
for the representation of quantum observables. In the
following discussion, we formulate the construction in the language of algebraic quantum eld
theory [85]. It should be emphasized that the following constructions can be generalized to all
background independent non-perturbative gauge eld theories with compact gauge groups.
3.1 General Programme for Algebraic Quantization
In the strategy of loop quantum gravity, a canonical programme is performed to quantize gen-
eral relativity, which has been cast into a dieomorphism invariant gauge eld theory, or more
generally, a dynamical system with constraints. The following is a summary for a general pro-
cedure to quantize a dynamical system with rst class constraints [133][4].
Algebra of Classical Elementary Observables
One starts with the classical phase space (A, , ) and R (R can be countable innity
2
)
rst-class constraints C
r
(r = 1...R) such that C
r
, C
s
=
R
t=1
f
t
rs
C
t
, where f
t
rs
is generally
a function on phase space, namely, the structure function of Poisson algebra. The algebra
of classical elementary observables P is dened as:
Denition 3.1.1: The algebra of classical elementary observables P is a collection of
functions f (m), m Aon the phase space such that
(1) f (m) Pshould separate the points of A, i.e., for any m m

, there exists f (m) P,


such that f (m) f (m

); (analogy to the p and q in A = T

R.)
(2) f (m), f

(m) P f (m), f

(m) P (closed under Poisson bracket);


(3) f (m) P

f (m) P (closed under complex conjugation).
So P forms a sub -Poisson algebra of C

(A). In the case of A = T

R, P is gener-
ated by the conjugate pair (q, p) with q, p = 1.
Quantum Algebra of Elementary Observables
Given the algebra of classical elementary observables P, a quantumalgebra of elementary
observables can be constructed as follows. Consider the formal nite sequences of classi-
cal observable ( f
1
... f
n
) with f
k
P. Then the operations of multiplication and involution
are dened as
( f
1
, ..., f
n
) ( f

1
, ..., f

m
) := ( f
1
, ..., f
n
, f

1
, ..., f

m
),
( f
1
, .., f
n
)

:= (

f
n
, ...,

f
1
).
One can dene the formal sum of dierent sequences with dierent number of elements.
Then the general element of the newly constructed free -algebra F(P) of P, is formally
2
This includes the case of eld theory with innite many degree of freedom, since one can introduce the
expression C
n,
=
_

d
3
x
n
(x)C

(x), where
n
(x)

n=1
forms a system of basis in L
2
(, d
3
x).
11
expressed as
_
N
k=1
( f
(k)
1
, ... f
(k)
n
k
), where f
(i)
n
i
P. Consider the elements of the form (se-
quences consisting of only one element)
( f + f

) ( f ) ( f

), (z f ) z( f ), [( f ), ( f

)] i( f , f

),
where z C is a complex number, and the canonical commutation bracket is dened as
[( f ), ( f

)] := ( f ) ( f

) ( f

) ( f ).
A 2-side ideal I of F(P) can be constructed from these element, and is preserved by the
action of involution . One thus obtains
Denition 3.1.2: The quantum algebra A of elementary observables is dened to be the
quotient -algebra F(P)/I.
Note that the motivation to construct a quantum algebra of elementary observables is
to avoid the problem of operators ordering in quantum theory so that the quantum algebra
A can be represented on a Hilbert space without ordering ambiguities.
Representation of Quantum Algebra
In order to obtain a quantum theory, we need to quantize the classical observables in the
dynamical system. The, so called, quantization is nothing but a -representation map
3

from the quantum algebra of elementary observable A to the collection of linear opera-
tors [(1) on a Hilbert Space 1. At the level of quantum mechanics, the well-known
Stone-Von Neumann Theorem concludes that in quantum mechanics, there is only one
strongly continuous, irreducible, unitary representation of the Weyl algebra, up to uni-
tary equivalence (see, for example, Ref.[113]). However, the conclusion of Stone-Von
Neumann cannot be generalized to the quantum eld theory because the latter has innite
many degrees of freedom (for detail, see, for example [157]). In quantum eld theory,
a representation can be constructed by GNS(Gelfand-Naimark-Segal)-construction for a
quantum algebra of elementary observable A, which is a unital -algebra actually. The
GNS-construction for the representation of quantum algebra A is briey summarized as
follows.
Denition 3.1.3: Given a positive linear functional (a state) on A, the null space
N

A with respect to is dened as N

:= a A(a

a) = 0, which is a left ideal


in A. Then a quotient map can be dened as [.]: A A/N

; a [a] := a + bb N

.
3
A map : A [(1) is a *-representation if and only if (1) there exists a dense subspace 1 of 1 contained
in
aA
[D((a)) D((a

))] where D((a)) is the domain of the operator (a) and (2) for every a, b A and C
the following conditions are satised in 1,
(a + b) = (a) + (b), (a) = (a),
(a b) = (a)(b), (a

) = (a)

.
Note that [(1) fails to be an algebra because the domains of unbounded operators cannot be the whole Hilbert
space. However, the collection of bounded operators on any Hilbert space is really a -algebra.
12
The GNS-representation for A with respect to is a -representation map:

: A
[(1

), where 1

:= (A/N

) and (.) denotes the completion with respect to the naturally


equipped well-dened inner product
< [a][b] >
1

:= (a

b)
on 1

. This representation map is dened by

(a)[b] := [a b], a A and [b] 1

,
where

(a) is an unbounded operator in general. Moreover, GNS-representation is a


cyclic representation, i.e.,

, such that ((a)

a A) = 1

and

is called
a cyclic vector in the representation space. In fact

:= [1] is a cyclic vector in 1

and
(

(a)

a A) = 1

. As a result, the positive linear functional with which we begin


can be expressed as
(a) =<

(a)

>
1

.
Thus a positive linear functional on A is equivalent to a cyclic representation of A, which
is a triple (1

). Moreover, every non-degenerate representation is an orthogonal


direct sum of cyclic representations ( for proof, see [58] ) .
So the kinematical Hilbert space 1
kin
= 1

for the system with constrains can be ob-


tained by GNS-construction. In the case that there are gauge symmetries in our dy-
namical system, supposing that there is a gauge group G acting on A by automorphisms

g
: A A, g G, it is preferred to construct a gauge invariant representation of A.
So we require the positive linear functional on A to be gauge invariant, i.e.,
g
= .
Then the group G is represented on the Hilbert space 1

as:
U(g)

(a)

(
g
(a))

,
and such a representation is a unitary representation of G. In loop quantum gravity, it is
crucial to construct an internal gauge invariant and dieomorphism invariant representa-
tion for the quantum algebra of elementary observables.
Implementation of the Constraints
In the Dirac quantization programme for a system with constraints, the constraints should
be quantized as some operators in a kinematical Hilbert space 1
kin
. One then solves
them at quantum level to get a physical Hilbert space 1
phys
, that is, to nd a quantum
analogy

C
r
of the classical constraint formula C
r
and to solve the general solution of
the equation

C
r
= 0. However, there are several problems in the construction of the
constraint operator

C
r
.
(i) C
r
is in general not in P, so there is a factor ordering ambiguity in quantizing C
r
to
be an operator

C
r
.
(ii) In quantum eld theory, there are ultraviolet(UV) divergence problems in construct-
ing operators. However, the UV divergence can be avoided in the background inde-
pendent approach.
13
(iii) Sometimes, quantum anomalies may appear when there are structure functions in
the Poisson algebra. Although classically we have C
r
, C
s
=
R
t=1
f
t
rs
C
t
, r, s, t =
1, ..., R, where f
t
rs
is a function on phase space, quantum mechanically it is possible
that [

C
r
,

C
s
] i
R
t=1

f
t
rs

C
t
due to the ordering ambiguity between

f
t
rs
and

C
t
. If one
sets [

C
r
,

C
s
] =
i
2

R
t=1
(

f
t
rs

C
t
+

C
t

f
t
rs
), for satisfying

C
r
= 0, we have
[

C
r
,

C
s
] =
i
2
R

t=1

C
t

f
t
rs
=
i
2
R

t=1
[

C
t
,

f
t
rs
]. (13)
However, [

C
t
,

f
t
rs
] are not necessary to equal to zero for all r, s, t = 1...R. If
not, the problem of quantum anomaly appears and the new quantum constraints
[

C
t
,

f
t
rs
] = 0 have to be imposed on physical quantum states, since the classical
Poisson brackets C
r
, C
s
are weakly equal to zero on the constraint surface A A.
Thus too many constraints are imposed and the physical Hilbert space 1
phys
would
be too small. Therefore this is not a satisfactory solution and one needs to nd a
way to avoid the quantum anomalies.
Solving the Constraints and Physical Hilbert Space
In general the original Dirac quantization approach can not be carried out directly, since
there is usually no nontrivial 1
kin
such that

C
r
= 0. This happens when the con-
straint operator

C
r
has generalized eigenfunctions rather than eigenfunctions. One then
develops the so-called Rened Algebraic Quantization Programme, where the solutions
of the quantum constraint can be found in the algebraic dual space of a dense subset in
1
kin
(see e.g. [84]). The quantum dieomorphism constraint in loop quantum gravity is
solved in this way. Another interesting way to solve the quantum constraints is the Mas-
ter Constraint Approach proposed by Thiemann recently [149], which seems especially
suited to deal with the particular feature of the constraint algebra of general relativity. A
master constraint is dened as M :=
1
2

R
r,s=1
K
rs
C
s

C
r
for some real positive matrix K
rs
.
Classically one has M = 0 if and only if C
r
= 0 for all r = 1...R. So quantum mechan-
ically one may consider solving the Master Equation:

M = 0 to obtain the physical
Hilbert space 1
phys
instead of solving

C
r
= 0, r = 1...R. Because the master con-
straint M is classically positive, one has opportunities to implement it as a self-adjoint
operator on 1
kin
. If it is indeed the case and 1
kin
is separable, one can use the direct in-
tegral representation of 1
kin
associated with the self-adjoint operator

M to obtain 1
phys
:
1
kin

_

R
d()1

,
< >
kin
=
_
R
d() < >
1

, (14)
where is a so-called spectral measure and 1

is the (generalized) eigenspace of



M with
the eigenvalue . The physical Hilbert space is then formally obtained as 1
phys
= 1

=0
with the induced physical inner product < >
1

=0
4
. Now the issue of quantum anomaly
4
One need to be careful for such a formal prescription, see the later discussion of master constraint or [63].
14
is represented in terms of the size of 1
phys
and the existence of sucient numbers of
semi-classical states.
Physical Observables
We denote Aas the original unconstrained phase space, Aas the constraint surface, i.e.,
A := m AC
r
(m) = 0, r = 1...R, and

A as the reduced phase space, i.e. the space
of orbits for gauge transformations generated by all C
r
. The concept of Dirac observable
is dened as the follows.
Denition 3.1.4:
(1) A function O on A is called a weak Dirac observable if and only if the function de-
pends only on points of

A, i.e., O, C
r

A
= 0 for all r = 1...R. For the quantum version,
a self-adjoint operator

O is a weak Dirac observable if and only if the operator can be
well dened on the physical Hilbert space.
(2) A function O on A is called a strong Dirac observable if and only if O, C
r

A
= 0
for all r = 1...R. For the quantum version, a self-adjoint operator

O is a strong Dirac
observable if and only if the operator can be dened on the kinematic Hilbert space 1
kin
and [

O,

C
r
] = 0 in 1
kin
for all r = 1...R.
A physical observable is at least a weak Dirac observable. While Dirac observables have
been found explicitly in symmetry reduced models, some even with an innite number
of degrees of freedom, it seems extremely dicult to nd explicit expressions for them
in full general relativity. Moreover the Hamiltonian is a linear combination of rst-class
constraints. So there is no dynamics in the reduced phase space, and the meaning of
time evolution of the Dirac observables becomes subtle. However, using the concepts of
partial and complete observables [121][115][122], a systematic method to get Dirac ob-
servables can be developed, and the problem of time in such system with a Hamiltonian
H =
R
r=1

r
C
r
may also be solved.
Classically, let f (m) and T
r
(m)
R
r=1
be gauge non-invariant functions (partial observables)
on phase space A, such that A
sr
C
s
, T
r
is a non-degenerate matrix on A. A system
of classical weak Dirac observables (complete observables) F

f ,T
labelled by a collection
of real parameters
r

R
r=1
can be constructed as
F

f ,T
:=

n
1
n
R

(
1
T
1
)
n
1
(
R
T
R
)
n
R
n
1
! n
R
!

X
n
1
1


X
n
R
R
( f ),
where

X
r
( f ) :=
R
s=1
A
1
rs
C
s
, f

C
r
, f . It can be veried that [

X
r
,

X
s
]
A
= 0 and
F

f ,T
, C
r

A
= 0, for all r = 1...R (for details see [61] and [62]).
The partial observables T
r
(m)
R
r=1
may be regarded as clock variables, and
r
is the time
parameter for T
r
. The gauge is xed by giving a system of functions T
r
(m)
R
r=1
and
corresponding parameters
r

R
r=1
, namely, a section in A is selected by T
r
(m) =
r
for
each r, and F

f ,T
is the value of f on the section. To solve the problem of dynamics, one
assumes another set of canonical coordinates (P
1
, , P
NR
,
1
, ,
R
; Q
1
, , Q
NR
, T
1
,
15
, T
R
) by canonical transformations in the phase space (A, , ), where P
s
and
r
are
conjugate to Q
s
and T
r
respectively. After solving the complete system of constraints
C
r
(P
i
, Q
j
,
s
, T
t
) = 0
R
r=1
, the Hamiltonian H
r
with respect to the time T
r
is obtained as
H
r
:=
r
(P
i
, Q
j
, T
t
). Given a system of constants (
0
)
r

R
r=1
, for an observable f (P
i
, Q
j
)
depending only on P
i
and Q
j
, the physical dynamics is given by [61][150]:
(

r
)
=
0
F

f ,T

A
= F

0
H
r
, f ,T

A
= F

0
H
r
,T
, F

0
f ,T

A
,
where F

0
H
r
,T
is the physical Hamiltonian function generating the evolution with respect to

r
. Thus one has addressed the problem of time and dynamics as a result.
Semi-classical Analysis
An important issue in the quantization is to check whether the quantum constraint oper-
ators have correct classical limits. This has to be done by using the kinematical semi-
classical states in 1
kin
. Moreover, the physical Hilbert space 1
phys
must contain enough
semi-classical states to guarantee that the quantum theory one obtains can return to the
classical theory when 0. The semi-classical states in a Hilbert space 1 should have
the following properties.
Denition 3.1.5: Given a class of observables S which comprises a subalgebra in the
space [(1) of linear operators on the Hilbert space, a family of (pure) states
m

mA
are said to be semi-classical with respect to S if and only if:
(1) The observables in S should have correct semi-classical limit on semi-classical states
and the uctuations should be small, i.e.,
lim
0

m
( a) a(m)
a(m)
= 0,
lim
0

m
( a
2
)
m
( a)
2

m
( a)
2
= 0,
for all a S.
(2) The set of cyclic vectors
m
related to
m
via the GNS -representation (

, 1

)
is dense in 1.
Seeking for semiclassical states are one of open issues of current research in loop quantum
gravity. Recent original works focus on the construction of coherent states of loop quan-
tum gravity in analogy with the coherent states for harmonic oscillator system [142][143]
[144][145][19][15].
The above is the general programme to quantize a system with constraints. In the following sub-
section, we will apply the programme to the theory of general relativity and restrict our view to
the representation with the properties of background independence and spatial dieomorphism
invariance.
16
3.2 Quantum Conguration Space
In quantum mechanics, the kinematical Hilbert space is L
2
(R
3
, d
3
x), where the simple R
3
is the
classical conguration space of free particle which has nite degrees of freedom, and d
3
x is the
Lebesgue measure on R
3
. In quantum eld theory, it is expected that the kinematical Hilbert
space is also the L
2
space on the conguration space of the eld, which is innite dimensional,
with respect to some Borel measure naturally dened. However, it is often hard to dene a
concrete Borel measure on the classical conguration space, since the integral theory on innite
dimensional space is involved [57]. Thus the intuitive expectation should be modied, and the
concept of quantum conguration space should be introduced as a suitable enlargement of the
classical conguration space so that an innite dimensional measure, often called cylindrical
measure, can be well dened on it. The example of a scalar eld can be found in the refer-
ences [20][24]. For quantum gravity, it should be emphasized that the construction for quantum
conguration space must be background independent. Fortunately, general relativity has been
reformulated as a dynamical theory of S U(2) connections, which would be great helpful for our
further development.
The classical conguration space for gravitational eld, which is denoted by 7, is a col-
lection of the su(2)-valued connection 1-form eld smoothly distributed on . The idea of the
construction for quantum conguration is due to the concept of holonomy.
Denition 3.2.1: Given a smooth S U(2) connection eld A
i
a
and an analytic curve c with the
parameter t [0, 1] supported on a compact subset (compact support ) of , the corresponding
holonomy is dened by the solution of the parallel transport equation [104]
d
dt
A(c, t) = [A
i
a
c
a

i
]A(c, t), (15)
with the initial value A(c, 0) = 1, where c
a
is the tangent vector of the curve and
i
su(2)
constitute an orthonormal basis with respect to the Killing-Cartan metric (, ) := 2Tr(),
which satisfy [
i
,
j
] =
k
i j

k
and are xed once for all. Thus the holonomy is an element in
S U(2), which can be expressed as
A(c) = /exp (
_
1
0
[A
i
a
c
a

i
] dt), (16)
where A(c) S U(2) and / is a path-ordering operator along the curve c (see the footnote at
p382 in [104]).
The denition can be well extended to the case of piecewise analytic curves via the relation:
A(c
1
c
2
) = A(c
1
)A(c
2
), (17)
where stands for the composition of two curves. It is easy to see that a holonomy is invariant
under the re-parametrization and is covariant under changing the orientation, i.e.,
A(c
1
) = A(c)
1
. (18)
17
So one can formulate the properties of holonomy in terms of the concept of the equivalent
classes of curves.
Denition 3.2.2: Two analytic curves c and c

are said to be equivalent if and only if they


have the same source s(c) (beginning point ) and the same target t(c) (end point ), and the
holonomies of the two curves are equal to each other, i.e., A(c) = A(c

) A 7. A equivalent
class of analytic curves is dened to be an edge, and a piecewise analytic path is an composition
of edges.
To summarize, the holonomy is actually dened on the set / of piecewise analytic paths with
compact supports. The two properties (17) and (18) mean that each connection in 7 is a ho-
momorphism from /, which is so-called a groupoid by denition [155], to our compact gauge
group S U(2). Note that the internal gauge transformation and spatial dieomorphism act co-
variantly on a holonomy as
A(e) g(t(e))
1
A(e)g(s(e)) and A(e) A( e), (19)
for any S U(2)-valued function g(x) on and spatial dieomorphism . All above discussion
is for classical smooth connections in 7. The quantum conguration space for loop quantum
gravity can be constructed by extending the concept of holonomy, since its denition does not
depend on an extra background. One thus obtains the quantum conguration space 7 of loop
quantum gravity as the following.
Denition 3.2.3: The quantum conguration space 7is a collection of all quantum connections
A, which are algebraic homomorphism maps without any continuity assumption from the col-
lection of piecewise analytic paths with compact supports, /, on to the gauge group S U(2),
i.e., 7 := Hom(/, S U(2))
5
. Thus for any A 7and edge e in /,
A(e
1
e
2
) = A(e
1
)A(e
2
) and A(e
1
) = A(e)
1
.
The transformations of quantum connections under internal gauge transformations and dieo-
morphisms are dened by Eq.(19).
The above discussion on the smooth connections shows that the classical conguration space
7 can be understood as a subset in the quantum conguration space 7. Moreover, the Giles
theorem [82] shows precisely that a smooth connection can be recovered from its holonomies
by varying the length and location of the paths.
On the other hand, it was shown in [155][146] that the quantum conguration space 7 can
be constructed via a projective limit technique and admits a natural dened topology. To make
the discussion precise, we begin with a few denitions:
Denition 3.2.4:
5
It is easy to see that the denition of 7does not depend on the choice of local section in S U(2)-bundle, since
the internal gauge transformations leave 7invariant.
18
1. A nite set e
1
, ..., e
N
of edges is said to be independent if the edges e
i
can only intersect
each other at their sources s(e
i
) or targets t(e
i
).
2. A nite graph is a collection of a nite set e
1
, ..., e
N
of independent edges and their
vertices, i.e. their sources s(e
i
) and targets t(e
i
). We denote by E() and V() respectively
as the sets of independent edges and vertices of a given nite graph . And N

is the
number of elements in E().
3. A subgroupoid () / can be generated from by identifying V() as the set of ob-
jects and all e E() together with their inverses and nite compositions as the set of
homomorphisms. This kind of subgoupoid in / is called tame subgroupoid. () is inde-
pendent of the orientation of , so the graph can be recovered from tame subgroupoid
up to the orientations on the edges. We will also denote by N

the number of elements


in E() where is recovered by the tame subgroupoid .
4. [ denotes the set of all tame subgroupoids in /.
One can equip a partial order relation on [
6
, dened by

if and only if is a sub-


groupoid in

. Obviously, for any two tame subgroupoids () and

) in [, there
exists

) [ such that ,

, where

. Dene X

Hom(, S U(2))
as the set of all homomorphisms from the subgroupoid () to the group S U(2). Note that an
element A

( = ()) is completely determined by the S U(2) group elements A(e) where


e E(), so that one has a bijection : X
()
S U(2)
N

, which induces a topology on X


()
such that is a topological homomorphism. For any pair

, one can dene a surjective


projection map P

from X

to X

by restricting the domain of the map A

from

to the sub-
groupoid , and these projections satisfy the consistency condition P

= P

. Thus
a projective family X

, P

is obtained by above constructions. Then the projective limit


lim

(X

) is naturally obtained.
Denition 3.2.5: The projective limit lim

(X

) of the projective family X

, P

is a subset
of the direct product space X

:=

[
X

dened by
lim

(X

) := A

[
P

= A

.
Note that the projection P

is surjective and continuous with respect to the topology of X

.
One can equip the direct product space X

with the so-called Tychonov topology. Since any X

is a compact Hausdor space, by Tychonov theorem X

is also a compact Hausdor space. One


then can prove that the projective limit, lim

(X

), is a closed subset in X

and hence a compact


Hausdor space with respect to the topology induced from X

. At last, one can nd the relation


between the projective limit and the prior constructed quantum conguration space 7. As one
might expect, there is a bijection between 7and lim

(X

) [146]:
: 7 lim

(X

);
A A

[
,
6
A partial order on [ is a relation, which is reective ( ), symmetric (

= ) and
transitive (

). Note that not all pairs in [ need to have a relation.


