Why Does Inflation Start at the Top of the Hill?

S.W. Hawking∗ , Thomas Hertog† ,

DAMTP, Centre for Mathematical Sciences, University of Cambridge
Wilberforce Road, Cambridge CB3 0WA, United Kingdom.
(February 1, 2008)

Abstract
arXiv:hep-th/0204212v1 25 Apr 2002

We show why the universe started in an unstable de Sitter state. The

quantum origin of our universe implies one must take a ‘top down’ approach
to the problem of initial conditions in cosmology, in which the histories that
contribute to the path integral, depend on the observable being measured.
Using the no boundary proposal to specify the class of histories, we study
the quantum cosmological origin of an inflationary universe in theories like
trace anomaly driven inflation in which the effective potential has a local
maximum. We find that an expanding universe is most likely to emerge in
an unstable de Sitter state, by semiclassical tunneling via a Hawking-Moss
instanton. Since the top down view is forced upon us by the quantum nature
of the universe, we argue that the approach developed here should still apply
when the framework of quantum cosmology will be based on M-Theory.

I. INTRODUCTION

Structure and complexity have developed in our universe, because it is out of equilib-
rium. This feature shows up in all known cosmological scenarios for the early universe,
which rely on gravitational instability to generate local inhomogeneities from an almost ho-
mogeneous and isotropic state for the universe. Inflation seems the best explanation for this
homogeneous and isotropic state because whatever drives the inflation will remove the local
instability and iron out irregularities. However the inflationary expansion has to be globally
unstable because otherwise it would continue forever and galaxies would never form.
The instability can be described as the evolution of an order parameter φ which can be
treated as a scalar field with effective potential V (φ). If V ′ /V is small, φ will roll slowly
down the potential and the universe will inflate by a large factor. However, this raises the
question: Why did the universe start with a high value of the potential? Why didn’t φ start
at the global minimum of V ?

∗ email: [email protected]

† email: [email protected]

1
 There have been various attempts to explain why φ started high on the potential hill. In
the old [1] and new [2,3] inflationary scenarios the universe was supposed to start with infinite
temperature at a singularity. As the universe expanded and cooled, thermal corrections
would make the effective potential time dependent. So even if φ started in the minimum of
V , it could still end up in a metastable false vacuum state (in old inflation) or at a local
maximum of V (in new inflation). The scalar field was then supposed to tunnel through
the potential barrier or just fall off the top of the hill and slowly roll down. However both
scenarios tended to predict a more inhomogeneous universe than we observe. They were also
unsatisfactory because they assumed an initial singularity and a fairly homogeneous and
isotropic pre-inflation hot big bang phase. Why not just assume the singularity produced
the standard hot big bang, since we don’t have a measure on the space of singular initial
conditions for the universe?
In the chaotic inflation scenario [5], quantum fluctuations of φ are supposed to drive
the volume weighted average φ up the potential hill, leading to everlasting eternal inflation.
However this effect is dependent on using the synchronous gauge: in other gauges the vol-
ume weighted average of the potential can go down. Looking from a 4 rather than 3+1
dimensional perspective, it is clear that the quantum fluctuations of a single scalar field are
insufficient to drive de Sitter like eternal inflation, if the de Sitter space is larger than the
Planck length. Eternal inflation may be possible at the Planck scale, but all our methods
would break down in this situation so it would mean that we could not analyze the origin
of the universe.
The aim of this paper however is to show that the universe can come into being and
start inflating without the need for an initial hot big bang phase or Planck curvature. It is
required that the potential V has a local maximum which is below the Planck density and
sufficiently flat on top, V ′′ /V > −4/3. This last condition means only the homogeneous
mode of the scalar field is tachyonic: the higher modes all have positive eigenvalues. It also
means there isn’t a Coleman-De Luccia solution [6] describing quantum tunneling from a
false vacuum on one side of the maximum to the true vacuum on the other side. Instead
there is only a homogeneous Hawking-Moss instanton [7] that sits on the top of the hill, at
the local maximum of V .
It has long been a problem to understand how the universe could decay from a false
vacuum in this situation. The Hawking-Moss instanton does not interpolate between the
false and true vacua, because it is constant in space and time. Instead, what must happen
is that the original universe can continue in the false vacuum state but that new completely
disconnected universes can form at the top of the hill via Hawking-Moss instantons. For
someone in one of these new universes, the universe in the false vacuum is irrelevant and
can be ignored.
The top of the hill might seem the least likely place for the universe to start. However we
shall show it is the most likely place for an inflationary universe to begin, if V ′′ /V > −4/3.
The reason is that although being at the top of the hill costs potential action, the saving of
gradient action from having a constant scalar field is greater. Thus inflation will start at the
top of the hill. In particular, this justifies Starobinsky’s scenario of trace anomaly inflation,
in which the universe starts in an unstable de Sitter state supported by the conformal
anomaly of a large number of conformally coupled matter fields [4].
The usual approach to the problem of initial conditions for inflation, is to assume some

2
initial configuration for the universe, and evolve it forward in time. This could be described
as the bottom up approach to cosmology. It is an essentially classical picture, because it
assumes there is a single well defined metric for the universe. By contrast, here we adopt a
quantum approach, based on the no boundary proposal [8], which states that the amplitude
for an observable like the 3-metric on a spacelike hypersurface Σ, is given by a path integral
over all metrics whose only boundary is Σ. The quantum origin of our universe and the no
boundary proposal naturally lead to a top down view of the universe, in which the histories
that contribute to the path integral, depend on the observable being measured.
We study the quantum cosmological origin of an expanding universe in theories like trace
anomaly inflation, by investigating the semiclassical predictions of the no boundary proposal
for the wave function of interest. One may argue that a clearer picture of the pre-inflationary
conditions can only emerge from a deeper understanding of quantum gravity at the Planck
scale. However, the amplitude of the cosmic microwave temperature anisotropies indicates
that the universe may always have been much larger than the Planck scale. This suggests
it might be possible to describe the origin of our universe within the semiclassical regime of
quantum cosmology. Correspondingly, the effective potential must have a local maximum
well below the Planck density, which is the case in the trace anomaly model.
The paper is organised as follows. In section 2 we review trace anomaly driven inflation,
since this provides an important theoretical motivation for inflation. We study the quantum
cosmology of the trace anomaly model and discuss the role of a special class of instanton
saddle-points of the no boundary path integral, which can be analytically continued to
Lorentzian universes. In section 3 we consider perturbations in anomaly-induced inflation
and show that the instability of the inflationary phase can be described by a scalar field
with an effective potential with a local maximum. We also discuss homogeneous fluctuations
about the instanton backgrounds and touch briefly on the effect of quantum matter on the
spectrum of microwave fluctuations predicted by anomaly-induced inflation. In section 4, we
consider a general model of inflation with an effective potential that has a local maximum.
We show that according to the no boundary proposal, provided the instability is sufficiently
weak, an expanding universe is most likely to start at the top of the hill, in a de Sitter state.
Finally, in section 5 we present our conclusions.

