ADP-95-11/M28, gr-qc/0101003
arXiv:gr-qc/0101003v2 3 Sep 2003
in B. Robson, N. Visvanathan and W.S. Woolcock (eds.), “Cosmology: The Physics
of the Universe”, Proceedings of the 8th Physics Summer School, Australian National University, Canberra, Australia, 16 January – 3 February, 1995, (World Scientific, Singapore, 1996), pp. 473-531.
Note added: These lecture notes, as published above, date from 1995. For more
recent work on the issue of the prediction of the period of inflation and the debate
about boundary condition proposals (§5.1), see:
N. Kontoleon and D.L. Wiltshire, Phys. Rev. D59 (1999) 063513 [= gr-qc/9807075].
D.L. Wiltshire, Gen. Relativ. Grav. 32 (2000) 515 [= gr-qc/9905090].
M.J.W. Hall, K. Kumar, and M. Reginatto, J. Phys. A36 (2003) 9779 [= hepth/0307259].
AN INTRODUCTION TO QUANTUM COSMOLOGY
DAVID L. WILTSHIRE
Department of Physics and Mathematical Physics,
University of Adelaide, S.A. 5005, Australia.
CONTENTS
1. Introduction
. . . 1.1 Quantum cosmology and quantum gravity
. . . 1.2 A brief history of quantum cosmology
2. Hamiltonian Formulation of General Relativity
. . . 2.1 The 3 + 1 decomposition
. . . 2.2 The action
3. Quantisation
. . . 3.1 Superspace
. . . 3.2 Canonical quantisation
. . . 3.3 Path integral quantisation
. . . 3.4 Minisuperspace
. . . 3.5 The WKB approximation
. . . 3.6 Probability measures
. . . 3.7 Minisuperspace for the Friedmann universe with massive scalar field
4. Boundary Conditions
. . . 4.1 The no-boundary proposal
. . . 4.2 The tunneling proposal
5. The Predictions of Quantum Cosmology
. . . 5.1 The period of inflation
. . . 5.2 The origin of density perturbations
. . . 5.3 The arrow of time
6. Conclusion
1. Introduction
These lectures present an introduction to quantum cosmology for an audience
consisting for a large part of astronomers, and also a number of particle physicists.
As such, the material covered will invariably overlap with that of similar introductory reviews1,2 , although I hope to emphasise, where possible, those aspects of
quantum cosmology which are of most interest to astronomers.
This is not an easy task – quantum cosmology is not often discussed at Summer
Schools such as this where there is a large emphasis on astrophysical phenomenology,
1
for the very good reason that the ideas involved are at present rather tentative, and
quantitative predictions are thus much more difficult to arrive at. Nevertheless
we should remember that not too long ago the whole enterprise of cosmology was
viewed as the realm of wild speculation. Vladimir Lukash remarked that when he
started out in research the advice he was given was that “cosmology was all right
for someone like Zeldovich to potter around with after he had already established
himself by producing a body of serious work, but it was inappropriate for a young
physicist embarking on a career”. Happily the situation is quite different today –
I am sure the earlier lectures in this School will have convinced you that modern
cosmology is a hardcore quantitative science, and that with the new technology
and techniques now being developed we can expect to accurately measure all the
important cosmological parameters within the next decade and thus enter into a
“Golden Era” for cosmological research.
Quantum cosmology, however, still enjoys the sort of status that all of cosmology
had until not so very long ago: essentially it is a dangerous field to work in if you
hope to get a permanent job. I hope to convince you nevertheless that quantum
cosmology represents a vitally important frontier of research, and that although it is
by nature somewhat speculative, such speculations are vital if we are to understand
the entire history of the universe.
On the face of it the very words “quantum” and “cosmology” do appear to some
physicists to be inherently incompatible. We usually think of cosmology in terms of
the very large scale structure of the universe, and of quantum phenomena in terms
of the very small. However, if the hot big bang is the correct description of the
universe – which we can safely assume given the overwhelming evidence described
in the earlier lectures – then the universe did start out incredibly small, and there
must have been an epoch when quantum mechanics applied to the universe as a
whole.
There are people who would take issue with this. In the standard Copenhagen
interpretation of quantum mechanics one always has a classical world in which the
quantum one is embedded. We have observers who make measurements – the observers themselves are well described by classical physics. If the whole universe is to
be treated as a quantum system one does not have such a luxury, and some would
argue that our conventional ideas about quantum physics cease to make sense. Yet
if quantum mechanics is a universal theory then it is inevitable that some form of
“quantum cosmology” was important at the earliest of conceivable times, namely
1/2
the Planck time, tPlanck = (G~) /c5/2 = 5.4 × 10−44 sec, (equivalent to 1019 GeV
as an energy, or 1.6 × 10−35 m as a length). At such scales, where the Compton
wavelength of a particle is roughly equal to its gravitational (Schwarzschild) radius,
irreducible quantum fluctuations render the classical concept of spacetime meaningless. Whether or not our current efforts at constructing a theory of quantum
cosmology are physically valid is therefore really a question of whether our current
understanding of quantum physics is adequate for considering the description of
processes at the very beginning of the universe, or whether quantum mechanics
2
itself has to be revised at some level. Such a question can really only be answered
by extensive work on the problem.
Setting aside the question of the fundamentals of quantum mechanics, let us
briefly review the problems which are left unanswered in the standard hot big bang
scenario. These are:
1. The value of η = Nbaryons /Nphotons : the exact value of this parameter is unexplained
in the hot big bang but is crucial in determining abundances of light elements
through primordial nucleosynthesis.
2. The horizon problem: the isotropy of the cosmic microwave background radiation
indicates that all regions of the sky must have been in thermal contact at some time
in the past. However, in the standard Friedmann-Robertson-Walker (FRW) models
regions separated by more than a couple of degrees have non-intersecting particle
horizons – i.e., they cannot have been in causal (and hence thermal) contact.
3. The flatness problem: given the range of possible values of the ratio of the density
of the universe to the critical density at the present epoch, Ω0 , then the FRW models
predict that Ω(t) must have been incredibly close to the spatially flat case Ω(t) ≃ 1
< −26 at the lepton era,
at early times; e.g., assuming Ω0 >
∼ 0.1 we find |Ω − 1| ∼ 10
−53
at the GUT era.
and |Ω − 1| <
∼ 10
4. The unwanted relic problem: models of the early universe which involve phase
transitions often produce copious amounts of topological defects, such as monopoles
produced at the GUT scale. If one puts the numbers in one finds that the density of
such relics is so great that they would exceed the critical density by such a margin
in the standard FRW models that the universe should have ended long ago!
5. The origin of density perturbations is unexplained.
6. The arrow of time is a physical mystery. On the one hand the laws of physics
are CPT-invariant, and on the other there is a thermodynamic arrow of time, as
prescribed by the Second Law of Thermodynamics, and it appears to match the
cosmological arrow of time, as prescribed by the expansion of the universe.
7. The initial conditions of the universe must be put in by hand, rather than being
physically prescribed.
The first four problems on the list are ones that can be explained without
appealing to quantum cosmology. The value of η (problem 1) is predicted by models
of baryogenesis, which typically take place at the GUT scale. Problems 2–4 are
solved by the inflationary universe scenario: an early phase of exponential expansion
of the universe drastically changes the past light cone, thereby removing the horizon
problem, while also driving Ω close to unity, and diluting unwanted relics to such
very low densities that they are close to unobservable.
Problems 5–7, on the other hand, are of a nature which is beyond the scope
of the inflationary universe scenario to satisfactorily explain. Inflation provides a
mechanism whereby initial small “quantum” perturbations are inflated to all length
scales to form a scale-free Harrison-Zeldovich spectrum, but it does not address
the question as to exactly how these perturbations arise. Furthermore, a typical
3
model of the very early universe might possess both inflationary and non-inflationary
solutions, so that the precise initial conditions of the universe can be crucial for
determining whether the universe undergoes a period of inflation sufficiently long to
be consistent with observation. The length of the period of inflation is precisely the
sort of quantitative result that we might hope quantum cosmology should provide.
Questions such as the origin of the arrow of time might appear to be of a more
philosophical nature – however, quantum cosmology should provide a calculational
framework in which such questions can begin to be addressed.
1.1. Quantum cosmology and quantum gravity
Quantum cosmology is perhaps most properly viewed as one attempt among
many to grapple with the question of finding a quantum theory of gravity. As a
field theory general relativity is not perturbatively renormalisable, and attempts to
reconcile general relativity with quantum physics have not yet succeeded despite
the attentions of at least one generation of physicists. It is perhaps not surprising
that the problem is such a difficult one since general relativity is a theory about the
large scale structure of spacetime, and to quantise it we have to quantise spacetime
itself rather than simply quantising fields that live in spacetime.
Many ideas have been considered in the quest for a fundamental quantum theory of gravity – whether or not these ideas have brought us closer to that goal is
difficult to say without the benefit of hindsight. However, such ideas have certainly
profoundly increased our knowledge about the nature of possible physical theories.
Some important areas of research have included:
1. Supergravity 3–5 . Using supersymmetry, a symmetry between fermions and
bosons, one can enlarge the gravitational degrees of freedom to include one or more
spin– 32 gravitinos, ψµ , in addition to the spin–2 graviton, gµν . Such a symmetry can
cure some but not all of the divergences of perturbative quantum general relativity∗.
In particular, while pure Einstein gravity is perturbatively non-renormalisable at
two loops7 , or at one loop if interacting with matter8–10 , in the case of supergravity
renormalisability fails only at the 3-loop level11◮ .
2. Superstring theory 13 . Progress can be made if in addition to using supersymmetry, one constructs a theory in which the fundamental objects have an extension
rather than being point-like: a theory of strings rather than particles. Much interest in string theory was generated in the mid 1980s with the discovery that certain
string theories appear to be finite at each order of perturbation theory. In some
∗
For details of the application of perturbative techniques to quantum cosmology, both with and without
supersymmetry, see [6].
◮
The status of the result concerning the 3-loop divergence of supergravity is not quite as rigorous as the
other examples mentioned, as the complete 3-loop calculation has not been done. However, it is known that
a 3-loop counterterm exists for all extended supergravities and there is no reason to expect the coefficient
of the counterterm to be zero, making 3-loop finiteness extremely unlikely. For a review of the ultraviolet
properties of supersymmetric field theories see [12].
4
sense stringy physics “smears out” the problems associated with pointlike interactions. The entire sum of all terms in the perturbation expansion diverges in the
case of the bosonic string14 , however, and it is believed that similar results should
apply to the superstring⊛ . Furthermore, despite what many see as the mathematical beauty of string theory, there has unfortunately as yet been no definitive success
in deriving concrete phenomenological predictions.
3. Non-perturbative canonical quantum gravity 16,17. The fact that general relativity
is not perturbatively renormalisable might simply be a failure of flat spacetime
quantum field theory techniques to deal with such an inherently non-linear theory,
rather than reflecting an inherent incompatibility between general relativity and
quantum physics. Given the divergence of the string perturbation series mentioned
above, a non-perturbative formulation of string theory would also be desirable. As a
starting point a systematic investigation of a non-perturbative canonical formalism
based on general relativity could provide deep insights into quantum gravity. Such a
programme has been investigated by Ashtekar and coworkers, mainly from the mid
1980s onwards. One principal difference from the canonical formalism which I will
describe in section 2 is that instead of taking the metric to be the fundamental object
to describe the quantum dynamics, one bases such a dynamics on a connection, in
this case an SL(2, C) spin connection18,19 . If one integrates this connection around
a closed loop one arrives at “loop variables”, which might be considered to be
analogous to the magnetic flux, Φ, obtained by integrating the electromagnetic
gauge potential, Aµ , around a closed loop. In the “loop representation” quantum
states are represented by functionals of such loops on a 3-manifold20,21 , rather
than by functionals of classical fields. Although a number of technical difficulties
remain, considerable progress has been made with the Ashtekar formulation, and
the reformulation of quantum cosmology in the Ashtekar framework22–25 could be
an area for some interesting future work.
4. Alternative models of spacetime. All the above approaches assume that the basic
quantum variables, be they a metric or a connection, are defined on differentiable
manifolds. Given that it is highly possible that “something strange” happens at the
Planck length it is plausible that one might have to abandon this assumption in order
to effectively describe the quantum dynamics of gravity. A number of ideas have
been considered on these lines. One framework which has been widely used both
for numerical relativity and studies of quantum gravity is that of Regge calculus 26△
in which one replaces smooth manifolds by spaces consisting of piecewise linear
simplicial blocks. Naturally, other possibilities for discretised spacetime structure
also exist – causal sets 28 being another example which has not yet been so widely
explored. A further possibility is topological quantisation whereby one replaces a
manifold by a set and quantises all topologies on that set29,30.
⊛
The question of finiteness of superstring perturbation theory is a difficult technical question, which
has still to be resolved – see, e.g., [15].
△
For a brief review and extensive bibliography see [27].
5
Many of the above alternatives fall into the category of being attempts to construct a fundamental theory of quantum gravity. The current quantum cosmology
programme is not quite as ambitious. One begins by making the assumption that
whatever the exact nature of the fundamental theory of quantum gravity is, in its
semiclassical limit it should agree with the semiclassical limit of a canonical quantum formalism based on general relativity alone. Thus we study the semiclassical
properties of quantum gravity based solely on Einstein’s theory, or some suitable
modification of it.
Clearly, any predictions made from such a foundation must be treated with
caution. In particular, a new fundamental theory of quantum gravity might introduce radically new physical processes at an energy scale relevant for cosmology.
String theory, for example, introduces its own fundamental scale which expressed
as a temperature (the Hagedorn temperature) is given by:
THagedorn =
√
~c
√
4πkB α′
,
(1.1)
the constant α′ being the Regge slope parameter, which is inversely proportional
to the string tension. It is not known exactly what the value of α′ is; however,
if THagedorn is comparable to or significantly lower than the Planck scale, then it
is clear that fundamentally “stringy” processes will be very important in the very
early universe, if string theory is indeed the “ultimate” theory.
Although new fundamental physics could drastically change the predictions of
quantum cosmology, I believe nevertheless that studying quantum cosmology based
even just on Einstein’s theory is an important activity. General relativity is a
remarkably successful theory; in seeking to replace it by something better it is important that we study processes at the limit of its applicability, thereby challenging
our understanding. General relativity is limited by the Planck scale – the physical
arenas in which this scale is approached include: (i) very small black holes; (ii) the
very early universe. Since it seems unlikely that we will ever be able to create energies of order 1019 GeV in the laboratory, the consequences of quantum gravity for
the physics of the very early universe will remain the one way of indirectly “testing”
it, at least for the foreseeable future.
It is thus important that we consider quantum gravity in a cosmological context.
Even if our current attempts do not fully reach the mark, in that we do not yet
have a fully-fledged quantum theory of gravity, they nonetheless constitute a vital
part of the process of trying to find such a theory.
1.2. A brief history of quantum cosmology
The quantum cosmology programme which I will describe in these lectures has
gone through three main identifiable phases to date:
6
1. Defining the problem. The canonical formalism, including the definition of the
wavefunction of the universe, Ψ, its configuration space – superspace – and its evolution according to the Wheeler-DeWitt equation, was set up in the late 1960s31–35 .
2. Boundary conditions. Quantum cosmology research went into something of a lull
during the 1970s but was revived in the mid 1980s when the question of putting
appropriate boundary conditions on the wavefunction of the universe was treated
seriously. The idea is that such boundary conditions should describe the “creation
of the universe from nothing”36,37 , where nothing means the absence of space and
time. A number of proposals for such boundary conditions emerged – two major
contenders being the “no-boundary” proposal of Hartle and Hawking38,39 and the
“tunneling” proposal advocated by Vilenkin40–42 .
3. Quantum decoherence. The mechanism of the transition from quantum physics to
classical physics (“quantum decoherence”) becomes vitally important when quantum physics is applied to the universe as a whole. The issues involved have begun
to occupy many researchers in the early 1990sH .
Two other important areas of quantum cosmology (or related) research have been:
(i) quantum wormholes and “baby universes”; (ii) supersymmetric quantum cosmology.
Quantum wormholes> were extremely fashionable in the particle physics community in the years 1988–1990. Such states arise when one considers topology
change in the path integral formulation of quantum gravity: quantum wormholes
are instanton solutions which play an important role in the Euclidean path integral. One deals directly with a “third quantised” formalism, (i.e., quantum field
theory over superspace), which includes operators that create and destroy universes
– so-called baby universes. Much of the excitement in the late 1980s was associated
with the idea that such processes could fix the fundamental constants of nature –
in particular, driving the cosmological constant to zero.
Supersymmetric quantum cosmology has emerged recently as one of the most active areas of current research⋔ . In considering the quantum creation of the universe
we are of course dealing with the very earliest epochs of the universe’s existence, at
which time it is believed that supersymmetry would not yet be broken. The inclusion of supersymmetry could therefore be vital from the point of view of physical
consistency.
Since the focus of this School is on cosmology, my intention in these lectures
is to cover topics 1 and 2 above, and then to proceed to discuss the predictions of
quantum cosmology. The third topic of quantum decoherence raises questions which
have not been resolved even in ordinary quantum mechanics, since the question of
decoherence really amounts to understanding what happens during the “collapse
of the wavefunction”. Although this is a fascinating issue it has more to do with
H
>
⋔
See [43] for a review and [44] for a collection of recent papers on the subject.