19
where A

means the restriction of the domain of the map A 7 = Hom(/, S U(2)). As a result,
the quantum conguration space is identied with the projective limit space and hence can be
equipped with the topology. In conclusion, the quantum conguration space 7 is constructed
to be a compact Hausdor topological space.
3.3 Cylindrical Functions on Quantum Conguration Space
Given the projective family X

, P

, the cylindrical function on its projective limit 7 is


well dened as follows.
Denition 3.3.1: Let C(X

) be the set of all continuous complex functions on X

, two func-
tions f

C(X

) and f

C(X

) are said to be equivalent or cylindrically consistent, denoted


by f

, if and only if P

= P

> ,

, where P

denotes the pullback map


induced from P

. Then the space Cyl(7) of cylindrical functions on the projective limit 7 is


dened to be the space of equivalent classes [ f ], i.e.,
Cyl(7) := [

C(X

)]/ .
One then can easily prove the following proposition by denition.
Proposition 3.3.1:
All continuous functions f

on X

are automatically cylindrical since each of them can generate


a equivalent class [ f

] via the pullback map P

for all

> , and the dependence of P

on the groups associated to the edges in

but not in is trivial, i.e., by the denition of the


pull back map,
(P

)(A(e
1
), ..., A(e
N

), ..., A(e
N

)) = f

(A(e
1
), ..., A(e
N

)), (20)
where N

denotes the number of independent edges in the graph recovered from the groupoid .
On the other hand, by denition, given a cylindrical function f Cyl(7) there exists a suitable
groupoid such that f = [ f

], so one can identify f with f

. Moreover, given two cylindrical


functions f , f

Cyl(7), by denition of cylindrical functions and the property of projection


map, there exists a common groupoid and f

, f

C(X

) such that f = [ f

] and f

= [ f

].
Let f , f

Cyl(7), there exists graph such that f = [ f

], and f

= [ f

], then the following


operations are well dened
f + f

:= [ f

+ f

], f f

:= [ f

], z f := [z f

],

f := [

f

],
where z Cand

f denotes complex conjugate. So we construct Cyl(7) as an Abelian -algebra.
In addition, there is a unital element in the algebra because Cyl(7) contains constant functions.
Moreover, we can well dene the sup-norm for f = [ f

] by
j f j := sup
A

(A

), (21)
20
which satises the C

property j f

f j = j f j
2
. Then Cyl(7) is a unital Abelian C

-algebra, after
the completion with respect to the norm.
From the theory of C

-algebra, it is known that a unital Abelian C

-algebra is identical to the


space of continuous functions on its spectrum space via an isometric isomorphism, the so-called
Gelfand transformation (see e.g. [146]). So we have the following theorem [17][18], which
nishes this section.
Theorem 3.3.1:
(1) The space Cyl(7) has a structure of a unital Abelian C

-algebra after completion with re-


spect to the sup-norm.
(2) Quantum conguration space 7 is the spectrum space of completed Cyl(7) such that
Cyl(7) is identical to the space C(7) of continuous functions on 7.
3.4 Loop Quantum Kinematics
In analogy with the quantization procedure of section 3.1, in this subsection we would like to
perform the background-independent construction of algebraic quantum eld theory for gen-
eral relativity. First we construct the algebra of classical observables. Taking account of the
future quantum analogs, we dene the algebra of classical observables P as the Poission -
subalgebra generated by the functions of holonomies (cylindrical functions) and the uxes of
triad elds smeared on some 2-surface. Namely, one can dene the classical algebra in analogy
with geometric quantization in nite dimensional phase space case by the so-called classical
Ashtekar-Corichi-Zapata holonomy-ux -algebra as the following [96].
Denition 3.4.1
The classical Ashtekar-Corichi-Zapata holonomy-ux -algebra is dened to be a vector space
P
ACZ
:= Cyl(7) +
C
(7), where +
C
(7) is the vector space of algebraic vector elds spanned
by the vector elds Y
f
(S ) Cyl(7), and their commutators, here the smeared ux vector
eld Y
f
(S ) is dened by acting on any cylindrical function:
Y
f
(S ) :=
_
S

abc

P
c
i
f
i
, ,
for any su(2)-valued function f
i
with compact supports on S and are cylindrical functions on
7. We equip P
ACZ
with the structure of an -Lie algebra by:
(1) Lie bracket , : P
ACZ
P
ACZ
P
ACZ
is dened by
(, Y), (

, Y

) := (Y

, [Y, Y

]),
for all (, Y), (

, Y

) P
ACZ
with ,

Cyl(7) and Y, Y

+
C
(7).
(2) Involution: p p p P
ACZ
is dened by complex conjugate of cylindrical functions
and vector elds, i.e., p := (, Y) p = (, Y) P
ACZ
, where Y := Y.
21
(3) P
ACZ
admits a natural action of Cyl(7) by

(, Y) := (

Y),
which gives P
ACZ
a module structure.
Note that the action of ux vector eld Y
f
(S ) on can be expressed explicitly on any cylindrical
function

C
1
(X
()
) via a suitable regularization[146]:
Y
f
(S )

=
_
S

abc

P
c
i
f
i
,

,
=

eE()

_
S

abc

P
c
i
f
i
, A(e)
mn


A(e)
mn

eE()
(S, e)
2
f
i
(S e)[
S e,s(e)
(A(e)
i
)
mn

S e,t(e)
(
i
A(e))
mn
]

A(e)
mn

vV()S

e at v
(S, e)
2
f
i
(v)X
(e,v)
i

,
where A(e)
mn
is the S U(2) matrix element of the holonomy along the edge e, X
(e,v)
i
is the
left(right) invariant vector eld L
(
i
)
(R
(
i
)
) of the group associated with the edge e if v is the
source(target) of edge e by denition:
L
(
i
)
(A(e)) :=
d
dt

t=0
(A(e) exp(t
i
)),
R
(
i
)
(A(e)) :=
d
dt

t=0
( exp(t
i
)A(e)),
and
(S, e) =
_

_
0, if e S = , or e lies in S ;
1, if e lies above S and e S = p;
1, if e lies below S and e S = p.
Since the surface S is oriented with normal n
a
, above means n
a
e
a

p
> 0, and below means
n
a
e
a

p
< 0, where e
a

p
is the tangent vector of e at p. And one should consider e S contained
in the set V() and some edges are written as the union of elementary edges which either lie in
S , or intersect S at their source or target. On the other hand, from the commutation relations
for the left(right) invariant vector elds, one can see that the commutators between ux vector
elds do not necessarily vanish when S S

. This unusual property is the classical origin


of the non-commutativity of quantum Riemannian structures [23].
The classical Ashtekar-Corichi-Zapata holonomy-ux -algebra serves as a classical alge-
bra of elementary observables in our dynamical system of gauge elds. Then one can construct
the quantum algebra of elementary observables from P
ACZ
in analogy with Denition 3.1.2.
Denition 3.4.2[96]
22
The abstract free algebra F(P
ACZ
) of the classical -algebra is dened by the formal direct sum
of nite sequences of classical observables (p
1
, ..., p
n
) with p
k
P
ACZ
, where the operations of
multiplication and involution are dened as
(p
1
, ..., p
n
) (p

1
, ..., p

m
) := (p
1
, ..., p
n
, p

1
, ..., p

m
),
(p
1
, .., p
n
)

:= ( p
n
, ..., p
1
).
A 2-sided ideal I can be generated by the following elements,
(p + p

) (p) (p

), (zp) z(p),
[(p), (p

)] i(p, p

),
((, 0), p) ( p),
where the canonical commutation bracket is dened by
[(p), (p

)] := (p) (p

) (p

) (p).
Note that the ideal I is preserved by the involution , and the last set of generators in the ideal
I cancels the overcompleteness generated from the module structure of P
ACZ
[4].
The quantum holonomy-ux -algebra is dened by the quotient -algebra A = F(P
ACZ
)/I,
which contains the unital element 1 := ((1, 0)). Note that a sup-norm has been dened by
Eq.(21) for the Abelian sub--algebra Cyl(7) in A.
For simplicity, we denote the one element sequences (equivalence classes)

((, 0)) and

((0, Y))
Cyl(7), Y +
C
(7) in A by

and

Y respectively, where the hat denotes the equiv-
alence class with respect to the quotient. In particular, for all cylindrical functions

and ux
vector elds

Y
f
(S ),

and

Y
f
(S )

=

Y
f
(S ).
It can be seen that the free algebra F(P
ACZ
) is simplied a great deal after the quotient, and every
element of the quantum algebra A can be written as a nite linear combination of elements of
the form

1


Y
f
11
(S
11
),

2


Y
f
21
(S
21
)

Y
f
22
(S
22
),
...

k


Y
f
k1
(S
k1
)

Y
f
k2
(S
k2
) ...

Y
f
kk
(S
kk
),
...
Moreover, given a cylindrical function and a ux vector eld Y
f
(S ), one has the relation from
the commutation relation:

Y
f
(S )

= i

Y
f
(S ) +



Y
f
(S ). (22)
23
Then the kinematical Hilbert space 1
kin
can be obtained properly via the GNS-construction for
unital -algebra A in the same way as in Denition 3.1.3. By the GNS-construction, a positive
linear functional, i.e. a state
kin
, on A denes a cyclic representation (1
kin
,
kin
,
kin
) for A. In
our case of quantum holonomy-ux -algebra, the state with both Yang-Mills gauge invariance
and dieomorphism invariance is dened for any

Cyl(7) and non-vanishing ux vector


eld Y
f
(S ) +
C
(7) as [96]:

kin
(

) :=
_
S U(2)
N
_
eE()
d
H
(A(e))

(A(e)
eE()
),

kin
( a

Y
f
(S )) := 0, a A,
where d
H
is the Haar measure on the compact group S U(2) and N

is the number of elements


in E(). This
kin
is called Ashtekar-Isham-Lewandowski state. The null space N
kin
A with
respect to
kin
is dened as N
kin
:= a A
kin
( a

a) = 0, which is a left ideal. Then a quotient


map can be dened as:
[.] : A A/N
kin
;
a [ a] := a +

b

b N
kin
.
The GNS-representation for A with respect to
kin
is a representation map:
kin
: A [(1
kin
)
such that
kin
( a

b) =
kin
( a)
kin
(

b), where 1
kin
:= (A/N
kin
) = (Cyl(7)) by straightforward
verication and the () denotes the completion with respect to the natural equipped inner product
on 1
kin
,
< [ a][

b] >
kin
:=
kin
( a



b).
To show how this inner product works, given any two cylindrical functions = [

],

=
[

] Cyl(7), the inner product between them is expressed as


< [

][

] >
kin
:=
_
X

(P

)(P

)d

, (23)
for any groupoid

containing both and

. The measure d

on X

is dened by the pull


back of the product Haar measure d
N

H
on the product group S U(2)
N

via the identication


bijection between X

and S U(2)
N

, where N

is number of maximal analytic edges generating


. In addition, a nice result shows that given such a family of measures

[
, a probability
measure is uniquely well-dened on the quantum conguration space 7 [17], such that the
kinematical Hilbert space 1
kin
coincides with the collection of the square-integrable functions
with respect to the measure on the quantum conguration space, i.e. 1
kin
= L
2
(7, d), just
as we expected at the beginning of our construction.
The representation map
kin
is dened by

kin
( a)[

b] := [ a

b], a A, and [

b] 1
kin
.
Note that
kin
( a) is an unbounded operator in general. It is easy to verify that

kin
(

Y
f
(S ))[

] = i[

Y
f
(S )]
24
via Eq.(22), which gives the canonical momentum operator. In the following context, we denote
the operator
kin
(

Y
f
(S )) by

P
f
(S ) on 1
kin
, and just denote the elements [

] in 1
kin
by for
simplicity.
Moreover, since
kin
:= 1 is a cyclic vector in 1
kin
, the positive linear functional which we
begin with can be expressed as

kin
( a) =<
kin

kin
( a)
kin
>
kin
.
Thus the Ashtekar-Isham-Lewandowski state
kin
on A is equivalent to a cyclic representation
(1
kin
,
kin
,
kin
) for A, which is the Ashtekar-Isham-Lewandowski representation for quantum
holonomy-ux -algebra of background independent gauge eld theory. One thus obtains the
kinematical representation of loop quantum gravity via the construction of algebraic quantum
eld theory. It is important to note that the Ashtekar-Isham-Lewandowski state is the unique
state on the quantum holonomy-ux -algebra A invariant under internal gauge transformations
and spatial dieomorphisms
7
, which are both automorphisms
g
and

on A and can be veri-


ed that
kin

g
=
kin
and
kin

=
kin
. So these gauge transformations are represented
as unitary transformations on 1
kin
, while the cyclic vector
kin
, representing no geometry vac-
uum state, is the unique state in 1
kin
invariant under internal gauge transformations and spatial
dieomorphisms. This is a very crucial uniqueness theorem for canonical quantization of gauge
eld theory [96]:
Theorem 3.4.1: There exists exactly one Yang-Mills gauge invariant and spatial dieomor-
phism invariant state (positive linear functional) on the quantum holonomy-ux -algebra. In
other words, there exists a unique Yang-Mills gauge invariant and spatial dieomorphism in-
variant cyclic representation for the quantumholonomy-ux -algebra, which is called Ashtekar-
Isham-Lewandowski representation. Moreover, this representation is irreducible with respect to
an exponential version of the quantum holonomy-ux algebra (dened in [130]), which is anal-
ogous to the Weyl algebra.
Hence we have nished the construction of kinematical Hilbert space for background inde-
pendent gauge eld theory and represented the quantum holonomy-ux algebra on it. Then
following the general programme presented in the last subsection, we should impose the con-
straints as operators on the kinematical Hilbert space since we are dealing with a gauge system.
3.5 Spin-network Decomposition of Kinematical Hilbert Space
The kinematical Hilbert space 1
kin
for loop quantum gravity has been well dened. In this
subsection, it will be shown that 1
kin
can be decomposed into the orthogonal direct sum of 1-
dimensional subspaces and nd a basis, called spin-network basis, in the Hilbert space, which
consists of uncountably innite elements. So the kinematic Hilbert space is non-separable. In
the following, we will show the decomposition in three steps.
Spin-network Decomposition on a Single Edge
7
The proof of this conclusion depends on the compact support property of the smear functions f
i
(see [96] for
detail).
25
Given a graph consisting of only one edge e, which naturally associates with a group
S U(2) = X
(e)
, the elements of X
(e)
are the quantum connections only taking non-
trivial values on e. Then we consider the decomposition of the Hilbert space 1
(e)
=
L
2
(X
(e)
, d
(e)
) = L
2
(S U(2), d
H
), which is nothing but the space of square integrable
functions on the compact group S U(2) with the natural L
2
inner product. It is natural
to dene several operators on 1
(e)
. First, the so-called conguration operator

f (A(e))
whose operation on any in a dense domain of L
2
(S U(2), d
H
) is nothing but multipli-
cation by the function f (A(e)), i.e.,

f (A(e))(A(e)) := f (A(e))(A(e)),
where A(e) S U(2). Second, given any vector su(2), it generates left invariant vector
eld L
()
and right invariant vector eld R
()
on S U(2) by
L
()
(A(e)) :=
d
dt

t=0
(A(e) exp(t)),
R
()
(A(e)) :=
d
dt

t=0
( exp(t)A(e)),
for any function C
1
(S U(2)). Then one can dene the so-called momentum operators
on the single edge by

J
(L)
i
= iL
(
i
)
and

J
(R)
i
= iR
(
i
)
,
where the generators
i
su(2) constitute an orthonormal basis with respect to the
Killing-Cartan metric. The momentum operators have the well-known commutation re-
lation of the angular momentum operators in quantum mechanics:
[

J
(L)
i
,

J
(L)
j
] = i
k
i j

J
(L)
k
, [

J
(R)
i
,

J
(R)
j
] = i
k
i j

J
(R)
k
, [

J
(L)
i
,

J
(R)
j
] = 0.
Third, the Casimir operator on 1
e
can be expressed as

J
2
:=
i j

J
(L)
i

J
(L)
j
=
i j

J
(R)
i

J
(R)
j
. (24)
The decomposition of 1
e
= L
2
(S U(2), d
H
) is provided by the following Peter-Weyl
Theorem.
Theorem 3.5.1 [53]:
Given a compact group G, the function space L
2
(G, d
H
) can be decomposed as an or-
thogonal direct sum of nite dimensional Hilbert spaces, and the matrix elements of the
equivalence classes of nite dimensional irreducible representations of G form an orthog-
onal basis in L
2
(G, d
H
).
Note that a nite dimensional irreducible representation of G can be regarded as a matrix-
valued function on G, so the matrix elements are functions on G. Using this theorem, one
can nd the decomposition of the Hilbert space:
L
2
(S U(2), d
H
) =
j
[1
j
1

j
],
26
where j, labelling irreducible representations of S U(2), are the half integers, 1
j
denotes
the carrier space of the j-representation of dimension 2 j +1, and 1

j
is its dual space. The
basis e
j
m
e
j
n
in 1
j
1

j
maps a group element g S U(2) to a matrix
j
mn
(g), where
m, n = j, ..., j. Thus the space 1
j
1

j
is spanned by the matrix element functions
j
mn
of equivalent j-representations. Moreover, the spin-network basis can be dened.
Proposition 3.5.1 [56]
The systemof spin-network functions on 1
(e)
, consisting of matrix elements
j
mn
in nite
dimensional irreducible representations labelled by half-integers j, satises

J
2

j
mn
= j( j + 1)
j
mn
,

J
(L)
3

j
mn
= m
j
mn
,

J
(R)
3

j
mn
= n
j
mn
,
where j is called angular momentum quantum number and m, n = j, ..., j magnetic
quantum number. The normalized functions
_
2 j + 1
j
mn
form an orthonormal basis in
1
(e)
by the above theorem and
_
7
e

j
mn
d
e
=
1
2 j + 1

n
,
which is called the spin-network basis on 1
(e)
. So the Hilbert space on a single edge has
been decomposed into one dimensional subspaces.
Note that the system of operators

J
2
,

J
(R)
3
,

J
(L)
3
forms a complete set of commutable op-
erators in 1
(e)
. There is a cyclic vacuum state in the Hilbert space, which is the
( j = 0)-representation
(e)
=
j=0
= 1, representing that there is no geometry on the
edge.
Spin-network Decomposition on Finite Graph
Given a groupoid generated by a graph with N oriented edges e
i
and M vertices,
one can dene the conguration operators on the corresponding Hilbert space 1

=
L
2
(X

, d

) = L
2
(S U(2)
N
, d
N
H
) by

f (A(e
i
))(A(e
1
), ..., A(e
N
)) := f (A(e
i
))(A(e
1
), ..., A(e
N
)).
The momentum operators

J
i
(e,v)
associated with a edge e connecting a vertex v are dened
as

J
i
(e,v)
:= (1 ...

J
i
... 1),
where we set

J
i
=

J
(L)
i
if v = s(e) and

J
i
=

J
(R)
i
if v = t(e), so

J
i
(e,v)
= iX
(e,v)
i
. Note that

J
i
(e,v)
only acts nontrivially on the Hilbert space associated with the edge e. Then one
can dene a vertex operator associated with vertex v in analogy with the total angular
momentum operator via
[

J
v
]
2
:=
i j

J
v
i

J
v
j
,
27
where

J
v
i
:=

e at v

J
(e,v)
i
.
Obviously, 1

can be rstly decomposed by the representations on each edge e of as:


1

=
e
1
(e)
=
e
[
j
(1
e
j
1
e
j
)] =
j
[
e
(1
e
j
1
e
j
)]
=
j
[
v
(1
v=s(e)
j(s)
1
v=t(e)
j(t)
)],
where j := ( j
1
, ..., j
N
) assigns to each edge an irreducible representation of S U(2), in
the fourth step the Hilbert spaces associated with the edges are allocated to the vertexes
where these edges meet so that for each vertex v,
1
v=s(e)
j(s)

s(e)=v
1
e
j
and 1
v=t(e)
j(t)

t(e)=v
1
e
j
.
The group of gauge transformations g(v) S U(2) at each vertex is reducibly represented
on the Hilbert space 1
v=s(e)
j(s)
1
v=t(e)
j(t)
in a natural way. So this Hilbert space can be
decomposed as a direct sum of irreducible representation spaces via Clebsch-Gordon
decomposition:
1
v=s(e)
j(s)
1
v=t(e)
j(t)
=
l
1
v
j(v),l
.
As a result, 1

can be further decomposed as:


1

=
j
[
v
(
l
1
v
j(v),l
)] =
j
[
l
(
v
1
v
j(v),l
)]
j
[
l
1
,j,l
]. (25)
It can also be viewed as the eigenvector space decomposition of the commuting operators
[

J
v
]
2
(with eigenvalues l(l + 1)) and [

J
e
]
2

i j

J
e
i

J
e
j
. Note that l := (l
1
, ..., l
M
) assigns
to each vertex(objective) of an irreducible representation of S U(2). One may also
enlarge the set of commuting operators to further rene the decomposition of the Hilbert
space. Note that the subspace of 1

with l = 0 is Yang-Mills gauge invariant, since the


representation of gauge transformations is trivial.
Spin-network Decomposition of 1
kin
Since 1
kin
has the structure 1
kin
= (
[
1

), one may consider to construct it as a


direct sum of 1

by canceling some overlapping components. The construction is pre-


cisely described as a theorem below.
Theorem 3.5.2:
Consider assignments j = ( j
1
, ..., j
N
) to the edges of any groupoid [ and assign-
ments l = (l
1
, ..., l
M
) to the vertices. The edge representation j is non-trivial on each
edge, and the vertex representation l is non-trivial at each spurious
8
vertex, unless it is
the base point of a close analytic loop. Let 1

be the Hilbert space composed by the


8
A vertex v is spurious if it is bivalent and e e

is itself analytic edge with e, e

meeting at v.
28
subspaces 1
,j,l
(assigned the above conditions) according to Eq.(25). Then 1
kin
can be
decomposed as the direct sum of the Hilbert spaces 1

, i.e.,
1
kin
=
[
1

C.
Proof:
Since the representation on each edge is non-trivial, by denition of the inner product, it
is easy to see that 1

and 1

are mutual orthogonal if one of the groupoids and

has
at leat an edge e more than the other due to
_
7
e

j
mn
d
e
=
_
7
e
1
j
mn
d
e
= 0
for any j 0. Now consider the case of the spurious vertex. An edge e with j-
representation in a graph is assigned the Hilbert space 1
e
j
1
e
j
. Inserting a vertex v
into the edge, one obtains two edges e
1
and e
2
split by v both with j-representations,
which belong to a dierent graph. By the decomposition of the corresponding Hilbert
space,
1
e
1
j
1
e
1

j
1
e
2
j
1
e
2

j
= 1
e
1
j
(
l=0...2 j
1
v
l
) 1
e
2

j
,
the subspace for all l 0 are orthogonal to the space 1
e
j
1
e
j
, while the subspace for
l = 0 coincides with 1
e
j
1
e
j
since 1
v
l=0
= C and A(e) = A(e
1
)A(e
2
). This completes
the proof.
Since there are uncountably many graphs on , the kinematical Hilbert 1
kin
is non-
separable. We denote the spin-network basis in 1
kin
by
s
, s = ((s), j
s
, m
s
, n
s
) and
vacuum
kin

0
= 1, where

s
:=
_
eE((s))
_
2 j
e
+ 1
j
e
m
e
n
e
( j
e
0),
which form a orthonormal basis with the relation <
s

s
>
kin
=
ss
. And Cyl

(7)
Cyl(7) denotes the linear span of the spin network functions
s
for (s) = .
The spin-network basis can be used to construct the so-called spin network representation
of loop quantum gravity.
Denition 3.5.1: The spin-network representation is a vector space

1 of complex valued func-
tions

: S C; s

(s),
where S is the set of the labels s for the spin network states.