II. TRACE ANOMALY DRIVEN INFLATION

A. Large N Cosmology

It has been argued that the theoretical foundations for inflation are weak, since it has
proven difficult to realise inflation in classical M-theory. A large class of supergravity theories
admit no warped de Sitter compactifications on a compact, static internal space [9,10] and
although some gauged N = 8 and N = 4 supergravities in D = 4 do permit de Sitter vacua
[11,12], these vacua are too unstable for a significant period of inflation to occur. However,
an appealing way to evade the no go theorems is to include higher derivative quantum
corrections to the classical supergravity equations, such as the trace anomaly.
Since we observe a large number of matter fields in the universe, it is natural to consider
the large N approximation [13]. In the large N approximation, one performs the path

3
integral over the matter fields in a given background to obtain an effective action that is a
functional of the background metric,
Z
exp(−W [g]) = d[φ] exp(−S[φ; g]). (2.1)

In the leading-order 1/N approximation, one can neglect graviton loops and look for a sta-
tionary point of the effective action for the matter fields combined with the gravitational
action. This is equivalent to solving the Einstein equations with the source being the ex-
pectation value of the matter energy-momentum tensor derived from W ,
1
Rµν − Rgµν = 8πGhTµν i. (2.2)
2
The expectation value of the energy-momentum tensor is generally non-local and depends
on the quantum state. However, during inflation, particle masses are small compared with
the spacetime curvature, R >> m2 , and in asymptotically free gauge theories, interactions
become negligible in the same limit. Therefore, at the high curvatures during inflation, the
energy-momentum tensor of a large class of grand unified theories is to a good approximation
given by the expectation value hTµν i of a large number of free, massless, conformally invariant
fields1 . The entire one-loop contribution to the trace of the energy-momentum tensor then
comes from the conformal anomaly [15], which is given for a general CFT by the following
equation,

g µν hTµν i = cF − aG + α′ ∇2 R, (2.3)

where F is the square of the Weyl tensor, G is proportional to the Euler density and the
constants a, c and α′ are given in terms of the field content of the CFT by
1
a= (NS + 11NF + 62NV ) , (2.4)
360(4π)2

1
c= (NS + 6NF + 12NV ) , (2.5)
120(4π)2

1
α′ = (NS + 6NF − 18NV ) , (2.6)
180(4π)2
with NS the number of real scalar fields, NF the number of Dirac fermions and NV the
number of vector fields2 .

1 Forsimplicity, it is assumed that scalar fields become conformally coupled at high energies, but
the contribution of the interaction terms to hTµν i is small at high curvature, as long as the couplings
don’t become very large [14].

2 We have quoted the value for α′ predicted by AdS/CFT, which agrees with point-splitting or
zeta function regularisation [16].

4
 The trace anomaly is entirely geometrical in origin and therefore independent of the
quantum state. In a maximally symmetric spacetime, the symmetry of the vacuum implies
that the expectation value of the energy-momentum tensor is proportional to the metric,
1
h0|Tµν |0i = gµν g ρσ h0|Tρσ |0i. (2.7)
4
Thus the trace anomaly acts just like a cosmological constant for these spacetimes, and a
positive trace anomaly permits a de Sitter solution to the Einstein equations.
The radius of the de Sitter solution is determined by the number of fields, N 2 , in the
CFT and is of order ∼ Nlpl . Therefore the one-loop contributions to the energy-momentum
tensor are ∼ 1/N 2 , which means they are of the same order as the classical terms in the
Einstein equations. On the other hand, the corrections due to graviton loops are ∼ 1/N 3 , so
for large N quantum gravitational fluctuations are suppressed, confirming the consistency
of the large N approximation.
For α′ = 0 in 2.3, the only O(3, 1) invariant solutions are de Sitter space and flat
space, which are the initial and final stages of the simplest inflationary universe. In order
for a solution to exist that interpolates between these two stages, one must have α′ < 0
in 2.3, as Starobinsky discovered [4]. Starobinsky showed that if α′ < 0, the de Sitter
solution is unstable, and decays into a matter dominated Friedman-Lemaitre Robertson-
Walker universe, on a timescale determined by α′ . The purpose of Starobinsky’s work was
to demonstrate that quantum effects of matter fields might resolve the Big Bang singularity.
From a modern perspective, it is more interesting that the conformal anomaly might have
been the source of a finite, but significant period of inflation in the early universe. Rapid
oscillations in the expansion rate at the end of inflation, would result in particle production
and (p)reheating.
Starobinsky showed that the de Sitter solution is unstable both to the future and to the
past, so it was not clear how the universe could have entered the de Sitter phase. This is the
problem of initial conditions for trace anomaly driven inflation, which should be addressed
within the framework of quantum cosmology, by combining inflation with a theory for the
wave function Ψ of the quantum universe. Hartle and Hawking suggested that the amplitude
for the quantum state of the universe described by 3-metric h and matter fields φ(x) on a
3-surface Σ, should be given by
XZ
Ψ[Σ, h, φΣ ] = N D[g]D[φ(x)]e−SE (g,φ) , (2.8)
M

where the Euclidean path integral is taken over all compact four geometries bounded only
by a 3-surface Σ, with induced metric h and matter fields φΣ . M denotes a diffeomorphism
class of 4-manifolds and N is a normalisation factor. The motivation to restrict the class
of manifolds and metrics to geometries with only a single boundary is that in cosmology, in
contrast with scattering calculations, one is interested in measurements in a finite region in
the interior of spacetime. The ‘no boundary’ proposal gives a definite ansatz for the wave
function Ψ[Σ, h, φΣ ] of the universe and in principle removes the initial singularity in the
hot Big Bang model. At least within the semiclassical regime, this yields a well-defined
probability measure on the space of initial conditions for cosmology.
One can appeal to quantum cosmology to explain how the de Sitter phase emerges in
trace anomaly inflation, since the no boundary proposal can describe the creation of an

5
inflationary universe from nothing. At the semiclassical level, this process is mediated by a
compact instanton saddle-point of the Euclidean path integral, which extrapolates to a real
Lorentzian universe at late times. To find the relative probability of different geometries
in the no boundary path integral, one must compute their Euclidean action. In the next
section, we consider a model of anomaly-induced inflation consisting of gravity coupled to
N = 4, U(N) super Yang-Mills theory, for which the AdS/CFT correspondence [17] provides
an attractive way to calculate the effective matter action on backgrounds without symmetry.
The fact that we are using N = 4, U(N) super Yang-Mills theory is probably not significant
since, as we shall describe, it is the large number of fields that matters in our discussion and
not the Yang-Mills coupling. Therefore, we expect our results to be valid for any matter
theory that is approximately massless during the de Sitter phase.