See [45] for a review.
See [46] and references therein.
7
the fundamentals of quantum mechanics than directly with cosmology. Likewise,
I will only briefly touch upon quantum wormholes and supersymmetric quantum
cosmology, as these areas are still in their infancy, and one is still at the point of
trying to resolve basic questions concerning quantum gravity. I hope the reader will
not be disappointed by this – however, given the vast scope of quantum gravity and
quantum cosmology one must necessarily be rather selective.
2. Hamiltonian Formulation of General Relativity
2.1. The 3 + 1 decomposition
In order to discuss quantum cosmology a fair amount of technical machinery
is required. In the canonical formulation we begin by making a 3 + 1-split of the
4-dimensional spacetime manifold, M, which will describe the universe, foliating it
into spatial hypersurfaces, Σt , labeled by a global time function, t. Thus we take
the 4-dimensional metric to be given by
ds2 = gµν dxµ dxν = −ω0 ⊗ ω0 + hij ωi ⊗ ωj ,
where
ω0 = N dt
(2.1)
(2.2)
ωi = dxi + N i dt.
Such a decomposition is possible in general if the manifold M is globally hyperbolic 47 .
The quantity N (t, xk ) is called the lapse function – it measures the difference between the coordinate time, t, and proper time, τ , on curves normal to the hypersurfaces Σt , the normal being nα = (−N , 0, 0, 0) in the above coordinates. The
quantity N i (t, xk ) is called the shift vector – it measures the difference between a
spatial point, p, and the point one would reach if instead of following p from one
hypersurface to the next one followed a curve tangent to the normal n. That is
to say, the spatial coordinates are “comoving” if N i = 0. Finally, hij (t, xk ) is the
intrinsic metric (or first fundamental form) induced on the spatial hypersurfaces by
the full 4-dimensional metric, gµν . In components we have
−N 2 + N k Nk Nj
,
(2.3)
(gµν ) =
Ni
hij
with inverse
(g
µν
)=
Nj
N2
i
j
ij
h − NNN2
−1
N2
Ni
N2
,
(2.4)
where hij is the inverse to hij , and the intrinsic metric is used to lower and raise
spatial indices: N k Nk ≡ hjk Nj Nk = hjk N j N k etc.
We use a (− + ++) Lorentzian metric signature, and natural units in which c = ~ = 1.
8
nµ
N i dt
dx i
Σt+dt
dτ = Ndt
Σt
xi
x i+dx i
Fig. 1: The 3 + 1 decomposition of the manifold, with lapse function, N , and shift
vector, N i .
One can construct an intrinsic curvature tensor 3Ri jkl (h) from the intrinsic
metric alone – this of course describes the curvature intrinsic to the hypersurfaces
Σt . One can also define an extrinsic curvature, (or second fundamental form), which
describes how the spatial hypersurfaces Σt curve with respect to the 4-dimensional
spacetime manifold within which they are embedded. This is given by
0
Kij ≡ − ni;j = −Γ ij n0
1
∂hij
=
Ni|j + Nj|i −
,
2N
∂t
(2.5)
where a semicolon denotes covariant differentiation with respect to the 4-metric, gµν ,
and a vertical bar denotes covariant differentiation with respect to the 3-metric, hij :
Ni|j ≡ Ni ,j −Γk ij Nk etc.
For a given foliation of M by spatial hypersurfaces, Σt , it is always possible to
choose Gaussian normal coordinates, in which
ds2 = −dt2 + hij dxi dxj .
(2.6)
These are comoving coordinates (N i = 0) with the additional property that t is
the proper time parameter (N = 1). This is the standard “gauge choice” that is
made in classical cosmology, and in such coordinates Kij = −ḣij , where dot denotes
9
differentiation with respect to t. In making the 3 + 1 decomposition, however, we
are only free to make a specific choice of coordinates such as (2.6) after variation of
the action if we want to be sure to obtain Einstein’s equations, and thus we must
retain the lapse and shift function for the time being.
2.2. The action
A relevant action for use in quantum cosmology is that of Einstein gravity plus
a possible cosmological term, Λ, and matter, given by
Z
Z
√
√
1
S = 2 d4 x −g 4R − 2Λ + 2
d3 x h K + Smatter ,
(2.7)
4κ
M
∂M
i
where κ2 = 4πG = 4πm−2
Planck , K ≡ K i is the trace of the extrinsic curvature,
and for many simple models the matter is specified by a single scalar field, Φ, with
potential, V(Φ),
Z
1 µν
4 √
(2.8)
Smatter = d x −g − g ∂µ Φ∂ν Φ − V(Φ) .
2
M
The boundary term48 in (2.7) does not of course contribute to the classical field
equations, and this term is usually omitted in a first course on general relativity.
However, in quantum physics we are often interested in phenomena which occur
when the classical field equations do not apply, that is “off-shell”, and thus it is
vitally important to retain the surface term. The simple matter action (2.8) given
here should simply be seen as being representative of the type of matter action
one might consider. Although the example given by (2.8) is sufficient for studying
many inflationary universe models, many other alternatives might also be of interest,
such as extra matter from a supergravity multiplet, or matter corresponding to the
low-energy limit of string theory. In the latter case, if one works in the “string
conformal frame” it is also necessary to alter the gravitational part of the action,
as one characteristic of string theory is the presence of the scalar dilaton, Φ, which
couples universally to matter (at least perturbatively). In that case (2.7), (2.8)
would be replaced by
Z
√
1
S= 2
d4 x −ge−2Φ 4R + 4g µν ∂µ Φ∂ν Φ − 8V(Φ) + . . .
4κ
M
(2.9)
Z
√ −2Φ
3
+2
d x he
K ,
∂M
where the ellipsis denotes any additional matter degrees of freedom, and we have
allowed for the possibility of the generation of a dilaton potential, V(Φ), via some
10
non-perturbative symmetry breaking mechanism. However, for simplicity the specific examples we will deal with here will be confined to models of the type (2.7),
(2.8).
We now wish to express (2.7), (2.8) in terms of the variables of the 3 + 1 split.
One can show that
4
R = 3R − 2Rnn + K 2 − K ij Kij ,
(2.10)
where 3R is the Ricci scalar of the intrinsic 3-geometry, and
Rnn ≡ Rαβ nα nβ = −K ij Kij + K 2 + (nα K + aα );α ,
(2.11)
with aα ≡ nβ nα;β . Combining (2.7), (2.10) and (2.11), and noting that the boundary integral involving aα vanishes identically since nα aα = 0, we obtain
Z
Z
√
1
(2.12)
S ≡ dt L = 2 dt d3 xN h Kij K ij − K 2 + 3R − 2Λ + Smatter .
4κ
As in the Hamiltonian formulation of field theory we define canonical momenta in
the standard fashion
√
h
δL
(2.13)
πij ≡
= − 2 K ij − hij K ,
4κ
δ ḣij
√
δL
h
Φ̇ − N i Φ,i ,
(2.14)
πΦ ≡
=
N
δ Φ̇
δL
π0 ≡
= 0,
(2.15)
δ Ṅ
δL
πi ≡
= 0.
(2.16)
δ Ṅi
The fact that the momenta conjugate to N and Ni vanish means that we are dealing
with primary constraints in Dirac’s terminology49,50.
If we use N , Ni , hij , Φ and their conjugate momenta as the basic variables we
obtain♦ a Hamiltonian
Z
H ≡ d3 x π0 Ṅ + πi Ṅi + πij ḣij + πΦ Φ̇ − L
Z
(2.17)
3
0
i
i
= d x π Ṅ + π Ṅi + N H + Ni H ,
where
♦
√
h
H = 2 G0̂0̂ − 2κ2 T 0̂0̂
2κ
Details of all missing steps in this section will be provided upon request in a plain brown envelope.
11
h
= 4κ Gijkl π π − 2
4κ
√
h
i
0̂ı̂
2 0̂ı̂
H = 2 G − 2κ T
2κ
= −2πij |j + hij Φ,j πΦ ,
2
and
√
ij kl
1 √ πΦ2
ij
R − 2Λ + 2 h
+ h Φ,i Φ,j +2V , (2.18)
h
3
(2.19)
Gijkl = 21 h−1/2 (hik hjl + hil hjk − hij hkl )
(2.20)
33
is the DeWitt metric . The hats in (2.18) and (2.19) denote orthonormal frame
components of the Einstein and energy-momentum tensors. In terms of these variables the action (2.12) then becomes
Z
S = dt d3 x π0 Ṅ + πi Ṅi − N H − Ni Hi .
(2.21)
If we vary (2.21) with respect to πij and πΦ we obtain their defining relations (2.13)
and (2.14). The lapse and shift functions now act as Lagrange multipliers; variation
of (2.21) with respect to the lapse function, N , yields the Hamiltonian constraint
H = 0,
(2.22)
while variation of (2.21) with respect to the shift vector, Ni , yields the momentum
constraint
Hi = 0.
(2.23)
From (2.18) and (2.19) it is clear that these constraints are simply the (00) and
(0i) parts of Einstein’s equations. In Dirac’s terminology these are secondary or
dynamical constraints.
The 3+1 decomposition of our spacetime looks to be very counterintuitive to the
usual ideas of general relativity. This is so by choice. Einstein described the l.h.s. of
his field equations which encodes the 4-dimensional geometry as a “hall of marble”,
which does not encourage people to tamper with it. However, to quantise spacetime
we must do just that: we must deconstruct spacetime and replace it by something
else. Thus far we have not done that – we have simply split our 4-dimensional
manifold into a sequence of spatial hypersurfaces, Σt . Time is the natural variable
to base this split upon since it plays a special role in quantum mechanics – it is a
parameter rather than an operator.
Classically the evolution of one spatial hypersurface to the next is completely
well-defined, (provided that the manifold, M, is globally hyperbolic), and given
initial data, hij , Φ, on an initial hypersurface, Σ, we can use the Cauchy development to stitch the hypersurfaces Σt together to recover the 4-dimensional manifold,
M. To quantise the theory, however, we want to perform a path integral over all
geometries, not just the classically allowed ones. Thus we must consider sequences
of geometries at the quantum level which cannot be stitched together in a regular
Cauchy development to form a 4-manifold which solves Einstein’s equations. (See
Fig. 2.) We must therefore abandon Einstein’s “hall of marble” – spacetime is no
longer a fundamental object.
12
Σ’
Σ’
Σ’
+
Σ
+
+ . . .
Σ
Σ
Fig. 2: Quantum geometrodynamics: in addition to the classical Cauchy development from Σ to Σ′ (left), the path integral includes a sum over all 4-manifolds
which interpolate between the initial and final configurations. The weighting by eiS
means that the greater the number of classically forbidden 3-geometries (shaded
slices) is in the interpolating 4-manifold, the smaller its contribution is to the path
integral.
As was mentioned in the Introduction the “deconstruction” of spacetime that
we adopt here is probably the most conservative choice we could make. Even though
we abandon the notion of spacetime in discussing the quantum dynamics of gravity,
our fundamental objects are still defined on regular 3-manifolds, Σ. A more radical
departure would be to replace these spatial hypersurfaces by some more general
set by using, for example, the ideas of Regge calculus, causal sets or topological
quantisation.
3. Quantisation
3.1. Superspace
As a prelude to the canonical quantisation of gravity let us first introduce the
relevant configuration space on which the quantum dynamics will be defined.
Consider the space of all Riemannian 3-metrics and matter configurations on
the spatial hypersurfaces, Σ,
Riem (Σ) ≡ {hij (x), Φ(x) | x ∈ Σ} .
(3.1)
This is an infinite-dimensional space, on account of the label x = {xi }, which
specifies the point on the hypersurface, but there are a finite number of degrees
of freedom at each point, x ∈ Σ. In fact, we are really interested in geometry
here and configurations which can be related to each other by a diffeomorphism,
i.e., a coordinate transformation, should be considered to be equivalent since their
intrinsic geometry is the same. Thus we factor out by diffeomorphisms on the spatial
13
hypersurfaces and identify superspace ⋆ as
Riem (Σ)
,
Diff0 (Σ)
where the subscript zero denotes the fact that we consider only diffeomorphisms
which are connected to the identity. This infinite-dimensional space will be the
basic configuration space of quantum cosmology.
The DeWitt metric (2.20) then provides a metric on superspace which we can
write as
GAB (x) ≡ G(ij)(kl) (x),
(3.2)
where the indices A, B run over the independent components of the intrinsic metric
hij :
A, B ∈ {h11 , h12 , h13 , h22 , h23 , h33 }.
The DeWitt metric has a (− + + + ++) signature at each point x ∈ Σ, regardless
of the signature of the spacetime metric, gµν . To incorporate all the degrees of
freedom, we also need to extend the range of the indices A, B to include the matter
fields, by appropriately defining GΦΦ (x) (and other components if more than one
matter field is present), thereby obtaining the full supermetric.
An inverse DeWitt metric, G (x) = G (ij)(kl) (x), can be defined by the requirement
G (ij)(kl) G(kl)(mn) = 12 δ im δ jn + δ in δ jm ,
(3.3)
AB
which gives
G (ij)(kl) =
3.2. Canonical quantisation
1
2
√
h hik hjl + hil hjk − 2hij hkl .
(3.4)
We will perform canonical quantisation by taking the wavefunction of the universe, Ψ[hij , Φ], to be a functional on superspace. Unlike the case of ordinary
quantum mechanics, the wavefunction, Ψ, does not depend explicitly on time here.
This is related to the fact that general relativity is an “already parametrised” theory
– the original Einstein-Hilbert action is time-reparametrisation invariant. Time is
contained implicitly in the dynamical variables, hij and Φ.
According to Dirac’s quantisation procedure49 we make the following replacements for the canonical momenta
πij → −i
⋆
δ
,
δhij
πΦ → −i
δ
,
δΦ
π0 → −i
δ
,
δN
πi → −i
δ
.
δNi
(3.5)
The use of the terminology “superspace” for the configuration space of quantum cosmology pre-
dates the discovery of supersymmetry, and of course is completely different from the combined manifold of
commuting and anticommuting coordinates which is called “superspace” in supersymmetric theories.
14
and then demand that the wavefunction is annihilated by the operator version of
the constraints. For the primary constraints we have
δΨ
= 0,
δN
δΨ
= 0,
π̂i Ψ = −i
δNi
π̂Ψ = −i
(3.6)
which implies that Ψ is independent of N and N i. The momentum constraint yields
δΨ
i
= 2κ2 T̂ 0̂ı̂ Ψ.
(3.7)
Ĥ Ψ = 0
⇒
i
δhij |j
Using
which
which
nition
(3.7) one can show51 that Ψ is the same for configurations {hij (x), Φ(x)}
are related by a coordinate transformation in the spatial hypersurface, Σ,
accords with the rationale for factoring out by diffeomorphisms in our defiof superspace. Finally, the Hamiltonian constraint yields
"
#
√
2
δ
h
ĤΨ = −4κ2 Gijkl
−3R + 2Λ + 4κ2 T̂ 0̂0̂ Ψ = 0,
(3.8)
+
δhij δhkl 4κ2
where for our scalar field example
T̂ 0̂0̂ =
−1 δ 2
+ 1 hij Φ,i Φ,j +V(Φ).
2h δΦ2 2
(3.9)
Eq. (3.8) is known as the Wheeler-DeWitt equation32,33. In fact, it is not a single
equation but is actually one equation at each point, x ∈ Σ. It is a second-order
hyperbolic functional differential equation on superspace. On account of factorordering ambiguities it is not completely well-defined, although there exist “natural”
choices33,52 of ordering for which the derivative pieces become a Laplacian in the
supermetric.
3.3. Path integral quantisation
An alternative to canonical quantisation which perhaps better accommodates
an intuitive understanding (c.f. Fig. 2) of the quantisation procedure is the path
integral approach. Path integral techniques in quantum gravity were pioneered in
the late 1970s53,54 . The starting point for this is Feynman’s idea that the amplitude
to go from one state with an intrinsic metric, hij , and matter configuration, Φ, on an
initial hypersurface, Σ, to another state with metric, h′ij and matter configuration
Φ′ on an final hypersurface, Σ′ , is given by a functional integral of eiS over all 4geometries, gµν , and matter configurations, φ, which interpolate between the initial
and final configurations
XZ
′
′
′
hhij , Φ , Σ |hij , Φ, Σi =
Dg Dφ eiS[gµν ,φ] .
(3.10)
M
15
In ordinary quantum field theory in flat spacetime the path integral suffers
from the difficulty that since the action S[gµν , φ] is finite the integral oscillates and
therefore fails to converge. Furthermore, the solution which extremises the action is
given by solving a hyperbolic equation between initial and final boundary surfaces –
which is not a mathematically well-posed problem, and may have either no solutions
or an infinite number of solutions. To deal with this problem one performs a “Wick
rotation” to “imaginary time” t → −iτ and considers a path integral formulated
in terms of the Euclidean action, I = −iS. The action is then positive-definite,
so that the path integral is exponentially damped and should converge. Also the
problem of finding the extremum becomes that of solving an elliptic equation with
boundary conditions, and this is well-posed.
