Sie sind auf Seite 1von 60

Propagating Scalar Modes in (2 + 1)-Dimensional

CDT Gravity
A Senior Thesis Presented to the Physics Faculty of the
University of California, Berkeley
Adam Lloyd Bruce
Contents
1 Why CDT Quantum Gravity? 3
1.1 Quantization and the Fundamental Forces . . . . . . . . . . . . . . . . . . . 3
1.2 Nonrenormalizability of Canonical Quantization . . . . . . . . . . . . . . . . 5
1.3 Causal Dynamical Triangulations . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Classical Gravity in (2 + 1)- Dimensions 9
2.1 general relativity in (2 + 1)- Dimensions . . . . . . . . . . . . . . . . . . . . 10
2.1.1 The Newtonian Limit . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 The [0, 1] Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 The ADM Formalism and Hamiltionian Formulation . . . . . . . . . . . . . 15
2.3.1 The ADM Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.2 The Hamiltonian and Canonical Constraints . . . . . . . . . . . . . 17
2.4 (2 + 1)- Path Integrals in the ADM Formalism . . . . . . . . . . . . . . . . 18
2.4.1 Equivalence to 2+1 Canonical Quantization . . . . . . . . . . . . . . 19
3 Nonperturbative Quantum Gravity with (2 + 1)-CDT 24
3.1 The Gravitational Path integral via Triangulation . . . . . . . . . . . . . . . 24
3.1.1 Continuum Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 The CDT Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1
3.2 Horava-Lifshitz Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Propagating Scalar Modes and Anisotropic Quantization . . . . . . . . . . . 30
4 Computational Approach 34
4.1 The Regge Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.1 The non-regularized measure space and associated transition amplitude 36
4.2 Monte-Carlo Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.1 Markov Chain Stepping . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.2 Sampling an Observable . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 The Metropolis Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.1 Equilibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3.2 Monte-Carlo in CDT: The Pachner Moves . . . . . . . . . . . . . . . 42
5 Simulation 44
5.1 Generation of the Seed Manifold . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.1 Adaptive Remeshing . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.1.2 Computation of Simplicial Geodesics . . . . . . . . . . . . . . . . . . 46
5.1.3 Extraction of the First Quadrant Wedge . . . . . . . . . . . . . . . . 48
6 Conclusion 49
2
List of Figures
2.1 Diagram of the triangulation used to compute ds
2
. . . . . . . . . . . . . . . 16
3.1 Disallowed Histories in Causal Dynamical Triangulations, such as wormhole
histories and branching points where the light cones are degenerate. . . . . 27
5.1 Triangulation by remeshing as presented in [20] . . . . . . . . . . . . . . . . 46
5.2 The triangulated manifold with simplicial great circles along the xz and yz
planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.3 The triangulated manifold with simplicial great circles along the xz and yz
planes, and positive quadrant removed . . . . . . . . . . . . . . . . . . . . . 48
3
List of Tables
1.1 The fundamental forces, their strengths, ranges, gauge particles and La-
grangians as in [2] and [3]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4
Abstract
Causal Dynamical Triangulations (CDT) quantum gravity is a relatively new quantization
scheme which uses simplicial latices to regularize the gravitational path integral. Recently
there has been much conjecture by scholars in the quantum gravity community that such
a quantization procedure may be equivalent to Horava-Lifshitz gravity in at least (2 + 1)-
dimensions. We outline a procedure for testing this numerically and generate a seed mani-
fold to perform such a simulation, using the fact, derived in the text, that the anisotropy of
Horava-Lifshitz gravity leads to at least one propagating scalar mode in (2+1)-dimensions
while traditional general relativity does not. We rst describe quantization from a historical
perspective, outline relevant results from (2 +1)-dimensional traditional general relativity,
discuss CDT and Horava-Lifshitz theory, discuss the general computational procedure of
CDT codes and nally describe the method for seed generation, including the adaptive
remeshing technique for generating suitable triangulations.
Acknowledgements
No one can accomplish such a monolithic task as this in isolation, and I have had the good
fortune to meet and hear the wisdom of many brilliant minds and kind hearts along the
way. First I must thank Dr. Patrick Zulkowski, who approached me with the idea on which
this thesis was written in early 2012, and whose patience, kindness, intelligence and general
good-spiritedness are certainly not surpassed at UC Berkeley and I suspect will be dicult
to surpass for anyone I will ever know. Next I must thank the UC Davis quantum gravity
group, who provided essential input and generously supplied us with their codes, previous
work, and have promised computing time for the remainder of the project. I also thank
Professor Michael DeWeese for agreeing to advice this thesis despite the fact that it was not
in his eld, simply because Dr. Zulkowski is a member of his biophysics group. A scientist
who stands so steadfastly and selessly by his students is by any standard an honorable
and kind man. Finally I must thank Professor Kai Hormann, who for no other reason
but his great generosity and kind-heartedness provided expert opinions, suggestions and
even results from previous work to understand the adaptive remeshing technique, including
supplying us with a basic triangulation to build the seed manifold. Without any of these
great people, this thesis would not be what it is, and might not have been anything at all.
No agency funded this work beyond those funding the afore mentioned scholars. It has
rather been a labor of love on my own part, and I suspect on the part of everyone else.
1
Dedication
Aig mAthair,
Bha ` thu a se`olta mi e`olas dean sinn sa` or,
agus se`olta mi a le`olas cha do chuir e cas am
br`oig a mise stad. Airson tha e`olas rud
cumhachdach s breatha, agus tha mi an
d`ochas gun gabh sinne am gliocas feum ` mh`ath.
2
Chapter 1
Why CDT Quantum Gravity?
1.1 Quantization and the Fundamental Forces
Nature, insofar as we currently know, is governed by four fundamental forces: the strong
force, the weak force, the electromagnetic force and the gravitational force. These form the
basis of all interactions which particles and elds engage in. Moreover, it is a reasonable
assumption that these forces should conform to the physics we understand to exist in the
experimental and theoretical regimes which we have studied. Hence we expect each force
to be consistent with special or general relativity and quantum mechanics. While each of
the forces conforms to the requirements of relativity, there is no quantum theory of gravity
which is currently accepted and experimentally tested.
It was Dirac who rst studied the quantization of the electromagnetic eld, developing
an approach to the theory known as traditional or canonical quantization. In canonical
quantization we assume the system to be governed by a Hamiltonian ] to which there is
associated a set of observables O
i
which obey Poisson bracket relations. The canonically
quantized version of the theory maps Poisson brackets to canonical commutation relations
3
according to
O
i
, O
j

1
i
[O
i
, O
j
], (1.1)
though this rule is not mathematically rigorous and therefore selection of proper observables
is the subject of ongoing debate, especially in quantum gravity. The electrodynamic eld
may for example be quantized via canonical quantization [1].
Another method of quantization which can be shown to be equivalent to the canonical
formalism is the path integral formalism. By considering the action over all possible state
space congurations, it is possible to compute a probability kernel for the particle or eld
to propagate to some state from some initial state
0
. By writing this as a path integral
over a nite Hamiltonian system, this kernel may be expressed as
K(, t,
0
, t
0
) =
_
T[] exp
_
i

_
t
t
0
L( , , t)d
n
x
_
, (1.2)
where the invariant measure on the space of trajectories is dened as
T[] := lim
N
_
i
2
_
N/2
_
_
N1

j=1
d
j
_
_
, (1.3)
with = it known as a Wick rotation[4], and L( , , t) is the Lagrangian density of the
system from which is related to the total Lagrangian via the integral relation
L :=
_
dtL( , , t). (1.4)
It is using this method that quantum electrodynamics can be tested to its furthest ex-
tremes, most noticeably to produce an accurate prediction of the electrons gfactor as
g 2.00231930419922 (1.5 10
12
) (see [6] or [7]).
Hence a quantum theory using path integral quantization may be entirely determined
4
Force Relative Strength Range (m) Gauge Boson Lagrangian
Strong 1 10
15
gluon L
QCD
=

i
(i(

)
ij
m
ij
)
j
1/4(G
a

a
)
Weak 10
6
10
18
W

, Z
0
L
EW
Electromagnetic 1/137 photon L
QED
= ic

mc
2

1/(4
0
)F

Gravitation 6 10
39
Graviton (posited) unknown
Table 1.1: The fundamental forces, their strengths, ranges, gauge particles and Lagrangians
as in [2] and [3].
by a given Lagrangian, or from a more fundamental standpoint a Hamiltonian density ],
which can be transformed uniquely into a Lagrangian, [2]. Lattice methods, which shall
be of particular importance to the current work, may be used to quantize the strong force
in the form of quantum chromodynamics.
Another important aspect of the quantization of a fundamental force is the production
of gauge bosons, resulting from the gauge group describing the interaction, which are
physically interpreted as the force mediating particles associated with the interaction. We
may therefore summarize the main aspects of the known quantization of the fundamental
forces by the entries in Table 1.1. Similar tables may be found in [2] or [3].
1.2 Nonrenormalizability of Canonical Quantization
There were attempts to quantize gravitational interaction almost immediately. It is known
as a result of classical general relativity that the theory can be derived from an action
known as the Einstein-Hilbert action,
S
EH
= M
2
p
_
d
4
x

gR, (1.5)
5
where we have dened M
p
1/

16G, g det(g

), and R = g

is the Ricci scalar.