1 is equipped with the scalar
product
<

,

>:=

sS

(s)

(s)

29
between square summable functions.
The relation between the Hilbert spaces

1 and 1
kin
is claried by the following proposition
[146].
Proposition 3.5.2:
The spin-network transformation
T : 1
kin


1;

(s) :=<
s
, >
kin
is a unitary transformation with inverse
T
1
=

sS

(s)
s
.
Thus the connection representation and the spin-network representation are Fourier trans-
forms of each other, where the role of the kernel of the integral is played by the spin-network
basis. Note that, in the gauge invariant Hilbert space of loop quantum gravity which we will
dene later, the Fourier transform with respect to the gauge invariant spin-network basis is the
so-called loop transform, which leads to the unitary equivalent loop representation of the theory
[118][73][122].
To conclude this subsection, we show the explicit representation of elementary observables
on the kinematical Hilbert space 1
kin
. The action of canonical momentum operator

P
f
(S ) on
dierentiable cylindrical functions

Cyl

(7) can be expressed as

P
f
(S )

(A(e)
eE()
) =

vV()S
f
i
(v)[

e at v
(S, e)

J
(e,v)
i
]

(A(e)
eE()
)
=

vV()S
f
i
(v)[

J
(S,v)
i(u)


J
(S,v)
i(d)
]

(A(e)
eE()
), (26)
where

J
(S,v)
i(u)


J
(e
1
,v)
i
+ ... +

J
(e
u
,v)
i
,

J
(S,v)
i(d)


J
(e
u+1
,v)
i
+ ... +

J
(e
u+d
,v)
i
, (27)
for the edges e
1
, ..., e
u
lying above S and e
u+1
, ..., e
u+d
lying below S . And it was proved that
the operator

P
f
(S ) is essentially self-adjoint on 1
kin
[146]. On the other hand, it is obvious to
construct conguration operators by spin-network functions:

(A(e)
eE()
) :=
s
(A(e)
eE((s))
)

(A(e)
eE()
).
Note that

s
may change the graph, i.e.,

s
: Cyl

(7) Cyl
(s)
(7). So far, the elementary
operators of quantum kinematics have been well dened on 1
kin
.
30
3.6 Quantum Riemannian Geometry
The well-established quantum kinematics of loop quantum gravity is now in the same status
as Riemannian geometry before the appearance of general relativity and Einsteins equation,
giving general relativity mathematical foundation and oering living place to the Einstein equa-
tion. Instead of classical geometric quantities, such as length, area, volume etc., the quantities
in quantum geometry are operators on the kinematical Hilbert space 1
kin
, and their spectrum
serve as the possible values of the quantities in measurements. So far, the kinematical quan-
tum geometric operators constructed properly in loop quantum gravity include length operator
[141], area operator [125][21], two dierent volume operators [18][125][22],

Q operator [100],
etc.. Recently, a consistency check was proposed for the dierent regularizations of the volume
operator [77][78]. We thus will only introduce the volume operator dened by Ashtekar and
Lewandowski [22], which is shown to be correct in the consistency check.
First, we dene the area operator with respect to a 2-surface S by the elementary operators.
Given a closed 2-surface or a surface S with boundary, we can divide it into a large number N
of small area cells S
I
. Taking account of the classical expression of an area, we set the area of
the 2-surface to be the limit of the Riemannian sum
A
S
:= lim
N
[A
S
]
N
= lim
N

I=1
_
P
i
(S
I
)P
j
(S
I
)
i j
.
Then one can unambiguously obtain a quantum area operator from the canonical momentum
operators

P
i
(S ) smeared by constant functions. Given a cylindrical function

Cyl

(7)
which has second order derivatives, the action of the area operator on

is dened in the limit


by requiring that each area cell contains at most only one intersecting point v of the graph and
S as

A
S

:= lim
N
[

A
S
]
N

= lim
N

I=1
_

P
i
(S
I
)

P
j
(S
I
)
i j

.
The regulator N is easy to remove, since the result of the operation of the operator

P
i
(S
I
) does
not change when S
I
shrinks to a point. Since the renement of the partition does not aect the
result of action of [

A
S
]
N
on

, the limit area operator



A
S
, which is shown to be self-adjoint
[21], is well dened on 1
kin
and takes the explicit expression as:

A
S

= 4
2
p

vV(S )
_
(

J
(S,v)
i(u)


J
(S,v)
i(d)
)(

J
(S,v)
j(u)


J
(S,v)
j(d)
)
i j

,
where

J
(S,v)
i(u)
and

J
(S,v)
i(d)
have been dened in Eq.(27). It turns out that the nite linear combinations
of spin-network basis in 1
kin
diagonalizes

A
S
with eigenvalues given by nite sums,
a
S
= 4
2
p

v
_
2j
(u)
v
( j
(u)
v
+ 1) + 2 j
(d)
v
( j
(d)
v
+ 1) j
(u+d)
v
( j
(u+d)
v
+ 1), (28)
where j
(u)
, j
(d)
and j
(u+d)
are arbitrary half-integers subject to the standard condition
j
(u+d)
j
(u)
j
(d)
, j
(u)
j
(d)
+ 1, ..., j
(u)
+ j
(d)
. (29)
31
Hence the spectrum of the area operator is fundamentally pure discrete, while its continuum
approximation becomes excellent exponentially rapidly for large eigenvalues. However, in fun-
damental level, the area is discrete and so is the quantum geometry. One can see that the
eigenvalue of

A
S
does not vanish even in the case where only one edge intersects the surface at
a single point, whence the quantum geometry is distributional.
The form of Ashtekar and Lewandowskis volume operator was introduced for the rst time
in [18], and its detailed properties are discussed in [22]. Given a region R with a xed coordinate
system x
a

a=1,2,3
in it, one can introduce a partition of R in the following way. Divide R into
small volume cells C such that, each cell C is a cube with coordinate volume less than and
two dierent cells only share the points on their boundaries. In each cell C, we introduce three
2-surfaces s = (S
1
, S
2
, S
3
) such that x
a
is constant on the surface S
a
. We denote this partition
(C, s) as /

. Then the volume of the region R can be expressed classically as


V
s
R
= lim
0

C
_
q
C,s
,
where
q
C,s
=
()
3
3!

i jk

abc
P
i
(S
a
)P
j
(S
b
)P
k
(S
c
).
This motivates us to dene the volume operator by naively changing P
i
(S
a
) to

P
i
(S
a
):

V
s
R
= lim
0

C
_
q
C,s
,
q
C,s
=
()
3
3!

i jk

abc

P
i
(S
a
)

P
j
(S
b
)

P
k
(S
c
).
Note that, given any cylindrical function

Cyl

(7), we require the vertexes of the graph


to be at the intersecting points of the triples of 2-surfaces s = (S
1
, S
2
, S
3
) in corresponding
cells. Thus the limit operator will trivially exist due to the same reason in the case of the area
operator. However, the volume operator dened here depends on the choice of orientations for
the triples of surfaces s = (S
1
, S
2
, S
3
), or essentially, the choice of coordinate systems. So it
is not uniquely dened. Since, for all choice of s = (S
1
, S
2
, S
3
), the resulting operators have
correct semi-classical limit, one settles up the problem by averaging dierent operators labelled
by dierent s [22]. The process of averaging removes the freedom in dening the volume
operator up to an overall constant
0
. The resulting self-adjoint operator acts on any cylindrical
function

Cyl

(7) as

V
R

=
0

vV()
_
q
v,

,
where
q
v,
= (8
2
p
)
3
1
48

e,e

,e

at v

i jk
(e, e

, e

)

J
(e,v)
i

J
(e

,v)
j

J
(e

,v)
k
,
32
here (e, e

, e

) sgn(
abc
e
a
e
b
e
c
)
v
with e
a
as the tangent vector of edge e and
abc
as the
orientation of . The only unsatisfactory point in the present volume operator is the choice
ambiguity of
0
. However, fortunately, the most recent discussion shows that the overall un-
determined constant
0
can be xed to be

6 by the consistency check between the volume
operator and the triad operator [77][78].
33
4 Implementation of Quantum Constraints
After constructing the kinematical Hilbert space 1
kin
of loop quantum gravity, one should im-
plement the constraints on it to obtain the physical Hilbert space which encodes the complete
information of quantum dynamics of general relativity, since the Hamiltonian of general rela-
tivity is a linear combination of the constraints. Recalling the constraints (7) in the Hamiltonian
formalism and the Poission algebra (9) among them, the subalgebra generated by the Gauss
constraints g() forms a Lie algebra and a 2-sided ideal in the constraints algebra. So in this
section, we rst solve the Gaussian constraints independently of the other two kinds of con-
straints and nd the solution space 1
G
, which is constituted by internal gauge invariant quan-
tum states. Then, although the subalgebra generated by the dieomorphism constraints is not
an ideal in the constraint algebra, we still would like to solve them independently of the scalar
constraints for technical convenience. After that, the quantum operator corresponding to the
Hamiltonian constraint(scalar constraint) is dened on the kinematical Hilbert space, and we
will also discuss an alterative for the implementation of the scalar constraint, which is called
the master constraint programme by modifying the classical constraint algebra.
4.1 Solutions of Quantum Gaussian Constraint
Recall that the classical expression of Gauss constraints reads
g() =
_

d
3
x
i
D
a

P
a
i
=
_

d
3
x

P
a
i
D
a

i
P(D),
where D
a

i
=
a

i
+
i
jk
A
j
a

k
. As the situation of triad ux, the Gauss constraints can be
dened as cylindrically consistent vector elds Y
D
on 7, which act on any cylindrical function
f

Cyl

(7) by
Y
D
f

(A(e)
eE()
) := P(D), f

(A(e)
eE()
).
Then the Gauss constraint operator can be dened in analogy with the momentum operator,
which acts on f

as:

g() f

(A(e)
eE()
) := iY
D
f

(A(e)
eE()
)
=

vV()
[
i
(v)

J
v
i
] f (A(e)
eE()
),
which is the generator of internal gauge transformations on Cyl

(7). The kernel of the opera-


tor is easily obtained in terms of the spin-network decomposition, which is the internal gauge
invariant Hilbert space:
1
G
=
,j
1

,j,l=0
C.
One then naturally gets the gauge invariant spin-network basis T
s
, s = ((s), j
s
, i
s
) in 1
G
via a
group averaging technique at each vertex[126][25][32](we will call T
s
spin-network state in the
following context):
T
s=(,j,i)
=

vV()
i
v

eE()

j
e
(A(e)), ( j
e
0)
34
assigning a non-trivial spin representation j on each edge and a invariant tensor i (intertwiner) on
each vertex. We denote the vector space of nite linear combinations of vacuum state and gauge
invariant spin-network states Cyl(7/g), which is dense in 1
G
. And Cyl

(7/g) Cyl(7/g)
denotes the linear span of the gauge invariant spin network functions T
s
for (s) = . All
Yang-Mills gauge invariant operators are well dened on 1
G
. However, the condition of acting
on gauge invariant states often changes the structure of the spectrum of quantum geometric
operators. For the area operator, the spectrumdepends on certain global properties of the surface
S (see [20][21] for details). For the volume operators, a non-zero spectrum arises from at least
4-valent vertices.
4.2 Solutions of Quantum Dieomorphism Constraint
Unlike the strategy in solving Gaussian constraint, one cannot dene an operator for the quan-
tum dieomorphism constraint as the innitesimal generator of nite dieomorphism trans-
formations (unitary operators since the measure is dieomorphism invariant) represented on
1
kin
. The representation of nite dieomorphisms is a family of unitary operators

U

acting on
cylindrical functions

by

:=

, (30)
for any spatial dieomorphism on . An 1-parameter subgroup
t
in the group of spatial
dieomorphisms is then represented as an 1-parameter unitary group

U

t
on 1
kin
. However,

t
is not weakly continuous, since the subspaces 1

()
and 1

(
t
)
are orthogonal to each
other no matter how small the parameter t is. So one always has
<

>
kin
<

>
kin
=<

>
kin
0, (31)
even in the limit when t goes to zero. Therefore, the innitesimal generator of

U

t
does not
exist. In the strategy to solve the dieomorphism constraint, due to the Lie algebra structure of
dieomorphism constraints subalgebra, the so-called group averaging technique is employed.
We now outline the procedure. First, given a colored graph (a graph and a cylindrical function
living on it), one can dene the group of graph symmetries GS

by
GS

:= Di f f

/TDi f f

,
where Di f f

is the group of all dieomorphisms preserving the colored , and TDi f f

is the
group of dieomorphisms which trivially acts on . We dene a projection map by averaging
with respect to GS

to obtain the subspace in Cyl

which is invariant under the transformation


of GS

P
Di f f ,

:=
1
n

GS

,
for all cylindrical functions

()
, where n

is the number of the nite elements of GS

.
Second, we average with respect to all remaining dieomorphisms which move the graph .
For each cylindrical function

Cyl

(7/g), there is an element (

) associated to it in the
35
algebraic dual space Cyl

of Cyl(7/g), which acts on any cylindrical function

Cyl

(7/g)
as:
(

)[

] :=

Di f f ()/Di f f

<

U


P
Di f f ,

>
kin
.
It is well dened since, for any given graph

, only nite terms are non-zero in the summation.


It is easy to verify that (

) is invariant under the group action of Di f f (), since


(

)[

U

] = (

)[

].
Thus we have dened a rigging map : Cyl(7/g) Cyl

Di f f
, which maps every cylindrical
function to a dieomorphism invariant one, where Cyl

Di f f
is spanned by vacuum state T
0
= 1
and rigged spin-network functions T
[s]
(T
s
), [s] = ([], j, i) associated with dieomor-
phism classes [] of graphs . Moreover a Hermitian inner product can be dened on Cyl

Di f f
by the natural action of the algebraic functional:
< (

)(

) >
Di f f
:= (

)[

].
The dieomorphism invariant Hilbert space 1
Di f f
is dened by the completion of Cyl

Di f f
with
respect to the above inner product < >
Di f f
. The dieomorphism invariant spin-network func-
tions T
[s]
form an orthonormal basis in 1
Di f f
. Finally, we have obtained the general solutions
invariant under both Yang-Mills gauge transformations and spatial dieomorphisms.
In general relativity, the problem of observables is a subtle issue due to the dieomorphism
invariance [116][119][120]. Now we discuss the operators as dieomorphism invariant ob-
servables on 1
Di f f
. We call an operator

O [(1
kin
) a strong observable if and only if

U
1

O

U

=

O, Di f f (). We call it a weak observable if and only if

O leaves 1
Di f f
invariant. Then it is easy to see that a strong observable

O must be a weak one. One notices that
a strong observable

O can rst be dened on 1
Di f f
by its dual operator

O

as
(

O

Di f f
)[] :=
Di f f
[

O],
then one gets
(

O

Di f f
)[

U

] =
Di f f
[

O

U

] =
Di f f
[

U
1

O

U

] = (

O

Di f f
)[],
for any
Di f f
1
Di f f
and 1
kin
. Hence

O

Di f f
is also dieomorphism invariant. In
addition, a strong observable also has the property of

O

) = (

O

) since,

1
kin
,
<

O

)(

) >
Di f f
= (

O

))[

] = (

)[

O

]
=

Di f f ()/Di f f

<

U


P
Di f f ,

>
kin
=
1
n

Di f f ()/Di f f

GS

<

U

>
kin
=
1
n

Di f f ()/Di f f

GS

<

U

>
kin
= < (

O

)(

) >
Di f f
.
36
Note that the Hilbert space 1
Di f f
is still non-separable if one considers the C
n
dieomorphisms
with n > 0. However, if one extends the dieomorphisms to be semi-analytic dieomotphisms,
i.e. homomorphisms that are analytic dieomorphisms up to nite isolated points (which can
be viewed as an extension of the classical concept to the quantum case), the Hilbert space 1
Di f f
would be separable [69][20].
4.3 Hamiltonian Constraint Operator
In the following, we consider the issue of scalar constraint in loop quantum gravity. One may
rst construct a Hamiltonian constraint (scalar constraint) operator in 1
kin
or 1
Di f f
, then at-
tempt to nd the physical Hilbert space 1
phys
by solving the quantum Hamiltonian constraint.
However, diculties arise here due to the special role played by the scalar constraints in the
constraint algebra (9). First, the scalar constraints do not form a Lie subalgebra. Hence the
strategy of group averaging cannot be used directly on 1
kin
for them. Second, modulo the
Gaussian constraint, there is still a structure function in the Poisson bracket between two scalar
constraints:
S(N), S(M) = +((N
b
M M
b
N)q
ab
), (32)
which raises the danger of quantum anomalyies in quantization. Moreover, the dieomorphism
constraints do not form an ideal in the quotient constraint algebra modulo the Gaussian con-
straints. This fact results in that the scalar constraint operator cannot be well dened on 1
Di f f
,
as it does not commute with the dieomorphism transformations

U

. Thus the previous con-


struction of 1
Di f f
does not appear very useful for the nal construction of 1
phys
, which is our
nal goal. However, one may still rst try to construct a Hamiltonian constraint operator in 1
kin
for technical convenience.
We recall the classical expression of Hamiltonian constraint:
S(N) :=

2
2
_

d
3
xN

P
a
i

P
b
j
_
det q
[
i j
k
F
k
ab
2(1 +
2
)K
i
[a
K
j
b]
]
= S
E
(N) 2(1 +
2
)7(N). (33)
The main idea of the construction is to rst express S(N) in terms of the combination of Poisson
brackets between the variables which have been represented as operators on 1
kin
, then replace
the Poisson brackets by canonical commutators between the operators. We will use the volume
functional for a region R and the extrinsic curvature functional dened by:
K :=
_

d
3
x

P
a
i
K
i
a
.
A key trick here is to consider the following classical identity of the co-triad e
i
a
(x) [134]:
e
i
a
(x) =
()
2
2

abc

i jk

P
b
j

P
c
k
_
det q
(x) =
2

A
i
a
(x), V
R
,
37
where V
R
is the volume functional for a neighborhood R containing x. And the expression of
the extrinsic curvature 1-form K
i
a
(x):
K
i
a
(x) =
1

A
i
a
(x), K.
Note that K can be expressed by a Poisson bracket between the constant-smeared Euclidean
Hamiltonian constraint and the total volume of the space :
K =
2
S
E
(1), V

. (34)
Thus one can obtain the equivalent classical expressions of S
E
(N) and 7(N) as:
S
E
(N) =

2
2
_

d
3
xN

P
a
i

P
b
j
_
det q

i j
k
F
k
ab
=
2

d
3
xN(x)
abc
Tr(F
ab
(x)A
c
(x), V
R
x
),
7(N) =

2
2
_

d
3
xN

P
a
i

P
b
j
_
det q
K
i
[a
K
j
b]
=
2

3
_

d
3
xN(x)
abc
Tr(A
a
(x), KA
b
(x), KA
c
(x), V
R
x
),
where A
a
= A
i
a

i
, F
ab
= F
i
ab

i
, Tr represents the trace of the Lie algebra matrix, and R
x

denotes an arbitrary neighborhood of x . In order to quantize the Hamiltonian constraint
as a well-dened operator on 1
kin
, one has to express the classical formula of S(N) in terms
of holonomies A(e) and other variables with clear quantum analogs. As a rst attempt [134],
this can be realized by introducing a triangulation T(), where the parameter describes how
ne the triangulation is, and the triangulation will ll out the spatial manifold when
0. Given a tetrahedron T(), we use s
i
()
i=1,2,3
to denote the three outgoing oriented
segments in with a common beginning point v() = s(s
i
()), and use a
i j
() to denote the
arc connecting the end points of s
i
() and s
j
(). Then several loops
i j
() are formed by

i j
() := s
i
() a
i j
() s
j
()
1
. Thus we have the identities:

_
s
i
()
A
a
s
a
i
(), V
R
v()
= A(s
i
())
1
A(s
i
()), V
R
v()
+ o(),

_
s
i
()
A
a
s
a
i
(), K = A(s
i
())
1
A(s
i
()), K + o(),
_
P
i j
F
ab
(x) =
1
2
A(
i j
())
1

1
2
A(
i j
()) + o(
2
),
where P
i j
is the plane with boundary
i j
. Note that the above identities are constructed by taking
account of internal gauge invariance of the nal formula of Hamiltonian constraint operator. So
we have the regularized expression of S(N) by the Riemannian sum [134]:
S

E
(N) =
2
3
2

T()
N(v())
i jk

38
Tr(A(
i j
())
1
A(s
k
())
1
A(s
k
()), V
R
v()
),
7

(N) =

2
6
4

T()
N(v())
i jk

Tr(A(s
i
())
1
A(s
i
()), KA(s
j
())
1
A(s
j
()), K
A(s
k
())
1
A(s
k
()), V
R
v()
),
S

(N) = S

E
(N) 2(1 +
2
)7

(N), (35)
such that lim
0
S

(N) = S(N). It is clear that the above regulated formula of S(N) is invari-
ant under internal gauge transformations. Since all constituents in the expression have clear
quantum analogs, one can quantize the regulated Hamiltonian constraint as an operator on 1
kin
(or 1
G
) by replacing them by the corresponding operators and Poisson brackets by canonical
commutators, i.e.,
A(e)

A(e), V
R


V
R
, ,
[ , ]
i
,
and K

K

2
i
[

S

E
(1),

V

].
Removing the regulator by 0, it turns out that one can obtain a well-dened limit operator
on 1
kin
(or 1
G
) with respect to a natural operator topology.
Now we begin to construct the Hamiltonian constraint operator in analogy with the classical
expression (57). All we should do is dene the corresponding regulated operators on dierent
1

separately, then remove the regulator so that the limit operator is dened on 1
kin
(or 1
G
)
cylindrically consistently. In the following, given a vertex and three edges intersecting at the
vertex in a graph of

Cyl

(7/g), we construct one triangulation of the neighborhood


of the vertex adapted to the three edges. Then we average with respect to the triples of edges
meeting at the given vertex. Precisely speaking, one can make the triangulations T() with the
following properties [134][146].
The chosen triple of edges in the graph is embedded in a T() for all , so that the vertex
v of where the three edges meet coincides with a vertex v() in T().
For every triple of segments (e
1
, e
2
, e
3
) of such that v = s(e
1
) = s(e
2
) = s(e
3
), there is a
tetrahedra T() such that v = v() = s(s
i
()), and s
i
() e
i
, i = 1, 2, 3. We denote
such a tetrahedra as
0
e
1
,e
2
,e
3
.
For each tetrahedra
0
e
1
,e
2
,e
3
one can construct seven additional tetrahedron

e
1
,e
2
,e
3
, =
1, ..., 7, by backward analytic extensions of s
i
() so that U
e
1
,e
2
,e
3
:=
7
=0

e
1
,e
2
,e
3
is a neigh-
borhood of v.
The triangulation must be ne enough so that the neighborhoods U(v) :=
e
1
,e
2
,e
3
U
e
1
,e
2
,e
3
(v)
are disjoint for dierent vertices v and v

of . Thus for any open neighborhood U

of the
graph , there exists a triangulation T() such that
vV()
U(v) U

.
The distance between a vertex v() and the corresponding arcs a
i j
() is described by the
parameter . For any two dierent and

, the arcs a
i j
(

) and a
i j
(

) with respect to
one vertex v() are semi-analytically dieomorphic with each other.
39
With the triangulations T(), the integral over is replaced by the Riemanian sum:
_

=
_
U

+
_
U

,
_
U

vV()
_
U(v)
+
_
U

v
U(v)
,
_
U(v)
=
1
E(v)

e
1
,e
2
,e
3
[
_
U
e
1
,e
2
,e
3
(v)
+
_
U(v)U
e
1
,e
2
,e
3
,(v)
],
where n(v) is the valence of the vertex v = s(e
1
) = s(e
2
) = s(e
3
), and E(v)
_
n(v)
3
_
de-
notes the binomial coecient which comes from the averaging with respect to the triples
of edges meeting at given vertex v. One then observes that
_
U
e
1
,e
2
,e
3
(v)
= 8
_

0
e
1
,e
2
,e
3
(v)
in the limit 0.
The triangulations for the regions
U(v) U
e
1
,e
2
,e
3
(v),
U


vV()
U(v),
U

, (36)
are arbitrary. These regions do not contribute to the construction of the operator, since the
commutator term [A(s
i
()), V
R
v()
]

vanishes for all tetrahedron in the regions (36).


Thus we nd the regulated expression of Hamiltonian constraint operator with respect to the
triangulations T() as [134]

E,
(N) =
16
3i
2

vV()
N(v)
E(v)

v()=v

i jk

Tr(

A(
i j
())
1

A(s
k
())
1
[

A(s
k
()),

V
U

v
]),

(N) =
4

2
3i
3

vV()
N(v)
E(v)

v()=v

i jk

Tr(

A(s
i
())
1
[

A(s
i
()),

K

]

A(s
j
())
1
[

A(s
j
()),

K

A(s
k
())
1
[

A(s
k
()),

V
U

v
]),

(N)

= [

S

E,
(N) 2(1 +
2
)

7

(N)]

vV()
N(v)

S

,
for any cylindrical function

Cyl

(7/g) is a nite linear combination of spin-network states


T
s
with (s) = .
40
By construction, the operation of

S

(N) on any

Cyl

(7/g) is reduced to a nite


combination of that of

S

v
with respect to dierent vertices of . Hence, for each > 0,

S

(N)
is a well-dened internal gauge invariant and dieomorphism covariant operator on Cyl(7/g).
The last step is to remove the regulator by taking the limit 0. However, the action of
the Hamiltonian constraint operator on

adds arcs a
i j
() with a
1
2
-representation with respect
to each v() of
9
, i.e. the action of

S

(N) on cylindrical functions is graph-changing. Hence


the operator does not converge with respect to the weak operator topology in 1
kin
when 0,
since dierent 1

()
with dierent graphs are mutually orthogonal. Thus one has to dene a
weaker operator topology to make the operator limit meaningful. By physical motivation and
the naturally available Hilbert space 1
Di f f
, the convergence of

S

(N) holds with respect to the


so-called Uniform Rovelli-Smolin Topology [124], where one denes

S

(N) to converge if and


only if
Di f f
[

S

(N)] converge for all


Di f f
Cyl

Di f f
and Cyl(7/g). Since the value
of
Di f f
[

S

(N)] is actually independent of by the fth property of the triangulations, the


sequence converges to a nontrivial result
Di f f
[

S

0
(N)] with arbitrary xed
0
> 0. Thus we
have dened a dieomorphism covariant, densely dened, closed but non-symmetric operator,

S(N) = lim
0

S

(N) =

S

0
(N), on 1
kin
(or 1
G
) representing the Hamiltonian constraint.
Moreover, a dual Hamiltonian constraint operator

S

(N) is also dened on Cyl

depending on
a specic value of
(

S

(N))[] := [

S

(N)],
for all Cyl

and Cyl(7/g). For


Di f f
Cyl

Di f f
Cyl

, one gets
(

S

(N)
Di f f
)[] =
Di f f
[

S

(N)].
which is independent of the value of .
Several remarks on the Hamiltonian constraint operator are listed in the following.
Finiteness of

S(N) on 1
kin
In ordinary quantum eld theory, the continuous quantum eld is only recovered when
one lets lattice spacing to approach zero, i.e., takes the continuous cut-o parameter to its
continuous limit. However, this will produce the well-known innities in quantum eld
theory and make the Hamiltonian operator ill-dened on the Fock space. So it seems
surprising that our operator

S(N) is still well dened, when one takes the limit 0
with respect to the Uniform Rovelli-Smolin Topology so that the triangulation goes to the
continuum. The reason behind it is that the cut-o parameter is essentially noneective
due to the dieomorphism invariance of our quantum eld theory. This is why there
is no UV divergence in the background independent quantum gauge eld theory with
dieomorphism invariance. On the other hand, from a convenient viewpoint, one may
think the Hamiltonian constraint operator as an operator dually dened on a dense domain
in 1
Di f f
. However, we will see that the dual Hamiltonian constraint operator cannot leave
1
Di f f
invariant.
9
The Hamiltonian constraint operator depends indeed on the choice of the representation j on the arcs a
i j
(),
which is known as one of the regularization ambiguities in the construction of quantum dynamics. For the simplic-
ity of the theory, one often choose the lowest label of representation j =
1
2
.
41
Implementation of Dual Quantum Constraint Algebra
One important task is to check whether the commutator algebra (quantum constraint al-
gebra) among the corresponding quantum operators of constraints both physically and
mathematically coincides with the classical constraint algebra by substituting quantum
constraint operators to classical constraint functionals and commutators to Poisson brack-
ets. Here the quantum anomaly has to be avoided in the construction of constraint opera-
tors (see the discussion for Eq.(13)). First, the subalgebra of the quantum dieomorphism
constraint algebra is free of anomaly by construction:


U
1


U
1

=

U

1 ,
which coincides with the exponentiated version of the Poisson bracket between two dif-
feomorphism constraints generating the transformations ,

Di f f ().
Second, the quantum constraint algebra between the dual Hamiltonian constraint opera-
tor S

(N) and the nite dieomorphism transformation



U

on dieomorphism-invariant
states coincides with the classical Poisson algebra between +(

N) and S(M). Given a


cylindrical function

associated with a graph and the triangulations T() adapted to


the graph , the triangulations T( ) T() are compatible with the graph .
Then we have by denition:
( ([

S(N),

U

])

Di f f
)[

]
= ([

S

(N),

U

]
Di f f
)[

]
=
Di f f
[

S

(N)

(N)

]
=

vV()
N(v)
Di f f
[

S

] N( v)
Di f f
[

S

]
=

vV()
[N(v) N( v)]
Di f f
[

S

]
= (

S

(N

N)
Di f f
)[

]. (37)
Thus there is no anomaly. However, Eq.(37) also explains why the Hamiltonian constraint
operator

S(N) cannot leave 1
Di f f
invariant.
Third, we compute the commutator between two Hamiltonian constraint operators. No-
tice that
[

S(N),

S(M)]

vV()
[M(v)

S(N) N(v)

S(M)]

S

vV()

V(

)
[M(v)N(v

) N(v)M(v

)]

S

,
where

is the graph changed from by the action of



S(N) or

S(M), which adds the arcs
a
i j
() on , T() is the triangulation adapted to and T(

) adapted to

. Since the newly


42
added vertices by

S

v
is planar, they will never contributes the nal result. So one has
[

S(N),

S(M)]

v,v

V(),vv

[M(v)N(v

) N(v)M(v

)]

S

=
1
2

v,v

V(),vv

[M(v)N(v

) N(v)M(v

)][

S

v


S

v

S

v
]

=
1
2

v,v

V(),vv

[M(v)N(v

) N(v)M(v

)][(

U

,v


U

v,v

)

S

v
]

,
(38)
where we have used the facts that [

S

v
,

S

] = 0 for v v

and there exists a dieomorphism

v,v
such that

S

v
=

U

,v

v
. Obviously, we have in the Uniform Rovelli-Smolin
Topology
([

S(N),

S(M)])

Di f f
= 0
for all
Di f f
Cyl

Di f f
. As we have seen in classical expression Eq.(32), the Poisson
bracket of any two Hamiltonian constraints is given by a generator of the dieomrophism
transformations. Therefore it is mathematically consistent with the classical expression
that two Hamiltonian constraint operators commute on dieomorphism invariant states,
as it is presented above. However, as it has been discussed in [72][95], the domain of
dual Hamiltonian constraint operator can be extended to a slightly larger space (habitat)
in Cyl

, whose elements are not necessary dieomorphism invariant. And it turns out that
the commutator between two Hamiltonian constraint operators continues to vanish on the
habitat, which seems to be problematic. Fortunately, the quantum operator corresponding
to the right hand side of classical Poisson bracket (32) also annihilates every state in the
habitat [72], so the quantum constraint algebra is consistent at this level. But it is not
clear that whether the quantum constraint algebra, especially the commutator between
two Hamiltonian constraint is consistent with the classical one (32) on some larger space
in Cyl

containing more dieomorphism variant states


10
. On the other hand, more works
on the semi-classical analysis are also needed to test the classical limit of Eq.(38) and
commutation relation (32). The way to do it is looking for some proper semi-classical
states for calculating the classical limit of the operators. But due to the graph-changing
property of the Hamiltonian constraint operator, the semi-classical analysis for the Hamil-
tonian constraint operator and the quantum constraint algebra is still an open issue so far.
General Regularization Scheme of the Hamiltonian Constraint
In [20], a general scheme of regulation is introduced for the quantization of the Hamilto-
nian constraint, and includes Thiemanns regularization we introduced above as a specic
choice. Such a general regularization can be summarized as follows: rst, we assign a
10
However, some scholars disagree with such an argument involving the habitat and consider the habitat to be
unphysical and completely irrelevant (see, e.g. Ref.[151]).
43
partition of into cells of arbitrary shape. In every cell of the partition we dene edges
s
J
, J = 1, ..., n
s
and loops
i
, i = 1, ..., n

, where n
s
, n

may be dierent for dierent cells.


We use to represent the scale of the cell Then x an arbitrary chosen representation
of S U(2). This structure is called a permissible classical regulator if the regulated
Hamiltonian constraint expression with respect to this partition has correct limit when
0.
Second, we assign the dieomorphism covariant property and let the partition adapted to
the choice of the graph. That is, given a cylindrical function

Cyl
3

(7/g), we make
the partition suciently rened that every vertex v V() is contained in exact one cell
of the partition. And if (, v) is dieomorphic to (

, v

) then, for every and

, the quin-
tuple (, v, , (s
J
), (
i
)) is dieomorphic to the quintuple (

, v

, (s

J
), (

i
)), where
and

are the cells in the partitions with respect to and

respectively, containing v
and v

respectively.
As a result, the Hamiltonian constraint operator in this general regularization scheme is
expressed as:

E,
(N) =

vV()
N(v)
i
2

i,J
C
iJ
Tr(([A(
i
)] [A(
1
i
)])[A(s
1
J
)][[A(s
J
)],

V
U

v
]),

(N) =

vV()
iN(v)

I,J,K
T
I JK
Tr([A(s
1
I
)][[A(s
I
)],

K][A(s
1
J
)][[A(s
J
)],

K]
[A(s
1
K
)][[A(s
K
)],

V
U

v
]),

(N)

= [

S

E,
(N) 2(1 +
2
)

7

(N)]

,
where C
iJ
and T
I JK
are xed constants independent of the value of , the values of them
are determined such that the above expressions have correct classical limits. After remov-
ing the regulator via dieomorphism invariance the same as we did above, we obtain a
well-dened dieomorphism covariant operator on 1
kin
(or 1
G
) in the sense of the Uni-
form Rovelli-Smolin Topology, or dual-dene the operator on some suitable domain in
Cyl

. Note that such a general scheme of construction exhibits that there is a great deal
of freedom in choosing the regulators, so that there are considerable ambiguities in our
quantization for seeking a proper quantum dynamics for gravity, which is also an open
issue today.
4.4 Master Constraint Programme
Although the Hamiltonian constraint operator introduced above is densely dened on 1
kin
and
dieomorphism covariant, there are still several unsettled problems which are listed below.
It is unclear whether the commutator between two Hamiltonian constraint operators re-
produces the classical Poisson bracket between two Hamiltonian constraints. Hence it is
unclear if the quantum Hamiltonian constraint produces the correct quantum dynamics
with correct classical limit [72][95].
44
The dual Hamiltonian constraint operator does not leave the Hilbert space 1
Di f f
invariant.
Thus the inner product structure of 1
Di f f
cannot be employed in the construction of
physical inner product.
Classically the collection of Hamiltonian constraints does not form a Lie algebra. So one
cannot employ group averaging strategy in solving the Hamiltonian constraint quantum
mechanically, since the strategy depends crucially on group structure.
One may see that all above issues come from the properties of the constraint algebra at classical
level. However, if one could construct an alternative classical constraint algebra, giving the
same constraint phase space, which is a Lie algebra (no structure functions), where the subalge-
bra of dieomorphism constraints forms an ideal, then the programme of solving the constraints
would be in a much better position. Such a constraint Lie algebra was rst introduced by Thie-
mann in [149]. The central idea is to introduce the master constraint:
M :=
1
2
_

d
3
x

C(x)
2
_
det q(x)
, (39)
where

C(x) is the scalar constraint in Eq.(8). One then gets the master constraint algebra:
+(

N), +(

) = +([

N,

N

]),
+(

N), M = 0,
M, M = 0.
The master constraint programme has been well tested in various examples [63][64][65]
[66][67]. In the following, we extend the dieomorphism transformations such that the Hilbert
space 1
Di f f
is separable. This separability of 1
Di f f
and the positivity and the dieomorphism
invariance of M will be working together properly and provide us with powerful functional
analytic tools in the programme to solve the constraint algebra quantum mechanically. The
regularized version of the master constraint can be expressed as
M

:=
1
2
_

d
3
y
_

d
3
x

(x y)

C(y)
_
V
U

C(x)
_
V
U

x
.
Introducing a partition / of the 3-manifold into cells C, we have an operator

H

C
acting on
any cylindrical function f

Cyl

(7/g) in 1
G
as

C
f

vV()

C
(v)
E(v)

v()=v

h
,
v
f

, (40)
via a family of state-dependent triangulations T() on as we did in the last section, where

C
(v) is the characteristic function of the cell C(v) containing a vertex v of the graph , and the
expression of

h
,
v
reads

h
,
v
=
16
3i
2

i jk
Tr(

A(
i j
())
1

A(s
k
())
1
[

A(s
k
()),
_

V
U

v
])
+2(1 +
2
)
4

2
3i
3

i jk
Tr(

A(s
i
())
1
[

A(s
i
()),

K

A(s
j
())
1
[

A(s
j
()),
_

V
U

v
]

A(s
k
())
1
[

A(s
k
()),

K

]). (41)
45
Note that

h
,
v
is similar to that involved in the regulated Hamiltonian constraint operator in
the last section, while the only dierence is that now the volume operator is replaced by its
quare-root in Eq.(41). Hence the action of

H

C
on f

adds arcs a
i j
() with 1/2-representation
with respect to each v() of . Thus, for each > 0,

H

C
is a Yang-Mills gauge invariant and
dieomorphism covariant operator dened on Cyl(7/g). The family of such operators can give
a limit operator

H
C
densely dened on 1
G
by the uniform Rovelli-Smollin topology. Then a
master constraint operator,

M, acting on any
Di f f
Cyl

Di f f
can be dened as [88]
(

M
Di f f
)[ f

] := lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

], (42)
for any f

is a nite linear combination of spin-network function. Note that



H

C
(

H

C
)

is
also a nite linear combination of spin-network functions on an extended graph with the same
skeleton of , hence the value of (

M
Di f f
)[ f

] is nite for any given


Di f f
. Thus

M
Di f f
lies in
the algebraic dual of the space of cylindrical functions. Furthermore, we can show that

Mleaves
the dieomorphism invariant distributions invariant. For any dieomorphism transformation
on ,
(

U


M
Di f f
)[ f

] = lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

]
= lim
/;,

Di f f
[

U

C/
1
2

H

1
()

1
(C)
(

H

1
(

1
(C)
)

]
= lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

], (43)
where in the last step, we used the fact that the dieomorphism transformation leaves the
partition invariant in the limit / and relabel (C) to be C. So we have the result
(

U


M
Di f f
)[ f

] = (

M
Di f f
)[ f

]. (44)
So given a dieomorphism invariant spin-network state T
[s]
, the result state

MT
[s]
must be a
dieomorphism invariant element in the algebraic dual of Cyl(7/g), which means that

MT
[s]
=

[s
1
]
c
[s
1
]
T
[s
1
]
,
then
lim
/;,

0
T
[s]
[

C/
1
2

H

C
(

H

C
)

T
s
2
] =

[s
1
]
c
[s
1
]
T
[s
1
]
[T
s
2
],
where the cylindrical function
_
C/
1
2

H

C
(

H

C
)

T
s
2
is a nite linear combination of spin-network
functions on some graphs

with the same skeleton of (s
2
) up to nite number of arcs. Hence
xing the dieomorphism equivalence class [s], only for spin-networks s
2
lies in nite number
of dieomorphism equivalence class the left hand side of the last equation is non-zero. So there
46
are also only nite number of classes [s
1
] in the right hand side such that c
[s
1
]
is non-zero. As a
result,

MT
[s]
is a nite linear combination of dieomorphism invariant spin-network states and
lies in the Hilbert space of dieomorphism invariant states 1
Di f f
for any [s]. And

M is densely
dened on 1
Di f f
.
Given two dieomorphism invariant spin-network functions T
[s
1
]
and T
[s
2
]
, one can give the
matrix elements of

M as [88][89]
< T
[s
1
]


MT
[s
2
]
>
Di f f
= (

MT
[s
2
]
)[T
s
1
[s
1
]
]
= lim
/;,

C/
1
2
T
[s
2
]
[

H

C
(

H

C
)

T
s
1
[s
1
]
]
= lim
/;,

C/
1
2
1
n
(s
2
)

Di f f ()/Di f f
(s
2
)

GS
(s
2
)
<

U

T
s
2
[s
2
]

C
(

H

C
)

T
s
1
[s
1
]
>
Kin
=

s
lim
/;,

C/
1
2
1
n
(s
2
)

Di f f ()/Di f f
(s
2
)

GS
(s
2
)
<

U

T
s
2
[s
2
]

C
T
s
>
Kin
< T
s
(

H

C
)

T
s
1
[s
1
]
>
Kin
=

[s]

vV((s[s]))
1
2
lim
,

0
T
[s
2
]
[

H

v
T
s,c[s,c]
]

s,c[s,c]
< T
s
(

H

v
)

T
s
1
[s
1
]
>
Kin
, (45)
where Di f f

is the set of dieomorphisms leaving the colored graph invariant, GS

denotes
the graph symmetry quotient group Di f f

/TDi f f

where TDi f f

is the dieomorphism which


is trivial on the graph , and n

is the number of elements in GS

. Note that we have used


the resolution of identity trick in the fourth step. Since only a nite number of terms in the
sum over spin-networks s, cells C /, and dieomorphism transformations are non-zero
respectively, we can interchange the sums and the limit. In the fth step, we take the limit
C v and split the sum
_
s
into
_
[s]
_
s[s]
, where [s, c] denotes the dieomorphism equivalent
class associated with s. Here we also use the fact that, given (s) and (s

) which are dierent up


to a dieomorphism transformation, there is always a dieomorphism transforming the graph
associated with

H

v
T
s
(v (s)) to that of

H

T
s
(v

(s

)) with (v) = v

, hence T
[s
2
]
[

H

v
T
s[s]
]
is constant for dierent s [s].
Since the term
_
s[s]
< T
s
(

H

v
)

T
s
1
[s
1
]
>
Kin
is independent of the parameter

, one can see


that by xing a arbitrary family of state-dependent triangulations T(

),

s[s]
< T
s
(

H

v
)

T
s
1
[s
1
]
>
Kin
=

< U

T
s
(

H

v
)

T
s
1
[s
1
]
>
Kin
=

<

H

v
U

T
s
T
s
1
[s
1
]
>
Kin
47
=

< U

1
(

1
(v)
T
s
T
s
1
[s
1
]
>
Kin
= T
[s
1
]
[

H

1
(

)
vV((s))
T
s
], (46)
where are the dieomorphism transformations spanning the dieomorphism equivalent class
[s]. Note that the kinematical inner product in above sum is non-vanishing if and only if ((s)))
coincides with the graph obtained from certain skeleton (s
1
) by the action of (

H

v
)

and v
V(((s))), i.e., the scale
1
(

) of the dieomorphism images of the tetrahedrons added by


the action coincides with the scale of certain tetrahedrons in (s) and
1
(v) is a vertex in (s).
Then we can express the matrix elements (83) as:
< T
[s
1
]


MT
[s
2
]
>
Di f f
=

[s]

vV((s[s]))
1
2
lim
,

0
T
[s
2
]
[

H

v
T
s[s]
]T
[s
1
]
[

H

v
T
s[s]
]
=

[s]

vV((s[s]))
1
2
(

H

v
T
[s
2
]
)[T
s[s]
](

H

v
T
[s
1
]
)[T
s[s]
]. (47)
From Eq.(85) and the fact that the master constraint operator

M is densely dened on 1
Di f f
,
it is obvious that

M is a positive and symmetric operator in 1
Di f f
. Therefore, the quadratic
form Q
M
associated with

M is closable [114]. The closure of Q
M
is the quadratic form of
a unique self-adjoint operator

M, called the Friedrichs extension of

M. We relabel

M to be

M for simplicity. From the construction of



M, the qualitative description of the kernel of the
Hamiltonian constraint operator in Ref.[136] can be transcribed to describe the solutions to the
equation:

M
Di f f
= 0. In particular, the dieomorphism invariant cylindrical functions based
on at most 2-valent graphs are obviously normalizable solutions. In conclusion, there exists a
positive and self-adjoint operator

M on 1
Di f f
corresponding to the master constraint (75), and
zero is in the point spectrum of

M.
Note that the quantum constraint algebra can be easily checked to be anomaly free. i.e.,
[

M,

U

] = 0, [

M,

M] = 0.
which is consistent with the classical master constraint algebra in this sense. As a result, the dif-
culty of the original Hamiltonian constraint algebra can be avoided by introducing the master
constraint algebra, due to the Lie algebra structure of the latter. Since zero is in the spectrum of

M [140], the further task is to obtain the physical Hilbert space 1


phys
which is the kernel of the
master constraint operator with some suitable physical inner product, and the issue of quantum
anomaly is represented in terms of the size of 1
phys
and the existence of semi-classical states.
Note that we will see in the next section that the master constraint programme can be straight-
forwardly generalized to include matter elds [89]. We list some open problems in the master
constraint programme for further research.
Kernel of Master Constraint Operator
Since the master constraint operator

M is self-adjoint, it is a practical problem to dene
DID of 1
Di f f
:
1
Di f f

_

d()1

,
48
< >
Di f f
=
_
R
d() < >
1

,
where () is the spectral measure with respect to the master constraint operator

M. It
is expected that we can identify 1

=0
with the physical Hilbert space. However, such a
prescription is ambiguous in the case that zero is only in the continuous spectrum, loses
physical information in the case that zero is an embedded eigenvalue and unambiguous
only if zero is an isolated eigenvalue in which case however the whole machinery of the
DID is not needed at all because 1

=0
1
Di f f
and the physical inner product coincide
with the kinematical (dieromorphism invariant) one [63]. There are some improved pre-
scriptions also presented in [63] by decomposing the measure with respect to the spectrum
types before direct integral decomposition, some ambiguities can be canceled by some
physical criterion, e.g., a complete subalgebra of bounded Dirac observables should be
represented irreducibly as self-adjoint operators on the physical Hilbert space, and the
resulting physical Hilbert space should admits a sucient number of semiclassical states.
Nonetheless, due to the complicated structure of the master constraint operator, it is di-
cult anyhow to manage the spectrum analysis and direct integral decomposition. On the
other hand, for the self-adjointness of the master constraint operator and the Lie-algebra
structure of the constraint algebra, a formal group averaging strategy was introduced in
[149] as a more concrete way to get the physical Hilbert space. It is realized by a formal
rigged map
phys
:

phys
: Cyl

Di f f

phys
f
phys
( f ) :=
_
R
dt
2
< e
i

Mt
f . >
Di f f
,
where e
i

Mt
is a one parameter continuous unitary group on 1
Di f f
by the self-adjointness
of

M, and
phys
is a subset of the algebraic dual of Cyl

Di f f
. It is trivial to see that
phys
( f )
is invariant under the (dual) transformation of e
i

Mt
. Thus a inner product can be formally
dened between two algebraic functionals
phys
( f ) and
phys
( f

) in
phys
via:
<
phys
( f )
phys
( f

) >
phys
:=
phys
( f )[ f

],
=
_
R
dt
2
< e
i

Mt
f f

>
Di f f
=
_
R
dt
2
_
R
d()e
it
< f () f

() >
1

=
_
R
d()() < f () f

() >
1

= [
_
R
d()()] < f (0) f

(0) >
1

=0
,
where we have used the spectrum decomposition with respect to the self-adjoint operator

M, the operator e
i

Mt
is represented by multiplication by a number e
it
on each 1

, and
the vector valued function f () is the spectrum decomposition representation of state
f 1
Di f f
. Although we can see from the above argument that the physical inner product
49
is proportional to the inner product in the ber Hilbert space 1

=0
, unfortunately, the
factor
_
R
d()() is divergent when has pure point part, e.g. zero is in the discrete
spectrum of

M. That is one reason why we claim that the above argument is formal.
On the other hand, the group averaging strategy and the formal physical inner product
we just dened has potential relationships with path-integral formulation and spin foam
models due to the positivity of the master constraint operator

M [149], and hopefully,
we may obtain the physical transition amplitude from this physical inner product in the
future. However, the whole technique of group averaging for solving the master constraint
is still formal so far, and the rigorous calculations for it has not done yet as far as we know.
Dirac Observables
Classically, one can prove that a function O C

(A) is a weak observable with respect


to the scalar constraint if and only if
O, O, M
A
= 0.
We dene O to be a strong observable with respect to the scalar constraint if and only if
O, M
A
= 0,
and to be a ultra-strong observable if and only if
O, S(N)
A
= 0.
In quantum version, an observable

O is a weak Dirac observable if and only if

O leaves
1
phys
invariant, while

Ois nowcalled a strong Dirac observable if and only if

Ocommutes
with the master constraint operator

M. Given a bounded self-adjoint operator

O dened
on 1
Di f f
, for instance, a spectral projection of some observables leaving 1
Di f f
invariant,
if the uniform limit exists, the bounded self-adjoint operator dened by group averaging

[O] := lim
T
1
2T
_
T
T
dt

U(t)
1

O

U(t)
commutes with

M and hence becomes a strong Dirac observable on the physical Hilbert
space.
Testing the Classical Limit of the Master Constraint Operator
One needs to construct spatial dieomorphism invariant semiclassical states to calculate
the expectation value and uctuation of the master constraint operator. If the results
coincide with the classical values up to corrections, one can go ahead to nish our
quantization programme with condence.
50
5 Quantum Matter Field on a Quantum Background
In ordinary quantum eld theory, the quantum eld is dened on a smooth background space-
time. However, it is expected that the smooth structure of a spacetime may break down at
Planck scale, so the present treatment of quantum eld theory is valid only in a semiclassical
sense. Thus we would like to modify the formulation of present quantum eld theory to make
it compatible with the quantum theory of gravity(spacetime) which we already established in
previous sections so as to explore the behavior of the quantum matter eld under Planck scale
and at extremely strong gravitational elds, e.g. inside the black hole or at the early age of the
universe.
In the following, an alternative quantization of scalar eld will be introduced, the advantage
of such a quantization scheme is that the quantumscalar eld doesnt depend on the background.
We will also see that the quantization technique for the previous Hamiltonian constraint can be
generalized to quantize the Hamiltonian of matter elds coupled to gravity. Then it is shown
that an operator corresponding to the Hamiltonian of the scalar eld can be well dened on
the coupled dieomorphism invariant Hilbert space. It is even positive and self-adjoint without
any divergence. Thus quantum gravity acts exactly as a natural regulator for the quantum scalar
eld in the polymer representation. Moreover, to study the whole dynamical systemof the scalar
eld coupled to gravity, a Hamiltonian constraint operator is dened in the coupled kinematical
Hilbert space. The contribution of the scalar eld to the Hamiltonian constraint can be promoted
to a positive self-adjoint operator. To avoid possible quantum anomalies and nd the physical
Hilbert space, we will also introduce the master constraint programme for the coupled system.
A self-adjoint master constraint operator is obtained in the dieomorphism invariant Hilbert
space, which assures the feasibility of the programme.
5.1 Polymer-like Representation of a Scalar Field
We begin with the total Hamiltonian of the gravity coupled with a massless real scalar eld
which is a linear combination of constraints:
1
tot
=
i
G
i
+ N
a
C
a
+ NC,
where
i
, N
a
and N are Lagrange multipliers, and the three constraints in the Hamiltonian are
expressed as [30][87]:
G
i
= D
a

P
a
i
:=
a

P
a
i
+
k
i j
A
i
a

P
a
k
, (48)
C
a
=

P
b
i
F
i
ab
A
i
a
G
i
+
a
, (49)
C =

2
2
_
det q

P
a
i

P
b
j
[
i j
k
F
k
ab
2(1 +
2
)K
i
[a
K
j
b]
]
+
1
_
det q
[

M
2

i j

P
a
i

P
b
j
(
a
)
b
+
1
2
M

2
], (50)
here the real number
M
is the coupling constant, and denotes the momentum conjugate to :
:=
[

M
N
_
det q(

N
a

a
).
51
Thus one has the elementary Poisson brackets
A
i
a
(x),

P
b
j
(y) =
a
b

i
j
(x, y),
(x), (y) = (x, y).
Note that the second term of the Hamiltonian constraint (50) is just the Hamiltonian of the real
scalar eld.
Then we look for the background independent representation for the real scalar eld coupled
to gravity, following the polymer representation of the scalar eld [27]. The classical cong-
uration space, 1, consists of all real-valued smooth functions on . Given a set of a nite
number of points X = x
1
, ..., x
N
in , a equivalence relation can be dened by: given two scalar
eld
1
,
2
1,
1