One may attempt to apply a similar approach to quantum gravity, replacing S
in (3.10) by the Euclidean action⊞ I[gµν , φ] = −iS[gµν , φ], and taking the sum in
(3.10) to be over all metrics with signature (+ + ++), which induce the appropriate
3-metrics hij and h′ij on the past and future hypersurfaces. This approach to
quantum gravity has had some important successes – most notably, it provides:
(i) an elegant way of deriving the thermodynamic properties of black holes55–57 ;
and (ii) a natural means for discussing the effects of gravitational instantons58–60 .
Gravitational instantons have been found to provide the dominant contribution to
the path integral in processes such as pair creation of charged black holes in a
magnetic field61–63 , and therefore provide a means of calculating the rates of such
processes semiclassically.
The problems associated with the Euclidean approach to quantum gravity are
considerable, however. Firstly, unlike ordinary field theories• such as Yang-Mills
theory the gravitational action is not positive-definite64 , and thus the path integral
does not converge if one restricts the sum to real Euclidean-signature metrics. To
make the path integral converge it is necessary to include complex metrics in the
sum64,65 . However, there is no unique contour to integrate along in superspace66–68
and the result one obtains for the path integral may depend crucially on the contour
that is chosen66 . Furthermore, the measure in (3.10) is ill-defined. It is really only
in the last ten years that mathematicians have succeeded in making path integration
in ordinary quantum field theory rigorously defined. Clearly, we may have to wait
some time before the same can be said of path integrals in quantum gravity.
The physicists’ approach is to set aside the issues involved in making the formalism rigorous and to see what can be learned nevertheless. We thus take the
⊞
Strictly speaking one should call this the Riemannian action, since “Euclidean” spaces are those which
are flat, whereas curved manifolds with (+ + ++) signature are known as Riemannian spaces. However, the
terminology “Euclidean” action which has been taken over from flat space quantum field theory seems to
have stuck, despite the fact that we are of course no longer dealing with R4 .
•
The action for fermi fields in ordinary quantum field theory is not positive-definite, but that is not a
problem since one can treat them as anticommuting quantities so that the path integral over them converges.
16
wavefunction Ψ of the universe on a hypersurface, Σ, with intrinsic 3-metric, hij ,
and matter configuration, Φ, to be defined38,39 by the functional integral
XZ
Ψ [hij , Φ, Σ] =
Dg Dφ e−I[gµν ,φ] .
(3.11)
M
where the sum is over a class of 4-metrics, gµν , and matter configurations, φ, which
take values hij and Φ on the boundary Σ. Alternative definitions of the wavefunction have been proposed. In particular, Linde69 has argued that one should Wick
rotate with the opposite sign, i.e., t → +iτ instead of t → −iτ as above, leading to a factor e+I instead of e−I in (3.11). Alternatively, one could stick with a
Lorentzian path integral70 , with eiS instead of e−I in (3.11). In any case, in order to achieve convergence of the path integral⊖ it is necessary to include complex
manifolds in the sum, which somewhat obscures the distinctions between these alternative proposed definitions of Ψ. The real distinction between the alternative
definitions arises when it comes to imposing boundary conditions, thereby restricting the 4-manifolds included in the sum in (3.11). For example, one could view the
Euclidean sector as being the appropriate sector of the quantum theory in which an
“initial” boundary condition on Ψ should be imposed, which would make (3.11) the
natural starting point, as is the case for the no-boundary proposal38,39. Alternative
boundary conditions would favour the Lorentzian path integral70 .
The path integral definition of the wavefunction (3.11) is consistent with the
earlier definition based on canonical quantisation to the extent that wavefunctions
defined according to (3.11) can be shown to satisfy the Wheeler-DeWitt equation
(3.8) provided that the action, the measure and the class of paths summed over are
invariant under diffeomorphisms71 .
In the canonical quantisation formalism any particular solution to the WheelerDeWitt equation will depend upon the specification of boundary conditions on
the wavefunction. In the path integral formulation the particular solution of the
Wheeler-DeWitt equation that one obtains will similarly depend on the contour of
integration chosen in superspace, and the class of 4-metrics one sums over in (3.11).
Unfortunately it is not known how the choice of contour and class of paths prescribes
the boundary conditions on the wavefunction in the general case, although it can
be answered for specific models. The question of boundary conditions is naturally
of prime importance for cosmology, and we shall return to this question in §4.
3.4. Minisuperspace
In practice to work with the infinite dimensions of the full superspace is not
possible, at least with the techniques that have been developed to date. One useful
⊖
Linde’s suggested modification [69] to (3.11) yields a convergent path integral for the scale factor,
which is all that one needs in the simplest minisuperspace models, but does not lead to convergence if one
includes matter or inhomogeneous modes of the metric.
17
approximation therefore is to truncate the infinite degrees of freedom to a finite
number, thereby obtaining some particular minisuperspace model. An easy way to
achieve this is by considering homogeneous metrics, since as was observed earlier
for each point x ∈ Σ there are a finite number of degrees of freedom in superspace.
The results we shall obtain by this approach will be somewhat satisfying in that
they do appear to have some predictive power. However, the truncation to minisuperspace has not as yet been made part of a rigorous approximation scheme to full
superspace quantum cosmology. As they are currently formulated minisuperspace
models should therefore be viewed as toy models, which we nonetheless hope will
capture some of the essence of quantum cosmology. Since we are simultaneously
setting most of the field modes and their conjugate momenta to zero, which violates
the uncertainty principle, this approach might seem rather ad hoc. However, in classical cosmology homogeneity and isotropy are important simplifying assumptions
which do have a sound observational basis. Therefore it is not entirely unreasonable to hope that a consistent truncation to particular minisuperspace models with
particular symmetries might be found in future♣ .
Let us thus consider homogeneous cosmologies for simplicity. Instead of having
a separate Wheeler-DeWitt equation for each point of the spatial hypersurface, Σ,
we then simply have a single Wheeler-DeWitt equation for all of Σ. The standard
FRW metrics, with
dr 2
2
i
j
2
2
2
2
hij dx dx = a (t)
,
k = −1, 0, 1,
(3.12)
+ r dθ + sin θdϕ
1 − kr 2
are of course one special example. In that case our model with a single scalar field
simply has two minisuperspace coordinates, {a, Φ}, the cosmic scale factor and the
scalar field. Many more general homogeneous (but anisotropic) models can also
be considered. Indeed all such models can be classified⊳ and apart from the FRW
models the other cases of interest are: (i) the Kantowski-Sachs models77,78 , which
have a 3-metric
hij dxi dxj = a2 (t)dr 2 + b2 (t) dθ 2 + sin2 θdϕ2 ,
(3.13)
and thus three minisuperspace coordinates, {a, b, Φ}; and (ii) the Bianchi models.
The Bianchi models are the most general homogeneous cosmologies with a 3dimensional group of isometries. These groups are in a one–to–one correspondence
with 3-dimensional Lie algebras, which were classified long ago by Bianchi79 . There
are nine distinct 3-dimensional Lie algebras, and consequently nine types of Bianchi
cosmology. The 3-metric of each of these models can be written in the form
hij dxi dxj = hij (t)ωi ⊗ ωj ,
♣
⊳
For some discussions of the validity of the minisuperspace approximation see [72–75].
See [76] for a review.
18
(3.14)
where ωi are the invariant 1-forms associated associated with the isometry group.
The simplest example is Bianchi I, which corresponds to the Lie Group R3 . In that
case we can choose ω1 = dx, ω2 = dy, and ω3 = dz, so that
hij dxi dxj = a2 (t)dx2 + b2 (t)dy 2 + c2 (t)dz 2 ,
(3.15)
and the minisuperspace coordinates are {a, b, c, Φ}. If we take the spatial directions
to be compact such a universe will have toroidal topology. In the special case that
a(t) = b(t) = c(t) we retrieve the spatially flat (k = 0) FRW universe.
The most complicated, and possibly the most interesting, Bianchi model is
Bianchi IX, associated to the Lie group SO(3, R). In this case the invariant 1-forms
may be written as
ω1 = − sin ψ dθ + sin θ cos ψ dϕ,
(3.16)
ω2 = cos ψ dθ + sin θ sin ψ dϕ,
3
ω = cos θ dϕ + dψ,
in terms of the Euler angles, (ψ, θ, ϕ), on the 3-sphere, S 3 . The spatial sections of
the geometry resulting from (3.14), (3.16) have the topology of S 3 , but are only
spherically symmetric in the special case that h11 (t) = h12 (t) = . . . h33 (t), which
corresponds to the k = +1 FRW universe. Geometrically the spatial hypersurfaces
can thus be thought of as squashed, twisted 3-spheres [see Fig. 3]. Bianchi IX has
played an important role in classical cosmological studies of the initial singularity –
it is the basis of the so-called “mixmaster universe”80,81 . As a classical dynamical
system Bianchi IX is extremely interesting because it appears to be chaotic, but
only just on the verge of being so. Over the years there has been much debate as
to whether Bianchi IX is or is not chaotic, and this seems to have been recently
resolved by an explicit demonstration that it is not integrable82 . The corresponding minisuperspace model will have six independent coordinates in addition to the
scalar field coordinate.
Fig. 3: Schematic geometry of spatial hypersurfaces in the Bianchi IX universe.
Technically speaking, what has been proved is the failure of integrability in the Painlevé sense. While
this does not guarantee the existence of chaotic regimes, it does make their existence extremely probable.
19
Let us now consider the minisuperspace corresponding to an arbitrary homogeneous cosmology. We will assume that the minisuperspace is of dimension n, and
A
will denote the minisuperspace coordinates by {q }. Since N i = 0 by assumption,
it follows from the definitions (2.5) and (3.4) that
√
G ijkl ḣij ḣkl = 4 hN 2 Kij K ij − K 2 ,
and consequently the Lorentzian action (2.12) now takes the form
Z
1
A B
S = dt
G (q)q̇ q̇ − N U(q) ,
2N AB
where
Z
√
1 ijkl
A
B
3
GAB dq dq = d x
G δhij δhkl + hδΦδΦ ,
8κ2
is the minisupermetric, which is now of finite dimension, n, and
Z
√
1
3
3
U= d x h
− R + 2Λ + V(Φ) .
4κ2
(3.17)
(3.18)
(3.19)
(3.20)
The action (3.18) is simply equivalent to that for a “point particle” moving in a
A
“potential” U. Variation of (3.18) with respect to q thus yields a geodesic equation
with force term,
1 d
N dt
A
q̇
N
+
1 A
B C
AB ∂U
Γ BC q̇ q̇ = −G
,
2
N
∂q B
(3.21)
A
where Γ BC are the Christoffel symbols determined from the minisupermetric, while
variation of (3.18) with respect to N yields the Hamiltonian constraint
1
A B
GAB (q)q̇ q̇ + U(q) = 0.
2
2N
(3.22)
The general solution to (3.21), (3.22) will have 2n − 1 independent parameters,
one of which is always trivial in the sense that it corresponds to a choice of origin
of the time parameter. In studying any particular minisuperspace model we must
take care to check that what we have done above is consistent, as it does not always
follow that substituting a particular ansatz into an action before varying it will yield
the same result as substituting the same ansatz into the field equations obtained
from variation of the original action. Eqs. (3.21) and (3.22) should correspond
respectively to the (ij) and (00) components of the original Einstein equations,
while the (0i) equation is trivially satisfied in the present case.
Quantisation is greatly simplified because now that our configuration space
is finite-dimensional we are effectively dealing with the quantum mechanics of a
20
constrained system. The canonical momenta and Hamiltonian are respectively given
by
B
GAB q̇
∂L
,
A =
∂ q̇
N
h
i
A
AB
H = πA q̇ − L =N 21 G πA πB + U(q) ≡ N H.
πA =
(3.23)
(3.24)
The πA are related to the canonical momenta (2.13)–(2.16) defined earlier by integration over the 3-volume of the hypersurfaces of homogeneity in (3.19). In terms
of the new variables the action (3.18) and Hamiltonian constraint (3.22) are respectively
Z
h
i
A
S = dt πA q̇ − N H ,
(3.25)
1 AB
G πA πB
2
+ U(q) = 0.
Under canonical quantisation (3.26) yields the Wheeler-DeWitt equation
ĤΨ = − 12 ∇2 + U(q) Ψ = 0,
where
i
h√
1
AB
∇ ≡√
−G G ∂B
∂A
−G
(3.26)
(3.27)
(3.28)
is the Laplacian operator of the minisupermetric. In arriving at (3.27) we have made
an explicit “natural choice” of factor ordering33,52 in order to accommodate the
factor-ordering ambiguity. This choice is favoured by independent minisuperspace
calculations of the prefactor, using zeta function regularisation and a scale invariant
measure, which can then be related to factor ordering dependent terms through a
semiclassical expansion of the Wheeler-DeWitt equation83 .
An alternative “natural choice”35,84–86 of factor ordering would yield a conformally invariant Wheeler-DeWitt equation,
n−2
1 2
ĤΨ = − 2 ∇ +
R + U(q) Ψ = 0,
(3.29)
8(n − 1)
where R is the scalar curvature obtained from the minisupermetric.
3.5. The WKB approximation
In view of the difficulties associated with solving the Wheeler-DeWitt equation
in general, the best we can realistically hope for in many minisuperspace models is
to look for appropriate approximate solutions in the semiclassical limit≖, in which
X
X
Ψ≃
Ψn ≡
An e−In ,
(3.30)
n
n
≖
The semiclassical limit of the full superspace Wheeler-DeWitt equation has been treated more formally
by a number of authors. See, e.g., [87], [88] and references therein.
21
where the sum is over saddle points of the path integral, the An being appropriate
(possibly complex) prefactors. In general we might expect to find regions in which
the wavefunction is exponential, Ψ ≃ e−I , and regions in which it is oscillatory,
Ψ ≃ eiS . The latter could be viewed as the wavefunction of a universe in the
classical “Lorentzian” or “oscillatory” region, while the former would correspond
to a universe in a classically inaccessible “Euclidean” or “tunneling” region. As
has already been mentioned, the sum in (3.30) will in general contain a number of
saddle points with an action, In , which is neither purely real nor purely imaginary.
Our own universe is of course Lorentzian at late times, and therefore the only
minisuperspace models which can be of direct physical relevance are those for which
the Wheeler-DeWitt equation does possess approximate solutions of the oscillatory
type. Approximate solutions of this type can be obtained by performing a WKB
expansion, for which purpose it is necessary to restore ~ in the minisuperspace
Wheeler-DeWitt equation (3.27). If we assume that each component Ψn satisfies
(3.27) separately, then
0 = ĤΨn = − 21 ~2 ∇2 + U An e−In /~
i
o
nh
2
2
2
−In /~
1
1
− 2 (∇In ) + U An + ~ ∇In · ∇An + 2 An ∇ In + O(~ ) ,
=e
(3.31)
AB
0
where the dot implies contraction with the minisupermetric G . The O(~ ) and
O(~) terms give two equations for In and An . If we decompose In into real and
imaginary parts according to In = Rn − iSn then the real and imaginary parts of
the O(~0 ) term in (3.31) give
2
− 12 (∇Rn )2 + 21 (∇Sn ) + U = 0,
∇Rn · ∇Sn = 0.
(3.32)
(3.33)
Provided that the imaginary part of the action varies much more rapidly than the
real part, i.e., (∇Rn )2 ≪ (∇Sn )2 , then (3.32) is the Lorentzian Hamilton-Jacobi
equation for Sn :
1 AB ∂Sn ∂Sn
+ U(q) = 0.
(3.34)
G
2
∂q A ∂q B
Comparison of (3.34) with (3.27) suggests a strong correlation between coordinates
and momenta, and invites the identification
πA =
∂Sn
.
∂q A
(3.35)
C
If we differentiate (3.34) w.r.t. q we obtain
1 AB
,C
2G
2
∂Sn ∂Sn
∂U
AB ∂Sn ∂ Sn
= 0.
A
B + G
A
B
C +
∂q ∂q
∂q ∂q ∂q
∂q C
22
(3.36)
If we define a minisuperspace vector field
d
AB ∂Sn ∂
≡G
,
(3.37)
ds
∂q A ∂q B
then combining (3.35), (3.36) and (3.37) we obtain
dπC 1 AB
∂U
+ 2 G ,C πA πB + C = 0,
(3.38)
ds
∂q
which after raising indices is the same geodesic equation (3.21) obtained earlier
provided we identify the parameter, s, with the proper time on the geodesics.
We can now solve the equation given by the O(~) term of (3.31). Since |∇Rn | ≪
|∇Sn | it follows that the terms involving Rn can be neglected, and thus
dAn
AB ∂Sn ∂An
∇Sn · ∇An ≡ G
= − 12 An ∇2 Sn,
(3.39)
A
B ≡
∂q ∂q
ds
which may be readily integrated. We thus obtain a first-order WKB wavefunction
Z
2
1
(3.40)
Ψn = Cn exp iSn − 2 ds ∇ Sn ,
where Cn is an arbitrary (complex) constant to be appropriately normalised, and
we have reverted to natural units in which ~ = 1.
The wavefunction (3.40) could be considered to be the analogue of the wavefunction for coherent states in ordinary quantum mechanics,
−(x − x̄n (t))2
ipn x
,
(3.41)
ψn (x, t) = cne
exp
σ2
which describes a wave packet which is “peaked” about a classical particle trajectory, x̄n (t), and which thus roughly “predicts” classical behaviour.