B. Effective Matter Action

We consider, in Euclidean signature, Einstein gravity coupled to a N = 4, U(N) super

Yang-Mills theory with large N,
1
Z
4 √ 1
Z √
S=− d x gR − d3 x hK + W, (2.9)
2κ κ
where W denotes the Yang-Mills effective action. The field content of the Yang-Mills theory
is NS = 6N 2 , NF = 2N 2 and NV = N 2 , yielding an anomalous trace

N2
g µν hTµν i = (F − G). (2.10)
64π 2
The one-loop result for the conformal anomaly is exact, since it is protected by supersymme-
try. Therefore, inflation supported by the trace anomaly of N = 4, U(N) super Yang-Mills
would never end. The presence of non-conformally invariant fields in realistic matter theo-
ries, however, necessarilly alters the value of α′ in the anomaly 2.3. Since the coefficient of
the ∇2 R term plays such an important role in trace anomaly driven inflation, we ought to
include this correction. As a first approximation, one can account for the non-conformally
invariant fields by adding a local counterterm to the action,

αN 2 Z 4 √ 2
Sct = d x gR . (2.11)
192π 2
This leads to an extra contribution to the conformal anomaly, which becomes

µν N2 αN 2 2
g hTµν i = (F − G) + ∇ R. (2.12)
64π 2 16π 2
For α < 0, the expansion now changes from exponential to the typical power law ∼ t2/3
of a matter dominated universe, on a time scale ∼ 12|α| log N. One can construct more
sophisticated models of anomaly driven inflation, by taking in account corrections from
particle masses and interactions in a more precise way. One could, for instance, consider
soft supersymmetry breaking during inflation. The coefficient α′ could then vary in time,
because the decoupling of massive sparticles at low energy [18] alters the number of degrees

6
of freedom that contribute to the quantum effective action. For our purposes, however, it is
sufficient to consider the theory above.
In no boundary cosmology, one is interested in solutions that describe a Lorentzian in-
flationary universe that emerges from a compact instanton solution of the Euclidean field
equations. These geometries provide saddle-points of the Euclidean path integral 2.8 for
the wave function of interest. Because our universe is Lorentzian at late times, it has been
suggested that the relevant instanton saddle-points of the no boundary path integral are
so-called ‘real tunneling’ geometries [19,20]. Cosmological real tunneling solutions are com-
pact Riemannian geometries joined to an O(3, 1) invariant Lorentzian solution of Einstein’s
equations, across a hypersurface of vanishing extrinsic curvature Kµν . Such instanton solu-
tions can then be used as background in a perturbative evaluation of the no boundary path
integral, to find correlators of metric perturbations during inflation, which in turn determine
the cosmic microwave anisotropies.
We now compute the effective matter action W on such perturbed instanton metrics.
After eliminating the gauge freedom, the perturbed metric on the spaces of interest can be
written as

ds2 = B 2 (χ)γµν dxµ dxν = B 2 (χ)((1 + ψ)γ̂µν + θµν )dxµ dxν , (2.13)

where γ̂µν is the metric on the unit S 4 and θµν is transverse and traceless with respect to
the four sphere.
In order to evaluate the no boundary path integral, we must first compute the quantum
effective action W [B, h] on the background 2.13. The effective action of the matter fields is
computed as an expansion around the homogeneous background with metric gµν = B 2 (χ)γ̂µν .
To second order in the metric perturbation, W [B, h] is determined by the one and two-point
function of the energy-momentum tensor on the unperturbed O(4) invariant background.
The one-point function is given by the conformal anomaly. Since the FLRW background is
conformal to the round four sphere, the two-point function can be calculated by a conformal
transformation from S 4 . On S 4 , the two-point function is determined entirely by symme-
try and the trace anomaly [23]. Therefore, since the energy-momentum tensor transforms
anomalously, the two-point function on 2.13 should be fully determined by the two-point
function on S 4 , the trace anomaly and the scale factor B(χ). For the matter theory we have
in mind, all these quantities are independent of the coupling, so it follows that the effective
action W [B, h] is independent of the coupling, to second order in the metric perturbation.
In [24], it was found how the effective action that generates a conformal anomaly of the
form 2.3, transforms under a conformal transformation. We can use this result to relate
W [B, h] on the perturbed FLRW space to W [r, h] on the perturbed four sphere with radius
r. Writing B(χ) = r eσ(χ) , where r is an arbitrary radius, the transformation is given by

N2 Z 4 √ 1

W [σ(χ), h] = W̃ [r, h] − 2
d x γ σ(Rµν Rµν − R2 ) + 2∇µ σ∇µ σ∇2 σ
32π 3
1

µν µν µ 2
+2(R − γ R)∇µ σ∇ν σ + (∇µ σ∇ σ) (2.14)
2

Here W̃ denotes the effective action on the perturbed four sphere of radius r with metric
γµν , and the Ricci scalar R and covariant derivative ∇µ refer to the same space.

7
 The generating functional W̃ [r, h] was computed in [25,23], by using the AdS/CFT cor-
respondence [17],
Z Z
Z[h] ≡ d[g] exp(−Sgrav [g]) = d[φ] exp(−SCF T [φ; h]) ≡ exp(−WCF T [h]), (2.15)

where Z[h] is the supergravity partition function on AdS5 . The AdS/CFT calculation is
performed by introducing a fictional ball of (Euclidean) AdS that has the perturbed sphere
as its boundary. In the classical gravity limit, the CFT generating functional can then be
obtained by solving the IIB supergravity field equations, to find the bulk metric g that
matches onto the boundary metric h, and adding a number of counterterms that depend
on the geometry of the boundary, in order to render the action finite as the boundary is
moved off to infinity. To second order in the perturbation h, the quantum effective action
(including the R2 counterterm) is given by
W̃ = W̃ (0) + W̃ (1) + W̃ (2) + . . . (2.16)
where
3βN 2 Ω4 3αN 2 Ω4 3N 2 Ω4
W̃ (0) = − + + (4 log 2 − 1) , (2.17)
8π 2 4π 2 32π 2
3N 2
Z q
W̃ (1) = d4 x γ̂ ψ, (2.18)
16π 2 r 2
3N 2
Z q h 
ˆ 2 + 2 ψ − αψ ∇ ˆ 4 + 4∇
ˆ2 ψ
   i
W̃ (2) = − d 4
x γ̂ ψ ∇
64π 2 r 4
N2 X
Z q 2
4 ′ µν ′ (p) ′
+ d x γ̂ θ (x )Hµν (x ) (Ψ(p) − 4αp(p + 3)) ,
256π 2 r 4 p
ˆ 2 on the round four sphere and
where p labels the eigenvalues of the Laplacian ∇
Ψ(p) = p(p + 1)(p + 2)(p + 3) [ψ(p/2 + 5/2) + ψ(p/2 + 2) − ψ(2) − ψ(1)]
+p4 + 2p3 − 5p2 − 10p − 6 + 2βp(p + 1)(p + 2)(p + 3), (2.19)
and we have allowed for a finite contribution, with coefficient β, of the third counterterm,
which is necessary to cancel a logarithmic divergence of the tensor perturbation. Inserting
the expression for W̃ in 2.14 and evaluating the terms depending on the scale factor σ(χ),
one obtains the quantum effective action of the Yang-Mills teory on a general, perturbed
FLRW geometry. For completeness, we also give the Einstein-Hilbert action of the perturbed
four sphere,
3Ω4 r 2 3
Z q
SEH = − − d4 x γ̂ψ
4πG Z4πG
1 q 3 1 µν ˆ 2

+ 4
d x γ̂ ˆ 2 µν
ψ ∇ ψ + 2θ θµν − θ ∇ θµν . (2.20)
16πGr 2 2 4
We shall use these results in section III, where we discuss the instability of anomaly-
induced inflation. But first, we return to the background evolution. In the next paragraph,
we discuss a class of O(4) invariant ‘real tunneling’ instanton solutions of the Starobinsky
model 2.9 and study their role in the no boundary path integral for the wave function of an
inflationary universe.