2
if and only if exp[i
i

1
(x
i
)] = exp[i
j

2
(x
j
)] for all x
i
X and all real
number
j
. Hence we obtain a bijection between 1/ and R
X
, which is N copies of the Bohr
compactication of R [132]. Since one can dene a projective family with respect to the set of
point (graph for scalar eld), thus a projective limit 1, which is a compact topological space,
is obtained as the quantum conguration space of scalar eld. Next, we denote by Cyl
X
(1) the
vector space generated by nite linear combinations of the following functions of :
T
X,
() :=
_
x
j
X
exp[i
j
(x
j
)],
where (
1
,
2
, ,
N
) are arbitrary non-zero real numbers assigned at each point. It is
obvious that Cyl
X
(1) has the structure of a -algebra. The vector space Cyl(1) of all cylindrical
functions on 1 is dened by the linear span of the linear span of T
0
= 1 and T
X,
. Completing
Cyl(1) with respect to the sup norm, one obtains a unital Abelian C*-algebra Cyl(1). Thus
one can use the GNS structure to construct its cyclic representations. A preferred positive linear
functional
0
on Cyl(1) is dened by

0
(T
X,
) =
_
1 if
j
= 0 j
0 otherwise,
which denes a dieomorphism-invariant faithful Borel measure on 1 as
_
1
d(T
X,
) =
_
1 if
j
= 0 j
0 otherwise.
(51)
Thus one obtains the Hilbert space, 1
KG
kin
which is dened by L
2
(1, d), of square integrable
functions on a compact topological measure space 1 with respect to . The inner product can
be expressed explicitly as:
< T
c
T
c
>
KG
kin
=
cc
, (52)
where the label c := (X, ) are called scalar-network.
As one might expect, the quantum conguration space 1 is just the Gelfand spectrum of
Cyl(1). More concretely, for a single point set X
0
x
0
, Cyl
X
0
(1) is the space of all almost
periodic functions on a real line R. The Gelfand spectrum of the corresponding C*-algebra
52
Cyl
X
0
(1) is the Bohr completion R
x
0
of R [27], which is a compact topological space such that
Cyl
X
0
(1) is the C*-algebra of all continuous functions on R
x
0
. Since R is densely embedded in
R
x
0
, R
x
0
can be regarded as a completion of R.
It is clear from Eq.(51) that an orthonomal basis in 1
KG
kin
is given by the scalar vacuum
T
0
= 1 and so-called scalar-network functions T
c
(), where c = (X, ) and (
1
,
2
, ,
N
)
are non-zero real numbers. So the total kinematical Hilbert space 1
kin
is the direct product of
the kinematical Hilbert space 1
GR
kin
for gravity and the kinematical Hilbert space for real scalar
eld, i.e., 1
kin
:= 1
GR
kin
1
KG
kin
. Then the spin-scalar-network state T
s,c
T
s
(A) T
c
()
Cyl
(s)
(7/g) Cyl
X(c)
(1) Cyl
(s,c)
is a gravity-scalar cylindrical function on graph (s, c)
(s) X(c). Note that generally X(c) may not coincide with the vertices of the graph (s). It is
straightforward to see that all of these functions constitutes an orthonormal basis in 1
kin
as
< T
s
(A) T
c
()T
s
(A) T
c
() >
kin
=
s

c
.
Note that none of 1
kin
, 1
GR
kin
and 1
KG
kin
is a separable Hilbert space.
Given a pair (x
0
,
0
), there is an elementary conguration for the scalar eld, the so-called
point holonomy,
U(x
0
,
0
) := exp[i
0
(x
0
)].
It corresponds to a conguration operator

U(x
0
,
0
), which acts on any cylindrical function
() Cyl
X(c)
(1) by

U(x
0
,
0
)() = U(x
0
,
0
)(). (53)
All these operators are unitary. But since the family of operators

U(x
0
, ) fails to be weakly
continuous in , there is no eld operator

(x) on 1
KG
kin
. The momentum functional smeared on
a 3-dimensional region R is expressed by
(R) :=
_
R
d
3
x(x).
The Poisson bracket between the momentum functional and a point holonomy can be easily
calculated to be
(R), U(x, ) = i
R
(x)U(x, ),
where
R
(x) is the characteristic function for the region R. So the momentum operator is dened
by the action on scalar network functions T
c=(X,)
as
(R)T
c
() := i(R), T
c
() = [

x
j
X

j
(x
j
)]T
c
().
Now we can impose the quantum constraints on 1
kin
and consider the quantum dynamics.
First, the Gauss constraint can be solved independently of 1
KG
kin
, since it only involves the
gravitational eld. It is also expected that the dieomorphism constraint can be implemented
by the group averaging strategy in the similar way as in the case of pure gravity. Given a spatial
53
dieomorphism transformation , a unitary transformation

U

was induced by in the Hilbert


space 1
kin
, which is expressed as

T
s=((s),j,i),c=(X(c),)
= T
s=(((s)),j,i),c=((X(c)),)
.
Then the dieromorphism invariant spin-scalar-network functions are dened by group averag-
ing as
T
[s,c]
:=
1
n
(s,c)

Di f f ()/Di f f
(s,c)

GS
(s,c)

T
s,c
, (54)
where Di f f

is the set of dieomorphisms leaving the colored graph invariant, GS

denotes
the graph symmetry quotient group Di f f

/TDi f f

where TDi f f

is the set of the dieomor-


phisms which is trivial on the graph , and n

is the number of elements in GS

. Following the
standard strategy in quantization of pure gravity, an inner product can be dened on the vector
space spanned by the dieomorphism invariant spin-scalar-network functions (and the vacuum
states for gravity, scalar and both respectively) such that they form an orthonormal basis as:
< T
[s,c]
T
[s

,c

]
>
Di f f
:= T
[s,c]
[T
s

,c

[s

,c

]
] =
[s,c],[s

,c

]
. (55)
After the completion procedure, we obtain the expected Hilbert space of dieomorphism in-
variant states for the scalar eld coupled to gravity, which is denoted by 1
Di f f
.
5.2 Dieomorphism Invariant Hamiltonian of a Scalar Field
In the following discussion, we consider the quantum scalar eld on a uctuating background.
A similar idea was considered in Ref.[139], where a Hamiltonian operator with respect to a U(1)
group representation of the scalar eld is dened on a kinematical Hilbert space 1
kin
of matter
coupled to gravity. Then an eective Hamiltonian operator of the scalar eld can be constructed
as a quadratic form via
<
matter
,

H
e f f
matter
(m)

matter
>
KG
kin

:= <
grav
(m)
matter
,

H
matter

grav
(m)

matter
>
kin
, (56)
where
grav
(m) 1
GR
kin
presents a semiclassical state of gravity approximating some classical
spacetime background m where the quantum scalar eld lives. Thus the eective Hamiltonian
operator

H
e f f
matter
(m) of scalar eld contains also the information of the uctuating background
metric. In the light of this idea, we will construct a Hamiltonian operator

S
KG
for scalar eld
in the polymer-like representation. It turns out that this Hamiltonian operator can be dened
in the Hilbert space 1
Di f f
of dieomorphism invariant states for scalar eld coupled to gravity
without UV-divergence. So the quantum dynamics of the scalar eld is obtained in a dieo-
morphism invariant way, which is expected in the programme of loop quantum gravity. Thus,
here an eective Hamiltonian operator of the scalar eld could be extracted in 1
Di f f
by den-
ing <
[m]
(A, ),

S
KG

[m]
(A, ) >
Di f f
to be its expectation value on dieomorphism invariant
states () of the scalar eld, where the dieomorphism invariant semiclassical state
[m]
(A)
represents certain uctuating geometry with spatial dieomorphism invariance, and the label
54
[m] denotes the classical geometry approximated by
[m]
(A). Moreover, the quadratic proper-
ties of the scalar eld Hamiltonian will provide powerful functional analytic tools in the quan-
tization procedure, such that the self-adjointness of the Hamiltonian operator can be proved by
a theorem in functional analysis.
Then the crucial point is to dene an operator corresponding to the Hamiltonian functional
S
KG
of the scalar eld, which can be decomposed into two parts
S
KG
= S
KG,
+ S
KG,Kin
,
where
S
KG,
=

M
2
_

d
3
x
1
_
det q

i j

P
a
i

P
b
j
(
a
)
b
,
S
KG,Kin
=
1
2
M
_

d
3
x
1
_
det q

2
.
We will employ the following identities:

P
a
i
=
1
2

abc

i jk
e
j
b
e
k
c
and e
i
a
(x) =
2

A
i
a
(x), V
U
x
,
where
abc
denotes the Levi-Civita tensor tensity and V
U
x
is the volume of an arbitrary neigh-
borhood U
x
containing the point x. By using the point-splitting strategy, the regulated version
of the Hamiltonian is obtained as:
S
KG,
=

M
2
_

d
3
y
_

d
3
x

(x y)
i j

1
_
V
U

P
a
i
(x)(
a
(x))
1
_
V
U

P
b
j
(y)
b
(y)
=
32
M

4
_

d
3
y
_

d
3
x

(x y)
i j

aec
(
a
(x))Tr(
i
A
e
(x), V
3/4
U

x
A
c
(x), V
3/4
U

x
)

bf d
(
b
(y))Tr(
j
A
f
(y), V
3/4
U

y
A
d
(y), V
3/4
U

y
),
S
KG,Kin
=
1
2
M
_

d
3
x(x)
_

d
3
y(y)
_

d
3
u
det(e
i
a
(u))
(V
U

u
)
3/2
_

d
3
w
det(e
i
a
(w))
(V
U

w
)
3/2

(x y)

(u x)

(w y)
=
1
2
M
2
8
9()
6
_

d
3
x(x)
_

d
3
y(y)
_

d
3
u
abc
Tr(A
a
(u),
_
V
U

u
A
b
(u),
_
V
U

u
A
c
(u),
_
V
U

u
)
_

d
3
w
de f
Tr(A
d
(w),
_
V
U

w
A
e
(w),
_
V
U

w
A
f
(w),
_
V
U

w
)

(x y)

(u x)

(w y),
55
where we denote by A
a
A
i
a

i
,

(x y) the characteristic function of a box containing x with


scale such that lim
0

(x y)/
3
= (x y), and V
U

x
is the volume of the box. In order
to quantize the Hamiltonian S
KG
as a well-dened operator in the polymer-like representation,
we have to express the classical formula of the Hamiltonian in terms of elementary variables
with clear quantum analogs by introducing a triangulation T() of , where the parameter
describes how ne the triangulation is. The quantity regulated on the triangulation is required
to have correct limit when 0. Given a tetrahedron T(), we use s
i
()
i=1,2,3
to denote
the three outgoing oriented segments in with a common beginning point v() = s(s
i
()) and
use a
i j
() to denote the arcs connecting the end points of s
i
() and s
j
(). Then several loops

i j
() are formed by
i j
() := s
i
() a
i j
() s
j
()
1
. Thus we have the identities:

_
s()
dt A
a
s
a
(t), V
3/4
U

s(s())
= A(s())
1
A(s()), V
3/4
U

s(s())
+ o(),
and
_
s()
dt
a
s
a
(t) =
1
i
U(s(s()), )
1
[U(t(s()), ) U(s(s()), )] + o()
for nonzero , where s(s()) and t(s()) denote respectively the beginning and end points of
segment s() with scale associated with a tetrahedron . Regulated on the triangulation, the
scalar eld Hamiltonian reads
S

KG,
=
4
M
9
4

T()

T()

(v() v(

))
i j

lmn
1

U(v(), )
1
[U(t(s
l
()), ) U(v(), )]
Tr(
i
A(s
m
())
1
A(s
m
()), V
3/4
U

v()
A(s
n
())
1
A(s
n
()), V
3/4
U

v()
)

kpq
1

U(v(

), )
1
[U(t(s
k
(

)), ) U(v(

), )]
Tr(
j
A(s
p
(

))
1
A(s
p
(

)), V
3/4
U

v(

)
A(s
q
(

))
1
A(s
q
(

)), V
3/4
U

v(

)
),
S

KG,Kin
=
16
81
M
()
6

T()

T()
()(

T()

imn
Tr(A(s
i
(

))
1
A(s
i
(

)),
_
V
U

v(

)

A(s
m
(

))
1
A(s
m
(

)),
_
V
U

v(

)

A(s
n
(

))
1
A(s
n
(

)),
_
V
U

v(

)
)

T()

jkl
Tr(A(s
j
(

))
1
A(s
j
(

)),
_
V
U

v(

)

A(s
k
(

))
1
A(s
k
(

)),
_
V
U

v(

)

A(s
l
(

))
1
A(s
l
(

)),
_
V
U

v(

)
)

(v() v(

))

(v(

) v())

(v(

) v(

)). (57)
56
Note that the above regularization is explicitly dependent on the parameter , which will lead to
a kind of quantization ambiguity of the real scalar eld dynamics in polymer-like representation.
Introducing a partition / of the 3-manifold into cells C, we can smear the essential square
roots of S

KG,
and S

KG,Kin
in one cell C respectively and promote them as regulated operators
in 1
kin
with respect to triangulations T() depending on spin-scalar-network state T
s,c
as

W
,C
,i
T
s,c
=

vV((s,c))

C
(v)
E(v)

v()=v

h
,
,v,i
T
s,c
,

W
,C
Kin
T
s,c
=

vV((s,c))

C
(v)
E(v)

v()=v

h
,
Kin,v
T
s,c
, (58)
where
C
(v) is the characteristic function of the cell C, and

h
,
,v,i
:=
i

lmn
1
(v)

U(v, (v))
1
[

U(t(s
l
()), (v))

U(v, (v))]
Tr(
i

A(s
m
())
1
[

A(s
m
()),

V
3/4
U

v
]

A(s
n
())
1
[

A(s
n
()),

V
3/4
U

v
]),

h
,
Kin,v
:=
1
(i)
3
(v)
lmn
Tr(

A(s
l
())
1
[

A(s
l
()),
_

V
U

v
]

A(s
m
())
1
[

A(s
m
()),
_

V
U

v
]

A(s
n
())
1
[

A(s
n
()),
_

V
U

v
]). (59)
Both operators in (58) and their adjoint operators are densely dened on 1
kin
. We now give
several remarks on their properties.
Removal of regulator
It is not dicult to see that the action of the operator

W
,C
,i
on a spin-scalar-network
function T
s,c
is graph-changing. It adds nite number of vertices with representation (v)
at t(s
i
()) with distance from the vertex v. Recall that the action of the gravitational
Hamiltonian constraint operator on a spin network function is also graph-changing. As
a result, the family of operators

W
,C
,i
also fails to be weakly converged when 0.
However, due to the dieomorphism covariant properties of the triangulation, the limit
operator can be well-dened via the uniformRovelli-Smolin topology, or equivalently, the
operator can be dually dened on dieomorphism invariant states. But the dual operator
cannot leave 1
Di f f
invariant.
Quantization ambiguity
As a main dierence of the dynamics in polymer-like representation from that in U(1)
group representation [138], a continuous label appears explicitly in the expression of
(58). Hence there is an one-parameter quantization ambiguity due to the real scalar eld.
Recall that the construction of gravitational Hamiltonian constraint operator also has a
similar ambiguity due to the choice of the representations j of the edges added by its
action. A related quantization ambiguity also appears in the dynamics of loop quantum
cosmology [50].
57
Since our quantum eld theory is expected to be dieomorphism invariant, we would like to
dene the Hamiltonian operator of polymer scalar eld in the dieomorphism invariant Hilbert
space 1
Di f f
. For this purpose we x the parameter to be a non-zero constant at every point.
Then what we will do is to employ the new quantization strategy developed in Refs. [149]
and [140]. We rst construct a quadratic form in the light of a new inner product dened in
Ref.[140] on the algebraic dual 1

of the space of cylindrical functions which is spanned by


spin-scalar-networks T
s,c
(where the family of labels s, c includes the vacuum states for gravity,
scalar and both). Then we prove that the quadratic form is closed. Note that, although the
calculation employing this inner product is formal, it can lead to a well-dened expression
of the desired quadratic form Eq.(65). Since an arbitrary element of 1

is of the form =
_
s,c
c
s,c
< T
s,c
>
kin
, one can formally dene an inner product < >

on 1

via
< ,

>

:= <

s,c
c
s,c
< T
s,c
>
kin

,c

,c
< T
s

,c
>
kin
>

:=

s,c;s

,c

c
s,c
c

,c

< T
s,c
T
s

,c
>
kin
1
_
([s, c])([s

, c

])
=

s,c
c
s,c
c

s,c
1
([s, c])
, (60)
where the Cantor aleph denotes the cardinal of the set [s, c]. Note that we exchange the
coecients on which the complex conjugate was taken in Ref.[140], so that the inner product
<
Di f f

Di f f
>

reduces to <
Di f f

Di f f
>
Di f f
for any
Di f f
,

Di f f
1
Di f f
. Completing the
quotient with respect to the null vectors by this inner product, one gets a Hilbert space 1

. Our
purpose is to construct a quadratic form associated to some positive and symmetric operator in
analogy with the classical expression of (57). So the quadratic form should rst be given in
a positive and symmetric version. It is then natural to dene two quadratic forms on a dense
subset of 1
Di f f
1

as:
Q
KG,
(
Di f f
,

Di f f
) := lim
/

C/
64
4
M
9
4

i j
<

W
C
,i

Di f f


W
C
, j

Di f f
>

,
Q
KG,Kin
(
Di f f
,

Di f f
) := lim
/

C/
8
4

16
81
M
()
6
<

W
C
Kin

Di f f


W
C
Kin

Di f f
>

,
(61)
for any two states
Di f f
and

Di f f
which are nite linear combinations of T
[s,c]
, where the dual
limit operator

W
C
of either family of

W
,C
,i
or

W
,C
Kin
in (58) is naturally dened on dieomor-
phism invariant states as

W
C

Di f f
[T
s,c
] = lim
0

Di f f
[

W
,C
T
s,c
]. (62)
To show that the quadratic forms are well dened, we write

W
C
,i

Di f f
=

s,c
w

,i,s,c
(C) < T
s,c
>

,i,s,c
(C) = (

W
C
,i

Di f f
)[T
s,c
],

W
C
Kin

Di f f
=

s,c
w

Kin,s,c
(C) < T
s,c
>

Kin,s,c
(C) = (

W
C
Kin

Di f f
)[T
s,c
].
58
Then, by using the inner product (60) the quadratic forms in (61) become
Q
KG,
(
Di f f
,

Di f f
)
:= lim
/

C/
64
4
M
9
4

i j

s,c
w

,i,s,c
(C)w

, j,s,c
(C)
1
([s, c])
= lim
/

C/
64
4
M
9
4

i j

[s,c]
1
([s, c])

s,c[s,c]
w

,i,s,c
(C)w

, j,s,c
(C),
Q
KG,Kin
(
Di f f
,

Di f f
)
:= lim
/

C/
8
4

16
81
M
()
6

s,c
w

Kin,s,c
(C)w

Kin,s,c
(C)
1
([s, c])
= lim
/

C/
8
4

16
81
M
()
6

[s,c]
1
([s, c])

s,c[s,c]
w

Kin,s,c
(C)w

Kin,s,c
(C).
(63)
Note that, since
Di f f
is a nite linear combination of the dieomorphism invariant spin-scalar-
network basis, taking account of the operational property of

W
C
there are only a nite number
of terms in the summation
_
[s,c]
contributing to (63). Hence we can interchange
_
[s,c]
and
lim
/
_
C/
in above calculation. Moreover, for a suciently ne partition such that each cell
contains at most one vertex, the sum over cells therefore reduces to nite terms with respect to
the vertices of (s, c). So we can interchange
_
s,c[s,c]
and lim
/
_
C/
to obtain:
Q
KG,
(
Di f f
,

Di f f
)
= 64
4
M
9
4

i j

[s,c]
1
([s, c])

s,c[s,c]
lim
/

C/
w

,i,s,c
(C)w

, j,s,c
(C)
= 64
4
M
9
4

i j

[s,c]
1
([s, c])

s,c[s,c]

vV((s,c))
(

W
v
,i

Di f f
)[T
s,c
](

W
v
, j

Di f f
)[T
s,c
],
Q
KG,Kin
(
Di f f
,

Di f f
)
= 8
4

16
81
M
()
6

[s,c]
1
([s, c])

s,c[s,c]
lim
/

C/
w

Kin,s,c
(C)w

Kin,s,c
(C)
= 8
4

16
81
M
()
6

[s,c]
1
([s, c])

s,c[s,c]

vV((s,c))
(

W
v
Kin

Di f f
)[T
s,c
](

W
v
Kin

Di f f
)[T
s,c
],
(64)
where the limit / has been taken so that C v. Since given (s, c) and (s

, c

) which
are dierent up to a dieomorphism transformation, there is always a dieomorphism trans-
forming the graph associated with

W
,v
T
s,c
(v (s, c)) to that of

W
,v

T
s

,c
(v

(s

, c

)) with
(v) = v

, (

W
v

Di f f
)[T
s,c[s,c]
] is constant for dierent (s, c) [s, c], i.e., all the ([s, c]) terms
in the sum over (s, c) [s, c] are identical. Hence the nal expressions of the two quadratic
forms can be written as:
Q
KG,
(
Di f f
,

Di f f
)
59
= 64
4
M
9
4

i j

[s,c]

vV((s,c))
(

W
v
,i

Di f f
)[T
s,c[s,c]
](

W
v
, j

Di f f
)[T
s,c[s,c]
],
Q
KG,Kin
(
Di f f
,

Di f f
)
= 8
4

16
81
M
()
6

[s,c]

vV((s,c))
(

W
v
Kin

Di f f
)[T
s,c[s,c]
](

W
v
Kin

Di f f
)[T
s,c[s,c]
].
(65)
Note that both quadratic forms in (65) have nite results and hence their form domains are
dense in 1
Di f f
. Moreover, both of them are obviously positive, and the following theorem will
demonstrate their closedness.
Theorem 5.2.1: Both Q
KG,
and Q
KG,Kin
are densely dened, positive and closed quadratic
forms on 1
Di f f
, which are associated uniquely with two positive self-adjoint operators respec-
tively on 1
Di f f
such that
Q
KG,
(
Di f f
,

Di f f
) = <
Di f f


S
KG,

Di f f
>
Di f f
Q
KG,Kin
(
Di f f
,

Di f f
) = <
Di f f


S
KG,Kin

Di f f
>
Di f f
.
Therefore the Hamiltonian operator

S
KG
:=

S
KG,
+

S
KG,Kin
(66)
is positive and also have a unique self-adjoint extension.
Proof: We follow the strategy developed in Refs.[140] and [88] to prove that both Q
KG,
and
Q
KG,Kin
are closeable and uniquely induce two positive self-adjoint operators

S
KG,
and

S
KG,Kin
.
One can formally dene

S
KG,
and

S
KG,Kin
acting on dieomorphism invariant spin-scalar net-
work functions via:

S
KG,
T
[s
1
,c
1
]
:=

[s
2
,c
2
]
Q
KG,
(T
[s
2
,c
2
]
, T
[s
1
,c
1
]
)T
[s
2
,c
2
]
, (67)

S
KG,Kin
T
[s
1
,c
1
]
:=

[s
2
,c
2
]
Q
KG,Kin
(T
[s
2
,c
2
]
, T
[s
1
,c
1
]
)T
[s
2
,c
2
]
. (68)
Then we need to show that both of the above operators are densely dened on the Hilbert
space 1
Di f f
. Given a dieomorphism invariant spin-scalar network function T
[s
1
,c
1
]
, there are
only a nite number of terms T
[s
1
,c
1
]
[

W
,v
T
s,c[s,c]
] which are nonzero in the sum over equivalent
classes [s, c] in (65). On the other hand, given one spin-scalar-network function T
s,c[s,c]
, there
are also only a nite number of possible T
[s
2
,c
2
]
such that the terms T
[s
2
,c
2
]
[

W
,v
T
s,c[s,c]
] are
nonzero. As a result, only a nite number of terms survive in both sums over [s
2
, c
2
] in Eqs.
(67) and (68). Hence both

S
KG,
and

S
KG,Kin
are well dened on spin-scalar-network basis.
Then it follows from Eqs. (65), (67) and (68) that they are positive and symmetric operators
densely dened in 1
Di f f
, whose quadratic forms coincide with Q
KG,
and Q
KG,Kin
on their form
domains. Hence both Q
KG,
and Q
KG,Kin
have positive closures and uniquely induce self-adjoint
60
(Friedrichs) extensions of

S
KG,
and

S
KG,Kin
respectively [114], which we denote by

S
KG,
and

S
KG,Kin
as well. As a result, the Hamiltonian operator

S
KG
dened by Eq.(66) is also positive
and symmetric. Hence it has a unique self-adjoint (Friedrichs) extension.

We notice that, from a dierent perspective, one can construct the same Hamiltonian oper-
ator

1
KG
without introducing an inner product on 1

. The construction is sketched as follows.