This becomes
P
problematic, however, if we consider a superposition, ψ = n ψn , of such states
since interference between different wave packets will in general destroy the classical
behaviour. In order to interpret the total wavefunction as saying that the particle
follows a roughly classical trajectory, x̄n (t), with probability |cn|2 , it is necessary
that a decoherence mechanism should exist which renders this quantum mechanical
interference negligible89 .
The issue of quantum decoherence is clearly also of great importance in quantum
cosmology, since in order to interpret Ψ in (3.30) in a similar fashion a similar mechanism must exist. It has been recently argued, furthermore, that decoherence is a
necessary feature of the WKB interpretation of quantum cosmology, since without
decoherence the existence of chaotic cosmological solutions would lead to a breakdown of the WKB approximation90. This is analogous to similar problems with
the commutativity of the limits t → ∞ and ~ → 0 in ordinary quantum mechanics
when applied to chaotic systems.
The issues involved in decoherence pose complex conceptual questions for the
fundamentals of quantum mechanics itself, quite apart from the problems specific
to quantum cosmology⊠ . Here we will merely assume that such a mechanism exists,
⊠
For further details see, e.g., [91]–[94] and references therein.
23
and we will take the view that Ψ can be considered to “predict” a classical spacetime
if there exist WKB-type solutions (3.40), which yield a strong correlation between
A
A
π and q according to (3.35). The sense in which the minisuperspace positions
and momenta are “strongly correlated” can be made more precise through the use
of quantum distribution functions, such as the Wigner function91,95 . By use of the
Wigner function one may show95 that wavefunctions of the oscillatory type, Ψ ∼ eiS ,
predict a strong correlation between coordinates and momenta, whereas wavefunctions of the type Ψ ∼ e−I , which are also typical minisuperspace solutions, do not.
Such exponential wavefunctions can thus be considered as describing universes in
a purely quantum “tunneling” regime, before the quantum to classical transition.
We will interpret wavefunctions, Ψ ∼ eiS , as corresponding to classical spacetime, or
rather a set of classical spacetimes as S is a first integral of the equations of motion.
3.6. Probability measures
Given a solution, Ψ, to the Wheeler-DeWitt equation it is necessary to construct a probability measure in order to make predictions. One central question in
quantum cosmology is how one should construct such a measure.
The minisuperspace Wheeler-DeWitt equation (3.27) is a second-order equation
very much like the Klein-Gordon equation in ordinary field theory, and it readily
follows from (3.27) that the current33
J = − 21 i Ψ∇Ψ − Ψ∇Ψ
(3.42)
is conserved: ∇ · J = 0. The similarity to the Klein-Gordon current extends to
the fact that the natural inner product33 constructed from J is not positive-definite
and so gives rise to negative probabilities. In quantum field theory this is not a
problem since one can split the wavefunction up into positive and negative frequency components which correspond to particles and anti-particles. However, as
has already been mentioned there is no well-defined notion of positive frequencies in
superspace on account of its lack of symmetries96 . A further problem is that many
natural wavefunctions would have zero norm with this definition. For example, the
no-boundary wavefunction is real and gives J = 0.
The similarity of Ψ to the Klein-Gordon field has suggested to many people that
one should turn Ψ into an operator, Ψ̂, thereby introducing quantum field theory on
superspace, or “third quantisation”. One then arrives at operators which create and
annihilate universes. However, as we do not perform measurements over a statistical
ensemble of universes it is not clear how we can arrive at sensible probabilities using
such a formulation.
The difficulties with the Klein-Gordon current of course led Dirac to introduce
the Dirac equation, and it is worth mentioning that a similar resolution of the
problem is available in supersymmetric quantum cosmology. In particular, one can
24
go to a theory which includes fermionic variables by considering quantum cosmology based on supergravity◭ rather than the purely bosonic Einstein theory. The
constraints of supergravity, which may be viewed as the Dirac square root of the
constraints of general relativity98,99, are reducible to first-order equations. Furthermore, this also translates into simplifications in homogeneous minisuperspace
models – the appropriate constraint equation which determines the quantum evolution of the wavefunction can be considered to be the Dirac square root of the
Wheeler-DeWitt equation100–103 . As a result it is possible to construct100 a Diractype probability density which is conserved by the equation QΨ = 0, where Q is
the supercharge.
Another alternative to the question of the probability measure is to use |Ψ|2
directly as a probability measure52,104 , by defining the probability of the universe
being in a region, A, of superspace by
Z
P(A) ∝
|Ψ|2 ∗ 1
(3.43)
A
∗
where 1 is the volume-element on superspace, ∗ being the Hodge dual in the
supermetric. This definition of a necessarily positive-definite probability density
works very well for homogeneous minisuperspaces, for which the volume form ∗ 1 is
independent of x ∈ Σ. This is perhaps not surprising since as was observed in §3.4
the assumption of homogeneity reduces the problem to one of quantum mechanics,
and |Ψ|2 is of course the probability density in conventional quantum mechanics.
Problems with the definition (3.43) do arise since even in some simple examples
the wavefunction is not normalisable, but instead hΨ|Ψi = ∞. One further problem
is that whereas in ordinary quantum mechanics |Ψ|2 describes a probability density
in configuration space – i.e., the space of particle positions – in quantum cosmology
the configuration space is (mini)superspace and time is implicitly contained in the
(mini)superspace coordinates. These coordinates cannot therefore be thought of as
the mere analogues of particle positions. As a result the recovery of the conservation
of probability and the standard interpretation of the quantum mechanics for small
subsystems is not necessarily straightforward in the approach based on (3.43), and
may involve understanding some subtle questions about the role of time in quantum
gravity.
Ultimately a formulation such as (3.43) which is based on absolute probabilities may not be required since it is impossible to measure statistical ensembles of
universes and thus all we can really test are conditional probabilities rather than
absolute probabilities. For example, a relevant testable probability might be the
probability, P(A|B), of finding Ψ in a region A of superspace given that Ψ started
in another region B of superspace. Page2,105 has explored the construction of conditional probabilities in quantum cosmology without the use of absolute probabilities.
◭
In fact, a naı̈ve first-order Hamiltonian formulation for the minisuperspaces of homogeneous cosmolo-
gies was found early on [97], but until the development of supergravity there was no natural interpretation
of the Dirac-type constraint equation obtained.
25
Clearly the issues surrounding the choice of probability measure involve some
deep conceptual problems which may perhaps get to the heart of the broader conceptual basis of quantum gravity. Such issues have been discussed by a number
of authors in the context of quantum cosmology2,43,104–107 and I will not address
them here in any detail.
For the purposes of examining how we might hope to make predictions from
the proposed boundary conditions of §4.1,4.2 we shall merely consider quantum
cosmology in the WKB limit. In this limit it follows from (3.40) that each of the
components Ψn in (3.30) has a conserved Klein-Gordon-type current (3.42) given
by
Jn ≃ |An |2 ∇Sn ,
(3.44)
which flows very nearly along the direction of the classical trajectories. The current
A
conservation law ∇A Jn = 0 implies that
A
dP = Jn dΣA ,
(3.45)
is a conserved probability measure on the set of trajectories with tangent ∇Sn ,
where dΣA is the element of a hypersurface, Σ, in minisuperspace which cuts across
the flow and intersects each curve in the congruence once and only once. One
finds that for a pencil of trajectories near the classical trajectory the probability
density (3.45) is positive-definite. Vilenkin has argued104 that positive-definiteness
of the probability measure is really only required in the semiclassical limit, as this
is the only limit in which we obtain a universe accessible to observation where the
conventional laws of physics apply. Therefore given that the current (3.42) works
in the WKB limit, this is all that is needed if we are content that “probability”
and “unitarity” are only approximate concepts in quantum gravity. One can also
show2,52 that the manifestly positive definition (3.43) yields essentially the same
result as (3.45) in the WKB limit.
3.7. Minisuperspace for the Friedmann universe with massive scalar field
Let us now apply our results to the particularly simple case of a homogeneous,
isotropic universe with a single scalar field, with a potential which allows for inflationary behaviour. A quadratic potential is possibly the simplest example with this
property, and thus has been much studied in quantum cosmology.
For convenience we will introduce a numerical normalisation factor σ 2 =
κ /(3V) into the metric, where V is the 3-volume of the unit hypersurface – e.g.,
V = 2π 2 for the 3-sphere, k = +1. We are considering closed universes only, which
requires making topological identifications for the k = 0, −1 cases, so that V remains
finite. In place of (2.1), (2.2) and (3.12) we then have a metric
dr 2
2
2 2
2
2
2
2
2
2
ds = σ −N dt + a (t)
.
(3.46)
+ r dθ + sin θdϕ
1 − kr 2
2
26
We also take the scalar field action to be normalised by
Z
V (φ)
1 µν
3
4 √
.
Smatter = 2 d x −g − g ∂µ φ∂ν φ −
κ
2
2σ 2
(3.47)
M
√
Thus N → σN , a → σa, Φ = 3φ/κ and V = 3V /(2κ2 σ 2 ) relative to our earlier
definitions, and a, φ and V are now dimensionless. As for the general homogeneous
minisuperspace discussed in the previous section, we use a gauge in which Ni = 0.
ȧ
From (3.46) it follows that 3R = σ6k
2 a2 and Kij = − σN a hij , and thus the action
takes the form (3.18) with a minisupermetric
A
B
GAB dq dq = −ada2 + a3 dφ2 ,
and potential
U=
(3.48)
a3 V (φ) − ka .
1
2
(3.49)
Alternatively, it is sometimes useful to express (3.48) in terms of a conformal gauge
A
B
GAB dq dq = e3α −dα2 + dφ2 ,
(3.50)
= −(4uv)−1/4 dudv,
(3.51)
where α = ln a, or alternatively in null coordinates
u = 12 e2(α−φ) = 12 a2 e−2φ ,
(3.52)
v = 21 e2(α+φ) = 12 a2 e2φ .
The canonical momenta are given by
π0 = 0,
πa =
−aȧ
,
N
πφ =
and the classical equations of motion yield
aȧ2 a3 φ̇2
− ka + a3 V
H = 21 − 2 +
N
N2
2aφ̇2
ȧ
1 d
+
− aV = 0,
N dt N
N2
!
1 d
φ̇
ȧφ̇
dV
+3
+ 12
= 0,
2
N dt N
aN
dφ
a3 φ̇
,
N
!
= 0,
(3.53)
(3.54)
(3.55)
(3.56)
which are equivalent to the geodesic equation (3.21) with metric (3.48) and potential
(3.49). The lapse function is not of physical relevance classically since we can choose
an alternate proper time parameter, dτ = N dt, or equivalently choose a gauge
N = 1 in (3.54)–(3.56) so that t is the proper time. One of these equations depends
27
on the other two by virtue of the Bianchi identity, as is always the case in general
relativity. We thus effectively have two independent differential equations in two
unknowns, one first order and one second order, or equivalently an autonomous
system of three first order differential equations. The solution therefore depends on
three free parameters, but as mentioned above one of these amounts to a choice of
the origin of time which is of no physical importance. Thus there is a two-parameter
family of physically distinct solutions.
The classical solutions cannot be written in a simple closed form except for
certain special values of k and the potential V (φ), which unfortunately does not
even include the quadratic potential⊲ V (φ) = m2 φ2 . The qualitative property of
the solutions may nonetheless be determined by studying the 3-dimensional phase
space. Instead of choosing a particular potential, however, let us suppose that we
are in a region of the phase space for which V can be approximated by a constant.
Such conditions are more or less met in the “slow-rolling approximation”109 of inflationary cosmology, in which |V ′ /V | ≪ 6 and |V ′′ /V | ≪ 9. In this approximation
the dynamics is described by setting φ̇ ≃ 0 in (3.54) and (3.55), and setting φ̈ ≃ 0
in (3.56). In approximating the nearly flat region of the potential V by a cosmological constant we ignore the slow change of the scalar field determined from (3.56).
We thus obtain a simplified model which possesses classical inflationary solutions,
provided the constant V is chosen to be positive. This will serve as a useful test
model for quantum cosmology.
In the case that V is constant, eq. (3.56) integrates to give φ̇ = Ca−3 , where
C is an arbitrary constant, and it therefore follows that the Friedmann equation
(3.54) can be written in terms of an elliptic integral in a2 (η),
Z a2
dz
1
√
η=2
,
(3.57)
V z 3 − kz 2 + C 2
0
where η is the conformal time parameter defined by dη = a−1 dt. It is thus possible
to express the general solution in terms of elliptic functions. Since the properties
of such functions are not very transparent perhaps, we can alternatively plot the
2-dimensional phase space – e.g., in terms of the variables φ̇ and α̇, as in Fig. 4 in
order to understandn
the qualitative
features of the
osolutions. There are four critical
√
√
points, at (φ̇, α̇) =
0, ±1/ V , ±1/ 2V , 0 , the first two being nodes, A± ,
which are endpoints for all values of the spatial curvature, k, and the latter two
saddle points, B± , for the k = +1 solutions.
The general solution for the spatially flat case (k = 0), which corresponds to
the bold separatrices shown in Fig. 4, is given by110
1/3
√ 1/3
√
a = Ca sinh 23 V t
cosh 23 V t
,
(3.58)
√
1
3
φ = 3 ln tanh 2 V t + Cφ ,
⊲
The Einstein equations for the FRW universe with a massive scalar field can be solved approximately
[108] in various limits, however.
28
Fig. 4: The 2-dimensional phase plot of α̇ = ȧ/a versus φ̇ for the simplified model
with constant potential, V .
in closed form, where Ca and Cφ are arbitrary constants.
√ At late times the solutions
(3.58) have an exponential scale factor, a → Ca exp( V t) as t → ∞, and constant
scalar field φ → Cφ . Furthermore, one can see from Fig. 4 that all the k = −1
solutions in the upper half-plane, and a number of the k = +1 solutions are also
attracted to the point A+ with a similar inflationary behaviour at late times. (The
corresponding point A− on the k = 0 separatrix corresponds to the time-reversed
solution, with an inflationary phase as t → −∞.) The simplified universe corresponding to Fig. 4 of course is far from being the complete picture, as the model
does not allow for any exit from inflation. However, Fig. 4 illustrates the typical
situation that a given model will possess regimes with inflationary behaviour and
regimes with non-inflationary behaviour. In Fig. 4 the k = +1 solutions to the right
and left of the separatrices that pass through B± fall into the latter category, for
example. The situation becomes even more involved when one considers the full
3-dimensional phase space for some particular potential V (φ).
The case of the k = +1 solutions in Fig. 4 illustrates the general feature that
classical dynamics are highly dependent on initial conditions. In order to obtain
a sufficiently long inflationary epoch to overcome the problems mentioned in the
Introduction, (of order 65 e-folds growth in the scale factor), the initial values of φ
29
and φ̇ must be restricted to a particular region of the phase space. In particular,
φ̇ must be small initially. Classically, there is no a priori reason for one choice of
initial conditions over any other choice, unless further ingredients are added. The
degree to which inflationary initial conditions are preferred relative to other initial
conditions – i.e., how probable is inflation? – is precisely the sort of question that
we might therefore hope quantum cosmology could answer.
It is possible to attempt to solve this question without resorting to quantum cosmology. To do this one must construct a measure on the set of all universes111,112 ,
and then compare the number of inflationary solutions with a sufficiently long exponential phase to the number of other solutions. Preliminary results112 seemed to
indicate that almost all models with a massive scalar field undergo a period of inflation. However, a more careful analysis108 revealed that the answer is ambiguous, as
both the set of inflationary solutions and the set of non-inflationary solutions have
infinite measure.
Alternatively, if as we expect the universe began in some sort of tunneling
process or similar transition from a quantum regime, then we could expect the
“initial” classical parameters to be determined, at least in a probabilistic fashion,
from more fundamental quantum processes. The question of the most probable
state of the universe is then pushed back a level and becomes: “what is a typical
wavefunction for the universe?”
In the context of the present minisuperspace model, therefore, we can proceed
by quantising the Wheeler-DeWitt equation (3.54), to obtain
ĤΨ =
− 21 ∇2
∂ ∂
∂2
3
a a
− ka + a V (φ) Ψ
−
+U Ψ=
∂a ∂a ∂φ2
2
∂2
4α
6α
1 −3α ∂
= 2e
−
− ke + e V (φ) Ψ
∂α2 ∂φ2
2
∂
k
1/4
1/2
2
= (4uv)
− + (uv) V Ψ
∂u∂v 2
=0
1
2
1
a3
(3.59)
in terms of the various sets of coordinates given earlier. In general boundary conditions will have to specified in order to solve (3.59). However, we can consider the
approximate form of the WKB solutions without considering boundary conditions
for the time being.
We will confine ourselves to regions in which the potential can be approximated
by a cosmological constant, as in the analysis of Fig. 4, so that we can drop the
term involving derivatives with respect to φ in (3.59), thereby obtaining a simple
1-dimensional problem which is amenable to a standard WKB analysis. The first
30
order WKB wavefunction (3.40) which solves (3.59) in this approximation is
Ψ(a, φ) ≃
a (a2 V
a (k −
B(φ)
1/4
(φ) − k)
C(φ)
a2 V
1/4
(φ))
3/2
±i
2
exp
a V (φ) − k
,
3V (φ)
3/2
±1
2
exp
k − a V (φ)
,
3V (φ)
a2 V > k, (3.60)
a2 V < k.
(3.61)
If V is positive, as was assumed above, then oscillatory type solutions will thus exist
for large values of the scale factor, while the exponential type solutions will exist
for small values of the scale factor if k = +1.