Now suppose this is constructed as a typical quantum eld theory. We rst write the
metric as the Minkowski space metric plus a perturbation h

, g

+h

. Hence the
Einstein-Hilbert action for the quantum eld must have the form
S = M
2
p
_
d
4
x(hh +hhh +h
2
hh +. . . ). (1.6)
From the second term in equation 1.6, we see that the second order Feynman term will go
as kkkk/k
2
k
2
= 1, and hence the integral diverges. Taking out two of the momentum
terms in the denominator in order to extract the hh dierential, the action now diverges
as k
2
. This process is easily seen to diverge more and more rapidly as one moves to
higher and higher order Feynman terms. The third order term will diverge as k
4
.
We may also include matter in this integral via the term
Matter Interation =
1
2
_
d
4
xh

, (1.7)
where
T

(x) =
2

g
S
M
g

(x)
,
is the stress-energy tensor for any non-graviton matter eld. This implies that 1.6 can be
written as
S =
_
d
4
x[M
2
p
(hh +hhh +h
2
hh +. . . ) + (hT +. . . )]. (1.8)
We can see that even by inclusion of matter the action still diverges as before. This is the
well-known failure of traditional quantization methods to yield a renormalizable quantum
theory of gravity. The use of Euclidean lattices, such as those successful for quantum
6
chromodynamics, also fails to achieve a renormalizable theory. It was shown in simulations
[31], that the seed universes produced by the computational methods could branch o
into a baby universe at each point in the spacetime. The degeneracy introduced by such
baby universe geometries prevents the continuum theory from being well-dened [30], [32].
This has lead to a continuing investigation to determine how to properly quantize the
gravitational eld.
1.3 Causal Dynamical Triangulations
Causal Dynamical Triangulations or CDT is a concrete research program which attempts
to renormalize gravity nonperturbatively via the use of a lattice structure, similar to that
employed in quantum chromodynamics but simplicial instead of Euclidean. This is origi-
nally described in [5] [8] as a method inspired by the Wilsonian Renormalization Group.
Using this method, described in detail in chapter 3, the classical Einstein-Hilbert action
can be written in terms of the lattice connectivity data, using the Regge Calculus, which
is discussed in detail in chapter four. The regularized path integral using the Reggeized
action can be computed numerically, using a Monte-Carlo simulation, the details of which
are also discussed in chapter four. After the calculation the lattice length can be taken
to zero
1
, and a result may be obtained which is invariant under rescaling of the primitive
length. CDT has gained some notoriety for demonstrations that de-Sitter CDT space-
times possesses a well-dened classical limit, and there has been work done claiming that
classical results are in agreement with known physics. It has also been speculated that
simplicial quantization via CDT might be related to Horava-Lifshitz quantization in at
1
It should be noted however that this limit cannot be taken arbitrarily, but rather at a second-order
phase transition. At a second order phase transition the discrete structure of the underlying spacetime stops
aecting the statistical model strongly, due to the rapid divergence of the correlation lengths, implying that
the continuum limit may be taken here safely, where it will be unique and well-dened. This shall be
understood at all other times when we take a 0 in this work.
7
least (2 + 1)-dimensions, which we shall provide a method for dierentiating from dieo-
morphism invariant quantization by searching for propagatin scalar modes which are riled
out in general relativity.
We shall see that for the purpose of the present study, it is computationally convenient
to use a (2 + 1)-dimensional spacetime rather than the full 3 + 1-dimensions. It is for this
reason that the next chapter is devoted to developing known classical results from general
relativity in (2 + 1)-dimensions.
8
Chapter 2
Classical Gravity in (2 + 1)-
Dimensions
There is no intrinsic nature to CDT gravity which species the dimensionality of the
underlying spacetime. Thus we are free to choose any dimensionality we wish so long as such
a dimensionality admits vacuum solutions to the classical Einstein Field Equations. For a
comparison to Horava-Lifshitz gravity however, it is necessary to study general relativity in
(2+1)-dimensions, where it shall be shown that traditional relativity admits no propagating
modes. In order to formulate (2 + 1)-dimensional relativity it is necessary to understand
the ADM decomposition of the metric. This and the former topic are the main subjects
of this chapter, along with several other signicant results from traditional relativity in
(2 + 1)-dimensions.
9
2.1 general relativity in (2 + 1)- Dimensions
By calculating the Euler-Lagrange equations of the Einstein-Hilbert action we may derive
the Einstein Field Equations
R


1
2
g

R + g

= 8GT

(2.1)
on the (2 + 1)manifold /, where may or may not be taken as zero without loss of
generality. As in 3 + 1 relativity the eld equations remain completely covariant, as can
be veried by showing they are invariant under the action of Di(/) which is therefore
identied as a gauge group on /.
Furthermore we may show that for the special case of (2 +1)- spacetime the curvature
tensor forms a linear relation with the Ricci tensor:
R

= g

+g

1
2
(g

)R. (2.2)
It can be shown that due to this decomposition, solutions to the eld equations take
two forms: = 0 when = 0 and = const when ,= 0.
For N-dimensional classical gravity it can be shown the phase space is characterized by
the spacial metric for a constant-time hypersurface and the conjugate momentum, which
have N(N 1)/2 degrees of freedom. The total dimensionality is reduced however by the
eld equations themselves which in general form N constraints on the initial conditions,
while a proper choice of gauge can reduce the dimensionality by another N. This gives the
entire dimensionality of the phase space as
D = N(N 1) 2N = N(N 3) (2.3)
10
per spacetime point. For N = 3 the phase space has dimension zero at each spacetime point,
implying the theory has no propagating eld degrees of freedom in (2 + 1)- dimensions.
Hence the geometry is completely determined by the constraints modulo some nite number
of global degrees of freedom.
2.1.1 The Newtonian Limit
Since the theory has no propagating degrees of freedom, the Newtonian limit is expected
to be novel. We shall conclude that the Newtonian limit of a classical (2 + 1)- theory has
no force between static point masses.
We make the ansatz that the metric can be separated into a homogeneous Minkowski
component plus a perturbation h

, making the total metric g

+ h

. It can be
shown there is some gauge for which the eld equations take the form

1
2

_
h


1
2

_
8GT

_
h

1
2

_
= 0.
(2.4)
The Newtonian limit is given by letting T
00
, with as the mass density, and we
ignore all non-00 components of the stress-energy tensor. Then by the rst equation of 2.4,

_
h
00
+
1
2

_
=
2
_
h
00
+
1
2

_
= 16G. (2.5)
We also neglect time derivatives, which shall be suppressed by some power of v/c so
that
2

2
. Furthermore, we recall the Newtonian potential (x) is dened such that
the Poisson equation

2
(x) = 4G(x) (2.6)
11
is satised. By combining 2.5 and 2.6 we have

2
_
h
00
+
1
2

_
=
2
(4), (2.7)
and thus we may conclude
h
00
+
1
2

= 4. (2.8)
When the assumptions of the Newtonian limit are applied to the Christoel symbol we
have

i
00
=
1
2
(1 +h
ii
)(
0
h
0i
+
0
h
00

i
h
00
)
1
2

i
h
00
, (2.9)
since the time derivatives are ignored and h
ii
<< 1. Therefore the geodesic equation may
be written as
d
2
x
i
dt
2