8
 C. Real Tunneling Geometries

It is easily seen that the total action is stationary under all perturbations hµν , if the
background is a round four sphere with radius
N 2G
rs2 = . (2.21)
4π
By slicing the four sphere at the equator χ = π/2 and writing χ = π2 − it, it analytically con-
tinues into the Lorentzian to the de Sitter solution mentioned above, with the cosmological
constant provided by the trace anomaly of the large N Yang-Mills theory.
Other compact, real instanton solutions of the form

ds2 = dτ 2 + b2 (τ )dΩ23 (2.22)

were found in [23], by numerically integrating the Einstein equations, which can be obtained
directly from the trace anomaly by using energy-momentum conservation. Imposing regular-
ity at the North Pole (at τ = 0) of the instanton leaves only the third derivative of the scale
factor at the North Pole as an adjustable parameter. It is convenient to define dimensionless
variables τ̃ = τ /rs and f (τ̃ ) = b(τ )/rs . For α < 0, there exists a second regular, compact
‘double bubble’ instanton, with f ′′′ (0) = −2.05, together with a one-parameter family of
instantons with an irregular South Pole. For f ′′′ (0) < −1, the scale factor of the latter
has two peaks. For −1 < f ′′′ (0) < 0 on the other hand, they are similar to the singular
Hawking-Turok instantons that have been considered in the context of scalar field inflation
[26].
The Lorentzian part of the real tunneling saddle-points is obtained by analytically contin-
uing the instanton metric across a hypersurface of vanishing extrinsic curvature. The double
bubble instanton can be continued across its ‘equator’ to give a closed FLRW universe, or
into an open universe by a double continuation across the South Pole. Our numerical studies
show that the closed universe rapidly collapses and that the open spacetime hyper-inflates,
with the scale factor blowing up at a finite time. Similarly, the singular instantons can be
continued into an open FLRW universe across τ = 0, by setting τ = it and Ω3 = iφ. For
f ′′′ (0) < −1 this again gives hyper-inflation, but for −1 < f ′′′ (0) < 0 one obtains a realistic
inflationary universe. The four sphere solution as well as the singular instantons that are
small perturbations of S 4 at the regular pole, are most interesting for cosmology, since they
yield long periods of inflation.
Using the expressions 2.14 and 2.17 for W [σ(χ)] and the relations
dτ ′
" !#
τ b(τ )
Z
−1
χ(τ ) = 2 lim tan tan(ǫ/2) exp , B(τ ) = , (2.23)
ǫ→0 ǫ b(τ ′ ) sin(χ)
one can numerically compute the action of the real tunneling geometries [27]. On an unper-
turbed FLRW background, conformal to the round four sphere, the total Euclidean action
becomes
3N 2 Ω3 1
Z 
(0)
S = 2
dχ sin3 χ (12(log 2 + σ − β) − 3 + 6σ ′2 − σ ′4 − 4σ ′3 cot χ)
32π 3 i
−e (σ + 2) + 2α(σ ′′ + 3σ ′ cot χ + σ ′2 − 2)2
2σ ′2
(2.24)

9
where σ = log(B/r). On the round four sphere, σ → 0, so the action reduces to

3N 2 Ω4
S (0) = (8α − 3 + 4(log 2 − β)) (2.25)
32π 2
We find that for all α < 0 the regular double bubble instanton has much lower action
than the four sphere. The singular double bubble instantons have divergent action, but the
Hawking-Turok type instantons have finite action3 . For given α, the action of the latter
class depends on the third derivative of the scale factor at τ = 0. This is the analogue of the
situation in scalar field inflation, where the action of the Hawking-Turok instantons depends
on the value of the inflaton field at the North Pole. The action of the singular instantons
tends smoothly to the S 4 action 2.25 as f ′′′ (0) → −1 and it decreases monotonically with
increasing f ′′′ (0).
To summarize, we found a one-parameter family of finite-action, compact solutions of the
Euclidean field equations that can be analytically continued across a spacelike surface Σ of
vanishing curvature, to Lorentzian geometries that describe realistic inflationary universes.
The condition on Σ guarantees that a real solution of the Euclidean field equations is contin-
ued to a real Lorentzian spacetime. The Euclidean region is essential, since there is no way
to round off a Lorentzian geometry without introducing a boundary. What is the relevance
then, in the context of the no boundary proposal, of these real tunneling geometries with
regard to the problem of initial conditions in cosmology?
At least at the semiclassical level, the no boundary proposal gives a measure on the space
of initial conditions for cosmology. The weight of each classical trajectory is approximately
|Ψ|2 ∼ e−2SR , where SR is the real part of the Euclidean action of the solution. For real
tunneling solutions this comes entirely from the part of the manifold on which the geometry
is Riemannian. The simplicity of this situation has led to the interpretation of the no
boundary proposal as a bottom up theory of initial conditions. In particular, it has been
argued that if a given theory allows different instantons, the no boundary proposal predicts
our universe to be created through the lowest-action solution, since this would give the
dominant contribution to the path integral. Applying this interpretation to trace anomaly
driven inflation, one must conclude that the no boundary proposal predicts the creation of a
hyper-inflating universe emerging from the double bubble instanton, or a nearly empty open
universe that occurs by semiclassical tunneling via a singular instanton with |f ′′′ (0)| small.
The situation is similar in many theories of scalar field inflation. Restricting attention to
real tunneling geometries, a bottom up interpretation of the no boundary proposal generally
favours the creation of large spacetimes. One typically obtains a probability distribution
that is peaked around instantons in which the field at the surface of continuation is near the
minimum of its potential, yielding very little inflation. Hence, the most probable universes
are nearly empty open universes or collapsing closed universes, depending on the analytic
continuation one considers. Weak anthropic arguments have been invoked to try to rescue
the situation [26], by weighing the a priori no boundary probability with the probability of

3 For completeness, we should mention that if α > 0 one must have f ′′′ (0) ≤ −1 in order for the
solution to be compact. For f ′′′ (0) < −1, the instantons have a singular South Pole but finite
action, and continue to hyper-inflating open universes.

10
the formation of galaxies. However, for the most natural inflaton potentials, this still predicts
a value of Ω0 that is far too low to be compatible with observations. Another attempt [28],
based on introducing a volume factor that represents the projection onto the subset of states
containing a particular observer, leads to eternal inflation at the Planck density, where the
theory breaks down. In fact, invoking conditional probabilities is contrary to the whole idea
of the no boundary proposal, which by itself specifies the quantum state of the universe.
Clearly the predictions of a bottom up interpretation of the no boundary proposal do
not agree with observation. This is because it is an essentially classical interpretation, which
is neither relevant nor correct for cosmology. The quantum origin of the universe implies its
quantum state is given by a path integral. Therefore, one must adopt a quantum approach
to the problem of initial conditions, in which one considers the no boundary path integral
2.8 for a given quantum state of the universe. We shall apply such a quantum approach in
section IV, to describe the origin of an inflationary universe, in theories like trace anomaly
inflation. It turns out that the relevant saddle-points are not exactly real tunneling geome-
tries. Instead, one must consider complex saddle points, in which the geometry becomes
gradually Lorentzian at late times.