The oscillatory solutions are of the form Ψ ∼ eiS (neglecting the prefactor),
where S satisfies the Hamilton-Jacobi equation (3.34). Comparing this to the Hamiltonian constraint (3.26) we find a strong correlation (3.35) between
√ momenta and
1 3
2
coordinates. For large scale factors, a V ≫ |k|, so that S ≃ ± 3 a V . In this limit
(3.23), (3.35) and (3.50) thus yield
∂S
∂α
∂S
πφ =
∂φ
πα =
⇒
√
α̇ ≃ ± V ,
⇒
φ̇ ≃ 0,
(3.62)
which correspond in fact to the inflationary points, A± , of Fig. 4. The oscillatory
wavefunction thus “picks out” classical inflationary universes.
Since the minisupermetric (3.50) is conformal to 2-dimensional Minkowski space
in the coordinates (α, φ), it is convenient to represent it by a Carter-Penrose conformal diagram (see Fig. 5). In each case we plot (p − q) horizontally and (p + q)
vertically, where tan p = α + φ, and tan q = α − φ. The boundary consists of
points corresponding to past timelike infinity, i− = {(a, φ) | a = 0, φ finite }, future timelike infinity, i+ = {(a, φ) | a = ∞, φ finite }, left and right spacelike infinity, i0L,R = {(a, φ) | a = finite , φ = ±∞}; and past and future null boundaries,
I−
= {(a, φ) | a = 0, φ = ±∞} and I+
= {(a, φ) | a = 0, φ = ±∞}. In each case
L,R
L,R
the subscript L (left) is associated with φ → −∞, and the subscript R (right)
with φ → +∞. The approximate region for which oscillatory WKB solutions exist
is shown in Fig. 5(a,b) for the approximate minisuperspace with a cosmological
constant, in Fig. 5(c,d) for V (φ) = m2 φ2 , and in Fig. 5(e,f) for potentials, V (φ),
typically found in higher-derivative gravity theories and in string theory with supersymmetry breaking.
Naturally it is of interest to know whether the inflationary WKB wavefunctions
are typical solutions to the Wheeler-DeWitt equation. To determine a typical wavefunction for the universe, we need to make a choice of boundary conditions for Ψ
when solving (3.59).
31
Fig. 5: Conformal diagrams of the 2-dimensional minisuperspace. The region where
oscillatory WKB solutions exist, as given by the rough criterion a2 V > 1, is shaded
for various potentials: (a) V = 0.25 (const); (b) V = 4 (const); (c) V = 0.25φ2 ;
2
(d) V = 25φ2 ; (e) V = 1 − e−φ/f with f = 1.5; (f) V = 4 sinh2 φ exp −f e−2φ
with f = 0.1.
32
4. Boundary Conditions
The specification of boundary conditions for the Wheeler-DeWitt equation may
seem a disappointment, as it might appear that we are just replacing an arbitrary
initial choice of parameters which describe the classical evolution of the universe
by an arbitrary initial choice of parameters which describe its quantum evolution.
However, if quantum mechanics is a universal theory then it must have applied at
the earliest epochs of the existence of the universe, in which case it is natural that
the quantum dynamics precedes the classical dynamics. This justifies a quantum
boundary condition for the universe as being more fundamental than a classical one.
In any case, the only alternative to choosing quantum boundary conditions would
be that mathematical consistency might be enough to guarantee a unique solution
to the Wheeler-DeWitt equation, as DeWitt originally hoped33 . If the experience
gained from the study of minisuperspace models translates to superspace, then this
would not appear to be the case, however.
The question naturally arises as to whether there should be some natural boundary condition, which once and for all determines the quantum evolution of the universe at early times, or alternatively whether the nature of the quantum dynamics
might be somewhat indifferent to such choices. Deep conceptual problems are involved in trying to make headway with this question. Unlike other situations in
quantum physics, where boundary conditions are readily specified by the symmetry of particular problems, such as spherical symmetry in the case of the hydrogen
atom, the origin of the universe poses a situation in which all intuition must be
abandoned and we can at best proceed on aesthetic grounds alone.
Having made a choice of boundary condition, we can of course solve the WheelerDeWitt equation and study the physical consequences for the evolution of the universe. However, without some additional principle we should by rights study many
different boundary conditions before we can begin to have any confidence about the
predictions made. To arrive at a principle which would circumvent this problem is
an immense challenge: it would more or less amount to an additional law of physics
which must be appended to the others which describe the quantum evolution of the
universe. The situation might be considered to be the same as trying to describe
the phase transition from gas to liquid if all the physical phenomena that we knew
about related to the gaseous phase only. The universe appears to have undergone
a phase transition when it was formed, but the only experience we have available
involves the “after” state of the universe alone.
Progress can of course only be made by attempting to define natural boundary
conditions for the wavefunction of the universe, and examining the consequences.
This became an important activity in the 1980s. I will only discuss the two most
studied boundary condition proposals, the “no-boundary proposal” and the “tunneling proposal”. However, other proposals have been put forward, including the
“all possible boundaries proposal” of Suen and Young113 and the “symmetric initial
condition” of Conradi and Zeh114,115 .
33
4.1. The no-boundary proposal
The proposal of Hartle and Hawking38,39 is that one should restrict the sum in
the definition of the wavefunction of the universe (3.11) to include only compact
Euclidean 4-manifolds, M, for which the spatial hypersurface Σ on which Ψ is
defined forms the only boundary, and only matter configurations which are regular
on these geometries. The universe then has no singular boundary to the past, as is
the case for the standard FRW cosmology. The sum (3.11) thus includes manifolds
such as those shown in Fig. 6, but not those shown in Fig. 2. As Hawking39 puts
it: the boundary conditions of the universe are that it has no boundary.
Σ
Σ
Σ
+
+
+ . . .
Fig. 6: Geometries allowed by the Hartle-Hawking no-boundary proposal.
Intuitively, what Hartle and Hawking had in mind in formulating this proposal
was to get rid of the initial singularity by “smoothing the geometry
of the universe off
√
in imaginary time”. For example, whereas a surface with h = 0 would be singular
in a Lorentzian signature metric, this is not necessarily the case if the metric is
of Euclidean signature, as can be seen from the example of S 4 shown in Fig. 7.
Ideally, the no-boundary√proposal should tell us what initial conditions to set when
we take manifolds with h → 0 or any similar limit consistent with the proposal.
If the limit is taken at an initial time τ = 0, the no-boundary proposal would
lead to conditions on hij (x, 0), φ(x, 0) and their derivatives. In practice, quantum
cosmology is rarely studied beyond the semiclassical approximation, in which Ψ ≃
Ae−Icl , where Icl is the classical (possibly complex) action evaluated along the
solution to the Euclidean field equations. In the semiclassical approximation one
therefore works only with boundary conditions on the metric and matter fields
which correspond to the no-boundary proposal at the classical level. In particular,
we demand: (i) that the 4-geometry is closed; and (ii) that the saddle points of the
functional integral correspond to regular solutions of the classical field equations
which match the prescribed initial data on Σ.
One question which is not explained by the no-boundary proposal is the choice
of a contour of integration for the path integral. As was mentioned earlier, the path
integral over real Euclidean metrics does not converge, and thus it is necessary to
include complex metrics to make the path integral converge. Such metrics will generally include ones which are neither truly Euclidean nor truly Lorentzian, and thus
34
x5
x5
(a)
(b)
R
Fig. 7: Slicing a 4-sphere of radius R embedded in flat 5-dimensional space:
(a) a surface x5 = a < R intersects the 4-sphere in a 3-sphere of non-zero radius;
(b) when x5 = R the 3-sphere shrinks to zero radius but there is no singularity of
the 4-geometry.
the naı̈ve picture of a compact Euclidean geometry sewn onto a Lorentzian one,
which is suggested by “smoothing the geometry of the universe off in imaginary
time”, is not completely accurate. In general, one might expect a truly complex
metric to interpolate the Euclidean and Lorentzian ones, and in general the initial geometry might be only approximately Euclidean and the final geometry only
approximately Lorentzian116 . Unfortunately the criteria for achieving convergence
of the path integral do not single out a unique contour of integration, and the noboundary proposal does not appear to offer any further clues as to how the contour
should be chosen117 .
Σ
Σ
(a)
(b)
Fig. 8: Euclidean solutions which correspond to matching a given 3-sphere hypersurface, Σ, to a 4-sphere which is: (a) less than half filled; (b) more than half
filled.
Non-uniqueness of the contour of integration is already a problem in the simplest
conceivable non-trivial minisuperspace model, namely the k = +1 FRW universe
with a cosmological term and no other matter – the “de Sitter minisuperspace”
model. At the semiclassical level one can calculate e−Icl by the steepest descents
method. Hartle and Hawking discussed this in their original “no boundary” paper38 ,
and argued heuristically that one particular saddle point would yield the dominant
35
contribution to the path integral, namely the saddle point corresponding to the
classical Euclidean solution which matches the S 3 hypersurface Σ to a less than
half filled 4-sphere, rather than the solution which matches S 3 to a more than
half filled 4-sphere (see Fig. 8). However, their argument did not stand up to a
more rigorous analysis. Halliwell and Louko66 found a means of evaluating the
minisuperspace path integral exactly, and thereby explicitly determined convergent
contours. They showed that this simple model possesses inequivalent contours for
which the path integral converges. These pass through different saddle points and
lead to different semiclassical wavefunctions, Ψ. There are thus many different
no-boundary wavefunctions, each corresponding to a different choice of contour.
The problem persists in more complicated models67,68 . Since different no-boundary
wavefunctions could lead to different physical predictions, the ambiguity associated
with the choice of contour would appear to be the most significant problem with
the no-boundary proposal which still needs to be resolved.
Let us return to the minisuperspace example of §3.7 and examine the implications of imposing the Hartle-Hawking boundary condition. For simplicity we will
specialise to k = +1 models. The other values of k have also been discussed in the
literature.2,105,118
The minisuperspace no-boundary wavefunction for the k = +1 models is given
by
ΨHH [a, φ] =
whereN
I=
1
2
Z
0
τf
dτ N
"
Z
−a
N2
(a,φ)
da
dτ
Da Dφ DN e−I[a,φ,N ] ,
2
a3
+ 2
N
dφ
dτ
2
(4.1)
#
− a + a3 V .
(4.2)
The integral is taken over a class of paths which match the values
a(τf ) = a,
φ(τf ) = φ,
(4.3)
on the final surface, and the origin of the Euclidean time coordinate, τ , has been
chosen to be zero.
There are two approaches we can take to solving the Wheeler-DeWitt equation (3.59): either (i) attempt to interpret the Hartle-Hawking boundary condition
directly in terms of boundary conditions of Ψ on minisuperspace; or (ii) take a
saddle-point approximation to the path integral (4.1). For consistency these two
approaches should agree.
Let us first consider (3.59) directly. Hawking and Page52,119 have argued that
one can approximate the Hartle-Hawking boundary condition in minisuperspace by
saying that in an appropriate measure one should have Ψ = 1 when a → 0 with φ
N
I shall use a different font to distinguish the minisuperspace coordinates in the functional integral,
(a, φ), from their boundary values, (a, φ), in the 3-geometry on the hypersurface Σ.
36
regular, and Ψ = 1 also along the past null boundaries, in order to provide sufficient
Cauchy data to solve (3.59) everywhere in the (α, φ) plane. It is possible to solve
(3.59) exactly for a massless scalar field86,120 , (i.e., V = 0), but exact solutions
are not known for the massive scalar field. Approximate solutions can be found in
various regimes52,119,121, however.
Firstly, since Ψ must be regular as a → 0, we see that Ψ must be independent of
−3
φ in this limit, ∂Ψ
factor (using
∂φ ≃ 0, in order to overcome the divergence of the a
2
coordinates (a, φ)) in (3.59). Furthermore, for a V ≪ 1 the approximation V ≃ 0
is a good one⊲⊳ , and for k = ±1 (3.59) becomes a Bessel equation in the variable
1 2
a . For k = +1, which is the case of interest to us here, we therefore obtain the
2
solution52,119
(4.4)
Ψ ≃ I0 ( 21 a2 ),
P∞
z 2n
1
is the zero order modified Bessel function, and the
where I0 (z) ≡ n=0 (n!)
2
2
normalisation has been fixed to satisfy the boundary condition Ψ → 1 as a → 0.
For large a (with a2 V ≪ 1),
1
Ψ∼ √
exp
πa
1 2
2a
1 + O(a−2 ) ,
(4.5)
i.e., the wavefunction is of exponential type. It therefore agrees with the WKB
approximation (3.61) in the limit a2 V ≪ 1 for large a provided we take the (−)
solution of (3.61) with normalisation
1
+1
.
(4.6)
C(φ) = √ exp
3V (φ)
π
Let us now consider the limit V (φ) ≫ 1 but avoid regimes in which the φ dependence
of (3.59) is significant by assuming that V is approximately constant, as in the
approximation of Fig. 4, so that the φ derivative can still be neglected. The term
involving the spatial curvature k in (3.59) is now negligible compared to the last
term and can also be neglected, yielding the solution52,119
√
Ψ ≃ c(φ) J0 ( 31 a3 V ),
(4.7)
n
P∞
z 2n
where J0 (z) ≡ n=0 (−1)
is the zero order ordinary Bessel function. Since
(n!)2
2
the spatial curvature term in (3.59) is always dominant for a → 0, the approximation
which led to (4.7) no longer applies in that limit, so the factor c cannot be normalised
by the boundary condition Ψ(0, φ) = 1. For large a, however,
c
π
Ψ∼ √
cos S −
,
(4.8)
4
2πS
⊲⊳
We assume that V (φ) grows less strongly than e6|φ| as |φ| → ∞, i.e., |V ′ /V | < 6, so that this
approximation remains valid for arbitrarily large |φ|.
37
√
where S = 13 a3 V , which is a superposition of the two oscillatory WKB modes
(3.60) for large S. Using the WKB connection formula to match the (−) solution of
(3.61) to the oscillatory region, we find agreement with the asymptotic limit (4.8),
provided
r
q
2π
1
2
.
(4.9)
C = 3 exp
c=
3
3V (φ)
The φ-dependent corrections to these wavefunctions, which result from perturbations in directions in which φ̇ 6≃ 0 have been discussed by Page2,119,122.
Let us now consider the alternative method of determining Ψ by making a
saddle-point approximation to the path integral. In the semiclassical approximation
the wavefunction takes the form
Ψ ∼ exp (−Icl (a, φ)) ,
(4.10)
where Icl denotes the Euclidean action (4.2) evaluated at a classical (possibly complex) solution to the Euclidean field equations
2
2
1
a2 dφ
da
− 2
− 1 + a2 V (φ) = 0,
(4.11)
N 2 dτ
N
dτ
2
1 da
2
dφ
1 d
+ 2
+ V (φ) = 0,
(4.12)
N a dτ N dτ
N
dτ
1 d
3 da dφ 1 dV
1 dφ
+
−2
= 0,
(4.13)
N dτ N dτ
N a dτ dτ
dφ
which follow from (3.54)–(3.56) by replacing t → −iτ . We will henceforth restrict
= 0, i.e., N = const, in which case the functional
ourselves to the gauge in which dN
dτ
integral over N in (4.1) must be replaced by an ordinary integral.
The Hartle-Hawking boundary condition demands that
dφ
dτ
a(0) = 0,
= 0.
(4.14)
0
To see this consider the Euclidean 4-metric which is given by
ds2 = N 2 dτ 2 + a2 (τ )dΩ32 .
(4.15)
The Hartle-Hawking boundary condition requires that we close this 4-geometry in a
regular fashion as τ → 0. This is achieved if a ∼ ±N τ as τ → 0, since (4.15) is then
the same as the metric of the 4-sphere in spherical polar coordinates in this limit.
This suggests that we demand a(0) = 0 and N1 da
dτ 0 = ±1. However, the second
condition is guaranteed by the constraint equation (4.11) if the first condition is
imposed. The second condition of (4.14) is obtained by noting that (4.13) will give
= 0, since the middle term
a regular solution for φ in the limit τ → 0 only if dφ
dτ
0
of (4.13) diverges otherwise.
38
If as before we make the simplifying assumption that V can be approximated
≃ 0, then it follows that there exist solutions
by a cosmological constant, and dφ
dτ
satisfying the boundary conditions (4.3)and (4.14), which may be written
√
a sin(N V τ )
√
a(τ ) ≃
,
(4.16)
sin(N V τf )
where
√
sin2 (N V τf ) = a2 V,
(4.17)
which follows from solving the constraint (4.11). If a2 V < 1 then (4.17) will give
an infinite number of real solutions for the constant N , which may be conveniently
parametrised
√ i
1 h
(4.18)
(n + 12 )π ± cos−1 (a V ) , n ∈ Z,
N = Nn± ≡ √
V τf
√
with cos−1 (a V ) taken in the principal range (0, π2 ). Substitution of (4.18) in (4.16)
then gives
√
a(τ ) ≃ (−1)n V −1/2 sin(N V τ ).
(4.19)
In addition to the real solutions (4.18) of (4.17), there will also be complex solutions
if a2 V > 1.
Using the solution (4.19) it is straightforward to evaluate the classical action
(4.2). For n = 0, for example, we find
3/2 i
−1 h
.