1
2

i
h
00
= 0. (2.10)
Now dene the variable

h

which satises the system


h
00
=

h
00
+
1
N 2

= h


1
2

.
(2.11)
It is simple to verify

h
00
= 4. Indeed, Carlip [12] takes this as the starting point in the
derivation from 2.4. By combining 2.11 and 2.10 we nd
d
2
x
i
dt
2
+ 2
_
N 3
N 2
_

i
= 0. (2.12)
In the case where N = 4, the familiar Newtonian equations of motion are readily obtained.
Similarly, for the case where N > 4 the familiar equations are merely scaled by a constant,
which will only aect the integer order of the coupling strength of the gravitational force.
12
For N = 2 the theory is degenerate and for N = 3 the Newtonian potential does not aect
test particles, and therefore renders no force, as was previously claimed.
2.2 The [0, 1] Topology
Any (2 + 1)manifold / which is relevant to the current study possesses two general
characteristics. These are
/must admit time-orientable Lorentzian metrics, viz. a metric for which the metric
tensor has signature (++) and there are two spacelike boundaries

and
+
which
may be called past and future respectively.
/ must also admit solutions to the Einstein Field Equations in vacuum, including
= 0.
It is easily shown that a noncompact / admits a time-orientable Lorentzian metric if
and only if it also admits a nonvanishing vector eld. This makes it possible to use the
Poincare-Hopf index theorem.
Suppose V is a vector eld on / and x is a point such that V (x) = 0. Now construct
x
i
, an unbounded sequence on /, and let
#
be a countable set whose elements are
sets of /dieomorphisms,
j

i

ij

#
i

#
, mapping V (x
i
) V (x
i+1
). Then

i
([
#
]
i
V (x
i
)) V (x = ), (2.13)
since x
i
is unbounded, where we have dened
[
#
]
i
:=

ij

#
i

ij
. (2.14)
13
The implicit limit in 2.13 will remove the zero of V in any arbitrary neighborhood of /
where we wish to perform computations. Since this may be repeated for any arbitrary zero
of V , / must admit a Lorentzian manifold.
If /is closed then by the Poincare-Hopf index theorem there is a nonvanishing vector
eld if and only if (/) = 0. But since we assume / to be odd-dimensional the Euler
number necessarily vanishes, thus / must admit a time-orientable Lorentzian metric.
Now consider a manifold with boundaries

and
+
. Then we may use a result due
to [9] which states that / may admit a time-orientable Lorentzian metric if and only if
(

) = (
+
). (2.15)
We may nd which of these are also candidates to admit vacuum solutions of the Field
Equations. Since all closed Lorentzian manifolds contain closed timelike curves, they are
severely limited in their utility from a physical standpoint. Furthermore, noncompact
manifolds are not well understood mathematically.
We make the most use of a theorem by Mess [11] as in [12]. For a compact manifold
with a at, time-orientable Lorentzian metric and a surely spacelike boundary, the manifold
must necessarily have the topology
T
M
= [0, 1] , (2.16)
Where is some closed two-surface which is homeomorphic to either

or
+
. This
combined with the previous result implies no solution to the eld equations can simultane-
ously admit a topology change. Hence for any manifold under consideration the topology
is xed by the surface

.
14
2.3 The ADM Formalism and Hamiltionian Formulation
In traditional treatments of 3 + 1 relativity, spatial and temporal dimensions are unied
into a single manifold structure in which the spatial and temporal characteristics of a
physical system are formally coupled. Despite this, it is often a matter of practical use
to generate a foliation of the spacetime manifold for which the spatial and temporal di-
rections are mathematically distinct. This may be accomplished by making use of the
Arnowitt-Deser-Miser (ADM) formalism. The ADM formalism is valid in both (3 + 1)
and (2 + 1)- relativity, and is a cornerstone of the theory of CDT gravity and numerical
relativity in general. Classically it leads to the introduction of a natural Hamiltonian for
classical gravity, suggesting a variety of alternative methods for canonical quantization of
the theory, which we shall largely not be concerned with. We shall however make use of
the phase space constraints generated by the Hamiltonian representation to study in depth
the equivalence and uniqueness of path integral quantization to canonical quantization in
(2 + 1)- dimensions. We present the details of the classical theory here.
2.3.1 The ADM Metric
The [0, 1] topology of the last section may be thought of as a continuum existing between
an initial spacelike surface

= 0 and a nal spacelike surface


+
= 1. Using
these as endpoints we may generate a foliation of the spacetime into discrete time intervals
dt creating a possibly innite set of constant-time manifolds
t
on which there exists a
coordinate chart = x
i
and the metric g
ij
(t, x
i
) naturally accompanying the chart. We
insist that any set of foliations necessarily recover the topology of the original spacetime if
we choose to reform the space, that is, up to isomorphic reorderings of the foliations, we
15
Figure 2.1: Diagram of the triangulation used to compute ds
2
.
insist that the
t
obey
[0, 1] =

_
_

t(0,1)

t
_
_

+
. (2.17)
In order to ensure this in practice it is necessary to understand how any given foliation
connects to its neighbors in a metric sense, which we develop presently.
Let some point q(x
0
, . . . , x
i
)
t
be displaced by some innitesimal distance dq(x
0
, . . . , x
i
)
with some nontrivial component normal to a two dimensional embedding of
t
. This change
induces changes on the metric
dq(x
0
, . . . , x
i
)
_

_
dt d = Ndt
x
i
(t) x
i
(t +dt) = x
i
(t) N
i
dt
(2.18)
where we have dened a contravariant vector N

(t, x
i
) = (N(t, x
i
), N
i
(t, x
i
)) where N
0
is
known as the lapse function and N
i
is known as the shift vector, preserving informally the
distinction between time and space coordinates.
We may now compute a Lorentzian invariant interval, using a triangulation which
preserves the directionality of the Lie derivative, L
X
T

, for an arbitrary vector eld on


16
the shift manifold
t+dt
, as shown in gure 2.3.1. The result of this calculation gives the
metric
ds
2
= N
2
dt
2
+g
ij
(dx
i
+N
i
dt)(dx
j
+N
j
dt). (2.19)
Here it is customary to take the raising and lowering convention with respect to the spatial
components alone, so that g
ij
g
ij
=
j
i
, but not necessarily the same for the Greek indices.
2.3.2 The Hamiltonian and Canonical Constraints
We wish to form a Hamiltonian for the ADM decomposition of the metric. This will allow
us to put the ADM version of the path integral on an more rigorous footing. It can be
shown that under the ADM decomposition, the Einstein Hilbert action takes the form
S
EH

_
dt
_

d
2
xN

g(R 2 +K
ij
K
ij
K
2
) (2.20)
where the boundary terms have been dropped and K
ij
is the spatial component of the
extrinsic curvature. Here this takes the form
K
ij
=
1
2N
(
t
g
ij

i
N
j

j
N
i
) (2.21)
for a surface with a normal vector n

= (N, 0, 0) and the full extrinsic curvature K

+ n

. We may now compute the generalized canonical momenta in the


typical way,

ij
=
L
(
t
g
ij
)
=

g(K
ij
g
ij
) (2.22)
and hence write the Einstein-Hilbert action as
S
EH
=
_
dt
_

d
2
x(
ij

t
g
ij
N]N
i
]
i
), (2.23)
17
with constraints
]

=
_
_
_
1/

g(
ij

ij

2
)

g(R 2)
2
j

ij
_
_
_. (2.24)
The rst element, denoted ], is the Hamiltonian constraint while the remaining ]
i
form
the momentum constraints. In a closed universe, viz. a universe for which all boundary
terms in the Einstein-Hilbert action are seen to vanish, these lead to trivial equations of
motion, implying as we would expect that the total energy of a closed universe is zero.
2.4 (2 + 1)- Path Integrals in the ADM Formalism
The ADM decomposition of the metric leads to a form of the path integral for which
calculations are easier. Since the path integral formalism is not guaranteed to produce a
unique quantization, it is important to check that path integral quantization via an ADM
decomposition leads to the same result as canonical quantization, which is shown in detail
below.
For notational uniformity with [12], dene the convention T[. . . ] = . . . ). In this case
the latter symbol will stand as a path integral measure and not an expectation value. In
general we may represent the path integral in a coordinate basis as
q
f
, t
f
[q
i
, t
i
) =
_
dq)dp) expiS
EH
[q, p], (2.25)
where the Einstein-Hilbert action takes the ADM form
S
EH
=
_
d
3
x

gR =
_
dt
_

d
2
x(
ij
g
ij
N
i
]
i
N]). (2.26)
18
Then we may write the full ADM path integral as
Z =
_
d
ij
)dg
ij
)dN
i
)dN) exp
_
i
_
dt
_

d
2
x(
ij
g
ij
N
i
]
i
N])[, g]
_
. (2.27)
The shift coecients appear as Lagrange multipliers in the action, and thus integrated
become delta functionals imposing the Hamiltonian constraints on the phase space. Thus
the path integral is now dened over a measure on the reduced phase space, which is
a simple nite-dimensional space of physical degrees of freedom. Evaluation of the path
integral over such a reduced phase space requires the introduction of a specialized Jacobian,
called the Faddeev-Popov determinant, which shall be treated in the next subsection.
2.4.1 Equivalence to 2+1 Canonical Quantization
The ADM form of the Einstein-Hilbert action has a gauge symmetry, which may be iden-
tied as the dieomorphism group on the measure space generated by the Hamiltonian
constraints, ]

) ], ]
i
). If we suppose there is a nondegenerate set of homogeneous
gauge equations

, the path integral becomes


Z =
_
d
ij
)dg
ij
)dN
i
)dN)[

] det(]

