The No-Boundary Measure of the Universe

James B. Hartle,1 S.W. Hawking,2 and Thomas Hertog3

1
Department of Physics, University of California, Santa Barbara, 93106, USA
2
DAMTP, CMS, Wilberforce Road, CB3 0WA Cambridge, UK
3
Laboratoire APC, Université Paris 7, 10 rue A.Domon et L.Duquet, 75205 Paris, France and
International Solvay Institutes, Boulevard du Triomphe, ULB – C.P. 231, 1050 Brussels, Belgium
We consider the no-boundary proposal for homogeneous isotropic closed universes with a cosmo-
logical constant and a scalar field with a quadratic potential. In the semi-classical limit, it predicts
classical behavior at late times if the initial scalar field is more than a certain minimum. If the
classical late time histories are extended back, they may be singular or bounce at a finite radius.
The no-boundary proposal provides a probability measure on the classical solutions which selects
inflationary histories but is heavily biased towards small amounts of inflation. This would not be
compatible with observations. However we argue that the probability for a homogeneous universe
arXiv:0711.4630v4 [hep-th] 8 Jun 2008

should be multiplied by exp(3N ) where N is the number of e-foldings of slow roll inflation to obtain
the probability for what we observe in our past light cone. This volume weighting is similar to that
in eternal inflation. In a landscape potential, it would predict that the universe would have a large
amount of inflation and that it would start in an approximately de Sitter state near a saddle-point
of the potential. The universe would then have always been in the semi-classical regime.

Introduction sulting probabilties favor an inflationary past and, in a

The string theory landscape is believed to contain a vast landscape potential, suggest a semiclassical origin.
ensemble of stable and metastable vacua that includes Classical Prediction in Quantum Cosmology
some with a small positive effective cosmological constant In quantum cosmology states are represented by wave
and the low energy effective field theory of the Standard functions on the superspace of three-geometries and
Model. But the landscape by itself does not explain why spatial matter field configurations. For the homoge-
we are in one vacuum rather than in some other. For neous, isotropic models considered here minisuperspace
that one has to turn to cosmology and to a theory of the is spanned by the scale factor b and the value χ of the
quantum state of the universe. homogeneous scalar field. Thus, Ψ = Ψ(b, χ).
A manifest feature of our quantum universe is the wide The no-boundary wave function [1] is defined by the
range of epoch and scale on which the laws of classi- sum-over-histories
Z
cal physics apply, including classical spacetime. Classi-
Ψ(b, χ) = δgδφ exp(−I[a(τ ), φ(τ )]/h̄). (1)
cal spacetime is a prerequisite for the construction of ef- C
fective theories, for cosmology, and for eternal inflation.
Here, a(τ ) and φ(τ ) are the histories of the scale factor
But classical spacetime is not a property of every state
and matter field and I[a(τ ), φ(τ )] is their Euclidean ac-
in quantum gravity. Rather it emerges only for certain
tion. The sum is over cosmological geometries that are
quantum states.
regular on a manifold with only one boundary at which
We calculate the probability measure on classical a(τ ) and φ(τ ) take the values b and χ. The integration
spacetimes predicted by the no-boundary wave function is carried out along a suitable complex contour C which
(NBWF) [1] to leading semiclassical order for homoge- ensures the convergence of (1) and the reality of the re-
neous and isotropic minisuperspace models with a cos- sult.
mological constant and a scalar field with a quadratic For some regions of minisuperspace the integral in (1)
potential. We find the NBWF severely restricts the pos- can be approximated by the method of steepest descents.
sible classical universes and argue that such classicality Then the wave function will be well approximated to
restrictions would act as a strong vacuum selection prin- leading order in h̄ by a sum of terms of the form
ciple in the string landscape.
Ψ(b, χ) ≈ exp{[−IR (b, χ) + iS(b, χ)]/h̄}, (2)
The NBWF predicts the probabilities of entire classi-
cal histories. But we are interested in the probability one term for each extremizing history. The functions
for our observations which are restricted to a (thickened) IR (b, χ) and −S(b, χ) are the real and the imaginary
light cone located somewhere in the universe and extend- parts of the action evaluated at the extremum. In sim-
ing over roughly a Hubble volume [2]. To calculate such ple cases these extremizing histories may describe the
probabilities we must sum the probabilities for classical nucleation of a Lorentzian spacetime by a Euclidean in-
histories over all those that contain our data at least once stanton. But in general they will be complex — “fuzzy
[3, 4]. This defines the probability for our data in a way instantons”.
that is gauge invariant and dependent only on informa- In order for wave functions of the form (2) to predict an
tion in our past light cone. We will argue that the re- ensemble of Lorentzian histories with high probabilities
 2