(4.20)
1 ± 1 − a2 V (φ)
I± =
3V (φ)
If we substitute (4.19) into (4.15) we obtain the metric of the 4-sphere, and thus the
classical solutions correspond to matching
√ a given 3-sphere to 4-sphere(s). Furtherda
more, for n = 0 we find dτ τ = ∓N 1 − a2 V , so that the (−) solution of (4.18)
f
da
dτ τf
and (4.20) has
> 0, which corresponds to matching the 3-geometry to a less
than half filled 4-sphere (Fig. 8(a)), while the (+) solution similarly corresponds to
the more than half filled 4-sphere case (Fig. 8(b)). Values of n > 0 would appear to
give cases in which the 4-geometry pinches off to zero a number of times and then
“bounces back” resulting in linear chains of contiguous 4-spheres123,124 . However,
this interpretation of the saddle points as corresponding to universes with bounce
solutions does not appear to remain valid116 once one considers complex solutions
to the field equations (4.11)–(4.13).
As was discussed earlier the no-boundary condition does not prescribe a unique
contour of integration, and thus it is not completely clear which of the above saddle points should be included in the semiclassical wavefunction (3.30). Halliwell
and Hartle have shown117 that the points with n < 0, which have a negative lapse
function, lead to difficulties with the recovery of quantum field theory in curved
39
spacetime from quantum cosmology, and thus one might hope that these points
should be avoided in the contour of integration. However, no clear grounds present
themselves for omitting the other saddle points. For the purpose of making predictions from the no-boundary proposal we shall therefore make a choice by assuming
that the contour is such that the solution corresponding to the less than half filled
4-sphere, with action I− , provides the dominant contribution. (This is the choice
that Hartle and Hawking originally made38 .) Neglecting the prefactor, we therefore
obtain a no-boundary wavefunction
3/2
−1
1
2
1 − a V (φ)
,
(4.21)
exp
ΨHH (a, φ) ∝ exp
3V (φ)
3V (φ)
in the region a2 V < 1. Using the WKB matching procedure one can show that the
corresponding solution in the region a2 V > 1 is
3/2 π
1
1
2
ΨHH (a, φ) ∝ exp
cos
a V (φ) − 1
−
,
(4.22)
3V (φ)
3V (φ)
4
which is the superposition of the two WKB components of (3.60):
ΨHH = Ψ− + Ψ+ ,
3/2 π
1
1
2
Ψ± ∝ exp
exp ±i
.
a V (φ) − 1
−
3V (φ)
3V (φ)
4
(4.23)
One may observe that (4.21) and (4.22) agree with the solutions (4.4) and (4.7)
found earlier by direct examination of the Wheeler-DeWitt equation in the appropriate limits (4.5) and (4.8). Thus although the saddle point corresponding to the
less than half filled 4-sphere does not appear to be picked out in any special way
by the path integral, it is favoured by the “approximate” boundary condition52,119
that Ψ → 1 as a → 0.
One must add the caveat that the approximate boundary condition should be
amended if one is to consider genuinely complex solutions of the Euclidean field
equations (4.11)–(4.13). This issue has been considered by Lyons116 . In general
one must analytically continue the boundary condition (4.14) demanded by the
no-boundary proposal. Although the simple picture of matching a real Euclidean
solution to a real Lorentzian solution at the junction is no longer maintained, one
nonetheless finds solutions which are initially approximately Euclidean and at late
times are approximately Lorentzian, with classical inflationary behaviour116 .
4.2. The tunneling proposal
An alternative approach advocated by Vilenkin is that the boundary condition
for the wavefunction, Ψ, should be such as to embody the notion that the universe
“tunnels into existence from nothing” without making such specific restrictions on
the “initial” geometry as the Hartle-Hawking proposal does. Conceivably there are
40
many possible ways in which such a notion could be translated mathematically into
a definition of the wavefunction, Ψ, and indeed many alternative formulations of the
tunneling proposal have been put forward≬ . Some early versions were phrased in
a similar fashion to the no-boundary proposal: in particular, Vilenkin70 proposed
defining the wavefunction by a functional integral over Lorentzian metrics which
interpolate between a given matter configuration, Φ, and 3-geometry, hij , and a
vanishing 3-geometry, ∅, lying to its past
Ψ [hij , Φ, Σ] =
XZ
M
(h,Φ)
Dg Dφ eiS[gµν ,φ] .
(4.24)
∅
Vilenkin has also given an alternative formulation of the tunneling proposal
in terms of a boundary condition on superspace40,41 rather than a restriction on
manifolds included in the path integral. In order to formulate boundary conditions
on superspace it is necessary to consider its boundary, which can be thought of as
consisting of 3-metric and matter configurations for which the 3-curvature is infinite,
or |Φ| → ∞ etc. As we have already seen from the example of S 4 (Fig. 7), not all
singular 3-geometries will correspond to singular 4-geometries, as it is possible to
obtain a singular 3-geometry by a degenerate slicing of the 4-geometry. Therefore
we should distinguish points on the boundary of superspace which correspond to
genuine singularities of the 4-geometry from those that correspond to degenerate
slicings⋆ . We call the former the singular boundary of superspace, and the latter
the non-singular boundary.
The tunneling proposal of Vilenkin41 is that the wavefunction, Ψ, should be
everywhere bounded, and at singular boundaries of superspace Ψ includes only outgoing modes, i.e., those that carry a flux out of superspace. Thus ingoing modes
can only enter at the nonsingular boundary. This definition is somewhat vague as
there is no obvious rigorous definition of positive and negative frequency modes in
superspace due to the fact that it possesses no Killing vectors96 , and thus there
is no clear notion which modes are “ingoing” and which are “outgoing”. Furthermore, the structure of superspace is not completely understood, and it has not been
rigorously shown that its boundary can be split into singular and non-singular parts.
The tunneling proposal has in fact been formulated with the minisuperspace
WKB approximation in mind, in which case the notion of the boundary of minisuperspace and the notions of ingoing and outgoing modes are more clearly defined.
Since each oscillatory WKB mode Ψ ∼ eiSn has a current (3.44), we can classify
the modes as ingoing or outgoing according to the direction of ∇Sn on the surface
in question. Heuristically, the idea underlying the Vilenkin boundary condition is
that the ensemble of universes described by Ψ should not include any universes
≬
Linde’s proposal [69] embodies a similar philosophy. However, it gives a different wavefunction to
Vilenkin’s “tunneling” wavefunction in simple minisuperspace models [125].
⋆
This distinction can be made more precise using Morse functions [42].
41
contracting down from infinite size40 , but only those that correspond to “tunneling
from nothing”. As we shall see, the “outgoing flux” condition41 accords with this
notion at least in the case of simple minisuperspace models.
The outgoing-flux version of the tunneling proposal agrees with the path integral formulation (4.24) in the case of the simplest de Sitter minisuperspace42,66 ,
but the two versions would not appear to be equivalent in general68 . Whereas the
no-boundary wavefunction fixes the initial data but leaves the contour of path integration ambiguous, the tunneling proposal fixes the contour of integration but leaves
some ambiguity in the specification of the initial data, even when the outgoing-flux
condition is imposed68,126 .
Consider the minisuperspace model of §3.7. In the diagrams of Fig. 5 all surfaces
I±
and the points i0L,R and i+ will be part of the singular boundary, whereas the
L,R
point i− , which corresponds to a → 0 (α → −∞) with φ finite, is the only point
of the nonsingular boundary. Of course, the condition that a → 0 is not enough to
guarantee regularity of the 4-geometry (4.15), and in general one might not expect
boundary points to cleanly fall into the category of the “singular boundary” or the
“nonsingular boundary”. However, as we have already observed in the last section,
in the present minisuperspace model the additional requirement for regularity that
1 da
N dτ 0 = ±1 is guaranteed by the constraint equation (4.11) if a(0) = 0 is imposed.
The oscillatory WKB region, as shaded in the conformal diagrams of Fig. 5
and Fig. 9, is always bounded by i0R , I+
and i+ , and in all cases except that of
R
Fig. 5(f)▽ the oscillatory region is bounded by i0L and I−
also. The wavefunction
L
iSn
in this region is given by a superposition of terms e , where Sn is a solution to
the Hamilton-Jacobi equation (3.34), which in terms of the variables (α, φ) may be
written
2 2
∂S
∂S
−
+
+ U(α, φ) = 0,
(4.25)
∂α
∂φ
where U(α, φ) ≡ 2e3α U(α, φ) = e4α e2α V (φ) − 1 , and for convenience in what
follows we have suppressed the index n. The characteristics of (4.25) satisfy
dα
dφ
dS
d(S,φ )
d(S,α )
=
=
=
.
=
2S,α
−2S,φ
2U
U,α
U,φ
(4.26)
Thus each S(α, φ) describes a congruence of classical paths with
S,φ
∓S,φ
dφ
.
=−
=q
dα
S,α
S,φ2 +U
▽
(4.27)
Unlike the other cases the potential of Fig. 5(f), which corresponds to a potential in which super-
symmetry is broken through gaugino condensation in string theory, vanishes as φ → −∞. This limit is the
“weak coupling limit” of string theory.
42
Fig. 9: Some schematic probability flows consistent with the tunneling proposal.
An indicative oscillatory WKB region is shown (shaded area) for the potential
V = φ2 .
Since U > 0 in the oscillatory region (assuming V (φ) > 0), it follows that these
dφ
integral curves satisfy dα
< 1, i.e., they are “timelike” in the minisuperspace
3α
coordinates and have an endpoint at i+ . Since πα ≡ − e N α̇ = S,α it follows that
the WKB modes correspond to expanding universes (α̇ > 0) with πα = S,α < 0,
or contracting universes with πα = S,α > 0, if we assume⊘ N > 0. Since all paths
originate at finite values of α the latter solutions will extend from i+ into the interior
of the minisuperspace, i.e., they are “ingoing modes” and are to be excluded on
account of the tunneling boundary condition. This boundary condition only allows
the expanding solutions which are “outgoing” at i+ . In the approximation in which
(4.5) applies it
proposal demands that only the modes
h follows that the tunneling
i
3/2
−i
with Ψ ∼ exp 3V (φ) e2α V (φ) − 1
are admitted in the oscillatory region. The
tunneling wavefunction is thus complex in the oscillatory region, in contrast to the
⊘
This choice is a matter of convention and the opposite choice would reverse the roles of the “ingoing”
and “outgoing” modes. Equivalently, the coordinate t is an arbitrary label from the point of view of general
relativity, and it is a matter of convention whether we choose t to increase or decrease towards the “future”;
the “future” being defined by the expansion of the universe (the cosmological arrow) or the increase of
entropy (the thermodynamic arrow). In terms of Fig. 4 it amounts to an arbitrary choice between A+ and
A− as representing the “late-time behaviour” of the inflationary solutions.
43
no-boundary wavefunction (4.22) which is real.
To extend the tunneling wavefunction into the “tunneling region” (the unshaded
regions of Fig. 5 and Fig. 9), we can use the WKB matching procedure. Of course,
there will also be additional solutions that remain entirely in the tunneling region
and never cross into the oscillatory region. In the case of the null boundaries, I−
,
L,R
which border this region we observe that as α → −∞ (3.59) becomes the wave
equation in the (α, φ) coordinates, and thus the solutions are asymptotically null
which leads to a notion of “ingoing” and “outgoing” modes, so that the tunneling
condition can be imposed. Some possible probability flows consistent with the
tunneling proposal are shown in the conformal diagram of Fig. 9.
5. The Predictions of Quantum Cosmology
Given the many problems and uncertainties in the quantum cosmology programme that have been discussed above, one could easily form the opinion that it
is premature to talk about the predictions that quantum cosmology makes. Nevertheless, it is important to investigate the types of predictions we might expect
quantum cosmology to make about the universe, as well as realistically evaluating
the limitations of the these predictions. I will concentrate on three key areas in this
section. Other topics, most notably the variability of the constants of nature, have
been discussed recently by Vilenkin125,127.
5.1. The period of inflation
The construction of a suitable probability measure should allow us to answer
questions such as whether inflation is a feature of a typical universe. As we saw
in §3.6, the construction of a general probability measure is problematic. Thus we
shall limit our discussion to the oscillatory WKB limit in which case the KleinGordon-type current (3.42) is adequate.
The issue of the duration of the period of inflation has been a point of some
debate between proponents of the no-boundary wavefunction and proponents of
the tunneling wavefunction. Consider the canonical minisuperspace model of §3.7.
We have seen that in the WKB limit the Hartle-Hawking boundary condition gives
rise to a wavefunction (4.22) in the oscillatory region, which is strongly peaked
about the
set of classical solutions (3.62) that correspond to an inflating universe:
√
V t
a(t) ∝ e
, φ(t) ≃ φ0 = const.
Vilenkin40–42,70 has also studied the minisuperspace model of §3.7, but with
a different choice of factor ordering in the Wheeler-DeWitt equation (3.59). He
concludes that similarly to (4.22) the tunneling boundary condition leads to a WKB
wavefunction
3/2 iπ
−i
−1
2
exp
(5.1)
a V (φ) − 1
+
ΨV (a, φ) ∝ exp
3V (φ)
3V (φ)
4
44
in the oscillatory region, which is also peaked about the classical trajectories of the
inflationary universes (3.62).
Both the no-boundary and tunneling wavefunctions thus predict inflation, at
least in the context of this simple minisuperspace model. The important question
is how much inflation do the models predict? This will be largely decided by the
value, φ0 , of the scalar field with which the universe “nucleates” in the semiclassical
regime.
In order to study the probability flux arising from (3.42) in the WKB limit
it is necessary to focus on one particular WKB component. The no-boundary
wavefunction (4.22) is of course real, and the resulting current (3.42) is identically
zero. However, ΨHH is the superposition (4.23) of two WKB components, which
will correspond to contracting and expanding universes. As discussed in §3.5 it
is assumed that a decoherence mechanism exists so that the interference between
the two components is negligible89 , and we can therefore assume that the universe
is peaked about one or other WKB component for the purposes of determining
the probability measure. The difference between the no-boundary and tunneling
wavefunctions may therefore not seem great once decoherence to a classical universe
is assumed. In particular, if we take the “outgoing” WKB component in the noboundary case then the only significant
between ΨHH and ΨT is the φ
difference
dependent part of the prefactor, exp 3V±1
(φ) , which is obtained from boundary
conditions set in the tunneling region. This difference will have ramifications for
the probability flux, however.
One question which we might hope to answer in our minisuperspace model
would be: given that a Lorentzian universe nucleates, what is the probability that
it inflates by a sufficient amount (∼ 65 e-folds) to solve the problems of the standard cosmology mentioned in §1? The answer to this would involve integrating
the probability flux (3.45) on the surface separating the tunneling and oscillatory
regions, which is roughly given by a2 V (φ) = 1. However, our discussion here is
limited by the fact that our WKB approximation applies to trajectories with φ̇ ≃ 0.
For such trajectories the probability current, J, points chiefly in the direction of the
a-coordinate in minisuperspace. We can therefore attempt to answer the question
approximately by evaluating the probability current (3.44) on a surface Σ in minisuperspace with a = const. For sufficiently large values of a such surfaces will lie
almost entirely within the oscillatory regime (see Fig. 10). On these surfaces we
obtain a probability flux
exp 3V+2
no-boundary wavefunction, ΨHH ,
(φ)
dP = J · dΣ ∝
(5.2)
exp −2
tunneling
wavefunction,
Ψ
.
V
3V (φ)
Although the integral of (5.2) may diverge for particular potentials V (φ) of interest,
this should not be viewed as a problem since, as was emphasised in §3.6, questions in
quantum cosmology can only refer to conditional rather than absolute probabilities.
45
In the present case let us assume that inflation occurs for large values of the
scalar field, as is the case for potentials of the “chaotic” type, V = λφ2p . There
will then be a minimum value of the scalar field, φsuff , for which sufficient inflation
is obtained. For V = λφ2p a universe with φ = φ0 initially will undergo
3 2 2
φ0 − 9 p(2p − 1)
(5.3)
Ne ≃
2p
e-folds of inflation⊡ , so that in the case of the quadratic potential (p = 1) we find
φsuff >
∼ 6.6, for example. The relevant conditional probability for sufficient inflation
on an a = const surface is then given by
Z φ
2
±2
dφ0 exp
3V (φ0 )
φsuff
(5.4)
P φ0 > φsuff | φ1 < φ0 < φ2 = Z φ
,
2
±2
dφ0 exp
3V (φ0 )
φ1
where the (+) case refers to the no-boundary wavefunction and the (−) case to the
tunneling wavefunction, and the values φ1 and φ2 are respectively lower and upper
cutoffs on the allowed values of φ, which must be determined by physical criteria.
A minimum cut-off might be expected for a variety of physical reasons, such as
avoiding classical universes which rapidly recollapse.
In their original investigations of the question of the duration of the period of
inflation for the quadratic scalar potential, Hawking and Page52 took φ2 = ∞, in
which case both integrals in (5.4) are dominated by large values of φ, leading to
a probability P ≃ 1, and thus a “prediction” of inflation. However, this result
9
has been criticised by Vilenkin41 since for k = +1 we have V = 16
m4Planck V so
that values of mφ >
∼ 4/3 are in excess of the Planck scale, and the semiclassical
approximation will no longer apply. Vilenkin suggested that an upper cutoff, φ2 ,
should be introduced at the Planck scale. If this is the case then provided the
lower cutoff φ1 is sufficiently close to zero, we would find that the integral in the
denominator of (5.4) becomes very large in the case of the no-boundary wavefunction
(+ sign), leading to P ≪ 1, whereas this would not be the case for the tunneling
wavefunction (− sign), and thus the latter would predict more inflation.