Using the two well-dened operators

W
,C
,i
and

W
,C
Kin
as in (58), as well as their adjoint operators
(

W
,C
,i
)

and (

W
,C
Kin
)

, one may dene two operators on 1


Di f f
corresponding to the two terms in
(57) by
(

S
KG,

Di f f
)[T
s,c
] = lim
,

0,/

Di f f
[

C/
64
4
M
9
4

i j

W
,C
,i
(

W

,C
, j
)

T
s,c
]
(

S
KG,Kin

Di f f
)[T
s,c
] = lim
,

0,/

Di f f
[

C/
8
4

16
81
M
()
6

W
,C
Kin
(

W

,C
Kin
)

T
s,c
],
(69)
for any spin-scalar-network T
s,c
. In analogy with the discussion about the master constraint
operator and Ref.[88], it can be shown that both above operators leave 1
Di f f
invariant and are
densely dened on 1
Di f f
. Moreover, the quadratic forms associated with them coincide with
the quadratic forms in (65). Thus the Hamiltonian operator

S
KG
:=

S
KG,
+

S
KG,Kin
coincides
with the one constructed in the quadratic form approach.
In summary, we have constructed a positive self-adjoint Hamiltonian operator on 1
Di f f
for
the polymer-like scalar eld, depending on a chosen parameter . Thus there is an 1-parameter
ambiguity in the construction. However, there is no UV divergence in this quantum Hamiltonian
without renormalization, since quantum gravity plays the role of a natural regulator for the
polymer-like scalar eld.
5.3 Hamiltonian Constraint Equation for the Coupled System
In this section we consider the whole dynamical system of scalar eld coupled to gravity. Re-
call that in perturbative quantum eld theory in curved spacetime, the denition of some basic
physical quantities, such as the expectation value of the energy-momentum, is ambiguous and
it is challenging dicult to calculate the back-reaction of quantum elds on the background
spacetime [157]. This is reected by the fact that the semi-classical Einstein equation,
R

[g]
1
2
R[g]g

= <

T

[g] >, (70)


are known to be inconsistent and ambiguous [70][146]. One could speculate that the di-
culty is related to the fact that the usual formulation of quantum eld theories are background
dependent. Following this line of thought, if the quantization programme is by construction
non-perturbative and background independent, it may be possible to solve the problems fun-
damentally. In loop quantum gravity, there is no assumption of a priori background metric at
all. The quantum geometry and quantum matter elds are coupled and uctuating naturally
with respect to each other on a common manifold. On the other hand, there exists the time
61
problem in quantum theory of pure gravity, since all the physical states have to satisfy certain
version of quantum Wheeler-DeWitt constraint equation. However, the situation could improve
when matter eld is coupled to gravity [54][122]. In the following construction, we impose the
quantum Hamiltonian constraint on 1
kin
, and thus dene a quantum Wheeler-DeWitt constraint
equation for the scalar eld coupled to gravity. Then one can gain an insight into the problem
of time from the coupled equation, and the back-reaction of the quantum scalar eld is included
in the framework of loop quantum gravity.
We now dene an operator in 1
kin
corresponding to the scalar eld part S
KG
(N) of the total
Hamiltonian constraint functional, which can be read out from Eqs. (35) and (50) as
S
KG
(N) = S
KG,
(N) + S
KG,Kin
(N),
where
S
KG,
(N) =

M
2
_

d
3
xN
1
_
det q

i j

P
a
i

P
b
j
(
a
)
b
,
S
KG,Kin
(N) =
1
2
M
_

d
3
xN
1
_
det q

2
.
In analogy with the regularization and quantization in the previous section, the regulated version
of quantum Hamiltonian constraint

S

KG
(N) of scalar eld is expressed by taking the limit C
v:

KG
(N)T
s,c
:=

vV((s,c))
N(v)[64
4
M
9
4

i j
(

W
,v
,i
)


W
,v
, j
+ 8
4

16
81
M
()
6
(

W
,v
Kin
)


W
,v
Kin
]T
s,c
, (71)
where for any v V((s, c)), the operators

W
,v
,i
T
s,c
=
1
E(v)

v()=v

h
,
,v,i
T
s,c
,

W
,v
Kin
T
s,c
=
1
E(v)

v()=v

h
,
Kin,v
T
s,c
,
and their adjoints are all densely dened in 1
kin
. Hence the family of Hamiltonian constraint
operators (71) is also densely dened, and the regulator can be removed via the Uniform
Rovelli-Smollin topology, or equivalently the limit operator dually acts on dieomorphism in-
variant states as
(

S

KG
(N)
Di f f
)[ f ] = lim
0

Di f f
[

S

KG
(N) f ], (72)
for any f Cyl(7/g) Cyl(1). Similar to the dual of

S
GR
(N), the operator

S

KG
(N) fails to
commute with the dual of nite dieomorphism transformation operators, unless the smearing
function N(x) is a constant function over . In fact, the dual Hamiltonian constraint opera-
tor smeared by N = 1 is just the dieomorphism invariant Hamiltonian we just dened in the
62
last subsection. From Eq.(71), it is not dicult to prove that for positive N(x) the Hamilto-
nian constraint operator

S
KG
(N) of a scalar eld is positive and symmetric in 1
kin
and hence
has a unique self-adjoint extension [89]. Our construction of

S
KG
(N) is similar to that of the
Higgs eld Hamiltonian constraint in Ref.[138]. However, like the case of

S
KG
, there is a
one-parameter ambiguity in our construction of

S
KG
(N) due to the real scalar eld, which is
manifested as the continuous parameter in the expression of

h
,
,v,i
in (59). Note that now is
not required to be a constant, i.e., its value can be changed from one point to another. Thus the
total Hamiltonian constraint operator of scalar eld coupled to gravity has been obtained as

S(N) =

S
GR
(N) +

S
KG
(N). (73)
Again, there is no UV divergence in this quantum Hamiltonian constraint. Recall that, in stan-
dard quantum eld theory the UV divergence can only be cured by a renormalization procedure,
in which one has to multiply the Hamiltonian by a suitable power of the regulating parameter
. However, now has naturally disappeared from the expression of (73). So renormalization is
not needed for the polymer-like scalar eld coupled to gravity, since quantum gravity has played
the role of a natural regulator. This result heightens our condence that the issue of divergences
in quantum eld theory can be cured in the framework of loop quantum gravity.
Now we have obtained the desired matter-coupled quantum Hamiltonian constraint equation
(

S

KG
(N)
Di f f
)[ f ] = (

S

GR
(N)
Di f f
)[ f ]. (74)
Comparing it with the well-known Sch ordinger equation for a particle,
i

t
(x, t) = H( x,

i

x
)(x, t),
where (x, t) L
2
(R, dx) and t is a parameter labeling time evolution, one may take the view-
point that the matter eld constraint operator

S

KG
(N) plays the role of i

t
. Then appears as
the parameter labeling the evolution of the gravitational eld state. In the reverse viewpoint, the
gravitational eld would become the parameter labeling the evolution of the quantum matter
eld. Note that such an idea has been successfully applied in a loop quantum cosmology model
to help us to understand the quantum nature of big bang in the deep Planck regime [28][29].
5.4 Master Constraint for the Coupled System
Recall that in order to avoid possible quantum anomalies and nd the physical Hilbert space
of quantum gravity, the master constraint programme was rst introduced in the last section.
The central idea is to construct an alternative classical constraint algebra, giving the same con-
straint phase space, which is a Lie algebra (no structure functions) and where the subalgebra of
dieomorphism constraints forms an ideal. A self-adjoint master constraint operator for loop
quantum gravity is then proposed on 1
Di f f
. The master constraint programme can be gener-
alized to matter elds coupled to gravity in a straightforward way. We now take the massless
real scalar eld to demonstrate the construction of a master constraint operator according to
the same strategy as we did in the last section. By this approach one not only avoids possible
quantum anomalies which might appear in the conventional canonical quantization method, but
63
also might give a qualitative description of the physical Hilbert space for the coupled system.
We introduce the master constraint for the scalar eld coupled to gravity as
M :=
1
2
_

d
3
x
C(x)
2
_
det q(x)
, (75)
where C(x) is the Hamiltonian constraint in (50). After solving the Gaussian constraint, one
gets the master constraint algebra as a Lie algebra:
+(

N), +(

) = +([

N,

N

]),
+(

N), M = 0,
M, M = 0, (76)
where the subalgebra of dieomorphism constraints forms an ideal. So it is possible to dene a
corresponding master constraint operator on 1
Di f f
. In the following, the positivity and the dif-
feomorphism invariance of M will be working together properly and provide us with powerful
functional analytic tools in the quantization procedure.
The regulated version of the master constraint can be expressed via a point-splitting strategy
as:
M

:=
1
2
_

d
3
y
_

d
3
x

(x y)
C(y)
_
V
U

y
C(x)
_
V
U

x
. (77)
Introducing a partition / of the 3-manifold into cells C, we have an operator

H

C
acting on
any spin-scalar-network state T
s,c
via a family of state-dependent triangulation T(),

C
T
s,c
=

vV((s,c))

C
(v)
E(v)

v()=v

h
,
GR,v
T
s,c
+

vV((s,c))

C
(v)
E(v)
[64
4
M
9
4

i j
( w
,v
,i
)

w
,v
, j
+ 8
4

16
81
M
()
6
( w
,v
Kin
)

w
,v
Kin
]T
s,c
, (78)
where

h
,
GR,v
=
16
3i
2

i jk
Tr(

A(
i j
())
1

A(s
k
())
1
[

A(s
k
()),
_

V
U

v
])
+ (1 +
2
)
8

2
3i
3

i jk
Tr(

A(s
i
())
1
[

A(s
i
()),

K


A(s
j
())
1
[

A(s
j
()),

K

]

A(s
k
())
1
[

A(s
k
()),
_

V
U

v
]),
w
,v
,i
=
i

v()=v

lmn
1

U(v, )
1
[

U(t(s
l
()), )

U(v, )]
Tr(
i

A(s
m
())
1
[

A(s
m
()),

V
5/8
U

v
]

A(s
n
())
1
[

A(s
n
()),

V
5/8
U

v
]),
64
w
,v
Kin
=
1
(i)
3

v()=v
(v)
lmn
Tr(

A(s
l
())
1
[

A(s
l
()),

V
5/12
U

v
]

A(s
m
())
1
[

A(s
m
()),

V
5/12
U

v
]


A(s
n
())
1
[

A(s
n
()),

V
5/12
U

v
]). (79)
Hence the action of

H

C
on a cylindrical function f

adds analytical arcs a


i j
() with
1
2
-representation
and points at t(s
i
()) with representation constant with respect to each vertex v() of . Thus,
for each > 0,

H

C
is a S U(2) gauge invariant and dieomorphism covariant operator dened
on Cyl(7/g) Cyl(1). The limit operator

H
C
is densely dened on 1
Kin
by the uniform
Rovelli-Smolin topology. And the same result holds for the adjoint operator (

H

C
)

Then a master constraint operator,



M, on 1
Di f f
can be dened by:
(

M
Di f f
)[T
s,c
] := lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

T
s,c
]. (80)
Since

H

C
(

H

C
)

T
s,c
is a nite linear combination of spin-scalar-network functions on an graph
with skeleton , the value of (

M
Di f f
)[T
s,c
] is nite for a given
Di f f
that is a nite linear
combination of T
[s,c]
. So

M
Di f f
is in the algebraic dual of the space of cylindrical functions.
Moreover, we can show that it is dieomorphism invariant. For any dieomorphism transfor-
mation ,
(

U


M
Di f f
)[ f

] = lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

]
= lim
/;,

Di f f
[

U

C/
1
2

H

1
()

1
(C)
(

H

1
(

1
(C)
)

]
= lim
/;,

Di f f
[

C/
1
2

H

C
(

H

C
)

], (81)
for any cylindrical function f

, where in the last step, we used the fact that the dieomorphism
transformation leaves the partition invariant in the limit / and relabel (C) to be C. So
we have the result
(

U


M
Di f f
)[ f

] = (

M
Di f f
)[ f

]. (82)
So given a dieomorphism invariant spin-scalar-network state T
[s,c]
, the result state

MT
[s,c]
must
be a dieomorphisminvariant element in the algebraic dual of Cyl(7/g)Cyl(1), which means
that

MT
[s,c]
=

[s
1
,c
1
]
c
[s
1
,c
1
]
T
[s
1
,c
1
]
,
then
lim
/;,

0
T
[s,c]
[

C/
1
2

H

C
(

H

C
)

T
s
2
,c
2
] =

[s
1
,c
1
]
c
[s
1
,c
1
]
T
[s
1
,c
1
]
[T
s
2
,c
2
],
65
where the cylindrical function
_
C/
1
2

H

C
(

H

C
)

T
s
2
,c
2
is a nite linear combination of spin-scalar-
network functions on some graphs

with the same skeleton of (s
2
, c
2
) up to nite number of
arcs and vertices. Hence xing the dieomorphism equivalence class [s, c], only for spin-scalar-
networks s
2
, c
2
lying in a nite number of dieomorphism equivalence class on the left hand
side of the last equation is non-zero. So there are also only nite number of classes [s
1
, c
1
] in
the right hand side such that c
[s
1
,c
1
]
is non-zero. As a result,

MT
[s,c]
is a nite linear combination
of dieomorphism invariant spin-network states so lies in the Hilbert space of dieomorphism
invariant states 1
Di f f
for any [s, c]. And

M is densely dened on 1
Di f f
.
We now compute the matrix elements of

M. Given two dieomorphism invariant spin-
scalar-network functions T
[s
1
,c
1
]
and T
[s
2
,c
2
]
, the matrix element of

M is calculated as
< T
[s
1
,c
1
]


MT
[s
2
,c
2
]
>
Di f f
= (

MT
[s
2
,c
2
]
)[T
s
1
,c
1
[s
1
,c
1
]
]
= lim
/;,

C/
1
2
T
[s
2
,c
2
]
[

H

C
(

H

C
)

T
s
1
,c
1
[s
1
,c
1
]
]
= lim
/;,

C/
1
2
1
n
(s
2
,c
2
)

Di f f /Di f f
(s
2
,c
2
)

GS
(s
2
,c
2
)
<

U

T
s
2
,c
2
[s
2
,c
2
]

C
(

H

C
)

T
s
1
,c
2
[s
1
,c
1
]
>
Kin
=

s,c
lim
/;,

C/
1
2
1
n
(s
2
,c
2
)

Di f f /Di f f
(s
2
,c
2
)

GS
(s
2
,c
2
)
<

U

T
s
2
,c
2
[s
2
,c
2
]

C
T
s,c
>
Kin
<
s,c
(

H

C
)

T
s
1
,c
1
[s
1
,c
1
]
>
Kin
=

[s,c]

vV((s,c[s,c]))
1
2
lim
,

0
T
[s
2
,c
2
]
[

H

v
T
s,c[s,c]
]

s,c[s,c]
< T
s,c
(

H

v
)

T
s
1
,c
1
[s
1
,c
1
]
>
Kin
, (83)
where Di f f

is the set of dieomorphisms leaving the colored graph invariant, GS

denotes
the graph symmetry quotient group Di f f

/TDi f f

where TDi f f

is the dieomorphisms which


is trivial on the graph , and n

is the number of elements in GS

. Note that we have used the


resolution of identity trick in the fourth step. Since only a nite number of terms in the sum
over spin-scalar-networks (s, c), cells C /, and dieomorphism transformations are non-
zero respectively, we can interchange the sums and the limit. In the fth step, we take the
limit C v and split the sum
_
s,c
into
_
[s,c]
_
s,c[s,c]
, where [s, c] denotes the dieomorphism
equivalence class associated with (s, c). Here we also use the fact that, given (s, c) and (s

, c

)
which are dierent up to a dieomorphism transformation, there is always a dieomorphism
transforming the graph associated with

H

v,(s,c)
T
s,c
(v (s, c)) to that of

H

(s

,c

)
T
s

,c
(v


(s

, c

)) with (v) = v

, hence T
[s
2
,c
2
]
[

H

v,(s,c)
T
s,c[s,c]
] is constant for dierent (s, c) [s, c].
Since the term
_
s,c[s,c]
< T
s,c
(

H

v
)

T
s
1
,c
1
[s
1
,c
1
]
>
Kin
is independent of the parameter

, one
can see that by xing a family of arbitrary state-dependent triangulations T(

),

s,c[s,c]
< T
s,c
(

H

v
)

T
s
1
,c
1
[s
1
,c
1
]
>
Kin
66
=

< U

T
s,c
(

H

v
)

T
s
1
,c
1
[s
1
,c
1
]
>
Kin
=

<

H

v
U

T
s,c
T
s
1
,c
1
[s
1
,c
1
]
>
Kin
=

< U

1
(

1
(v)
T
s,c
T
s
1
,c
1
[s
1
,c
1
]
>
Kin
= T
[s
1
,c
1
]
[

H

1
(

)
vV((s,c))
T
s,c
], (84)
where are the dieomorphismtransformations spanning the dieomorphismequivalence class
[s, c]. Note that the kinematical inner product in the above sum is non-vanishing if and only
if ((s, c))) coincides with the extended graph obtained from certain skeleton (s
1
, c
1
) by the
action of (

H

v
)

and v V(((s, c))), i.e., the scale


1
(

) of the dieomorphism images of the


tetrahedrons added by the action coincides with the scale of certain tetrahedrons in (s, c) and

1
(v) is a vertex in (s, c). Then we can express the matrix elements (83) as:
< T
[s
1
,c
1
]


MT
[s
2
,c
2
]
>
Di f f
=

[s,c]

vV((s,c[s,c]))
1
2
lim
,

0
T
[s
2
,c
2
]
[

H

v
T
s,c[s,c]
]T
[s
1
,c
1
]
[

H

v
T
s,c[s,c]
]
=

[s,c]

vV((s,c[s,c]))
1
2
(

H

v
T
[s
2
,c
2
]
)[T
s,c[s,c]
](

H

v
T
[s
1
,c
1
]
)[T
s,c[s,c]
]. (85)
From Eq.(85) and the result that the master constraint operator

M is densely dened on 1
Di f f
,
it is obvious that

M is a positive and symmetric operator on 1
Di f f
. Hence, it is associated with
a unique self-adjoint operator

M, called the Friedrichs extension of

M. We relabel

M to be

M for simplicity. In conclusion, there exists a positive and self-adjoint operator



M on 1
Di f f
corresponding to the master constraint (75). It is then possible to obtain the physical Hilbert
space of the coupled system by the direct integral decomposition of 1
Di f f
with respect to

M.
Note that the quantum constraint algebra can be easily checked to be anomaly free. Eq.(82)
assures that the master constraint operator commutes with nite dieomorphism transforma-
tions, i.e.,
[

M,

U

] = 0. (86)
Also it is obvious that the master constraint operator commutes with itself,
[

M,

M] = 0. (87)
So the quantum constraint algebra is precisely consistent with the classical constraint algebra
(76) in this sense. As a result, the diculty of the original Hamiltonian constraint algebra can
be avoided by introducing the master constraint algebra, due to the Lie algebra structure of the
latter.
67
6 The Semiclassical Limit of Quantum Dynamics
As shown in previous chapters, both the Hamiltonian constraint operator

S(N) and the master
constraint operator

M can be well dened in the framework of loop quantum gravity. However,
since the Hilbert spaces 1
kin
and 1
Di f f
, the operators

S(N) and

M are constructed in such
ways that are drastically dierent from usual quantum eld theory, one has to check whether
the constraint operators and the corresponding algebras have correct semiclassical limits with
respect to suitable semiclassical states.
6.1 The Construction of Coherent States
In order to nd the proper semiclassical states and check the classical limit of the theory, the
idea of a non-normalizable coherent state dened by a generalized Laplace operator and its
heat kernel was introduced for the rst time in [26]. Recently, kinematical coherent states were
constructed in two dierent approaches. One leads to the so-called complexier coherent states
proposed by Thiemann et al [142][143][144][145]. The other was proposed by Varadarajan
[152][153][154] and further developed by Ashtekar et al [19][15].
The complexier approach is motivated by the coherent state construction for compact Lie
groups [86]. One begins with a positive function C (complexier) on the classical phase space
and arrives at a coherent state
m
, which more possibly belongs to the dual space Cyl

rather
than 1
kin
. However, one may consider the so-called cut-o state of
m
with respect to a
nite graph as a graph-dependent coherent state in 1
kin
[146]. By construction, the coherent
state
m
is an eigenstate of an annihilation operator coming also from the complexier C and
hence has the desired semiclassical properties [143][144]. We now sketch the basic idea of
its construction. Given the Hilbert space 1 for a dynamical system with constraints and a
subalgerba of observables S in the space [(1) of linear operators on 1, the semiclassical
states with respect to S are dened in Denition 3.1.5. Kinematical coherent states
m

mA
are semiclassical states which in addition satisfy the annihilation operator property [142][146],
namely there exists a certain non-self-adjoint operator z = a + i

b with a,

b S and a certain
squeezing parameter , such that
z
m
= z(m)
m
. (88)
Note that Eq.(88) implies that the minimal uncertainty relation is saturated for the pair of ele-
ments ( a,

b), i.e.,

m
([ a
m
( a)]
2
) =
m
([

b
m
(

b)]
2
) =
1
2

m
([ a,

b]). (89)
Note also that coherent states are usually required to satisfy the additional peakedness property,
namely for any m A the overlap function <
m
,
m
> is concentrated in a phase volume
1
2

m
([ q, p]), where q is a conguration operator and p a momentum operator. So the central
element in the construction is to dene a suitable annihilation operator z in analogy with the
simplest case of harmonic oscillator. A powerful tool named as complexier is introduced in
Ref.[142] to dene a meaningful z operator which can give rise to kinematical coherent states
for a general quantum system.
68
Denition 6.1.1: Given a phase space A = T

( for some dynamical system with conguration


coordinates q and momentum coordinates p, a complexier, C, is a positive smooth function on
A, such that
(1) C/ is dimensionless;
(2) lim
p
C(m)
p
= for some suitable norm on the space of the momentum;
(3) Certain complex coordinates (z(m), z(m)) of Acan be constructed from C.
Given a well-dened complexier C on phase space A, the programme for constructing co-
herent states associated with C can be carried out as the following.
Complex polarization
The condition (3) in Denition 7.3.1 implies that the complex coordinate z(m) of A can
be constructed via
z(m) :=

n=0
i
n
n!
q, C
(n)
(m), (90)
where the multiple Poisson bracket is inductively dened by q, C
(0)
= q, q, C
(n)
=
q, C
(n1)
, C. One will see that z(m) can be regarded as the classical version of an
annihilation operator.
Dening the annihilation operator
After the quantization procedure, a Hilbert space 1 = L
2
((, d) with a suitable measure
d on a suitable conguration space ( can be constructed. It is reasonable to assume that
C can be dened as a positive self-adjoint operator

C on 1. Then a corresponding oper-
ator z can be dened by transforming the Poisson brackets in Eq.(90) into commutators,
i.e.,
z :=

n=0
i
n
n!
1
(i)
n
[ q,

C]
(n)
= e


C/
qe

C/
, (91)
which is called as an annihilation operator.
Constructing coherent states
Let
q
(q) be the -distribution on ( with respect to the measure d. Since

C is assumed
to be positive and self-adjoint, the conditions (1) and (2) in Denition 7.3.1 imply that
e


C/
is a well-dened smoothening operator. So it is quite possible that the heat kernel
evolution of the -distribution, e


C/

q
(q), is a square integrable function in 1, which is
even analytic. Then one may analytically extend the variable q

in e


C/

q
(q) to complex
values z(m) and obtain a class of states

m
as

m
(q) := [e


C/

q
(q)]
q

z(m)
, (92)
such that one has
z

m
(q) := [e


C/
q
q
(q)]
q

z(m)
= [q


C/

q
(q)]
q

z(m)
= z(m)

m
(q). (93)
69
Hence

m
is automatically an eigenstate of the annihilation operator z. So it is natural to
dene coherent states
m
(q) by normalizing

m
(q).
One may check that all the coherent state properties usually required are likely to be satised
by the above complexier coherent states
m
(q) [146]. As a simple example, in the case of one-
dimensional harmonic oscillator with Hamiltonian H =
1
2
(
p
2
2m
+
1
2
m
2
q
2
), one may choose the
complexier C = p
2
/(2m). It is straightforward to check that the coherent state constructed
by the above procedure coincides with the usual harmonic oscillator coherent state up to a phase
[146]. So the complexier coherent state can be considered as a suitable generalization of the
concept of usual harmonic oscillator coherent state.
The complexifer approach can be used to construct kinematical coherent states in loop quan-
tum gravity. Given a suitable complexier C, for each analytic path e one can dene
A
C
(e) :=

n=0
i
n
n!
A(e), C
(n)
, (94)
where A(e) S U(2) is assigned to e. As the complexier C is assumed to give rise to a
positive self-adjoint operator

C on the kinematical Hilbert space 1
kin
, one further supposes that

C/T
s
=
s
T
s
, where is a so-called classicality parameter, T
s
(A)
s
form a basis in 1
kin
and are analytic in A 7. Moreover the -distribution on the quantum conguration space 7
can be formally expressed as
A
(A) =
_
s
T
s
(A