)
exp
_
i
_
dt
_

d
2
x(
ij
g
ij
N
i
]
i
N])[, g]
_
.
(2.28)
We may simplify the calculation via a procedure outlined in [12]. Choose a family of
constant curvature metrics g
ij
on the moduli space, and thus parametrized by moduli m

,
so that g
ij
(m

) = e
2
f

g
ij
(m

). Now set
g
ij
= 2()g
ij
+ (

)
ij
+m

()ij

ij
=
1
2
g
ij
+

g(

PY )
ij
+

g p

ij
()
,
(2.29)
19
where the operator

P may be dened via the test function
(

P)
ij
=
i

j
+
j

i
g
ij

k
. (2.30)
T

is a multilinear operator, which maps to the orthogonal projection of a given modular


deformation of the spatial metric onto ker(

P

), and may be written


T

=
()
,
()
)
()
,
()
)
1
(2.31)
for which
()
ij
g
ij
/m

, and the
()
are chosen to form a basis of the kernel of

P

.
The inner product in 2.31 shall be dened shortly.
Supposing = 0, the Hamiltonian constraints become
]
i
=

g(

P


PY )
i

] =
1
2

g
e
2

2
+ 2
_
g
_


k
2
_
+
_
g g
ik
g
jl
e
2
_
(

PY )
ij
+ p

()ij
_ _
(

PY )
kl
+ p

()kl
_
(2.32)
which is an orthogonal decomposition of the integration variables, implying that all sec-
ondary Jacobians will be equal to unity.
We now wish to dene some map, J : g
ij
,
ij
) m

, , , p

, , Y
i
), from the native
measure space into the reduced phase space of the moduli. For such a map we may dene
an inner product from the tangent space to the space of all -metrics,
q, q) :=
_

_
_

d
2
x

gg
ij
g
kl
g
ik
g
il
q = g
_

d
2
x

gg
ij

j
q =
_

d
2
x

g()
2
q = ,
(2.33)
20
and consider the integral
_
d(g
ij
))e
ig,g
=
_
d(m))d())d())J
g
e
i,

P

P
e
8i,
e
imm

()
,
()

1,
(2.34)
which when evaluated leads to the result
J
g
= det [


P[
1/2
det [T[ det [
()
,
()
)[
1/2
. (2.35)
We may evaluate a similar integral for d(
ij
)) to give the result
J

= det [


P[
1/2
det [
()
,
()
)[
1/2
. (2.36)
The total Faddeev-Popov determinant is obtained by a convolution of the component parts
J = J
g
J

= det [


P[
1/2
det [
()
,
()
)[
1/2
. (2.37)
Though the map was originally dened from the tangent space to phase space, we may
apply this result to calculate the change of coordinates in the general path integral, which
allows a change of coordinates into the ADM shift coecients N

, and integrate over these.


This allows us to represent the path integral as
Z =
_
d
n
p)d
n
m) det [
()
,
()
)[

_
d(/

g))d)dY )d) det [


P[[

][]

g] det []

[
exp
_
i
_
dt
_

d
2
x(
ij
g
ij
N
i
]
i
N])
_
.
(2.38)
21
The spatial component of Y becomes unity when integrated against the delta function
[]
i
/g] and the determinant of

P


P. The same is true for the integral if we may
separate the Hamiltonian containing Poisson bracket terms into their spatial and temporal
components, making such a decomposition desirable.
Suppose the gauge condition
0
= /

g T = 0, the so-called York time gauge,


is chosen. It may be shown that the determinant det []

[ factors evenly into the


temporal and spatial components det []

[ = det []
i
,
i
[ det []
0
,
0
[. Moreover,
by considering the simplicial structure,
=
_

d
2
x
ij
g
ij
=
_

d
2
x(p
ij
g
ij
+ 2
2 g
ij
[e
2
_
g
_

+
k
2
_
Y
i
+
1
2

i
(e
2
)]
i
(2.39)
discussed thoroughly in [13], the Poisson brackets in the determinant may be evaluated.
From this it be shown that
det []
0
,
0
[ = det

e
2
_

+
T
2
2
e
2
+e
2
p

g
ij
g
kl

()ik

()jl
_

. (2.40)
The path integral is now simplied to
Z =
_
d
n
p)d
n
m) det [
()
,
()
)[
_
)[]/

g] det []
0
,
0
[
exp
_
i
_
dt
_

d
2
x(
ij
g
ij
N
i
]
i
N])
_
.
(2.41)
The integral is nontrivial but only leaves a term cancelling ]
0
,
0
, making the
integral
Z =
_
d
n
p)d
n
m) det [
()
,
()
)[ expi

S
EH
, (2.42)
22
where we now evaluate the Einstein-Hilbert action at the solution of the constraints,

S
EH
.
By using the variable substitution
p

=
_

d
2
xe
2
_

ij

1
2
g
ij

_
g
ij
m

=
()
,
()
) p

, (2.43)
we may make such a coordinate substitution p

()
,
()
) p

and reduce the path


integral into the nal form
Z =
_
d
n
p)d
n
m) expi

S
EH
[p, m], (2.44)
which is an ordinary quantum mechanical path integral of the form 2.25. This form of
standard quantum mechanical path integral can be proven [1] to be equivalent to canonical
quantization of the theory and therefore unique.
23
Chapter 3
Nonperturbative Quantum
Gravity with (2 + 1)-CDT
Although we have described classical gravity at length in (2 + 1)- dimensions, we have
yet to describe in detail a quantization. This chapter takes up a general discussion of
CDT quantization in an arbitary number of dimensions. Due to the nature of the CDT
formalism, it is simple to go from a model in an arbitrary number of dimensions to one in
(2 + 1)- dimensions.
3.1 The Gravitational Path integral via Triangulation
CDT quantum gravity is typically introduced in terms of the gravitational path integral,
or sum over histories. The general idea is to use the regularized version of the path integral
on a dierentiable manifold /,
Z(G
N
, ) =
_
gG(M)
T[g

] expiS
EH
, (3.1)
24
where G
N
is the Newtonian constant and we have dened the Einstein-Hilbert action,
S
EH
=
1
G
N
_
d
3
_
det g(R 2). (3.2)
Here is the cosmological constant. It should also be noted that a similar expression may
be written in 1 + 1 and 3 + 1 dimensions at this point without any loss of generality.
For 3.1 to be well-dened it is necessary to specify the measure over which a nite value
of the integral may be obtained. An obvious choice is the measure of causal Lorentzian
geometries on /, Lor(/), but this space alone is not well-dened as two Lorentzian
manifolds may be dieomorphic, and should not add separately to the measure. Hence
a measure of Lorentzian manifolds modulo dieomorphisms may be used as the measure
space, implying
((/) =
Lor(/)
Di(/)
. (3.3)
Since the path integral is regularized, so too is the measure space ( (
a,N
where (
a,N
is
the regularized version of the measure space of N-dimensional simplicial manifolds and a
is the atomic edge length. From quantum eld theory, we know that this is expected to
be highly singular with the set of smooth, classical congurations corresponding to a set
of measure zero. This is easily shown for the space identied in 3.3.
The atomic edge length a is not seen as fundamental, and in CDT results are only
considered in the case of a 0, and hence N , which is identical to a continuum
solution. The Regge calculus (discussed in section 4.1) provides for the regularization of
this path integral with respect to the simplicial lattice as
Z
CDT
a,N
=

TG
a,N
1
C
T
expiS
Regge
[T], (3.4)
25
where T is a given simplex in the regularized measure and S
Regge
is the Reggeized Einstein-
Hilbert action. We then will have Z
CDT
= lim
(N,a)(,0)
Z
CDT
a,N
, where
Z
CDT
= lim
(N,a)(,0)