Introducing a cutoff at the Planck scale might be deemed a rather arbitrary
procedure, since without any knowledge of Planck scale physics it is impossible
to be sure whether the “real” answer is better approximated by the introduction
of a cutoff or not. However, it has been argued on the basis of investigations at
the 1-loop level that the wavefunction is damped at large values of φ by quantum
corrections128,129 . This renders the wavefunction normalisable and would justify a
cutoff, φ2 , near the Planck scale.
⊡
In the slow-rolling approximation [109] it follows from (3.54) and (3.56) (with N = 1) that the number
Rφ
of e-folds is Ne = 6 φ 0 dφ V /V ′ , where φ0 is the initial value and φe the final value of the scalar field at the
e
3
end of the inflationary epoch. For V = λφ2p we have Ne = 2p
φ02 − φe2 . The value of φe can be estimated
from the limit set by |V ′ /V | ≪ 6 and |V ′′ /V | ≪ 9.
46
Fig. 10: Conformal diagram for V = 0.04φ2 . The oscillatory region, given roughly
by a2 V > 1, is lightly shaded. Lines a = const are superimposed. For very large
values of φ these lie almost entirely in the oscillatory region. The region of φ-values
excluded by a Planck scale cutoff is darkly shaded.
Given that the no-boundary wavefunction apparently yields a wavefunction
peaked around the lower cutoff, φ1 , it is important to determine what a reasonable
value of this cutoff should be. This issue has been considered by Grishchuk and
Rozhansky130 , and also more recently by Lukas131 , who have conducted numerical
investigations to analyse the behaviour of the caustic132 in the (a, φ) plane which
separates the Euclidean and Lorentzian solutions. This is illustrated in Fig. 11,
where classical solutions to the Euclidean field equations (3.54)–(3.56) (with N = 1)
are shown. The solutions beginning at a = 0 initially follow lines with φ̇ ≃ 0, but
the approximation eventually breaks down when the trajectories curl back and recollapse with a → 0 and |φ| ∝ − ln a → ∞. They thus represent a flow from i− to
I− in the conformal diagrams. The solutions cross each other on the caustic, which
for large values of φ corresponds to a2 V (φ) = 1, in accordance with our expectation
from the WKB approximation.
To the right of the Euclidean solutions in Fig. 11 we would find Lorentzian
solutions with φ̇ ≃ 0 sufficiently far away from the caustic132 . These solutions
are not depicted here. The “nucleation of a universe” would thus correspond to
a solution of the Wheeler-DeWitt equation which was initially peaked about a
classical Euclidean trajectory with φ̇ ≃ 0, and which then crossed over the caustic
to be peaked about a corresponding trajectory with φ̇ ≃ 0 in the Lorentzian region.
47
Trajectories
a2 V=1
Improved Caustic
4.4
4.2
4
3.8
φ
3.6
3.4
3.2
3
2.8
2.6
0
0.02
0.04
0.06
a
0.08
0.1
0.12
0.14
Fig. 11: Classical Euclidean trajectories for the potential V = φ4 . The approximate
caustic a2 V = 1 is indicated by a dashed line, and the improved caustic by a dotted
line. (From [131].)
While crossing the caustic, the wavefunction would be peaked about a complex
solution which was neither truly Euclidean nor truly Lorentzian.
It is evident from Fig. 11 that for small values of φ the caustic begins to deviate
from the curve a2 V = 1. In fact, one finds130,131 that for sufficiently small values of
φ Lorentzian universes never nucleate at all. Accordingly, a portion of the shaded
region in the conformal diagrams should be excised about the φ = 0 axis in the
vicinity of i+ . The cutoff, φ∗ , at which this occurs has been estimated by Lukas131
as being
n h
io1/2
φ∗ ≃ p + 2p 1 − (1 + e−1 )−1/2
,
(5.5)
in the case of the chaotic potentials, V = λφ2p . Provided that the interpretation
of (5.4) remains valid♠ , we should therefore set φ1 = φ∗ as the lower bound in
the integrals. In the case of the quadratic potential, V = m2 φ2 , we then have
φ1 ≃ 1.5, and using the values of φsuff and φ2 found earlier it is straightforward to
♠
As a cautionary note, one should observe that since the approximation of the caustic by the curve
2
a V = 1 breaks down for small φ, the use of the surfaces a = const which was assumed in (5.4) may not be
appropriate for small φ, and ideally one should determine the probability flux across the caustic itself. Of
course, in a more careful analysis one would consider complex solutions to the field equations in line with
[116], rather than real Euclidean and real Lorentzian solutions with a junction condition. One would hope
that this would not alter the conclusions of the analysis much.
48
check that for typical values of m ≪ 1, (5.4) gives P ≪ 1 for ΨHH and P ∼ 1 for ΨV
in accordance with the earlier discussion.
The above analysis would appear to indicate that the no-boundary wavefunction
effectively “predicts” a value of φ ≃ φ∗ , which in the case of the quadratic potential
is unfortunately less than φsuff . However, the calculation is model-dependent, and
if one could find a potential for which φ∗ >
∼ φsuff then both ΨHH and ΨV would yield
P ∼ 1. If we compare (5.3) and (5.5) (for Ne ≃ 65) we see that in the case of the
chaotic potentials, V = λφ2p , this requirement is equivalent to p >
∼ 62. However, it
requires an enormously small value of λ to keep such a V below the Planck scale
and this is not promising. Lukas131 has also estimated the value of φ∗ for some
other potentials but did not find any candidates with φ∗ >
∼ φsuff . However, given
the model-dependence of the calculations it cannot be ruled out that some other
potential, or the coupled effects of two or more scalar fields, might give φ∗ >
∼ φsuff
and thereby make inflation a “prediction” of the no-boundary wavefunction.
5.2. The origin of density perturbations
It is expected that the anisotropies in the cosmic microwave background radiation, which were first definitively observed in 1992, have their origin in quantum
fluctuations in the very early universe. Such fluctuations can be described using
the formalism of quantum field theory in curved spacetime. Since the era of quantum cosmology is in a sense prior to that in which quantum field theory in curved
spacetime is applicable, one would hope to trace the origin of the primordial perturbations back to quantum cosmology. Indeed, this can be done133–135 and I will
very briefly outline the main results.
With homogeneous minisuperspaces as a starting point, one can add small inhomogeneous perturbations to the metric and matter fields:
hij (x, t) = a2 (t) (Ωij + εij ) ,
Φ(x, t) = Φ0 (t) + δΦ(x, t),
N (x, t) = N0 (t) + δN (x, t),
Ni (x, t) = 0 + δNi (x, t).
(5.6)
(5.7)
(5.8)
(5.9)
Here we have restricted attention to the k = +1 FRW minisuperspace model, the
subscript zero denotes the unperturbed quantities, and Ωij is the unperturbed standard round metric on the 3-sphere. The perturbations can be expanded in terms
of spherical harmonics on the 3-sphere133 . If one substitutes the ansatz (5.6)–(5.9)
into the classical action (2.12) and expands to quadratic order one obtains an action
which can be split into the original minisuperspace action, S0 , and an additional
action, S2 , quadratic in the perturbations,
A
A
S = S0 [q , N0 ] + S2 [q , N0 , εij , δΦ, δN , Ni],
49
(5.10)
and a corresponding Hamiltonian
Z
Z
Z
3
3
H = N0 H0 + d x H2 + d x δN H1 + d3 x δNi Hi ,
(5.11)
where H0 is the unperturbed Hamiltonian (3.54), and the terms H1 and H2 are
linear and quadratic in the perturbations respectively. There is now a non-trivial
momentum constraint at each point x ∈ Σ, Hi (x) = 0, while the Hamiltonian constraint splits into a piece linear in the perturbations, H1 (x) = 0, plus a homogeneous
piece
Z
H0 + H2 ≡ H0 + d3 x H2 = 0.
(5.12)
We may quantise (5.12) in the standard fashion to obtain a modified WheelerDewitt equation
i
h
(5.13)
ĤΨ = − 12 ∇2 + U(q) + Ĥ2 Ψ = 0,
in place of (3.27), where the Laplacian is still defined in terms of the original minisuperspace coordinates according to (3.28), and Ĥ2 is a second order differential
operator which results from quantisation of the homogeneous part of the perturbations. In the case of pure scalar field modes, for example, the perturbations may be
decomposed as
1X
fnlm (t)Qnlm (x),
(5.14)
δΦ(x, t) =
σ
nml
where the
Qnlm
satisfy the 3-dimensional Laplace equation on S 3 ,
(3)
∆Qnlm = −(n2 − 1)Qnlm ,
(5.15)
and one finds1,89,133
Ĥ2 =
1
2
X −1 ∂ 2
nml
a3 ∂f2nlm
+ (n2 − 1)a + m2 a3 f2nlm
(5.16)
in the case of the quadratic scalar potential.
It is possible to find solutions to the Wheeler-DeWitt equation (5.13) in which
A
the minisuperspace coordinates, q , are treated semiclassically in the WKB approximation, while the perturbations are treated quantum mechanically. One can
show1,89,133,134 that the solutions take the form
Ψ = C(q, Φ) eiS0 (q,Φ) ψ̃,
(5.17)
where S0 satisfies the unperturbed Hamilton-Jacobi equation (3.34), the prefactor
C depends only on the unperturbed minisuperspace coordinates, and the functions
ψ̃ satisfy the functional Schrödinger equation
i
∂ ψ̃
= Ĥ2 ψ̃.
∂t
50
(5.18)
The different modes of the scalar perturbations (5.14) do not interact, for example,
and in this case
Y
ψ̃ =
ψnlm (t, fnlm ),
(5.19)
nml
where each mode ψnlm separately satisfies (5.17) with Ĥ2 given by (5.16).
The wavefunction (5.17) is thus peaked about classical trajectories, with corrections, ψ̃, which satisfy the functional Schrödinger equation along these trajectories.
This is in fact precisely the starting point for the treatment of matter modes in a
curved spacetime background using the formalism of quantum field theory in curved
spacetime. Thus the above result is important in that it demonstrates that quantum cosmology is consistent with the standard approach to the quantum treatment
of cosmological perturbations. As an added bonus the imposition of a boundary
condition on Ψ, such as the no-boundary or tunneling condition, will result in the
choice of particular solutions of the functional Schrödinger equation, and consequently a particular vacuum state for the matter modes. Both the no-boundary
condition and the tunneling condition pick out41,136–139 a de-Sitter invariant state
known as the “Euclidean” or Bunch-Davies vacuum140,141 , which is the state that
is often assumed in cosmological calculations of density perturbations. In fact, this
state is picked out by many boundary conditions137,138, and thus could be regarded
as a natural quantum state for matter in quantum cosmology.
One may solve the functional Schrödinger equation and study the growth133 of
the modes. As discussed above, the results obtained are the same as those which
are found if one begins with quantum matter fields in de-Sitter invariant vacuum
states in background inflationary spacetimes142,143 . In particular, one obtains a
scale-free spectrum of density perturbations which act as seeds for the formation
of galaxies and other structures in the universe. Such a spectrum accords well
with the spectrum deduced from the COBE measurements of cosmic microwave
anisotropies144 .
5.3. The arrow of time
The question of the origin of the arrow of time in the face of the CPT-invariance
of the laws of physics is one of the deepest unresolved conceptual issues in theoretical
physics, and it has been the subject of more than one major conference44,145 . The
question of the nature of time assumes prime importance in quantum gravity and
quantum cosmology♥ . One might argue that if the basic laws of physics are time
symmetric, then the arrow of time could have its origin in a boundary condition for
the wavefunction of the universe.
There are many different arrows of time observed in the universe. Some of
these such as the psychological arrow of time (we remember only the past), and
♥
The conceptual issues surrounding the nature of time in quantum cosmology would entail a series of
lectures in themselves. For further discussion see, e.g., [24], [146]–[148] and references therein.
51
the electromagnetic arrow of time (the choice of retarded as opposed to advanced
solutions of Maxwell’s equations), could well be argued to be consequences of the
thermodynamic arrow of time, which arises from the second law of thermodynamics.
However, the expansion of the universe provides an alternate cosmological arrow of
time which is not obviously directly related to the increase of entropy. Whether
such a relationship does exist is a question which might potentially be resolved by
quantum cosmology.
In 1985 Hawking149 proposed that the thermodynamic arrow and cosmological
arrows of time were correlated: that is to say if the universe were spatially closed
then entropy would decrease in the contracting phase⊚ . In arriving at this proposal
he was influenced by the fact that the no-boundary wavefunction is CPT-invariant,
and also by some early studies of simple minisuperspace models which possessed
quasi-periodic solutions151 which “bounced” instead of recontracting to a singularity as a → 0. However, Hawking soon changed his mind on the issue, which he
terms his “greatest mistake”152. This change of mind was brought about by a number of factors. Firstly, Page pointed out that CPT-invariance of the no-boundary
wavefunction does not imply CPT-invariance for an individual WKB component of
the wavefunction, which would correspond to the history of a classical universe153 .
Furthermore, minisuperspace models which were subsequently studied, such as that
of the Kantowski-Sachs universe, were found not to admit bounce solutions but always possessed singularities to the future154,155 . Finally, as was mentioned in §4.1,
the bounce solutions do not feature even in simple minisuperspace models once
one considers the contribution of complex metrics116 . In general, the approximate
minisuperspace boundary condition that Ψ → 1 as a → 0 must be altered to allow for approximately Euclidean metrics which also contribute a rapidly oscillating
component to the wavefunction116,156 .
Hawking, Laflamme and Lyons156 have recently argued that a thermodynamic
arrow of time results from the imposition of the Hartle-Hawking boundary condition. More precisely, they have considered the evolution of primordial fluctuations
as outlined in §5.2, but accounting for the “approximately Euclidean” geometries116
which appear to be required if one considers complex metrics. They find156 that
the gravitational wave perturbations have an amplitude that remains in the linear
regime and is roughly time-symmetric about the time of maximum expansion. Such
perturbations cannot be said to give rise to an arrow of time. Density perturbations behave differently, however. They start out small but grow large and become
non-linear as the universe expands, and moreover this growth continues during the
contracting phase of the spatially closed universe. This growth of inhomogeneity
therefore provides an arrow of time which could be considered to be an essentially
thermodynamic arrow. Since it does not match the cosmological arrow in the contracting phase, the only reason for the coincidence of the two arrows in the present
epoch would appear to be an anthropic one. In particular, the conditions that would
⊚
This idea was first suggested by Gold [150], and has resurfaced a number of times in different contexts.
52
prevail in a contracting phase would appear to preclude the existence of life152,156.
Thus the fact that we are around to make observations means we must find ourselves
in a cosmological epoch in which the two arrows coincide.
It should be added that the debate about whether the cosmological and thermodynamic arrows of time coincide has not yet been closed, however, and it is
still maintained by Kiefer and Zeh157 that the boundary condition on the WheelerDeWitt equation must be such that the thermodynamic arrow would reverse in a
recontracting universe.
6. Conclusion
I hope to have shown you that although research in quantum cosmology is still
rather speculative and open-ended, its framework nonetheless has the potential not
only to provide answers to questions surrounding the origin and early evolution of
the universe, but also to help us unravel the mysteries of quantum gravity. Quantum
cosmology is a field in which a great deal remains to be done, and the results of
§5 should be regarded as a hopeful indication of the types of predictions we might
hope to make. It is too early to draw definitive conclusions about the relative merits
of the various boundary condition proposals. The results of §5.1 ostensibly favour
the “tunneling proposal”. However, this conclusion is only based on a few simple
models, and therefore some caution must be exercised.
As a final note, it is worth mentioning that recent results show that supersymmetry provides a means of restricting possible boundary conditions for the
Wheeler-DeWitt equation, or the corresponding Dirac square-root equation. The
supergravity constraint equations for various homogeneous minisuperspace models
appear to be so restrictive that they only pick out the most symmetric quantum
states102,103,158. In particular, simple analytic solutions for Ψ in the supersymmetric Bianchi-IX minisuperspace have been found102 in the empty and filled fermion
sectors, which have a natural interpretation as158 wormhole states121,159, or as160
Hartle-Hawking no-boundary states. Some doubts were initially cast on the relevance of these solutions, as the states appear to have no counterpart in 4-dimensional
supergravity161,162. However, more recent work46 on supersymmetric minisuperspaces corresponding to Bianchi “class A”76 models indicates that infinitely many
physical states with finite (even) fermion number can be found, and these are direct
analogues of physical states in full supergravity. This result shows that such minisuperspace models are likely to be physically very important for quantum cosmology.
Acknowledgement I would like to thank A. Lukas for giving me permission to
use Fig. 11, which originally appeared in [131].
53
References
1. J. J. Halliwell, in Quantum Cosmology and Baby Universes, eds. S. Coleman,
J. B. Hartle T. Piran and S. Weinberg, (World Scientific, Singapore, 1991),
p. 159.
2. D. N. Page, in Proceedings of the Banff Summer Institute on Gravitation,
August 1990, eds. R. B. Mann and P. S. Wesson, (World Scientific, Singapore, 1991).