)T
s
(A). Thus by applying Eq.(92) one obtains
coherent states

A
C
(A) = (e


C/
)
A
(A)
A

A
C =

s
e

s
T
s
(A
C
)T
s
(A). (95)
However, since there are an uncountably innite number of terms in the expression (95), the
norm of

A
C
(A) would in general be divergent. So

A
C
(A) is generally not an element of 1
kin
but rather a distribution on a dense subset of 1
kin
. In order to test the semiclassical limit of
quantum geometric operators on 1
kin
, one may further consider the cut-o state of

A
C
(A)
with respect to a nite graph as a graph-dependent coherent state in 1
kin
[146]. So the
key input in the construction is to choose a suitable complexifer. There are vast possibilities
of choice. For example, a candidate complexier C is considered in Ref.[148] such that the
corresponding operator acts on cylindrical functions f

by
(

C/) f

=
1
2
(

eE()
l(e)

J
2
e
) f

, (96)
where

J
2
e
is the Casimir operator dened by Eq.(24) associated to the edge e, the positive num-
bers l(e) satisfying l(e e

) = l(e) + l(e

) and l(e
1
) = l(e) serves as a classicalization parameter.
Then it can be shown from Eq.(94) that A
C
(e) is an element of S L(2, C). So the classical inter-
pretation of the annihilation operators is simply the generalized complex S U(2) connections. It
has been shown in Refs. [143] and [144] that the cut-o state of the corresponding coherent
state,

A
C
,
(A) =

A
C
,
(A)/

A
C
,
(A), (97)
70
with

A
C
,
(A) :=

s,(s)=
e

1
2
_
eE((s))
l(e) j
e
( j
e
+1)
T
s
(A
C
)T
s
(A). (98)
has desired semiclassical properties in testing the kinematical operators (e.g. holonomy and
ux). But unfortunately, these cut-o coherent states cannot be directly used to test the semi-
classical limit of the Hamiltonian constraint operator

S(N), since

S(N) is graph-changing so that
its expectation values with respect to these cut-o states are always zero! So further work in this
approach is expected in order to overcome the diculty. Anyway, the complexier approach
provides a clean construction mechanism and manageable calculation method for semiclassical
analysis in loop quantum gravity.
We now turn to the second approach. As we have seen, loop quantum gravity is based on
quantum geometry, where the fundamental excitations are one-dimensional polymer-like. On
the other hand, low energy physics is based on quantum eld theories which are constructed
in a at spacetime continuum. The fundamental excitations of these elds are 3-dimensional,
typically representing wavy undulations on the background Minkowskian geometry. The core
strategy in this approach is then to relate the polymer excitations of quantum geometry to Fock
states used in low energy physics and to locate Minkowski Fock states in the background inde-
pendent framework. Since the quantum Maxwell eld can be constructed in both Fock repre-
sentation and polymer-like representation, one rst gains insights from the comparison between
the two representations, then generalizes the method to quantum geometry. A Laplacian oper-
ator can be dened on 1
kin
[26][19], from which one may dene a candidate coherent state
0
,
also in Cyl

, corresponding to the Minkowski spacetime. To calculate the expectation values of


kinematical operators, one considers the so-called shadow state of
0
, which is the restriction
of
0
to a given nite graph. However, the construction of shadow states is subtly dierent from
that of cut-o states.
We will only describe the simple case of the Maxwell eld to illustrate the ideas of the
construction [152][153][20]. Following the quantum geometry strategy discussed in Sec.4, the
quantum conguration space A for the polymer representation of the U(1) gauge theory can be
similarly constructed. A generalized connection A A assigns each oriented analytic edge in
an element of U(1). The space A carries a dieomorphism and gauge invariant measure
0
induced by the Haar measure on U(1), which gives rise to the Hilbert space, 1
0
:= L
2
(A, d
0
),
of polymer states. The basic operators are holonomy operators

A(e) labeled by one-dimensional
edges e, which act on cylindrical functions by multiplication, and smeared electric eld opera-
tors

E(g) for suitable test one-forms g on , which are self-adjoint. Note that, since the gauge
group U(1) is Abelian, it is more convenient to smear the electric elds in 3 dimensions [20].
The eigenstates of

E(g), so-called ux network states A
,n
, provide an orthonormal basis in 1
0
,
which are dened for any nite graph with N edges as:
A
,n
(A) := [A(e
1
)]
n
1
[A(e
2
)]
n
2
[A(e
N
)]
n
N
, (99)
where n (n
1
, , n
N
) assigns an integer n
I
to each edge e
I
. The action of

E(g) on the ux
network states reads

E(g) A
,n
= (

I
n
I
_
e
I
g)A
,n
. (100)
71
In this polymer-like representation, cylindrical functions are the nite linear combinations of
ux network states and span a dense subspace of 1
0
. Denote by Cyl the set of cylindrical
functions and by Cyl

its algebraic dual. One then has a triplet Cyl 1


0
Cyl

in analogy
with the case of loop quantum gravity.
The Schr odinger or Fock representation of the Maxwell eld, on the other hand, depends on
the Minkowski background metric. Here the Hilbert space is given by 1
F
= L
2
(S

, d
F
), where
S

is the appropriate space of tempered distributions on and


F
is the Gaussian measure. The
basic operators are connections

A( f ) smeared in 3 dimensions with suitable vector densities
f and smeared electric elds

E(g). But

A(e) fail to be well dened. To resolve this tension
between the two representations, one proceeds as follows. Let x be the Cartesian coordinates of
a point in = R
3
. Introduce a test function by using the Euclidean background metric on R
3
,
f
r
(x) =
1
(2)
3/2
r
3
exp(x
2
/2r
2
), (101)
which approximates the Dirac delta function for small r. The Gaussian smeared form factor for
an edge e is dened as
X
a
(e,r)
(x) :=
_
e
ds f
r
(e(s) x) e
a
. (102)
Then one can dene a smeared holonomy for e by
A
(r)
(e) := exp[i
_
R
3
X
a
(e,r)
(x)A
a
(x)], (103)
where A
a
(x) is the U(1) connection one-form of the Maxwell eld on . Similarly one can
dene Gaussian smeared electric elds by
E
(r)
(g) :=
_
R
3
g
a
(x)
_
R
3
f
r
(y x)E
a
(y). (104)
In this way one obtains two Poission bracket algebras. One is formed by smeared holonomies
and electric elds with
A
(r)
(e), A
(r)
(e

) = 0 = E(g), E(g

) (105)
A
(r)
(e), E(g) = i(
_
R
3
X
a
(e,r)
g
a
) A
(r)
(e).
The other is formed by unsmeared holonomies and Gaussian smeared electric elds with
A(e), A(e

) = 0 = E
(r)
(g), E
(r)
(g

) (106)
A(e), E
(r)
(g) = i(
_
R
3
X
a
(e,r)
g
a
) A(e).
Obviously, there is an isomorphism between them,
I
r
: (A
(r)
(e), E(g)) (A(e), E
(r)
(g)). (107)
72
Using the isomorphism I
r
, one can pass back and forth between the polymer and the Fock
representations. Specically, the image of the Fock vacuum can be shown to be the following
element of Cyl

[152][153],
(V =

,n
exp[

I J
G
I J
n
I
n
J
] (A
,n
, (108)
where (A
,n
Cyl

maps the ux network function A


,n
) to one and every other ux net-
work functions to zero. While the states (A
,n
do not have any knowledge of the underlying
Minkowskian geometry, this information is coded in the matrix G
I J
associated with the edges
of the graph , given by [20]
G
I J
=
_
e
I
dt e
a
I
(t)
_
e
J
dt

e
J
b
(t

)
_
d
3
x
ab
(x) [ f
r
(x e
I
(t))

1
2
f (x, e
J
(t

))], (109)
where
ab
is the at Euclidean metric and its Laplacian. Therefore, one can single out the Fock
vacuum state directly in the polymer representation by invoking Poincar e invariance without
any reference to the Fock space. Similarly, one can directly locate in Cyl

all coherent states


as the eigenstates of the exponentiated annihilation operators. Since Cyl

does not have an


inner product, one uses the notion of shadow states to do semiclassical analysis in the polymer
representation. From Eq.(108), the action of the Fock vacuum (V on A
,n
reads
(VA
,n
) =
_
A

d
0

A
,n
, (110)
where the state V

is in the Hilbert space 1

for the graph and given by


V

(A) =

n
exp[

I J
G
I J
n
I
n
J
] A
,n
(A). (111)
Thus for any cylindrical functions

associated with ,
(V

) = (V

), (112)
where the inner product in the right hand side is taken in 1

. Hence V

(A) are referred to as


shadows of (V on the graphs . The set of all shadows captures the full information in (V.
By analyzing shadows on suciently rened graphs, one can introduce criteria to test if a given
element of Cyl

represents a semi-classical state [20]. It turns out that the state (V does satisfy
this criterion and hence can be regarded as semi-classical in the polymer representation.
The mathematical and conceptual tools gained from simple models like the Maxwell elds
are currently being used to construct semiclassical states of quantum geometry. A candidate
kinematical coherent state corresponding to the Minkowski spacetime has been proposed by
Ashtekar and Lewandowki in the light of a Laplacian operator [19][20]. However, the detailed
structure of this coherent state is yet to be analyzed and there is no a priori guarantee that it is
indeed a semiclassical state.
One may nd comparisons of the two approaches from both sides [147][20]. It turns out that
Varadarajans Laplacian coherent state for the polymer Maxwell eld can also be derived from
73
Thiemanns complexier method. However, one cannot nd a complexier to get the coherent
state proposed by Ashtekar et al. for loop quantum gravity. Both approaches have their own
virtues and need further developments. The complexier approach provides a clear construc-
tion mechanism and manageable calculation method, while the Laplacian operator approach is
related closely with the well-known Fock vacuum state. One may also expect that a judicious
combination of the two approaches may lead to signicant progress in the semiclassical analysis
of loop quantum gravity.
6.2 Algebraic Quantum Gravity Approach
As we have shown in the last subsection, although Thiemanns complexier coherent state has
a clear calculable mechanism and correct semi-classical properties in testing kinematical opera-
tors, it fails to be a qualied semi-classical state for the quantum dynamics since the semiclassi-
cal limit of the Hamiltonian constraint operator

S(N) or master constraint operator

M is clearly
not correct, both

S(N) and

M are graph-changing so that their expectation values with respect
to these cut-o coherent states are always zero. So a possible way to avoid such a problem is to
dene a non-graph-changing version of Hamiltonian constraint operator or similarly, a master
constraint operator. However, such a modication is hard to make in the framework of loop
quantum gravity since the action of Hamiltonian constraint operator always adds several arcs
on certain graphs. But if the framework of loop quantum gravity is suitably modied then it
turns out that a version of non-graph-changing Hamiltonian constraint operator can be proposed
and the semi-classical analysis for the quantum dynamics can be carried out with the complex-
ier coherent states dened previously. Such a modication is recently made by Thiemann in
[79][80][81] and is called algebraic quantum gravity (AQG) approach. We describe it briey in
what follows.
Algebraic quantum gravity is a new approach to canonical quantum gravity suggested by
loop quantum gravity. But in contrast to loop quantum gravity, the quantum kinematics of al-
gebraic quantum gravity is determined by an abstract -algebra generated by a countable set
of elementary operators labeled by a single algebraic graph with countably innite number of
edges, while in loop quantum gravity the elementary operators are labeled by a collection of
embedded graphs with nite number of edges. Thus one can expect that in algebraic quantum
gravity, we lose the information of the topological and dierential structure of the manifold in
all the quantization procedure before we do semi-classical analysis. Hence the quantum the-
ory will be of course independent of the topology and dierential structure of the manifold but
based only on an algebraic graph, which only contains the information of the number of vertices
and their oriented valence.
Denition 6.2.1: An oriented algebraic graph is an abstract graph specied by its adjacency
matrix , which is an N N matrix. One of its entries
I J
stand for the number of edges that
start at vertex I and end at vertex J. The valence of the vertex I is given by v
I
=
_
J
(
I J
+
JI
).
We also use V() and E() to denote the sets of vertices and edges respectively.
In our quantization procedure, we x a specic cubic algebraic graph with a countably in-
nite number of edges N = and the valence of each vertex v
I
= 2 dim(). Such a specic
74
choice, although it detracts from the generality of the theory, is practically sucient for our use
in the semiclassical analysis.
Given the algebraic graph , we dene a quantum -algebra by associating with each edge
e an element A(e) of a compact, connected, semisimple Lie group G and an element E
j
(e) take
value in its Lie algebra g. These elements are subject to the commutation relations
[

A(e),

A(e

)] = 0,
[

E
j
(e),

A(e

)] = iQ
2

e,e

j
/2

A(e),
[

E
j
(e),

A(e

)] = iQ
2

e,e
f
jkl

E
l
(e

),
and -relations

A(e)

= [

A(e)
1
]
T
,

E
j
(e)

=

E
j
(e),
where Q stands for the coupling constant,
j
is the generators in the Lie algebra g and f
jkl
is the
structure constant of g. We denote the abstract quantum -algebra generated by above elements
and relations by A.
A natural representation of A is the innite tensor product Hilbert space 1

=
e
1
e
where
1
e
= L
2
(G, d
H
)[145], whose element is denoted by
f

e
f
e
. Two elements
f
and
f
in
1

are said to be strongly equivalent if


_
e
< f
e
, f

e
>
1
e
1 converges. We denote by [ f ] the
strongly equivalence class containing
f
. It turns out that two elements in 1

are orthogonal
if they lie in dierent strongly equivalence classes. Hence the innite tensor Hilbert space 1

can be decomposed as a direct sum of the Hilbert subspaces (sectors) 1

[ f ]
which are the closure
of strongly equivalence classes [ f ]. Furthermore, although each sector 1

[ f ]
is separable and
has a natural Fock space structure, the whole Hilbert space 1

is non-separable since there are


uncountably innite number of strongly equivalence classes in it. Our basic elements in the
quantum algebra are represented on 1

in an obvious way

A(e)
f
:= [A(e) f
e
] [
e

e
f
e
],

E
j
(e)
f
:= [iQ
2
X
e
j
f
e
] [
e

e
f
e
].
As one might have expected, all these operators are densely dened and E
j
(e) is essentially
self-adjoint. Given a vertex v V(), the volume operator can be constructed by using the
operators we just dened

V
v
:=
3
p
_

1
48

e
1
e
2
e
3
=v

v
(e
1
, e
2
, e
3
)
i jk

E
i
(e
1
)

E
j
(e
2
)

E
k
(e
3
),
where the values of
v
(e
1
, e
2
, e
3
) should be assigned once for all for each vertex. When we
embed the algebraic graph into some manifold, the embedding should be consistent with the
assigned values of
v
(e
1
, e
2
, e
3
).
Then we discuss the quantum dynamics. By the regularization methods frequently used in
the last two sections, the half densitized constraints can be quantized to be composite operators
as we list below.
75
Gauss constraint

G
j
(v) :=

Q
(1/2)
v

e at v

E
j
(e);
Spatial dieomorphism constraint

D
j
(v) :=
1
E(v)

e
1
e
2
e
3
=v

v
(e
1
, e
2
, e
3
)
L(v, e
1
, e
2
)

L(v,e
1
,e
2
)
Tr(
j
[

A()

A()
1
]

A(e
3
)[

A(e
3
)
1
,
_

V
v
]);
Euclidean Hamiltonian constraint (up to an overall factor)

H
(r)
E
(v) :=
1
E(v)

e
1
e
2
e
3
=v

v
(e
1
, e
2
, e
3
)
L(v, e
1
, e
2
)

L(v,e
1
,e
2
)
Tr([

A()

A()
1
]

A(e
3
)[

A(e
3
)
1
,

V
(r)
v
]);
Lorentzian Hamiltonian constraint (up to an overall factor)

T(v) :=
1
E(v)

e
1
e
2
e
3
=v

v
(e
1
, e
2
, e
3
)
Tr(

A(e
1
)[

A(e
1
)
1
, [

H
(1)
E
,

V]]

A(e
2
)[

A(e
2
)
1
, [

A(e
3
)
1
, [

H
(1)
E
,

V]]


A(e
3
)[

A(e
3
)
1
,
_

V
v
]),

H(v) =

H
(1/2)
E
(v) +

T(v); (113)
where

V :=
_
v

V
v
,

H
(1)
E
:=
_
v

H
(1)
E
(v) and

Q
(r)
v
:=
1
E(v)

e
1
e
2
e
3
=v

v
(e
1
, e
2
, e
3
)
Tr(

A(e
1
)[

A(e
1
)
1
,

V
(r)
v
]

A(e
2
)[

A(e
2
)
1
,

V
(r)
v
]

A(e
3
)[

A(e
3
)
1
,

V
(r)
v
]).
L(v, e
1
, e
2
) denotes the set of minimal loops starting at v along e
1
and ending at v along e
1
2
. And
a loop L(v, e
1
, e
2
) is said to be minimal provided that there is no other loop within sat-
isfying the same restrictions with fewer edges traversed. Note that since we only have a single
cubic algebraic graph, the dieomorphism constraint can only be implemented by dening the
operators corresponding to dieomorphism generators because a nite dieomorphism trans-
formation is not meaningful in our algebraic treatment unless the algebraic graph is embedded
in a manifold. As a result, the (extended) master constraint can be expressed as a quadratic
combination:

M :=

vV()
[

G
j
(v)


G
j
(v) +

D
j
(v)


D
j
(v) +

H(v)


H(v)].
76
It is trivial to see that all the above operators are non-graph-changing and embedding indepen-
dent because we have only worked on a single algebraic graph so far. However, when we test
the semiclassical limit of these operators, especially the master constraint operators, we should
specify an embedding map X which map a algebraic graph to be an embedded one. With this
specic embedding, we can see the correspondence between the classical algebra of elementary
observables and the quantum -algebra. We dene the holonomy and suitably modied ux by
A(e) := A(X(e)) := /exp(
_
X(e)
A),
E
j
(e) := 2Tr[
j
_
S
e

abc
dx
a
dx
b
A(
e
(x))E
c
(x)A(
e
(x))
1
],
where S
e
is a face which intersects the edge X(e) only at an interior point p
e
of both S
e
and X(e).
We choose a system of paths
e
(x)
x
for all x S
e
, such that
e
(x) starts at s(X(e)) along X(e)
until p
e
and then runs within S
e
until x. As one might expect, the quantum-algebra we dened
previously is just consistent with the classical Poisson algebra generated by these holonomis
and uxes:
A(e), A(e

) = 0,
E
j
(e), A(e

) = Q
2

e,e

j
/2A(e),
E
j
(e), A(e

) = Q
2

e,e
f
jkl
E
l
(e

).
Then we consider the coherent states. By employing the Laplacian complexier on each
edge
C
e
:=
1
2Q
2
a
2
e
E
j
(e)E
j
(e),
the coherent state is obtained as it was in the last section:

t
e
e;(A,E)
(A)
t
e
e;g(A,E)
(A(e)) =

dim()e
t

(g(A, E)A(e)),
where

denotes the eigenvalue of the Laplacian on G and t =


2
p
/a
2
e
represents the classical-
ization parameter. The coherent state peaks at the complexied classical phase space point
g(A, E) :=

n=0
(i)
n
n!
C
e
, A(e)
n
= exp(iE(e)/a
2
e
)A(e),
note that the parameter a
e
is specied such that E(e)/a
2
e
is dimensionless. Hence the coherent
state on the whole graph is represented by an innite tensor product state:

t
A,E
(A) :=

eE()

t
e
e;(A,E)
(A)

t
e
e;(A,E)
(A)
.
The peakness, uctuation and other semiclassical properties of these states have been checked
in [143][144] in which the most important part is that
<
A,E


A(e)
A,E
>= A(e) <
A,E


E(e)
A,E
>= E(e)
77
up to terms which vanish faster than any power of t
e
as t
e
0. And the uctuations are small.
With the semiclassical state we just constructed, the expectation value of the above (ex-
tended) master constraint operator can be calculated and its semiclassical limit can be tested. In
the following, we summarize the result of the calculation. In [80], a semiclassical calculation
for the master constraint operator is carried out based on a cubic algebraic graph. The calcula-
tion makes use of a simplifying assumption: we substitute the gauge group for gravity S U(2)
by U(1)
3
. And the result of the calculation shows that in U(1)
3
case the (extended) master
constraint operator has correct semiclassical limit
lim
t0
<
t
m


M
t
m
>= M
cubic
[m] M[m] ( 0)
where m represents a phase space point and is the lattice parameter such that the lattice be-
come continuum as 0. In addition, it is shown that the next-to-leading order terms which
contribute to the uctuation of

M are nite.
Moreover, the calculation in [81] shows that the result of the exact non-Abelian calculation
matches precisely the results of the Abelian approximation, provided that we replace the clas-
sical U(1)
3
terms h
j
e
, p
e
j