TG
a,N
1
C
T
expiS
Regge
[T]. (3.5)
Here C
T
is the simplicial analogue of the order of the automorphism group of the Lorentzian
metric g.
3.1.1 Continuum Limit
By taking the discretization to the continuum case, we may postulate the existence of a well-
dened continuum limit. If this exists, it will not depend on details of the regularization
by construction. That is to say, the gluing rules and geometry of the simplices will not
ultimately aect the classical limit. This implies a robustness of Planck-scale physics, as
is discussed in [5]. This robustness is similar to the universality exhibited by statistical
mechanics, whereby the behavior of a system may be independent of the details of the
system, [5], [33].
3.1.2 The CDT Lattice
It was stated before that the simplicial building blocks are 3-simplices for the (2 + 1)-
theory, while the (3 + 1) theory uses four-simplices. By making assumptions about the
temporal component of the spacetime, in particular by introducing periodic boundary
conditions (see [8]), regularizing the spacetime reduces down to the problem of triangulating
a given Lorentzian manifold. This gives a number of so called gluing rules which must be
followed. In general, gluing rules which are too liberal would cause the number of allowed
gluings to grow too fast as a function of the volume of the triangulations to allow the
path integral to be well-dened, while gluing rules which are too restrictive will destroy
26
Figure 3.1: Disallowed Histories in Causal Dynamical Triangulations, such as wormhole
histories and branching points where the light cones are degenerate.
local curvature degrees of freedom. Hence we wish [(
a,N
[ e
cN
simultaneously with
the existence of curvature degrees of freedom. A conclusion gained from investigating
such models, with gluing rules within this window, is that they do not correspond to a
good classical limit. In these cases the large scale dimensionality of the spacetime becomes
dynamical despite the constant dimensionality of the simplices, a problem discovered during
research into Euclidean Dynamical Triangulation, where there were found Hausdor phases
of d = 2 and d = .
The only available solution to this comes through CDT. By restricting the histories
allowed in the regularized path integral to those which are causal, forbidding the presence
of branching points or any structure with a classically singular light cone, shown in (b)
of 3.1.2. This is in contrast to purely Euclidean path integral approaches, which are
not necessarily causal. The causality constraint does not generically suppress large-scale
curvature uctuations, and thus will not suppress curvature degrees of freedom, as desired.
Causality is implemented in the gluing rules by creating a layered geometric structure
27
labelled by an integer-valued proper time t. A Wick rotation t it ([4]), maps this into a
subset of Euclidean piecewise at geometries. Since Wick rotations map into a strict subset
of these geometries, the sum over geometries is changed, leading to a dierent continuum
limit than the purely Euclidean theory.
3.2 Horava-Lifshitz Gravity
Horava-Lifshitz gravity is a gravity model which includes with a dynamic spacetime metric
a preferred foliation of the spacetime manifold, /, by spacelike hypersurfaces. The natu-
ral mathematical description of this foliation is given by the group of dieomorphisms of
/ preserving the foliation. This leads to an anisotropy between space and time dimen-
sions measured with respect to an anisotropic scaling parameter z, such that for a locally
Minkowski metric,
_
_
_
t
x
_
_
_
_
_
_
t
x
_
_
_
_
_
_
b
z
t
bx
_
_
_, (3.6)
where b > 0 and the ultraviolet divergences of the theory are suppressed for z > 1. In a
general spacetime,
_
_
_
t
x
_
_
_
_
_
_
t
x
_
_
_
_
_
_
f(t)
(x, t)
_
_
_, (3.7)
for an arbitrary f and .
The preferred foliation of Horava-Lifzhitz gravity makes it a natural setting to use the
ADM formalism from classical relativity as discussed in chapter 2, with the shift vector
and the lapse function playing the role of gauge elds associated with the dieomorphism
group. Although not true in general, reparametrizations of the space coordinates may be
independent of the time coordinates, implying that the relative gauge eld, N(t, X) might
be only a function of the time coordinate, viz. N(t, X) N(t). This leads to a version of
28
the (2 + 1)- theory which is projectable.
By writing the action for such a theory in d spatial dimensions, we can include higher
curvature terms by constructing such an action as the sum of a kinetic term quadratic in
temporal derivatives and a potential term of mass dimension 2d in spatial derivatives. In
this case z = d. [25] give a general kinetic term to be
1
16G
_
M
dtd
d
x
_
(t, x)N(t)[K
ij
(t, x)K
ij
(t, x) K
2
(t, x)], (3.8)
where is a parameter of the relevant DeWitt supermetric as described in [26], [27], and
we have dened
K
ij
(t, x) =
1
2N(t)
[
t

ij
(t, x)
i
N
j
(t, x)
j
N
i
(t, x)]. (3.9)
The potential term is given by
1
16G
_
M
dtd
d
x
_
(t, x)N(t)V [(t, x)]. (3.10)
Here V [(t, x)] is a scalar functional of the spatial metric tensor and spatial derivatives up
to order 2d. This leads to a total action,
S
HL
=
1
16G
_
M
dtd
d
x
_
(t, x)N(t)[K
ij
(t, x)K
ij
(t, x) K
2
(t, x) +V [(t, x)]], (3.11)
and G G
N
in the low energy limit.
We nd that the lapse function, being only a function of time, introduces a Hamiltonian
constraint, ]

= 0, into the theory with


]

=
_

d
2
X
_
(t, x)[K
ij
(t, x)K
ij
(t, x) K
2
(t, x) +R
2
2
(t, x) R
2
(t, x) +2], (3.12)
29
where R
2
is a total derivative described by the Gauss-Bonnet theorem in the integral of
over .
The nal constraint we may make upon the Horava-Lifshitz theory to compare it with
CDT gravity is a choice between noncompact and compact spatial topologies, which dictate
the allowed values of the local Hamiltonian constraint. Noncompact spatial topologies are
typical in many theories of quantum gravity, but in Horava-Lifshitz gravity it is shown by
[25] that the ]

= 0 constraint implies ]

may be thought of as a globally conserved


charge, which is far more natural in CDT than more typical implications for noncompact
spatial topologies.
3.3 Propagating Scalar Modes and Anisotropic Quantiza-
tion
In order to determine the number of propagating scalar modes of the theory, it is necessary
to consider V : the part of the Lagrangian which does not contain any temporal derivatives.
In (2 + 1)- dimensions the following are true: (1) R
abcd
= 1/2(g
ac
g
db
g
ad
g
cb
)R, (2) any
potential terms will dier only by total divergences, and (3) a
i
(as dened below) is the
gradient of a scalar. This allows us to write V as
V = R +a
i
a
i
+g
1
R
2
+g
2

2
R +g
3
(a
i
a
i
)
2
+g
4
Ra
i
a
i
+g
5
a
2
( a)
+g
6
( a)
2
+g
7
(
i
a
j
)(a
i
a
j
).
(3.13)
30
The g
i
coupling constants have dimensions of M
2
, since the mass scale has been absorbed
into the coupling constants. This gives a general (2 + 1)-dimensional action
S =
M
2
pl
2
_
d
2
xdtN

g[K
ij
K
ij
K
2
+Ra
i
a
i
g
1
R
2
+g
2

2
R
+g
3
(a
i
a
i
)
2
+g
4
Ra
i
a
i
+g
5
a
2
( a) +g
6
( a)
2
+g
7
(
i
a
j
)(a
i
a
j
)].
(3.14)
Field equations can be computed by varying the lapse, shift, and induced metric. This is
shown in [28]. The equations of motion can be stated as follows:
The equation of motion due to variation with respect to the lapse is given by
K
ij
K
ij
K
2
+R[2 a +a
i
a
i
] g
1
R
2
g
2

2
R +g
3
[4
i
(a
j
a
j
a
i
) + 3(a
i
a
i
)
2
]
+g
4
[2
i
(Ra
i
) +Ra
i
a
i
] +g
5
a
i
a
i
a + 2 (a[ a]) [
2
(Na
i
a
i
)]/N
g
6
[3( a)
2
+ 2
2
( a) + 2a
2
( a) + 4(a )( a)]
g
7
[(
i
a
j
)(
i
a
j
) 2(
i
a
j
)a
i
a
j
4
i
(a
j
(
i
a
j
)) 2
i

2
a
i
] = 0.
(3.15)
The equation of motion due to variation with respect to the shift is given by

ij

i
K
ij
Kg
ij
= 0. (3.16)
31
and the equation of motion due to the spatial metric is given by
1

t
(

g
ij
+ 2N(K
l
i
K
lij
KK
ij
)
1
2
N(K
ij
K
ij
K
2
)g
ij
N
ij
(
m
N
m
)
L
N
(N
ij
) + 2NN
(i

j)m
a
m

1
N
[
i

j
g
ij

2
]N +
_
a
i
a
j

1
2
a
l
a
l
g
ij
_
+
1
2
g
1
R
2
g
ij
2g
1
1
N
[
i

j
g
ij

2
](NR) g
2
1
N
[
i

j
g
ij

2
](
2
N) g
2
a
(i

j)
R +g
2

m
(NRa
m
)
2N
g
ij
+ 2g
3
a
2
a
i
a
j

1
2
g
3
(a
2
)
2
g
ij
+g
4
Ra
i
a
j
g
4
1
N
[
i

j
g
ij

2
](Na
i
a
i
)
1
2
g
4
Ra
2
g
ij
+g
5
a
i
a
j
[( a) a
2
]
1
2
g
5
(
i
a
2
)a
j
+ (
j
a
2
)a
i
+
1
2
g
5
(a
2
)
2
+a
m

m
a
2
g
ij
2g
6
( a)a
i
a
j
g
6
(
i
[ a])a
j
+ (
j
[ a]a
i
+g
6
_
a
2
( a) +a ( a) +
1
2
( a)
2
_
g
ij