3. P. van Nieuwenhuizen, Phys. Rep. 68 (1981) 189.
4. P. West, Introduction to Supersymmetry and Supergravity, (World Scientific,
Singapore, 1986).
5. A. Salam and E. Sezgin (eds.), Supergravities in Diverse Dimensions, (World
Scientific, Singapore, 1989).
6. G. Esposito, Quantum Gravity, Quantum Cosmology and Lorentzian Geometries, (Springer Lecture Notes in Physics m12, Berlin, 1992).
7. M. H. Goroff and A. Sagnotti, Phys. Lett. 160B (1985) 81; Nucl. Phys.
B266 (1986) 709.
8. G. ’t Hooft and M. Veltman, Ann. Inst. H. Poincaré 20 (1974) 69.
9. S. Deser and P. van Nieuwenhuizen, Phys. Rev. D10 (1974) 411.
10. S. Deser, P. van Nieuwenhuizen and H. S. Tsao, Phys. Rev. D10 (1974)
3337.
11. S. Deser, J. H. Kay and K. S. Stelle, Phys. Rev. Lett. 38 (1977) 527.
12. P. S. Howe and K. S. Stelle, Int. J. Mod. Phys. A4 (1989) 1871.
13. M. B. Green, J. H. Schwarz and E. Witten, Superstring theory, (Cambridge
University Press, 1987).
14. D. J. Gross and V. Periwal, Phys. Rev. Lett. 60 (1988) 2105.
15. S. Davis, Preprint DAMTP-R/94/27-rev (1994), hep-th/9503231.
16. A. Ashtekar, Lectures on Non-perturbative Canonical Gravity, (World Scientific, Singapore, 1991).
17. C. Rovelli, Class. Quantum Grav. 8 (1991) 1613.
18. A. Sen, J. Math. Phys. 22 (1981) 1781.
19. A. Ashtekar, Phys. Rev. Lett. 57 (1986) 2244; Phys. Rev. D36 (1987) 1587.
20. C. Rovelli and L. Smolin, Phys. Rev. Lett. 61 (1989) 1155; Nucl. Phys.
B331 (1990) 80.
21. M. P. Blencowe, Nucl. Phys. B341 (1990) 213.
22. H. Kodama, Prog. Theor. Phys. 80 (1988) 1024; Phys. Rev. D42 (1990)
2548.
23. C. Soo and L. N. Chang, Int. J. Mod. Phys. D3 (1994) 529.
54
24. L. Smolin and C. Soo, Nucl. Phys. B449 (1995) 289.
25. J. Louko, Phys. Rev. D51 (1995) 586.
26. T. Regge, Nuovo Cimento 19 (1961) 558.
27. R. M. Williams and P. A. Tuckey, Class. Quantum Grav. 9 (1992) 1409.
28. L. Bombelli, J. Lee, D. Meyer and R. D. Sorkin, Phys. Rev. Lett. 59 (1987)
521.
29. C. J. Isham, Class. Quantum Grav. 6 (1989) 1509.
30. C. J. Isham, Y. Kubyshin and P. Renteln, Class. Quantum Grav. 7 (1990)
1053.
31. J. A. Wheeler, in Relativity, Groups and Topology, eds. C. DeWitt and B.
S. DeWitt, (Gordon and Breach, New York, 1963), p. 315.
32. J. A. Wheeler, in Batelles Rencontres, eds. C. DeWitt and J. A. Wheeler,
(Benjamin, New York, 1968), p. 242.
33. B. S. DeWitt, Phys. Rev. 160 (1967) 1113.
34. C. W. Misner, Phys. Rev. 186 (1969) 1319; in Relativity, eds. M. Carmeli,
Finkler and L. Witten, (Plenum, New York, 1970), p. 55.
35. C. W. Misner, in Magic Without Magic: John Archibald Wheeler, ed. J.
Klauder, (W.H. Freeman, San Francisco, 1972), p. 441.
36. S. W. Hawking, in Astrophysical Cosmology, eds. H. A. Brück, G. V. Coyne
and M. S. Longair, (Pontifica Academia Scientarium Scripta Varia 48, Vatican City, 1982) p. 563.
37. A. Vilenkin, Phys. Lett. 117B (1982) 25; Phys. Rev. D27 (1983) 2848.
38. J. B. Hartle and S. W. Hawking, Phys. Rev. D28 (1983) 2960.
39. S. W. Hawking, Nucl. Phys. B239 (1984) 257.
40. A. Vilenkin, Phys. Rev. D33 (1986) 3560.
41. A. Vilenkin, Phys. Rev. D37 (1988) 888.
42. A. Vilenkin, Phys. Rev. D50 (1994) 2581.
43. J. B. Hartle, in Quantum Cosmology and Baby Universes, eds. S. Coleman,
J. B. Hartle T. Piran and S. Weinberg, (World Scientific, Singapore, 1991),
p. 65.
44. J. J. Halliwell, J. Perez-Mercader, and W. H. Zurek (eds.), The Physical
Origins of Time Asymmetry, (Cambridge University Press, 1994).
45. A. Strominger, in Quantum Cosmology and Baby Universes, eds. S. Coleman, J. B. Hartle, T. Piran and S. Weinberg, (World Scientific, Singapore,
1991), p. 269.
46. A. Csordás and R. Graham, Phys. Rev. Lett. 74 (1995) 4129.
47. R. M. Wald, General Relativity, (Chicago University Press, 1984).
55
48. J. W. York, Phys. Rev. Lett. 28 (1972) 1082.
49. P. A. M. Dirac, Proc. R. Soc. Lond. A246 (1958) 326; 333.
50. A. Hanson, T. Regge and C. Teitelboim, Constrained Hamiltonian Systems,
(Accademia Nationale dei Lincei, Rome, 1976).
51. P. W. Higgs, Phys. Rev. Lett. 1 (1958) 373.
52. S. W. Hawking and D. N. Page, Nucl. Phys. B264 (1986) 185.
53. G. W. Gibbons and S. W. Hawking, Phys. Rev. D15 (1977) 2752.
54. S. W. Hawking, in General Relativity: An Einstein Centenary Survey (Cambridge University Press, 1979), p. 746.
55. J. B. Hartle and S. W. Hawking, Phys. Rev. D13 (1976) 2188.
56. G. W. Gibbons and S. W. Hawking, Phys. Rev. D15 (1977) 2738.
57. G. W. Gibbons and M. J. Perry, Proc. R. Soc. Lond. A358 (1978) 467.
58. T. Eguchi and A. J. Hanson, Phys. Lett. 74B (1978) 249.
59. G. W. Gibbons and S. W. Hawking, Phys. Lett. 78B (1978) 430; Commun.
Math. Phys. 66 (1979) 291.
60. G. W. Gibbons and C. N. Pope, Commun. Math. Phys. 66 (1979) 267.
61. G. W. Gibbons, in Fields and Geometry, ed. A. Jadczyk, (World Scientific,
Singapore, 1986).
62. D. Garfinkle and A. Strominger, Phys. Lett. 256B (1991) 146.
63. H. F. Dowker, J. P. Gauntlett, S. B. Giddings and G. T. Horowitz, Phys.
Rev. D50 (1994) 2662.
64. G. W. Gibbons, S. W. Hawking and M. J. Perry, Nucl. Phys. B138 (1978)
141.
65. K. Schleich, Phys. Rev. D36 (1987) 2342.
66. J. J. Halliwell and J. Louko, Phys. Rev. D39 (1989) 2206.
67. J. J. Halliwell and J. Louko, Phys. Rev. D40 (1989) 1868.
68. J. J. Halliwell and J. Louko, Phys. Rev. D42 (1990) 3997.
69. A. D. Linde, Zh. Eksp. Teor. Fiz. 87 (1984) 369 [Sov. Phys. JETP 60 (1984)
211]; Lett. Nuovo Cimento 39 (1984) 401; Rep. Prog. Phys. 47 (1984) 925.
70. A. Vilenkin, Phys. Rev. D30 (1984) 509; Nucl. Phys. B252 (1985) 141.
71. J. J. Halliwell and J. B. Hartle, Phys. Rev. D43 (1991) 1170.
72. K. V. Kuchař and M. P. Ryan, Phys. Rev. D40 (1989) 3982.
73. S. Sinha and B. L. Hu, Phys. Rev. D44 (1991) 1028.
74. F. D. Mazzitelli, Phys. Rev. D46 (1992) 4758.
75. A. Ishikawa and T. Isse, Mod. Phys. Lett. A8 (1993) 3413.
56
76. M. A. H. MacCallum, in General Relativity: An Einstein Centenary Survey
eds. S. W. Hawking and W. Israel, (Cambridge University Press, 1979), p.
533.
77. A. S. Kompaneets and A. S. Chernov, Zh. Eksp. Teor. Fiz. 47 (1964) 1939
[Sov. Phys. JETP 20 (1965) 1303].
78. R. Kantowski and R. K. Sachs, J. Math. Phys. 7 (1966) 443.
79. L. Bianchi, Mem. di Mat. Soc. Ital. Sci. 11 (1897) 267.
80. C. W. Misner, Phys. Rev. Lett. 22 (1969) 1071.
81. V. A. Belinskii, E. M. Lifschitz and I. M. Khalatnikov, Usp. Fiz. Nauk. 102
(1970) 463 [Sov. Phys. Usp. 13 (1970) 745].
82. A. Latifi, M. Musette and R. Conte, Phys. Lett. 194A (1994) 83.
83. J. Louko, Ann. Phys. (N.Y.) 181 (1988) 318.
84. J. J. Halliwell, Phys. Rev. D38 (1988) 2468.
85. I. G. Moss, Ann. Inst. H. Poincaré 49 (1988) 341.
86. D. N. Page, J. Math. Phys. 32 (1991) 3427.
87. T. Banks, Nucl. Phys. B249 (1985) 332.
88. S. P. Kim, Phys. Rev. D52 (1995) 3382.
89. J. J. Halliwell, Phys. Rev. D39 (1989) 2912.
90. E. Calzetta and J. J. Gonzalez, Phys. Rev. D51 (1995) 6821.
91. S. Habib and R. Laflamme, Phys. Rev. D42 (1990) 4056.
92. E. Calzetta, Phys. Rev. D43 (1991) 2498.
93. B. L. Hu, J. P. Paz and S. Sinha, in Directions in General Relativity, Vol. I,
eds. B. L. Hu, M. P. Ryan and C. V. Vishveshwara, (Cambridge University
Press, 1993), p. 145.
94. B. L. Hu, in [44], p. 475.
95. J. J. Halliwell, Phys. Rev. D36 (1987) 3626.
96. K. V. Kuchař, J. Math. Phys. 22 (1981) 2640.
97. M. P. Ryan, Hamiltonian Cosmology (Springer Lecture Notes in Physics 13,
Berlin, 1972).
98. C. Teitelboim, Phys. Rev. Lett. 38 (1977) 1106.
99. R. Tabensky and C. Teitelboim, Phys. Lett. 69B (1977) 453.
100. A. Macı́as, O. Obregón and M. P. Ryan, Class. Quantum Grav. 4 (1987)
1477.
101. P. D. D’Eath and D. I. Hughes, Phys. Lett. 214B (1988) 498; Nucl. Phys.
B378 (1992) 381.
102. R. Graham, Phys. Rev. Lett. 67 (1991) 1381; Phys. Lett. 277B (1992) 393.
57
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
P. D. D’Eath, S. W. Hawking and O. Obregón, Phys. Lett. 300B (1993) 44.
A. Vilenkin, Phys. Rev. D39 (1989) 1116.
D. N. Page, Phys. Rev. D34 (1986) 2267.
J. B. Hartle, Phys. Rev. D37 (1988) 2818, D43 (1991) 1434.
D. N. Page, Phys. Rev. Lett. 70 (1993) 4034.
S. W. Hawking and D. N. Page, Nucl. Phys. B298 (1988) 789.
E. W. Kolb and M. S. Turner, The Early Universe, (Addison-Wesley, Reading, Mass., 1990).
D. L. Wiltshire, Phys. Rev. D36 (1987) 1634.
M. Henneaux, Lett. Nuovo Cimento 38 (1983) 609.
G. W. Gibbons, S. W. Hawking and J. M. Stewart, Nucl. Phys. B281 (1987)
736.
W. M. Suen and K. Young, Phys. Rev. D39 (1989) 2201.
H. D. Conradi and H. D. Zeh, Phys. Lett. 154A (1991) 321.
H. D. Conradi, Phys. Rev. D46 (1992) 612; Class. Quantum Grav. 12 (1995)
2423.
G. W. Lyons, Phys. Rev. D46 (1992) 1546.
J. J. Halliwell and J. B. Hartle, Phys. Rev. D41 (1990) 1815.
G. W. Gibbons and L. P. Grishchuk, Nucl. Phys. B313 (1989) 736.
D. N. Page, in Quantum Concepts in Space and Time, eds. R. Penrose and
C. J. Isham, (Oxford University Press, 1986) p. 274.
D. J. Kaup and A. P. Vitello, Phys. Rev. D9 (1974) 1648.
S. W. Hawking and D. N. Page, Phys. Rev. D42 (1990) 2655.
D. N. Page, Class. Quantum Grav. 7 (1990) 1841.
J. J. Halliwell and R. C. Myers, Phys. Rev. D40 (1989) 4011.
I. Klebanov, L. Susskind, and T. Banks, Nucl. Phys. B317 (1989) 665.
A. Vilenkin, in Proceedings of the International School of Astrophysics ‘D.
Chalonge’, Eriche, 1995, ed. N. Sanchez, (Kluwer, Dordrecht), to appear.
J. Louko and T. Vachaspati, Phys. Lett. 223B (1989) 21.
A. Vilenkin, Phys. Rev. Lett. 74 (1995) 846.
A. O. Barvinsky and A. Yu. Kamenshchik, Class. Quantum Grav. 7 (1990)
L181.
A. O. Barvinsky, Phys. Rep. 230 (1993) 237.
L. P. Grishchuk and L. V. Rozhansky, Phys. Lett. 234B (1990) 9.
A. Lukas, Phys. Lett. 347B (1995) 13.
L. P. Grishchuk and L. V. Rozhansky, Phys. Lett. 208B (1988) 369.
58
133.
134.
135.
136.
137.
138.
139.
140.
141.
142.
143.
144.
145.
146.
147.
148.
149.
J. J. Halliwell and S. W. Hawking, Phys. Rev. D31 (1985) 1777.
S. Wada, Nucl. Phys. B276 (1986) 729.
J. J. Halliwell and P. D’Eath, Phys. Rev. D35 (1985) 1100.
R. Laflamme, Phys. Lett. 198B (1987) 156.
S. Wada, Phys. Rev. Lett. 59 (1987) 2375.
T. Vachaspati, Phys. Lett. 217B (1989) 228.
T. Vachaspati and A. Vilenkin, Phys. Rev. D37 (1988) 898.
T. S. Bunch and P. C. W. Davies, Proc. R. Soc. Lond. A360 (1978) 117.
B. Allen, Phys. Rev. D32 (1985) 3136.
S. W. Hawking, Phys. Lett. 115B (1982) 295.
A. H. Guth and S. Y. Pi, Phys. Rev. Lett. 49 (1982) 1110; Phys. Rev. D32
(1985) 1899.
K. M. Gorski, G. Hinshaw, A. J. Banday, C. L. Bennett, E. L. Wright, A.
Kogut, G. F. Smoot, and P. Lubin, Astrophys. J. 430 (1994) L89.
T. Gold (ed.), The Nature of Time, (Cornell University Press, Ithaca, 1967).
A. Higuchi and R. M. Wald, Phys. Rev. D51 (1995) 544.
S. C. Beluardi and R. Ferraro, Phys. Rev. D52 (1995) 1963.
D. Marolf, Class. Quantum Grav. 12 (1995) 2469.
S. W. Hawking, Phys. Rev. D32 (1985) 2489.
150. T. Gold, in La Structure at l’Évolution de l’Univers, 11th Solvay Conference,
(Edition Stoops, Brussels, 1958), p. 81; Am. J. Phys. 30 (1962) 403.
151. S. W. Hawking and Z. C. Wu, Phys. Lett. 151B (1985) 15.
152. S. W. Hawking, Vist. Astr. 37 (1993) 559; and in [44], p. 346.
153. D. N. Page, Phys. Rev. D32 (1985) 2496.
154. R. Laflamme and E. P. S. Shellard, Phys. Rev. D35 (1987) 2315.
155. R. Laflamme, Time and Quantum Cosmology, (Ph.D. thesis, University of
Cambridge, 1988); in [44], p. 358.
156. S. W. Hawking, R. Laflamme and G. W. Lyons, Phys. Rev. D47 (1993)
5342.
157. C. Kiefer and H. D. Zeh, Phys. Rev. D51 (1995) 4145.
158. P. D’Eath, Phys. Rev. D48 (1993) 713.
159. S. W. Hawking, Phys. Rev. D37 (1988) 904.
160. R. Graham and H. Luckock, Phys. Rev. D49 (1994) R4981.
161. B. de Wit, H. Nicolai and H. J. Matschull, Phys. Lett. 318B (1994) 115.
162. S. M. Carroll, D. Z. Freedman, M. E. Ortiz and D. N. Page, Nucl. Phys.
B423 (1994) 661.
59