III. INSTABILITY OF ANOMALY-INDUCED INFLATION

A. Metric Perturbations

Two-point functions of metric perturbations can be computed directly from the no

boundary path integral. One perturbatively evaluates the path integral around an O(4)
invariant instanton background to obtain the real-space Euclidean correlator, which is then
analytically continued into the Lorentzian universe, where it describes the quantum fluctu-
ations of the graviton field in the primordial de Sitter phase [21,22]. The quantum state
of the Lorentzian fluctuations is uniquely determined by the condition of regularity on the
instanton [23]. Both scalar and tensor perturbations are given by a path integral of the form
Z
′
hhµν (x)hµ′ ν ′ (x )i ∼ d[h] exp(−S (2) )hµν (x)hµ′ ν ′ (x′ ), (3.26)

where S (2) denotes the second order perturbation of the action

S = SEH + SGH + SR2 + W̃ , (3.27)

with W̃ given by 2.16. For the scalars, eliminating the remaining gauge freedom introduces
Faddeev-Popov ghosts. These ghosts supply a determinant (∇ ˆ 2 + 4)−1 , which cancels a
similar factor in the scalar action, rendering it second order4 . The action for the tensors θµν
on the other hand is non-local and fourth order. Nevertheless, the metric perturbation and its
first derivative should not be regarded as two independent variables, since this would lead
to meaningless probability distributions in the Lorentzian [29]. Instead the path integral

4 The gauge freedom also leads to closed loops of Faddeev-Popov ghosts but they can be neglected
in the large N approximation.

11
should be taken over the fields θµν only5 , to compute correlators of the form 3.26. The
Euclidean action for θµν is positive definite, so the path integral over all θµν converges and
determines a well-defined Euclidean quantum field theory. One might worry that the higher
derivatives would lead to instabilities in the Lorentzian. This is not the case, however, since
the no boundary prescription to compute Lorentzian propagators by Wick rotation from the
Euclidean, implicitly imposes the final boundary condition that the fields remain bounded,
which eliminates the runaways [23,29].
The path integral 3.26 is Gaussian, so the correlation functions can be read off from the
perturbed action, equation 2.19 and 2.20:

32π 2 rs4  ˆ 2 −1

hψ(x)ψ(x′ )i = −∇ + 1/2α , (3.28)
3|α|N 2

and
(p)
′
∞
128π 2 rs4 X Wµνµ′ ν ′ (x, x′ )
hθµν (x)θµ′ ν ′ (x )i = , (3.29)
N 2 p=2 p2 + 3p + 6 + Ψ(p) − 4αp(p + 3)

(p)
where the bitensor Wµνµ′ ν ′ (x, x′ ) is defined as the usual sum over degenerate rank-2 harmon-
ics on the four sphere and Ψ(p) is given by 2.19.
The scalar two-point function 3.28 is just the propagator of a particle with physical
mass m2 = (2αrs2 )−1 . Since we are assuming α < 0, we have m2 < 0 so this particle is a
tachyon, which is the perturbative manifestation of the Starobinsky instability. Making α
more negative, makes the tachyon mass squared less negative, and therefore weakens the
instability. Indeed, the number of efoldings in the primordial de Sitter phase emerging from
the four sphere instanton is given by Nef olds ∼ 12|α|(log N −1). Therefore, in the interesting
regime, we have −m2 << m2pl , so semiclassical gravity should be a good approximation.
This result sheds light on the problem of initial conditions in trace anomaly inflation.
One can think of the non derivative term in the scalar correlator as a potential V (ψ), with
the unperturbed de Sitter solution at ψ = 0 at the maximum. If |α| is not too small, then
the top of the potential is sufficiently flat, so that the lowest-action regular instanton is a
homogeneous Hawking-Moss instanton [7], with ψ constant at the top. Since the instability
of the de Sitter phase is characterised entirely by the coefficient α of the R2 counterterm.
this means the problem of initial conditions in anomaly-induced inflation is similar to the
corresponding problem in many theories of scalar field inflation, where one ought to explain
why the inflaton starts initially at the top of the hill. We study the origin of these inflationary
universes in section IV. Before doing so, however, we comment on the homogeneous mode
in the scalar spectrum, which has given rise to some controversy in the literature.

5 Thismeans one loses unitarity. However, probabilities for observations tend towards those of the
second order theory, as the coefficients of the fourth order terms in the action tend to zero. Hence
unitarity is restored at the low energies that now occur in the universe.

12
 B. Homogeneous Fluctuations

The most interesting instantons in both trace anomaly driven inflation as well as most
theories of scalar field inflation possess a homogeneous fluctuation mode which decreases
their action [23,31]. The presence of such a negative mode is the perturbative manifestation
of the conformal factor problem. Indeed, since the conformal factor problem is closely
related to the instability of gravity under gravitational collapse, one expects instantons that
are appealing from a cosmological perspective, to possess a negative mode.
Writing the scalar propagator 3.28 on the four sphere instanton in momentum space
gives
∞
32π 2 rs4 X W (p) (µ(x, x′ ))
hψ(x)ψ(x′ )i = , (3.30)
3|α|N 2 p=0 p(p + 3) + m2

where the biscalar W (p) equals the usual sum over degenerate scalar harmonics on the four
sphere with eigenvalue λp = −p(p + 3) of the Laplacian.
There are many negative modes if −1/8 < α < 0. This is usually the perturbative
indication of the existence of a lower-action instanton solution. For instance, in scalar field
inflation with a double well potential, the Hawking-Moss instanton possesses several negative
modes if V,φφ /H 2 < −4, which is precisely the condition for the existence of a lower-action
Coleman-De Luccia instanton that straddles the maximum. On the other hand, if α < −1/8
in 3.30 then only the homogeneous (p = 0) negative mode remains, which is again similar
to the well-known negative homogeneous mode of the Hawking-Moss instanton in theories
with a scalar potential that is sufficiently flat.
The presence of a physical negative mode supports the interpretation of an instanton as
describing the decay of an unstable state through semiclassical tunneling [6]. On the other
hand, it has been argued that it questions its use in the no boundary path integral to define
the initial quantum state of the universe6 [31]. Within the semiclassical approximation,
however, it is more appropriate to project out the negative mode, since the semiclassical
approach is based on the assumption that the path integral can be expanded around solutions
of the classical field equations.
The conclusions of [31] are based on a perturbation calculation around compact, real
instanton backgrounds, that does not take in account the wave function of interest. One
expects, however, the configuration specifying the quantum state of the Lorentzian universe
to project out the negative mode from the perturbation spectrum. Consider for example
the wave function of a universe described by a 3-sphere with radius R2 = V0 /3 and field
φ = 0, in a theory of gravity coupled to a single scalar field with potential V0 (1 − φ2 )2 . In
the semiclassical approximation, this is given by half of a Hawking-Moss instanton with the

6 In scalar field inflation, one can view the singular Hawking-Turok instantons as constrained in-
stantons, with additional data specified on an internal boundary. For some theories, the constraint
introduced in [32] to resolve the singularity, also removes the negative mode, at least perturba-
tively [31]. However, it does not remove the instability non-perturbatively and for the most obvious
potentials, the lowest-action constrained instanton gives very little inflation.

13
field constant at the top of the potential. Obviously, this solution has no negative mode,
since the boundary condition on the 3-sphere Σ removes the lowest eigenvalue solution of
the Schrödinger equation for the perturbations. Since the negative mode corresponds to
a homogeneous fluctuation, this is probably true also for large 3-spheres in the Lorentzian
regime. Therefore, one expects that in the top down approach to cosmology, where the
quantum state of the universe is taken in account, the negative mode is automatically
projected out.