1
2
g
7
(
m
a
n
)(
m
a
n
)g
ij
2g
7
(a a
(i
)a
j)
2g
7
(
2
a
(i
)a
j)
+g
7
a
2
(
i
a
j
) +g
7
([
i
a
j
]a) = 0,
(3.17)
where L
N
is the Lie Derivative along the vector N
i
. The only one of these equations which
is dynamical is 3.17, with the dynamical variable g
ij
. Applying the uniformization theorem
as in [29], and since the theory is invariant under the transformation x
i
x
i
(t, x
i
) we may
set
g
ij
=
2
g
c
ij
, (3.18)
where g
c
ij
is the metric of a constant curvature, spherical, Euclidean or hyperbolic 2-
dimensional space. This makes 3.17 into a dynamical equation for the conformal factor
.
This is a crucial dierence between quantizing gravity isotropically or anisotropically.
For isotropic, or a traditional gravity model, we have shown in chapter 2 that there are
no propagating modes of freedom. We have just seen however that quantizing a Horava-
Lifshitz model via a CDT like simulation, it is plausible the quantization might yield at
least one propagating mode of freedom. Hence by simulation we can determine whether
CDT quantization is a numerical quantization of classical or Horava-Lifshitz gravity. This
32
requires generating a seed manifold, which is taken up on chapter 5. The next chapter
describes the general computational procedures in CDT quantum gravity and the one after
that the procedures taken to generate the seed manifold.
33
Chapter 4
Computational Approach
4.1 The Regge Calculus
The lattice formulation of general relativity was taken up nearly a half century ago, primar-
ily by Regge [14]. The central idea of the Regge procedure is to tessellate the underlying
spacetime with simplices, and in (2 + 1)- dimensions the relevant simplex is the 2-simplex
or triangle for the spatial slice, where the only extrinsic curvature is at the edges. The in-
trinsic curvature is centred entirely at the vertices. These are canonical points
i
described
by the decit angles
i
. It is shown in texts with devoted chapters on the Regge calculus,
such as [12] that such an intrinsic angle contributes 2
i
per vertex to the action. Thus we
may make the nite scale-independent approximation
S
spatial slice
=
_
d
2
x

gR 2

i
(M)

i
S
Regge
, (4.1)
where (/) is the set of all angles with where the curvature is concentrated in the sim-
plicial manifold, counting multiplicities.
Ths may be written in terms of the (N 2)-dimensional hinge of the simplex, making
34
the Ndimensional Reggeized action now
S
Regge
= 2

V
i
V(M)

i
V
i
, (4.2)
where 1(/) is the set of all simplicial hinges on the manifold. The hinge of a 2-simplex
may be identied as the side of the triangle, making the (2 + 1)- Regge action
S
2+1
Regge
= 2

eV(M)

e
, (4.3)
where
e
is the length of the eth edge. It is to be noted that alternatively dening units
such that 8G 1 leads to a normalization of the Regge action where the factor of two is
absent.
The Reggeized action may be used to regularize the path integral quantization proce-
dure described in chapter 2. Suppose /has triangulated boundaries
0
and
1
, such that
/ =
0

1
. Then a quantum state on
0
or
1
may be dened as some function of
elements of 1(
0
) or 1(
1
). This implies we may construct a Reggeized representation of
the transition amplitude between a state on
0
to one on
1
or vice versa, which may be
expressed as path integral
Z(1(
0
), 1(
1
)) =
_
eV(M)/V(M)
d
e
_
_

V(M)
exp2
e

_
_
. (4.4)
One great benet of using Causal Dynamical Triangulations is that the regularized path
integral is coordinate independent, implying that the rigorous mathematical issues with
the measure space assumed in 4.4 are curtailed. The following subsection contains a short
discussion on the mathematical issues that arise when CDT is not used to regularize the
path integral and how to form the transition amplitude in such circumstances. Although
35
not of direct signicance to this work, it is important to realize the advantages oered by
the CDT approach and hence the author has decided to included it here for completeness
and to satiate the curious reader. It may however be skipped without aecting overall
understanding of this works content.
4.1.1 The non-regularized measure space and associated transition am-
plitude
Note that the invariant measure over the Simplicial manifold is not mathematically clear.
For instance, while such a measure over a Euclidean lattice (QCD) could be deduced from
reducing a 3 + 1 representation of the path integral, the non-uniformity of the simplicial
paths make the choices of appropriate gauge ambiguous and the proper form of the Faddeev-
Popov determinant is not understood.
Suppose we associate with the angular momentum in the full (2 + 1)- dimensions the
geometry of a 3-simplex. Then we may describe the coloring of this simplex via the Wiger-
Racah 6j-symbol [15],
J
6

_

_
j
1
j
2
j
3
j
4
j
5
j
6
_

_
. (4.5)
It can be shown [16] the approximation
exp
_
i
6

i=1
j
i
_
J
6

1

6V
_
exp
_
i
_
S
Regge
+

4
__
+ exp
_
i
_
S
Regge
+

4
___
(4.6)
holds for large j, where S
Regge
is the Regge action for a 3-simplex with side length
i
=
1/2(j
i
+ 1/2), given by
S
3+1
Regge
=
6

i=1

i
_
j
i
+
1
2
_
, (4.7)
with
i
dened as the angle between the outward normals of any two faces meeting at the
36
ith vertex and V is the volume of the 3-simplex with side lengths
i
. Since 4.6 holds, we
may conclude that the Reggeized action per tetrahedra can be written as a function Wiger-
Racah 6j-symbols. Furthermore, this implies that the integrand of 4.4 may be expressed
as some product of Wiger-Racah 6j-symbols. Hence the entire Reggeized action may be
seen as a sum over such 6j-products.
If /has boundary /with a triangulation (/), then some triangulation (///)
may be chosen such that (/) := (///) (/) forms a simplicial manifold. If
the interior and exterior edges of some 3-simplex T are labelled with discreet coordinates
x
i
and j
i
respectively and j
i
(T) is the spin-coloring of T, then [16] [17] claim the transition
amplitude may be expressed according to
Z
(M)
[j
i
] = lim
L

xeL
(

i
L(M)
(1)
2j
i
_
2j
i
+ 1

V
i
V(M/M)
(L)
1

l
L(M/M)
(2x
l
+ 1)

TT (M)
(1)

6
i=1
j
i
(T)
J
6
(T)),
(4.8)
where L is the set of edges on some manifold, 1 the set of vertices on some manifold, T is
the set of 3-simplices on some manifold, and
(L) :=

jL
(2j + 1)
2
. (4.9)
It can be shown that 4.8 is invariant under a subdivision or rearrangement of the 3-
simplices in T , and moreover it can be proven that the Reggeized quantization procedure
in (2 + 1)- dimensions forms a topological eld theory. This is further suggested by the
Newtonian limit of the theory described in chapter 2.
37
4.2 Monte-Carlo Method
Monte-Carlo simulation may be reduced to the problem of approximately evaluating an
integral
I =
_
V
f(x)d
n
x, (4.10)
where V is a vector space of dimension n and f : V is a function on V . While
traditional quadrature methods use deterministic schemes to discretize the subspace form-
ing the domain of f based on either geometry or interpolating polynomials, the basis of
Monte-Carlo simulation is to use a randomly sampled set of vectors x
i

M
i=1
to esti-
mate the function at M points within the domain. The integral may then be approximated
according to the formula
I
1
M
M

i=1
f(x
i
). (4.11)
In general a Monte-Carlo estimate is improved for a larger number of trials, however this
may not always be the case. From elementary statistics it follows that the error on a
Monte-Carlo estimate is given by
E (I) =

f
2
) f)
2
M 1
, (4.12)
where it can be simply shown that
f
k
)
1
M
M

i=1
f(x
i
)
k
. (4.13)
4.2.1 Markov Chain Stepping
It is typically desirable to choose a sampling set ergodically but not necessarily completely
at random. The concept of Markov Chain stepping is that the outcome of a step can aect
38
its successor. Let S = s
i
be a set of distinct states of a system. Then, if the steps
are assumed to be ergodic, there is in principle a probability p
ij
of the system in state
s
i
transitioning into a state s
j
. From such transition probabilities we may construct a
transition matrix P whose elements are p
ij
.
In itself this operator is not clearly useful. However it may be shown that determining
the probability a state s
i
has transitioned into a state s
j
after n iterations is just the ijth
entry of the matrix P
n
. This allows a certain degree of determinism for a chain of steps
such as those which might constitute a Monte-Carlo sampling.
We may also dene a stepping length, p, as the radius of the neighborhood from which
the sampling vector is taken. For a smaller sampling length, the approximation is improved
at the cost of a larger convergence time for the simulation.
Although Markov Chain sampling has the advantage of a lower degree of randomisation,
it also introduces an unwanted autocorrelation, which causes many moves to be discarded.
Markov Chain stepping is thus signicantly more desirable for the generation of complex
states, which are far more dicult to generate completely at random, but not necessarily
always the most appropriate option.
4.2.2 Sampling an Observable
It is a result from statistical physics that the expectation value of any observable, O, is
given by
O) =
1
Z(H, )