j=1,2,3
by Tr(
j
A(e)), Tr(
j
E(e))
j=1,2,3
, which means that the theory
of algebraic quantum gravity admits a semiclassical limit whose innitesimal gauge symmetry
agrees with that of general relativity.
78
7 Conclusion and Discussion
As it was shown in the previous sections, loop quantum gravity oers a conceptually clear
and mathematically rigorous approach to quantize general relativity. In this approach, we are
seeking new physics deeply below the Planck scale. In the kinematical framework, a quantum
Riemannian geometry is established and some geometrical operators, e.g. area, volume, are
well-dened, and their spectrum are shown to be discrete, which means that the structure of the
space may be discrete below the Planck scale. Such a new phenomena sheds light on quantum
eld theory, lattice gauge theory and their renormalization. Moreover, the program in the quan-
tum dynamics of loop quantum gravity represents signicant progress in the research area of
quantum gravity. Before loop quantum gravity, the quantum Wheeler-DeWitt equation was only
a formal equation and from concrete calculations. However, in the framework of loop quantum
gravity, we already have a well-dened quantum Hamiltonian constraint operator which has
an explicit action on kinematical states, so that the quantum Wheeler-DeWitt equation is well-
dened in loop quantum gravity. On the other hand, the matter eld can also be quantized in
this framework and we show that the matter Hamiltonian is free of UV-divergence and dont
need a renormalization process. Furthermore, with the coupled matter eld, a matter coupled
Hamiltonian constraint operator is obtained so that the problem of time may be solved and, such
an idea is being translated into a new understanding of the early universe in the context of loop
quantum cosmology.
Although great progress has been made, as an unnished framework, loop quantum gravity
still has many issues to be solved in the future research. To conclude this thesis, we list some of
those in the following:
First of all, we dont have the complete solutions for either Hamiltonian constraint equa-
tion or master constraint equation. Thus one cannot explicitly construct the physical
Hilbert space for loop quantum gravity. So the quantum dynamics of gravity is essen-
tially unknown so far.
To make contact with experimental results, one should know the observables in the quan-
tum theory which have to be invariant under gauge transformation. However, some of the
Dirac observables that have been constructed involve an innite number of derivatives
and extremely hard to manage [150][61][62].
The semiclassical limit of loop quantum gravity is unknown so far, although a great deal
of progress has been made in the context of algebraic quantum gravity. And in alge-
braic quantum gravity, further research work is needed to show the uctuation of master
constraint operator should be small.
As it was shown at the end of section 4.3, the regularization process for the Hamiltonian
constraint operator is ambiguous and there is a list of free parameters. Thus it is also a
research project to remove as many ambiguities as possible. And some work has been
recently done in this direction [110].
The Immirzi parameter is another free parameter in the framework of loop quantum grav-
ity, which comes in with the classical formulation. In the classical theory, dierent values
79
of the Immirzi parameter label equivalent classical theories since they are connected by
canonical transformations. However, in quantum theory, it is problematic because the
representation with dierent Immirzi parameter are not unitarily equivalent.
The construction of loop quantum gravity crucially depends on the compactness of its
gauge group S U(2), which comes from an internal partial gauge xing. And it is ar-
gued that the internal Lorentz symmetry is broken in a non-natural way [131]. So it
seems to be better to switch back to the complex Ashtekar variables which are free
of the internal gauge xing and preserve the internal Lorentz symmetry and, will also
greatly simplify the Hamiltonian constraint. However, the price is that we should work
on non-compact gauge group S L(2, C), and there is no satisfactory quantization pro-
gramme for the S L(2, C)-gravity so far, although some work has been done in this di-
rection [71][105][106].
As it was shown in section 6.2, algebraic quantum gravity provides a clear way to make
many calculations accessible and the semiclassical analysis can be carried out in this
framework. And it seems that the construction of algebraic quantum gravity admits the
non-compact internal gauge group (or non-compact reduced conguration space) since
there is only one graph in quantization process. However, in this framework, the restric-
tion of the possible representation is so loose that there even exists a possible representa-
tion in which the spectrum of geometrical operators are continuous.
The transition amplitude calculation for loop quantum gravity is accessible in the so-
called spin foam model and depends on a unclear conjecture of GFT/spinfoam duality.
On the other hand, it is still not clear how to build a path-integral formulation and connect
with spin foam models from the canonical approach, and such a connection may give the
needed support for GFT/spinfoam duality.
We have constructed the dynamics of matter quantum eld theory on a quantum back-
ground in section 5. However, it is not clear how we can make the connection with the
ordinary quantum eld theory on curved spacetime. Moreover, it is still an problem how
to nd the non-perturbative correspondence of Hadamard states in perturbative quantum
eld theory in curved spacetime, although some hints of a connection may come out in
spin foam calculations [127][42].
80
References
[1] M. Arnsdorf and S. Gupta, Loop quantum gravity on non-compact spaces, Nucl. Phys. B
577, 529 (2000).
[2] A. Ashtekar, New variables for classical and quantum gravity, Phys. Rev. Lett. 57, 2244
(1986).
[3] A. Ashtekar, New Hamitonian formulation of general relativity, Phys. Rev. D 36, 1587
(1987).
[4] A. Ashtekar, Lectures on Non-perturbative Canonical Gravity, (World Scientic, Singa-
pore, 1991).
[5] A. Ashtekar, J. C. Baez, and K. Krasnov, Quantum geometry of isolated horizon and black
hole entropy, Adv. Theor. Math. Phys. 4, 1 (2000).
[6] A. Ashtekar, C. Beetle, and J. Lewandowski, Mechanics of rotating isoloted horizons,
Phys. Rev. D 64, 044016 (2001).
[7] A. Ashtekar, C. Beetle, and J. Lewandowski, Geometry of generic isolated horizons, Class.
Quantum Grav. 19, 1195 (2002).
[8] A. Ashtekar and M. Bojowald, Black hole evaporation: Aparadigm, Class. QuantumGrav.
22, 3349 (2005).
[9] A. Ashtekar and M. Bojowald, Quantum geometry and the Schwarzschild singularity,
Class. Quantum Grav. 23, 391 (2006).
[10] A. Ashtekar, M. Bojowald, and J. Lewandowski, Mathematical structure of loop quantum
cosmology, Adv. Theor. Math. Phys. 7, 233 (2003).
[11] A. Ashtekar, L. Bombelli, and A. Corichi, Semiclassical states for constrained systems,
Phys. Rev. D 72, 025008 (2005).
[12] A. Ashtekar, J. Engle, and C. van der Broeck, Quantum horizons and black hole entropy:
Inclusion of distortion and rotation, Class. Quantum Grav. 22, L27 (2005).
[13] A. Ashtekar, J. Engle, T. Powlowski, and C. van der Broeck, Mutipole moments of isolated
horizons, Class. Quantum Grav. 21, 2549 (2004).
[14] A. Ashtekar, S. Fairhurst, and B. Krishnan, Isolated horizons: Hamiltonian evolution and
the rst law, Phys. Rev. D 62, 104025 (2000).
[15] A. Ashtekar, S. Fairhurst, and J. Willis, Quantum gravity, shadow state and quantum me-
chanics, Class. Quantum Grav. 20, 1031 (2003).
[16] A. Ashtekar and C. J. Isham, Representation of the holonomy algebras of gravity and
non-Abelian gauge theories, Class. Quantum Grav. 9, 1433 (1992).
81
[17] A. Ashtekar and J. Lewandowski, Projective techniques and functional integration, J.
Math. Phys. 36, 2170 (1995).
[18] A. Ashtekar and J. Lewandowski, Dierential geometry on the space of connections via
graphs and projective limits, J. Geom. Phys. 17, 191 (1995).
[19] A. Ashtekar and J. Lewandowski, Relation between polymer like and Fock excitations,
Class. Quantum Grav. 18, L117 (2001).
[20] A. Ashtekar and J. Lewandowski, Background independent quantum gravity: A status
report, Class. Quantum Grav. 21, R53 (2004), (preprint: gr-qc/0404018).
[21] A. Ashtekar and J. Lewandowski, Quantum theory of geometry I: Area operators, Class.
Quantum Grav. 14, A55 (1997), (preprint: gr-qc/9602046).
[22] A. Ashtekar and J. Lewandowski, Quantum theory of geometry II: Volume operators, Adv.
Theor. Math. Phys. 1, 388 (1997), (preprint: gr-qc/9711031).
[23] A. Ashtekar, A. Corich, and J. A. Zapata, Quantum theory of geometry III: Non-
commutativity of Riemannian structures, Class. Quantum Grav. 15, 2955 (1998).
[24] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mour ao, and T. Thiemann, A manifest gauge-
invariant approach to quantum theories of gauge elds, (preprint: hep-th/9408108).
[25] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mour ao, and T. Thiemann, Quantization of
dieomorphism invariant theories of connections with local degrees of freedom, J. Math.
Phys. 36, 6456 (1995).
[26] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mour ao, and T. Thiemann, Coherent state
transforms for spaces of connections, J. Funct. Anal. 135, 519 (1996).
[27] A. Ashtekar, J. Lewandowski, and H. Sahlmann, Polymer and Fock representations for a
scalar eld, Class. Quantum Grav. 20, L11 (2003).
[28] A. Ashtekar, T. Powlowski and P. Singh, Quantum nature of the big bang, Phys. Rev. Lett.
96, 141301 (2006).
[29] A. Ashtekar, T. Powlowski and P. Singh, Quantum nature of the big bang: improved dy-
namics, Phys. Rev. D 74, 084003(2006).
[30] A. Ashtekar, J. Romano, R. Tate, New variables for gravity: Inclusion of matter, Phys.
Rev. D 40, 2572 (1989).
[31] A. Ashtekar, C. Rovelli, and L. Smolin, Weaving a classical geometry with quantum
threads, Phys. Rev. Lett. 69, 237 (1992).
[32] J. C. Baez, Spin networks and quantum gravity, in The Interface of Knots and Physics, ed.
L. Kauman (Providence, RI: American Mathematical Society, 1996).
82
[33] J. C. Baez, An introduction to spin foam models of quantum gravity and BF theory, Lect.
Notes Phys. 543, 25 (2000).
[34] J. Barbero, Real Ashtekar variables for Lorentzian signature spacetimes, Phys. Rev. D 51,
5507 (1995).
[35] J. W. Bardeen, B. Carter, and S. W. Hawking, The four laws of black hole mechanics,
Commun. Math. Phys. 31, 161 (1973).
[36] J. W. Barrett and L. Crane, Relativistic spin networks and quantum gravity, J. Math. Phys.
39, 3296 (1998).
[37] J. W. Barrett and L. Crane, A Lorentzian signature model for quantum general relativity,
Class. Quantum Grav. 17, 3101 (2000).
[38] D. Di Bartolo, R. Gambini, J. Griego, and J. Pullin, Consistent quantization of general
relativity in the space of Vassiliev knot invariants, Phys. Rev. Lett. 84, 2314 (2000).
[39] D. Di Bartolo, R. Gambini, J. Griego, and J. Pullin, Canonical quantum gravity in the
Vassiliev knot invariants arena: I. Kinematical structure, Class. Quantum Grav. 17, 3211
(2000).
[40] D. Di Bartolo, R. Gambini, J. Griego, and J. Pullin, Canonical quantum gravity in the
Vassiliev knot invariants arena: II. Constraints, habitats and consistency, Class. Quantum
Grav. 17, 3239 (2000).
[41] J. D. Bekenstein, Black holes and entropy, Phys. Rev. D 7, 2333 (1973).
[42] E. Bianchi, L. Modesto, C. Rovelli, S. Speziale, Graviton propagator in loop quantum
gravity, Class.Quantum Grav. 23, 6989, (2006).
[43] M. Bojowald, Absence of singularity in loop quantum cosmology, Phys. Rev. Lett. 86,
5227 (2001), (preprint: gr-qc/0102069).
[44] M. Bojowald, Dynamical initial conditions in quantum cosmology, Phys. Rev. Lett. 87,
121301 (2001).
[45] M. Bojowald, Ination from quantum geometry, Phys. Rev. Lett. 89, 261301 (2002).
[46] M. Bojowald, Initial conditions for a universe, Gen. Rel. Grav. 35, 1877 (2003).
[47] M. Bojowald, Spherically symmetric quantum geometry: States and basic operators,
Class. Quantum Grav. 21, 3733 (2004).
[48] M. Bojowald, Nonsingular black holes and degrees of freedom in quantum gravity, Phys.
Rev. Lett. 95, 061301 (2005).
[49] M. Bojowald, Elements of loop quantum cosmology, preprint: gr-qc/0505057.
[50] M. Bojowald, Loop quantum cosmology, Living Rev. Relativity, 8, (2005), 11.
83
[51] M. Bojowald, R. Goswami, R. Maartens, and P. Singh, A black hole mass threshold from
non-singular quantum gravitational collapse, Phys. Rev. Lett. 95, 091302 (2005).
[52] M. Bojowald and H. A. Kastrup, Symmetry reduction for quantized dieomorphism in-
variant theories of connections, Class. Quantum Grav. 17, 3009 (2000).
[53] T. Br ocker and T. Dieck, Representations of Compact Lie Group, (Springer-Verlag, 1985).
[54] J. D. Brown, K. V. Kuchar, Dust as a standard of space and time in canonical quantum
gravity, Phys. Rev. D51, 5600 (1995), (preprint: gr-qc/9409001).
[55] S. Carlip, Quantum gravity: a progress report, Rept. Prog. Phys. 64, 885 (2001).
[56] M. Carmeli and S. Malin, Theory of Spinor: An Introduction, (World Scientic, 2000).
[57] Y. Choquet-Bruhat, C. Dewitt-Morette, M. Dillard-Bleick, Analysis, Manifold, and
Physics, (North-Holland Publishing Company, 1977).
[58] J. B. Conway, A Course in Functional Analysis, Second Edition, (Springer-Verlag, 1990).
[59] L. Crane, A. Perez, and C. Rovelli, Perturbative niteness in spin-foam quantum gravity,
Phys. Rev. Lett. 87, 181301 (2001).
[60] G. Date and G. M. Hossain, Genericity of ination in isotropic loop quantum cosmology,
Phys. Rev. Lett. 94, 011301 (2005).
[61] B. Dittrich, Partial and complete observable for Hamiltonian contrained system, (preprint:
gr-qc/0411013).
[62] B. Dittrich, Partial and complete observable for canonical general relativity, (preprint: gr-
qc/0507106).
[63] B. Dittrich and T. Thiemann, Testing the master constraint programme for loop quan-
tum gravity I. General framework, Class. Quantum Grav. 23, 1025 (2006), (preprint: gr-
qc/0411138).
[64] B. Dittrich and T. Thiemann, Testing the master constraint programme for loop quantum
gravity: II. Finite dimensional systems, Class. Quantum Grav. 23, 1067 (2006), (preprint:
gr-qc/0411139).
[65] B. Dittrich and T. Thiemann, Testing the master constraint programme for loop quan-
tum gravity: III. S L(2, R) models, Class. Quantum Grav. 23, 1089 (2006), (preprint: gr-
qc/0411140).
[66] B. Dittrich and T. Thiemann, Testing the master constraint programme for loop quantum
gravity: IV. Free eld theories, Class. Quantum Grav. 23, 1121 (2006), (preprint: gr-
qc/0411141).
84
[67] B. Dittrich and T. Thiemann, Testing the master constraint programme for loop quantum
gravity: V. Interecting eld theories, Class. Quantum Grav. 23, 1143 (2006), (preprint:
gr-qc/0411142).
[68] M. Domagala and J. Lewandowski, Black hole entropy from quantum gravity, Class.
Quantum Grav. 21, 5233 (2004).
[69] W. Fairbain and C. Rovelli, Separable Hilbert space in loop quantum gravity, J. Math.
Phys. 45, 2802 (2004).
[70] E. E. Flanagan and R. M. Wald, Does backreaction enforce the averaged null energy con-
dition in semiclassical gravity, Phys. Rev. D54, 6233 (1996).
[71] L. Freidel and E. Livine, Spin networks for non-Compact groups, J. Math. Phys. 44, 1322
(2003).
[72] R. Gambini, J. Lewandowski, D. Marolf, and J. Pullin, On the consistency of the constraint
algebra in spin network quantum gravity, Int. J. Mod. Phys. D 7, 97 (1998).
[73] R. Gambini and J. Pullin, Loops, Knots, Gauge Theories and Quantum Gravity, (Cam-
bridge University Press, Cambridge, 1996).
[74] R. Gambini and J. Pullin, Canonical quantization of general relativity in discrete space-
times, Phys. Rev. Lett. 90, 021301 (2003).
[75] R. Gambini and J. Pullin, Consistent discretization and loop quantum geometry, Phys.
Rev. Lett. 94, 101302 (2005).
[76] R. Gambini and J. Pullin, Discrete spacetime, Preprint: gr-qc/0505023.
[77] K. Giesel and T. Thiemann, Consistency check on volume and triad operator quantisation
in loop quantum gravity I, (preprint: gr-qc/0507036).
[78] K. Giesel and T. Thiemann, Consistency check on volume and triad operator quantisation
in loop quantum gravity II, (preprint: gr-qc/0507037).
[79] K. Giesel and T. Thiemann, Algebraic quantum gravity (AQG) I. Conceptual setup,
preprint: gr-qc/0607099.
[80] K. Giesel and T. Thiemann, Algebraic quantum gravity (AQG) II. Semiclassical analysis,
preprint: gr-qc/0607100.
[81] K. Giesel and T. Thiemann, Algebraic quantum gravity (AQG) III. Semiclassical pertur-
bation theory, preprint: gr-qc/0607101.
[82] R. Giles, Reconstruction of gauge potentials from Wilson loops, Phys. Rev. D 24, 2160
(1981).
[83] J. Glimm and A. Jae, Quantum Physics, (Springer-Verlag, New York, 1987).
85
[84] N. Guilini and D. Marolf, On the generality of rened algebraic quantization, Class. Quan-
tum Grav. 16, 2479, (1999).
[85] R. Haag, Local Quantum Physics, 2nd ed. (Springer-Verlag, Berlin, 1996).
[86] B. C. Hall, The Segal-Bergmann coherent state transformation for compact Lie groups, J.
Funct. Anal. 122, 103 (1994).
[87] M. Han, Y. Ma, Y. Ding, and L. Qin, Hamiltonian Analysis of n-dimensional Palatini
Gravity with Matter, Mod. Phys. Lett. A20 725 (2005), (preprint: gr-qc/0503024).
[88] M. Han and Y. Ma, Master constraint operators in loop quantum gravity, Phys. Lett. B635,
225 (2006), (preprint: gr-qc/0510014).
[89] M. Han and Y. Ma, Dynamics of a scalar eld in a polymer-like representation, Class.
Quantum Grav. 23, 2741 (2006).
[90] M. Han, W. Huang and Y. Ma, Fundamental structure of loop quantum gravity, preprint:
gr-qc/0509064.
[91] S. W. Hawking, and G. F. R. Ellis, The large scale structure of space-time, (Cambridge
University Press, 1973).
[92] S. Hollands, R. M. Wald, On the renormalization group in curved spacetime, Commun.
Math. Phys. 237 123 (2003).
[93] S. Holst, Barberos Hamiltonian derived from a generalized Hilbert-Palatini action, Phys.
Rev. D 53, 5966 (1996) (preprint: gr-qc/9511026).
[94] R. K. Kaul and P. Majumdar, Logarithmic corrections to the Bekenstein-Hawking entropy,
Phys. Rev. Lett. 84, 5255 (2000).
[95] J. Lewandowski and D. Marolf, Loop constraints: a habitate and their algebra, Int. J. Mod.
Phys. D 7, 299 (1998).
[96] J. Lewandowski, A. Okolow, H. Sahlmann, T. Thiemann, Uniqueness of dieomorphism
invariant states on holonomy-ux algebras, (preprint: gr-qc/0504147).
[97] C. Liang, Introductory dierential geometry and general relativity I, II, (Beijing Normal
University Press, 2000, in Chinese).
[98] Y. Ma, Quantization of static spacetimes, Phys. Rev. D 65, 064012 (2002).
[99] Y. Ma and Y. Ling, Classical geometry from a physical state in canonical quantum gravity,
Phys. Rev. D 62, 064030 (2000).
[100] Y. Ma and Y. Ling,

Q operator for canonical quantum gravity, Phys. Rev. D 62, 104021
(2000).
86
[101] L. Modesto, Disappearance of the black hole singularity in loop quantum gravity, Phys.
Rev. D 70, 124009 (2004).
[102] L. Modesto, The Kantowski-Sachs space-time in loop quantum gravity, preprint: gr-
qc/0411032.
[103] L. Modesto, Quantum gravitational collapse, (preprint: gr-qc/0504043).
[104] M. Nakahara, Geometry, Topology, and Physics, Second Edition (Institute of Physics
Publishing, 2003).
[105] A. Okolow, Hilbert space built over connections with a non-compact structure group,
Class. Quant. Grav. 22, 1329 (2005).
[106] A. Okolow, Quantization of dieomorphism invariant theories of connections with a non-
compact structure group - an example, (preprint: gr-qc/0605138).
[107] P. Peld` an, Actions for gravity, with generalizations: A review, Class. Quantum Grav. 11,
1087 (1994), (preprint: gr-qc/9305011).
[108] A. Perez, Finiteness of a spin foam model for Euclidean quantum general relativity, Nucl.
Phys. B 599, 427, (2001).
[109] A. Perez, Spin foam models for quantum gravity, Class. Quantum Grav. 20, R43 (2003).
[110] A. Perez, On the regularization ambiguities in loop quantum gravity, Phys.Rev. D 73,
044007 (2006).
[111] A. Perez and C. Rovelli, Spin foam model for Lorentzian general relativity, Phys. Rev. D
63, 041501 (2001).
[112] A. Perez and C. Rovelli, Physical eects of the Immirzi parameter, Phys.Rev. D 73,
044013 (2006).
[113] M. Reed, and B. Simon, Methods of Modern Mathematical Phycics I: Functional Analy-
sis, (Academic Press, 1973).
[114] M. Reed and B. Simon, Methods of Modern Mathematical Physics II: Fourer Analysis,
Self-adjointness, (Academic Press, 1975), Page 177.
[115] M. Reisenberger and C. Rovelli, Spacetime states and covariant quantum theory, Phys.
Rev. D 65, 125016 (2002).
[116] C. Rovelli, What is observable in classical and quantum gravity, Class. Quantum Grav.
8, 1895 (1991).
[117] C. Rovelli, Black hole entropy from loop quantum gravity, Phys. Rev. Lett. 14, 3288
(1996).
[118] C. Rovelli, Loop quantum gravity, Living Rev. Relativity 1 (1998), 1.
87
[119] C. Rovelli, A note on the foundation of relativistic mechanics. I: Relativistic observables
and relativistic states, (preprint: gr-qc/0111037).
[120] C. Rovelli, A note on the foundation of relativistic mechanics. II: Covariant Hamiltonian
general relativity, (preprint: gr-qc/0202079).
[121] C. Rovelli, Partial observables, Phys. Rev. D 65, 124013 (2002).
[122] C. Rovelli, Quantum Gravity, (Cambridge University Press, 2004).
[123] C. Rovelli and L. Smolin, Loop representation for quantum general relativity, Nucl. Phys.
B 331, 80 (1990).
[124] C. Rovelli and L. Smolin, The physical Hamiltonian in nonperturbative quantum gravity,
Phys. Rev. Lett 72, 446 (1994).
[125] C. Rovelli and L. Smolin, Discreteness of area and volume in quantum gravity, Nucl.
Phys. B 442, 593, (1995).
[126] C. Rovelli and L. Smolin, Spin networks and quantum gravity, Phys. Rev. D 52, 5743,
(1995).
[127] C. Rovelli, Graviton propagator from background-independent quantum gravity,
Phys.Rev.Lett. 97, 151301, (2006).
[128] H. Sahlmann and T. Thiemann, Towards the QFT on curved spacetime limit of QGR: I.
A general scheme, Class. Quantum Grav. 23, 867 (2006), (preprint: gr-qc/0207030).
[129] H. Sahlmann and T. Thiemann, Towards the QFT on curved spacetime limit of QGR:
II. A concrete implementation, Class. Quantum Grav. 23, 909 (2006), (Preprint: gr-
qc/0207031).
[130] H. Sahlmann and T. Thiemann, Irreducibility of the Ashtekar-Isham-Lewandowski rep-
resentation, (preprint: gr-qc/0303074).
[131] J. Samuel, Is Barberos Hamiltonian formulation a gauge throry of Lorentzian gravity?
Class. Quantum Grav. 17, L141 (2000), (Preprint: gr-qc/0005095).
[132] S. Sternberg, Group Theory and Physics, (Cambridge University Press, 1994).
[133] lecture notes given by T. Thiemann at Beijing Normal University.
[134] T. Thiemann, Quantum spin dynamics (QSD), Class. Quantum Grav. 15, 839 (1998).
[135] T. Thiemann, Quantum spin dynamics (QSD): II. The kernel of the Wheeler - DeWitt
constraint operator, Class. Quantum Grav. 15, 875 (1998).
[136] T. Thiemann, Quantum spin dynamics (QSD): III. Quantum constraint algebra and phys-
ical scalar product in quantum general relativity, Class. Quantum Grav. 15, 1207 (1998).
88
[137] T. Thiemann, Quantum spin dynamics (QSD): IV. Euclidean quantum gravity as a model
to test Lorentzian quantum gravity, Class. Quantum Grav. 15, 1249 (1998).
[138] T. Thiemann, Quantum spin dynamics (QSD): V. Quantum gravity as the natural regula-
tor of the Hamiltonian constraint of matter quantum eld theories, Class. Quantum Grav.
15, 1281 (1998).
[139] T. Thiemann, Quantumspin dynamics (QSD): VI. QuantumPoincare algebra and a quan-
tum positivity of energy theorem for canonical quantum gravity, Class. Quantum Grav. 15,
1463 (1998).
[140] T. Thiemann, Quantum spin dynamics (QSD): VIII. The master constraint, Class. Quan-
tum Grav. 23, 2249 (2006).
[141] T. Thiemann, A length operator for canonical quantum gravity, J. Math. Phys. 39, 3372
(1998).
[142] T. Thiemann, Gauge eld theory coherent states (GCS): I. General properties, Class.
Quantum Grav. 18, 2025 (2001).
[143] T. Thiemann and O. Winkler, Gauge eld theory coherent states (GCS): II. Peakedness
properties, Class. Quantum Grav. 18, 2561 (2001).
[144] T. Thiemann and O. Winkler, Gauge eld theory coherent states (GCS): III. Ehrenfest
theorems, Class. Quantum Grav. 18, 4629 (2001).
[145] T. Thiemann and O. Winkler, Gauge eld theory coherent states (GCS): IV. Innite tensor
product and thermodynamical limit, Class. Quantum Grav. 18, 4997 (2001).
[146] T. Thiemann, Modern Canonical Quantum General Relativity, (Cambridge University
Press, in press), Draft preprint: gr-qc/0110034.
[147] T. Thiemann, Complexier coherent states for quantum general relativity, Class. Quan-
tum Grav. 23, 2063 (2006), (preprint: gr-qc/0206037).
[148] T. Thiemann, Lectures on loop quantum gravity, Lect. Notes Phys. 631 (2003) 41,
(preprint: gr-qc/0210094).
[149] T. Thiemann, The phoenix project: Master constraint programme for loop quantum grav-
ity, Class. Quantum Grav. 23, 2211 (2006), (preprint: gr-qc/0305080).
[150] T. Thiemann, Reduced phase space quantization and Dirac observables, Class. Quantum
Grav. 23, 1163 (2006), (preprint: gr-qc/0411031).
[151] T. Thiemann, Loop quantum gravity: an inside view, preprint: hep-th/0608210.
[152] M. Varadarajan, Fock representation from U(1) holonomy agebras, Phys. Rev. D 61,
104001 (2000).
89
[153] M. Varadarajan, Photons from quantized electric ux representations, Phys. Rev. D 64,
104003 (2001).
[154] M. Varadarajan, Gravitons from a loop representation of linearized gravity, Phys. Rev. D
66, 024012 (2002).
[155] J. M. Velhinho, A groupoid approach to spaces of generalized connections, J. Geom.
Phys. 41, 166 (2002).
[156] R. M. Wald, General Relativity, (The University of Chicago Press, 1984).
[157] R. M. Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermody-
namics, (The University of Chicago Press, 1994).
[158] S. Weinberg, The Quantum Theory of Fields I, II, III, (Cambridge University Press, New
York, 1995.)
[159] C. M. Will, The Confrontation between general relativity and experiments, Living Rev.
Relativity 4 (2001), 4.
90
Vita
Muxin Han was born in Beijing, Peoples Republic of China. He studied as an undergraduate
student in the Department of Physics at Beijing Normal University from 2001 to 2005, and
obtained his Bachelor of Science degree at Beijing Normal University in 2005. Muxin came
to the United States and began his graduate studies at Louisiana State University in August of
2005. His major is physics.
91