C. Quantum Matter and the Microwave Background

Before discussing the top down approach in more detail, we pause to briefly comment on
some of the characteristic predictions for observations of trace anomaly inflation.
To extract accurate predictions for the cosmic microwave anisotropies, one must evolve
the perturbations through the Starobinsky instability, to obtain initial conditions for the
inhomogeneities during the radiation and matter eras. Details of this calculation will
be presented elsewhere [36], but some interesting features of the microwave temperature
anisotropies predicted by anomaly-induced inflation, can be extracted from the correlators
3.28 and 3.29 in the primordial de Sitter era. Obviously, as can be seen from 3.29, the quan-
tum matter couples to the tensors. Starobinsky [33] and Vilenkin [34] assumed that the
amplitude of primordial gravity waves was not significantly altered by the quantum matter
loops. This assumption can now be examined using AdS/CFT, which has allowed us to
include the effect of the Weyl2 counterterm and the non-local part of the matter effective
action. We find that at small scales, matter fields dominate the tensor propagator and make
it decay like p4 log p. In other words, the CFT appears to give spacetime a rigidity on small
scales, an example of how quantum loops of matter can change gravity at short distances. In
fact, this suppression should occur even if inflation is not driven by the trace anomaly, since
we observe a large number of matter fields, whose effective action is expected to dominate
the propagator on small scales.
Secondly, both the higher derivative counterterms and the matter fields introduce
anisotropic stress, which is an important difference with scalar field inflation. This can
be seen from decomposing the tensors θµν into a scalar φ and tensor tij under O(4). The
former is the difference between the two potentials in the Newtonian gauge and corresponds
to anisotropic stress. Typically reheating at the end of anomaly-induced inflation leads to
creation of particles that are not in thermal equilibrium with the photon-baryon fluid, so one
expects some anisotropic stress to survive during the radiation era. To make more precise
predictions, however, a better understanding is required of the (probably time-dependent)
values of the coefficients α and β of the higher derivative counterterms in the theory.
Finally, we should mention that for the tensor propagator the higher derivative terms
also give rise to poles in the complex p-plane. These are harmless, however, since the contour
obtained from the Euclidean goes around the complex poles [23]. In other words, defining
our theory in the Euclidean, implicitly removes the instabilities associated with the complex
poles, like a final boundary condition removes the runaway solution of the classical radiation
reaction force [29].

14
 IV. ORIGIN OF INFLATION

We have seen that the predictions of the bottom up approach to the problem of initial
conditions in inflation do not agree with observation. This is because it is based on an
essentially classical picture, in which one assumes some initial condition for the universe
and evolves it forward in time. The quantum origin of our universe, however, means that its
wave function is determined by a path integral, in which one sums over all possible histories
that lead to a given quantum state, together with some suitable boundary conditions on
the paths. This naturally leads to a top down view of the universe. In a top down context,
rather than comparing the relative probabilities of different semiclassical geometries, one
looks for the most probable evolution that leads to a certain outcome.
We now apply the quantum top down interpretation of the no boundary proposal to study
the origin of an inflationary universe, in theories where the instability of the inflationary
phase can be described in terms of a single scalar field with an effective potential that has a
local maximum. As shown in section III, this includes trace anomaly driven inflation, since
the emergence of an anomaly driven inflationary universe is very similar to the creation of
an exponentially expanding universe in theories of new inflation.
We consider a model consisting of gravity coupled to a single scalar field, with a double
well potential V (φ) = A(1 − C2 φ2 )2 (with A, C > 0). For C < 2/3, the potential has a
maximum at φ = 0 with V,φφ /V sufficiently low so that there exists no Coleman-De Luccia
instanton, but only a Hawking-Moss instanton with φ = 0 everywhere on top of the hill.
Implementing a top down approach, we consider the quantum amplitudes Φ[Σ, h̃, K, φΣ ]
for different conformal 3-geometries h̃ with trace K of the second fundamental form, on
an expanding surface Σ during inflation7 . According to the no boundary proposal, the
defining path integral should be taken over all compact Riemannian geometries that induce
the prescribed configuration on Σ.
In the K-representation, the Euclidean action is given by
1
Z
√ 1
Z √ Z
√ 1 µν
 
S=− d4 x gR − d3 x hK + d4 x g g ∂µ φ∂ν φ + V (φ) , (4.31)
2κ 3κ 2
The usual wave function Ψ[h, φΣ ] is obtained from Φ[h̃, K, φΣ ] by an inverse Laplace trans-
form,
Z 
K

2
 Z √ 
Ψ[h, φΣ ] = d exp d3 x hK Φ[h̃, K, φΣ ] (4.32)
Γ 4iκ 3κ
where the contour Γ runs from −i∞ to +i∞.
Within the semiclassical approximation, the no boundary wave function is approximately
given by the saddle-point contributions. Restricting attention to saddle-points that are

7 Inprinciple we should consider the amplitude for a conformal 3-geometry on a surface Σ just
inside our past light cone, with K equal to the present Hubble rate and given values for all other
observables. However, it is sufficient to consider the quantum amplitude for a configuration on
an expanding surface in the inflationary period, since this can then be accurately evolved to the
future using classical laws.

15
invariant under the action of an O(4) isometry group, the instanton metric can be written
as

ds2 = dτ 2 + b2 (τ )dΩ23 , (4.33)

and the Euclidean field equations read

φ′′ = −Kφ′ + V,φ (4.34)

K ′ + K 2 = −(φ2,τ + V ) (4.35)

where φ′ = φ,τ and K = 3b,τ /b. The Lorentzian trace KL = −3ȧ/a is obtained by analytic
continuation. We first calculate the wave function for real K, and then analytically continue
to imaginary, or Lorentzian KL = −iK.
At the semiclassical level, there are two contributions to the given amplitude. For small
φΣ and any Euclidean K, there always exists a non-singular, Euclidean O(4) invariant so-
lution of the field equations, with the prescribed boundary conditions. This solution is part
of a deformed sphere, or Hawking-Turok instanton. In the approximation K = 3H cot(Hτ ),
with H 2 = A/3, and V (φ) ∼ A(1 − Cφ2 ), the solution of 4.34 is given by

2 F1 (3/2 + q, 3/2 − q, 2, z(K))

φ = φΣ (4.36)
2 F1 (3/2 + q, 3/2 − q, 2, z(KΣ ))

where
" #
q 1 K
q= 9/4 + 6C, z(K) = 1− 2 (4.37)
2 (A + K 2 )1/2

At the South Pole K → +∞, so in the instanton the scalar field slowly rolls up the hill from
its value at the regular South Pole to the prescribed value φΣ on the 3-sphere with trace
KΣ . The weight of the Hawking-Turok geometry in the no boundary path integral for the
wave function Φ[K, φΣ ] is approximately given by
1 Z 3 √ Z
√
S[KΣ , φΣ ] = − d x hK − d4 x gV (φ)
3κ "
12π 2 24π 2 C
#
K
=− 1− 2 − φ2 z 2 (KΣ ) ×
A (A + K 2 )1/2 A2 (1 − C) Σ
2 − 3C F ′
"   #
1 − 2z(KΣ ) + 3C(1 − z(KΣ )) z(KΣ ) + [(z(KΣ )] (4.38)
1 + 3C F