s
O(s) expH(s)/, (4.14)
where H is the Hamiltonian of the system, kT and Z(H, )

s
expH(s)/ is
the equilibrium partition function for the system. We may write this using the Boltzmann
39
distribution, T as
O) =

s
O(s) expH(s)/

s
expH(s)/
=

s
[O(s) expH(s)//T(s)]T(s)

s
[expH(s)//T(s)]T(s)
. (4.15)
It is shown in [18] that the far right side of equation 4.15 may be written in a condensed
form as
O)
1
M

i
O(s
i
), (4.16)
where the s
i
are points sampled according to the Boltzmann distribution.
4.3 The Metropolis Algorithm
The Metropolis algorithm was developed in 1953 at Los Alamos National Laboratory. The
fundamental idea is to compute the transition amplitude T (s

s) from state s to s

.
The probability of a transition is compared to a randomly generated uniformly distributed
number between 0 and 1, and if the probability exceeds this number, it is accepted and if
it does not, it is rejected.
First construct a Markov Chain, s
1
s
2
s
3
. . . , with probability P(s
i
s
i
+ 1)
for any given state to transition into its successor. Coupling this with the transition
probability we can compute the total probability of the (k + 1)th step from the kth as
T
k+1
(s) = T
k
(s) +

T (s

s)T
k
(s

) T (s s

)T
k
(s). (4.17)
As k the Markov Chain approaches an equilibrium value and T
k
(s) T
eq
(s). This
implies
T (s

s)T
eq
(s

) = T (s s

)T
eq
(s), (4.18)
40
which may be rewritten in a more suggestive form as
T (s s

)
T (s

s)
= e
H(s,s

)/
, (4.19)
where we have dened H(s, s

) = H(s

) H(s). Now it is necessary to formulate some


model for T (s s

). Although such a model is not unique, we will use the Metropolis


model, described by [18] to be
T (s s

) =
_

_
for H 0
e
H(s,s

)/
for H 0
(4.20)
where
1
is the Monte-Carlo time. This allows us to state formally the Metropolis Algo-
rithm, which is shown schematically below.
Algorithm (Metropolis):
1. Generate a starting state, s
0
.
2. O = 0.
3. while iterations < total iterations:
(a) generate trial state, s

,
(b) compute T(s s

, ),
(c) compute random x [0, 1]:
(d) if T > x: accept trial state, else: continue.
(e) O+ = O(s

),
4. return O/steps.
41
As before, in the limit as steps , the observable approaches its exact value. We may
extend this basic algorithm to take into account the equilibration of the simulation, which
is discussed below. Other eects not extremely pertinent to this study are found in the
vast literature on Monte-Carlo methods.
4.3.1 Equilibration
It is not necessarily true that a given sampling of the system will be in equilibrium. Since
initial states may not necessarily be in equilibrium, achieving such a conguration takes
a given time, known as the equilibrium time,
eq
which is a function of the system time
N = L
d
. The equilibrium time also varies inversely with the temperature of the system.
It typically takes a given number of Monte-Carlo sweeps, where we dene 1 MCS = N
spin updates. In any practical Monte-Carlo simulation observables are monitored as a
function of MCS to determine that the system is in an equilibrium state. Although some
observables have a small equilibrium time, others do not, and each variable it typically
sampled to determine whether the entire simulation has equilibrated.
4.3.2 Monte-Carlo in CDT: The Pachner Moves
The connection of the Monte-Carlo method to the actual simulation is crucial, but may
not be immediately clear at this point. Synoptically, we may say that the evaluation of the
regularized Einstein-Hilbert action over a suitable simplicial manifold is perfomed via the
means of a Monte-Carlo simulation with a particular set of moves, known in the literature
as the Pachner moves, and acting the same way as including something like the chain
stepping previously considered might act on the Monte-Carlo. The Pachner moves can be
found in nearly any review article on CDT gravity and many specic articles containing
new work (such as [10]). In (2 + 1)-dimensions the Pachner moves correspond to drawing
42
and replacing various diagonal elements on the unit 3-simplices constituting the simplicial
spacetime. The nal requirement of such a set of moves is that they be ergodic, meaning
that they are random but that the average state of each individual slice will also be that of
the entire ensemble. Although the Pachner moves have not yet been proven to be ergodic,
it is assumed in nearly all CDT literature that this is the case.
43
Chapter 5
Simulation
In order to run simulations it is necessary rst to evaluate a seed manifold. For the problem
at hand, considering the propagation of gravitons across the space time and from that the
number of propagating modes, the best seed manifold to use is one which resembles (an
American) football shape, since this will most closely resemble the initial state of the
process we wish to simulate.
5.1 Generation of the Seed Manifold
Properly evaluating the regularized path integral entails the generation of a seed manifold to
be used in the Monte-Carlo simulation. The typical starting place for a (2+1)-dimensional
CDT simulation is a 2-sphere, tessellated by equilateral triangles (simplices) of order 6. It
is a mathematical result given by [19] that there exists no such perfect triangulation of a
sphere a priori, so it is commonly considered equivalent to use a sphere triangulated by
simplices mostly of order 6 and a much smaller number of orders 5 or even 7. It has been
a major problem in previous work in CDT quantum gravity to successfully and eciently
generate a spherical triangulation with the desired characteristics. During the course of
44
this work we successfully resolved this issue by the use of a graphics technique known as
adaptive remeshing, discussed in detail presently.
5.1.1 Adaptive Remeshing
Adaptive remeshing is a common technique in computer graphics, and has been in common
use for decades in that eld. [20] provides an excellent introduction to the general tech-
niques, as well as several advanced results which we shall not be concerned with here. We
shall follow this exposition, though it should be known that [21] [22] [23] [24] also provide
references on the technique of remeshing and adaptive remeshing.
A complex triangulation may be considered as the result of an iterative subdivision al-
gorithm starting from a less complex triangulation. In this case it is said that the complex
triangulation, which for this chapter shall be called a mesh for consistency with the com-
puter graphics terminology, exhibits subdivision connectivity with respect to the simpler
mesh. To be perfectly clear about the terminology, we shall call the simplest mesh in this
iterative procedure the base mesh, and the most complex mesh the nal mesh.
Given a base mesh, S
0
, and topological manifold, the general problem of remeshing may
be stated as follows: if M is the set of all meshes and S is the set of all meshes exhibiting
subdivision connectivity, we wish to nd some operator U : M S subject to the con-
straint that the resulting nal mesh, S S, is the closest possible to its preimage U
1
[S].
This map can be approximated by a series of maps terminating in a map U

, whose image
corresponds to the nal subdivision connectivity mesh of the desired remeshing.
The operator U

is dicult to construct computationally. This can be resolved by


rst projecting the mesh without subdivision connectivity, M

, into some planar domain


R
2
. Consider projection operator f : , for which M

:= f(M

) is the mesh
dened by vertices p
i
, viz. the projected vertices P
i
such that the connectivity
45
Figure 5.1: Triangulation by remeshing as presented in [20]
of the set P
i
is preserved in the projection. It is obvious that an inverse map may be
applied F := f
1
, which shall be called the ination of the mesh. Now the problem has
been reduced to that of nding an R
2
-morphism from the planar mesh without subdivision
connectivity to one with such a connectivity, subject to the closeness condition as before.
Then the projection and ination operators may be used to construct the equivalent map
U

= FU

f
1
. The general idea of the mathematical structure of the adaptive remeshing
algorithm may be summarized by the commutative diagram below.
M