For small φΣ , there is a second semiclassical contribution to the wave function, coming
from universes that are created via an O(5) symmetric Hawking-Moss instanton with φ
constant at the top of the hill, but in which a quantum fluctuation disturbs the field, causing
it to run down to its prescribed value φΣ at the 3-sphere boundary with trace KΣ . Neglecting
prefactors, the action of the Hawking-Moss geometry is given by the first term in 4.38. It
follows that for KΣ = 0, the action of the Hawking-Turok geometry is more negative than
the action of the Hawking-Moss instanton. This would seem to suggest that the universe is

16
 Re dS
2
1
kl
0.5 1 1.5 2
-1
-2
-3
-4
FIG. 1. ℜ[δS] is proportional to the difference ∆S between the action of the Hawking-Moss
and Hawking-Turok type geometries discussed in the text.
The no boundary proposal predicts an expanding universe to be created in an unstable de Sitter
state, by semiclassical tunneling via a Hawking-Moss instanton.

least likely to start at the top of the hill. However, we are not interested in the amplitude
for a Euclidean spacetime, but in the no boundary wave function of a Lorentzian expanding
universe.
Within the regime in which φ remains small over the whole geometry, one can derive the
amplitude in the Lorentzian from our result for S[KΣ , φΣ ], by analytic continuation into the
complex K-plane. In a Lorentzian universe, Euclidean K is pure imaginary, KL = −iK.
Since the action is invariant under diffeomorphically related contours in the complex τ -plane,
we may deform the contour into one with straight sections, along the real and imaginary
K-axis. It follows immediately from 4.38 that the real part of the action for the Hawking-
Moss instanton is constant on the imaginary K-axis, unlike the action for the Hawking-
Turok geometry. According to the no boundary proposal, the relative probability of both
geometries is given by
A2HM −2ℜ[∆S]
P [KL , φΣ ] = e (4.39)
A2HT
where ∆S = SHM − SHT . The prefactors account for small fluctuations around the classical
solutions and can be neglected for small φ.
In figure 1 we plot ℜ[δS(kl )], which is proportional to the real part of the difference
∆S between the action of both geometries, as a function of Lorentzian kl = √ K2 L 2 , with
A −KL
A = 1 and C = 1/3. This shows that the real part of the action for the Hawking-Turok
geometry increases on the imaginary axis away from K = 0 and soon becomes larger than
the O(5) action. In addition, within our approximation φΣ enters only in the prefactor of
∆S. Therefore, the dominant contribution to the no boundary path integral for a Lorentzian
inflating universe comes from spacetimes which are created by semiclassical tunneling via
a Hawking–Moss instanton and which inflate for a long time before a quantum fluctuation
causes the field to roll down to its final value φΣ . This means that in an inflationary universe,
the scalar field is more likely to start at the top of the hill and roll down, than to start lower
down. The reason is that although being at the top of the hill costs potential energy, it saves

17
gradient energy, by having a scalar field that is constant in space and time. If the maximum
of the potential is fairly flat, the gradient energy is dominant, and the universe starts with a
constant scalar, at the top of the hill. Therefore, one does not need an initial hot Big Bang
phase, to explain why inflation began at a local maximum of the potential.
As mentioned above, this scenario is realised in trace anomaly driven inflation. The
unperturbed de Sitter solution 2.21 in anomaly-induced inflation emerges from the Hawking-
Moss geometry, while the inhomogeneous Hawking-Turok evolution corresponds to one of
the singular instantons discussed in section II. The field configuration on Σ determines the
third derivative of the scale factor at the regular South Pole, or equivalently the initial value
of the order parameter φ governing the instability. For α < −1/8, the instability of the de
Sitter phase is sufficiently weak, so that the universe is most likely to start at the top, in an
unstable de Sitter state. This result also justifies our calculation of metric perturbations,
which were based on a perturbative expansion of the path integral about the round four
sphere.
Finally, we should mention that because we are interested in real matter fields on Σ, the
analytic continuation into the complex K-plane means φ must be complex in the bulk of the
instanton8 . More precisely, at the South Pole, we must have ℑ[φ] = φΣ ℑ[F (zΣ )]/ℜ[F (zΣ )].
This has no physical meaning though, since the stationary phase approximation is just a
mathematical construction to evaluate the path integral over real φ.

V. DISCUSSION

We have studied the problem of initial conditions in cosmology. Because our universe
has a quantum origin, one must adopt a top down approach, in which one considers the path
integral over a class of histories that lead to a given quantum state of the universe. A top
down view is naturally implemented in the context of the no boundary proposal, which states
that the amplitude for the quantum state of the universe on a 3-surface Σ is given by a path
integral over all geometries that induce the prescribed configuration on their only boundary
Σ. We have investigated the no boundary predictions for the quantum cosmological origin of
a large class of inflationary universes. In particular, we have considered theories of inflation
where the global instability can be described in terms of a single scalar field φ with an
effective potential that has a sufficiently flat local maximum. This includes Starobinsky’s
trace anomaly model, since the nature of the initial instability in anomaly-induced inflation is
the same as the instability that occurs in new inflation. Trace anomaly driven inflation has a
sound motivation in fundamental particle physics, since we observe a large number of matter
fields in the universe, which may be expected to behave like a CFT in the early universe.
The no boundary proposal predicts an inflationary universe to be created in an unstable de

8 Complex instanton solutions have previously been considered in [35]. Physical constraints on
complex contours of 4-geometries in the no boundary path integral were discussed in [20]. In this
context, we should mention that in the absence of an extension of Bishop’s theorem to complex
geometries, it is not entirely clear whether the O(4) invariant geometries considered here are in
fact the lowest-action saddle-points that contribute to the wave function of interest.

18
Sitter state, by semiclassical tunneling via a Hawking-Moss instanton. The universe first
inflates, before a quantum fluctuation causes the field to roll down and inflation to end.
Provided −4 < V,φφ /H 2 < 0, the maximum of the potential is sufficiently flat, so that this
geometry has lower action than an inhomogeneous Hawking-Turok type evolution.
One could think of the no boundary proposal as describing the creation of universes with
different radii, like the formation of bubbles of steam. If the bubbles are small, they collapse
again, but there is a critical size above which they are more likely to grow. In theories where
the amplitude for an expanding universe is dominated by geometries that start in a de Sitter
state, this naturally leads to the interpretation of the round Hawking-Moss instanton on top
of the hill as corresponding to that critical size.
Correlators of observables on a spacelike hypersurface Σ should be computed directly
from the no boundary path integral, by summing over histories to the past of that surface.
In the semiclassical approximation, the dominant instanton saddle-point solution should be
used as background in a perturbative evaluation of the no boundary path integral, to find
correlators of metric perturbations during inflation. Therefore, our result justifies the per-
turbation calculations performed in section III and in [23], in which we computed Euclidean
propagators assuming a four sphere instanton background. Evolving the spectrum of pri-
mordial perturbations through the Starobinsky instability determines the cosmic microwave
anisotropies. Hence the boundary condition on the fluctuation modes imposed by the instan-
ton background may provide an observational discriminant between different saddle-points,
hereby connecting quantum cosmology and the top down approach to falsifiabble predictions
for observation [37].
We have argued that because our universe has a quantum origin, one must adopt a top
down approach to the problem of initial conditions in cosmology, in which the histories
that contribute to the path integral, depend on the observable being measured. There is
an amplitude for empty flat space, but it is not of much significance. Similarly, the other
bubbles in an eternally inflating spacetime are irrelevant. They are to the future of our
past light cone, so they don’t contribute to the action for observables and should be excised
by Ockham’s razor. Therefore, the top down approach is a mathematical formulation of
the weak anthropic principle. Instead of starting with a universe and asking what a typical
observer would see, one specifies the amplitude of interest. In the context of the no boundary
proposal, however, a top down description of the universe is not necessarily less ‘complete’
or less predictive. We believe that if we are to explain why the universe is the way we
observe it to be, a top down view is forced upon us by the quantum nature of the universe.
Therefore, although future developments in M-Theory will provide us with new insights
in how a theory of boundary conditions in cosmology should be formulated, the approach
developed here should still apply when the framework of quantum cosmology will be based
on M-Theory.