_f

F
M

Details of the procedure and several algorithms are described at length in [20]. This
group also had generated a triangulation of a sphere following the adaptive meshing pro-
cedure which was used with permission for the purposes of this work. The rst few steps
presented in their work is shown in gure 5.1.1.
5.1.2 Computation of Simplicial Geodesics
Once the triangulation was accomplished by the process described above, it was necessary
to compute the geodesics of the triangulation in order to pinpoint the region to be extracted
in the construction of the seed manifold. For a continuous sphere the geodesics are easily
46
Figure 5.2: The triangulated manifold with simplicial great circles along the xz and yz
planes
shown to be the great circles along a given equatorial plane. An algorithm was developed
to generalize this concept to the triangulated manifold. First a given number of discrete
points were calculated along the great circle in the continuum case, via the simple geometric
formula for an eective radius along the circle:
R
e,i
(g
i
) = [[g
i
[[ sin
_
arccos
_
1
[[g
i
[[
__
, (5.1)
where g
i
is the vector from the origin to the specied point on the continuum great circle.
The algorithm then found the vertex V
m
S

subject to the condition that


g
E
(V
m
, g
i
) = min
V S

g
E
(V, g
i
), (5.2)
47
where g
E
is the canonical euclidean metric in R
2
. This was repeated for each point in the
continuum great circle. The set of all V
m,i
was considered to constitute the simplicial
great circle along the chosen axis. This procedure was carried out for great circles along
the orthogonal axis corresponding to the xz and yz planes in the euclidean coordinates of
the remeshed sphere. These are shown in the gure 5.1.2.
5.1.3 Extraction of the First Quadrant Wedge
The second step of the generation of the seed manifold involved the extraction of a given
wedge, isolated by the boundaries of the great circles. The manifold and great circles gen-
erated in this case were well geometrized enough to make coarse cuts and replace individual
point. The extraction is shown in the gure 5.1.3.
Figure 5.3: The triangulated manifold with simplicial great circles along the xz and yz
planes, and positive quadrant removed
48
Chapter 6
Conclusion
We have outlined a concrete investigation into the possibility of using Causal Dynamical
Triangulations (CDT) to distinguish between Horava-Lifshitz quantization and quantiza-
tion of traditional general relativity via the generation of a football shaped seed manifold
and use of such a manifold in a (2 + 1)-dimensional CDT simulation, looking for evi-
dence of propagating scalar modes. We showed analytically in chapters two and four these
quantization procedures can be distinguished by the fact that while quantizing traditional
gravity in (2 + 1)-dimensions yields no propagating scalar modes, (2 + 1)-quantization of
Horava-Lifshitz gravity predicts at least one, dynamical in the spatial components of the
metric. We saw that this is intimately related to the fact that Horava-Lifshitz gravity
scales anisotropically in the temporal dimension, and hence the dierence can be thought
of as the dierence between isotropic and anisotropic quantizations of gravity.
After a cursory description of quantization in a historical context, we described in detail
the derivation and scientic content of results from traditional (2 +1)-dimensional general
relativity. This was followed by an in depth description of the theory of CDT, as well as
Horava-Lifshitz gravity and the derivation of propagating scalar modes. We then presented
49
a description of the computational process, outlining the general computational theory of
CDT simulations, and nally described the method by which a suitable seed manifold
was generated, where in particular we introduced a heretofore unused computer graphics
techniqueadaptive remeshingas a method of creating a high quality approximation of the
triangulation for any seed manifold, implying that in the future many other seed universes
may be generated easily by employing this technique. The UC Davis CDT code, on which
simulations will take place, is currently in preparation for handling specic topological
aspects of the seed as generated, and are expected to be functional in the near future.
Results, when available, will give insight into whether or not Horava-Lifshitz gravity
has deep underlying similarities to CDT-gravity. This will provide important evidence
to be considered in further studies of Horava-Lifshitz and CDT gravity, and will thus
help advance the understanding of various theories of quantum gravity, arguably the most
important problem in all of science today.
50
Bibliography
[1] Zee, A. (2010). Quantum Field Theory in a Nutshell Princeton University Press,
Princeton.
[2] Peskin, M. E., Schroeder, D. V. (1995). An Introduction to Quantum Field Theory.
Westview Press.
[3] Griths, D. J. (1987). Introduction to Elementary Particles. John Wiley & Sons, Inc.
[4] Wick, G. C. (1954). Properties of Bethe-Salpeter Wave Functions. Physical Review 96
(4): 1124-1134.
[5] Loll, R. (2008). The Emergence of Spacetime or Quantum Gravity on Your Desktop,
Class. Quantum Grav. 25 114006.
[6] Gabrielse, G., Henneke, D. (2006). Percision pins down the electrons magnetism.
CERN Courier 46 (8):35-37.
[7] Odom, B., Hanneke, D., dUrso, B, Gabrielse, G. (2006). New measurement of the
electron magnetic moment using a one-electron quantum cyclotron. Physical Rebiew
Letters 97 (3):030801.
[8] Ambjorn, J., Jurkiewicz, J. and R. Loll (2012). Causal Dynamical Triangulations and
the Quest for Quantum Gravity. Foundations of Space and time, pp. 321-337.
51
[9] Sorkin, R. D. (1986). Non-time-orientable Lorentzian cobordism allows for pair cre-
ation, Int. J. Theor. Phys 25, 877-881.
[10] Loll, R., Jordan, S. (2013) De Sitter Universe from Causal Dynamical Triangulations
without Preferred Foliation. arXiv:1307.5469v1
[11] Mess, G. (1990). Lorentz spacetimes of constant curvature, Institut des Hautes Etudes
Scientiques preprints IHES/M/90/28.
[12] Carlip, S. (1998). Quantum gravity in (2 + 1)- Dimensions, (Cambridge University
Press, Cambridge).
[13] Moncrief, V. (1989). Reduction of the einstein equations in (2 + 1)-dimensions to a
Hamiltonian system over Teichm uller space, J. Math. Phys. 30, 2907-2914.
[14] Regge, T. (1961). general relativity without coordinates. Nuovo Cimento 19, 558-571.
[15] Edmonds, A. R. (1957). Angular Momentum in Quantum Mechanics Princeton Uni-
versity Press, Princeton.
[16] Ponzano, G. and Regge, T. (1968). Semiclassical limit of Racah coecients in Spec-
troscopic and group theoretical methods in physics, ed. F. Bloch, S. G. Cohen, A.
De-Shalit, S. Sambursky, and I. Talmi (North-Holland, Amsterdam).
[17] Ooguru, H. (1992). Partition functions and topology-changing amplitudes in the three-
dimensional lattice gravity of Ponzano and Regge, Nucl. Phys. B382, 276-304.
[18] Katzgraber, H. G. (2009) Introduction to Monte-Carlo Methods Modern Computation
Science Eds. A. K. Hartmann and R. Leid l, BIS-Verlag Oldenburg, Germany.
52
[19] Kenmoku, M., Uchida, S., Matsuyama, T. (2002). Conical Singular Solutions in (2+1)-
Dimensional Gravity Employing the ADM Canonical Formalism. Int. Journal of Mod.
Phys. D, 12 677-687.
[20] Hormann, K., Labsik, U., Greiner, G. (2001). Remeshing triangulated surfaces with
optimal parameterizations. Computer-Aided Design 33, 779-788.
[21] Eck, M., DeRose, T., Duchamp, T., Hoppe, H., Lounsbery, M., and Stuetzle, W.
(1995). Multiresolution analysis of arbitrary meshes. In ACM Computer Graphics
SIGGRAPH 95 Proceddings, 173-182.
[22] Kobbelt, L., Vorsatz J., Labsik, U., and Seidel, H. P. (1999). A shrink wrapping
approach to remeshing polygonal surfaces. In Computer Graphics Forum (EURO-
GRAPHICS 99 Proceedings), 199-130.
[23] Lee, A., Sweldens, W., Schroder, P., Coswar, L., and Dobkin, D. (1998). Multirelosu-
tion adaptive parametrization of surfaces. In ACM Computer Graphics (SIGGRAPH
98 Proceedings), 95-104.
[24] Loop, C. (1987). Smooth subdivsion surfaces based on triangles. Masters thesis, Utah
University.
[25] Anderson, C., Carlip, S. J., Cooperman, J. H., Horava, P., Kommu, R. K., Zulkowski,
P. R. (2012). Phys. Rev. D 85, 044027.
[26] J. Ambjorn, A. Gorlich, J. Jurkiewicz, R. Loll, J. Gizbert-Studnicki, and T. Trzes-
niewski, (2011). Nucl. Phys. B849, 144.
[27] J. Ambjorn, S. Jordan, J. Jurkiewicz, and R. Loll, (2011). Phys. Rev. Lett. 107,
211303.
53
[28] Sotiriou, T.P., Visser, M., Weinfurtner, S. (2011). Lower-Dimensional Horava-Lifshitz
Gravity. arXiv:1103.3013 (http://arxiv.org/abs/1103.3013).
[29] Poincare, H. (1908). Sur luniformisation des fonctions analytiques, Acta Mathematica
31, 1.
[30] Gorlich, A. (2010). Causal Dynamical Triangulations in Four Dimensions. PhD Thesis.
arXiv:1111.6938.
[31] Ambjorn, J., Jain, S., Jurkiewicz, J., Kristjansen, C.F. (1993). Observing 4D Baby
Universes in Quantum Gravity. Phys. Lett. B 305, 208.
[32] Zohren, S. (2007). Causal Dynamical Triangulations. Enrage Newsletter, Number 2,
June 2007.
[33] Parunak, H.V.D., Brueckner, W., Savit, R. (2004). Universality in Multi-Agent Sys-
tems. Third International Joint Conference on Autonomous Agents and Multi-Agent
Systems. AAMAS 2004, pp. 930-937.
54

Das könnte Ihnen auch gefallen