Acknowledgments
It is a pleasure to thank Gary Gibbons, Harvey Reall and Toby Wiseman for helpful
discussions.

19
 REFERENCES
[1] A.H. Guth, The Inflationary Universe: A Possible Solution to the Horizon and Flatness
Problems, Phys. Rev. D23, 347 (1981).
[2] A.D. Linde, A New Inflationary Universe Scenario: A Possible Solution of the Horizon,
Flatness, Homogeneity, Isotropy and Primordial Monopole Problems, Phys. Lett. B108,
389 (1982).
[3] A. Albrecht and P. Steinhardt, Cosmology for Grand Unified Theories with Radiatively
Induced Symmetry Breaking, Phys. Rev. Lett. 48, 1220 (1982).
[4] A.A. Starobinsky, A New Type of Isotropic Cosmological Models without Singularity,
Phys. Lett. 91B, 99 (1980).
[5] A.D. Linde, Chaotic Inflation, Phys. Lett. B129, 177 (1983).
[6] S. Coleman and F. De Luccia, Gravitational Effects on and of Vacuum Decay, Phys.
Rev. D21, 3305 (1980).
[7] S.W. Hawking and I.G. Moss, Supercooled Phase Transitions in the Very Early Universe,
Phys. Lett. B110, 35 (1982).
[8] J.B. Hartle and S.W. Hawking, The Wave Function of the Universe, Phys. Rev. D28,
2960 (1983).
[9] G.W. Gibbons, Aspects of Supergravity Theories, in Supersymmetry, Supergravity, and
Related Topics, eds. F. del Aguila, J.A. de Azcarraga and L.E. Ibanez (World Scientific
1985), pp. 346-351.
[10] J. Maldacena and C. Nunez, Supergravity Description of Field Theories on Curved
Manifolds and No Go Theorem, Int. J. Mod. Phys. A16, 822 (2001), [hep-th/0007018].
[11] S.J. Gates and B. Zwiebach, Gauged N = 4 Supergravity Theory with a New Scalar
Potential, Phys. Lett. 123B, 200 (1983).
[12] C.M. Hull, New Gauging of N = 8 Supergravity, Phys. Rev. D30, 760 (1984);C.M. Hull,
De Sitter Space in Supergravity and M-Theory, JHEP 0111, 012 (2001)
[13] E. Tomboulis, 1/N Expansion and Renormalization in Quantum Gravity, Phys. Lett.
70B, 361 (1977).
[14] L. Parker and D.J. Toms, Renormalization-group Analysis of Grand Unified Theories
in Curved Spacetime, Phys. Rev. D29, 1584 (1984).
[15] D.M. Capper, M.J. Duff, Trace Anomalies in Dimensional Regularization, Nuovo Ci-
mento 23A, 173 (1974).
[16] N.D. Birrell and P.C.W. Davies, Quantum Fields in Curved Space, Cambridge University
Press, Cambridge, UK (1982).
[17] J. Maldacena, The Large N Limit of Superconformal Field Theories and Gravity, Adv.
Theor. Math. Phys. 2, 231 (1998), [hep-th/9711200].
[18] I.L. Shapiro and J. Sola, Massive Fields Temper Anomaly-Induced Inflation:The Clue
to Graceful Exit?, hep-th/0104182.
[19] G.W. Gibbons and J.B. Hartle, Real Tunneling Geometries and the Large-Scale Topology
of the Universe, Phys. Rev. D42, 2458 (1990).
[20] J.J. Halliwell and J.B. Hartle, Integration Contours for the No-Boundary Wave Function
of the Universe, Phys. Rev. D41, 1815 (1990).
[21] S. Gratton and N. Turok, Cosmological Perturbations from the No Boundary Path In-
tegral, Phys. Rev. D60, 123507 (1999), [astro-ph/9902265].

20
[22] T. Hertog and N. Turok, Gravity Waves from Instantons, Phys. Rev. D62, 083514
(2000), [astro-ph/9903075].
[23] S.W. Hawking, T. Hertog, H.S. Reall, Trace Anomaly Driven Inflation, Phys. Rev. D63,
083504 (2001), [hep-th/0010232].
[24] R.J. Riegert, Non-Local Action for the Trace Anomaly, Phys. Lett. 134B, 56 (1984).
[25] S.W. Hawking, T. Hertog, H.S. Reall, Brane New World, Phys. Rev. D62, 043501
(2000), [hep-th/0003052].
[26] S.W. Hawking and N. Turok, Open Inflation without False Vacua, Phys. Lett. B425,
25 (1998), [hep-th/9802030].
[27] T. Hertog, The Origin of Inflation, Ph.D. dissertation, University of Cambridge (2002).
[28] N. Turok, Before Inflation, Lecture at CAPP2000, [astro-ph/0011195].
[29] S.W. Hawking and T. Hertog, Living with ghosts, [hep-th/0107088].
[30] G.W. Gibbons, S.W. Hawking, M.J. Perry, Path Integrals and the Indefiniteness of the
Gravitational Action, Nucl. Phys. B138, 141 (1978).
[31] S. Gratton and N. Turok, Homogeneous Modes of Cosmological Instantons, Phys. Rev.
D63, 123514 (2001), [hep-th/0008235].
[32] K. Kirklin, N. Turok, T. Wiseman, Singular Instantons Made Regular, Phys. Rev. D63,
083509 (2000), [hep-th/0005062].
[33] A.A. Starobinsky, The Perturbation Spectrum Evolving from a Nonsingular, Initially de
Sitter cosmology, and the Microwave Background Anisotropy, Sov. Astron. Lett. 9, 302
(1983).
[34] A. Vilenkin, Classical and Quantum Cosmology of the Starobinsky Inflationary Model,
Phys. Rev. D32, 2511 (1985).
[35] G.W. Lyons, Complex Solutions for the Scalar Field Model of the Universe, Phys. Rev.
D46, 1546 (1992).
[36] S.W. Hawking and T. Hertog, in preparation (2002).
[37] S. Gratton, T. Hertog, N. Turok, An Observational Test of Quantum Cosmology, Phys.
Rev. D62, 063501 (2000), [astro-ph/9907212].

21