for classical correlations in time further conditions must -IR

8
be satisfied. A necessary one is the classicality constraint
7
|(∇S)2 | ≫ |(∇IR )2 |, (3) 6

where gradients and inner products are defined with 5

the minisuperspace metric. When (3) holds the ac- 4
tion S satisfies the Lorentzian Hamilton-Jacobi equa- 3
tion. The NBWF then predicts the corresponding en- 2
semble of Lorentzian histories. Their probabilities are
1
exp[−2IR (b, χ)]/h̄] to leading order in h̄.
Two key points should be noted: (1) The no-boundary Φ0
Φ0 cΦ0 s 2 3 4 5
wave function provides probabilities for entire classical
histories. (2) The histories in the classical ensemble are FIG. 1: The values of IR of the one-parameter set of clas-
not the same as the extremizing histories that provide sical histories predicted by the no-boundary proposal in a
the steepest descents approximation to the integral (1). quadratic potential minisuperspace model with µ = 3 and
The classical histories are real and Lorentzian and may Λ = .03. There are no classical histories for φ0 below a crit-
have two large large regions. The extrema are generally ical value φc0 at about 1.2. The universe therefore requires
complex with only one large region. a minimum amount of matter to behave classically at late
times. A critical value φs0 at about 1.5 separates large φ0 his-
Scalar Field Model
tories that bounce at a finite radius when extrapolated back
We have applied this prescription for classical prediction from singular histories for smaller φ0 .
to homogeneous isotropic closed universes with a cosmo-
logical constant Λ and a scalar field Φ with a quadratic
potential V (Φ) = (1/2)m2 Φ2 . We write the complex ho- semiclassical approximation. These are constant along
mogeneous isotropic metrics that provide the steepest- the integral curves. It is convenient to take φ0 ≡ |φ(0)|
descent approximation to the no-boundary path integral to be the parameter labeling different histories in this
(1) as classical ensemble.
ds2 = (3/Λ) dτ 2 + a2 (τ )dΩ23 .
 
(4) The classicality constraint (3) is not satisfied for all
integral curves of S. Specifically, in the interesting regime
The Euclidean action I then takes the form where µ > 3/2 we find the NBWF requires the universe to
contain a minimum amount of scalar field energy at early
9π
Z
dτ −aȧ2 − a + a3 times to behave classically at late times. (The value of µ

I[a(τ ), φ(τ )] =
4Λ C(0,υ) based on today’s Λ would be very much larger.) From
 i
+a3 φ̇2 + µ2 φ2 . (5) now on we restrict to this range. Similar conclusions were
reached in [6] for the Λ = 0 case. This is illustrated in
We use units where h̄ = c = G = 1 and define the mea- Figure 1, where we show IR (b, χ) for all members of the
sures φ = (4π/3)1/2 Φ and µ ≡ (3/Λ)1/2 m. The contour ensemble of classical histories predicted by the NBWF in
C(0, υ) in the complex τ -plane connects the South Pole a µ = 3 model with Λ ≈ .03. There is a critical value φc0
τ = 0 with an endpoint τ = υ where a and φ take real below which there are no classical solutions. The lower
values b and χ. bound φc0 implies a lower bound on the scalar field in the
We evaluated the NBWF in the semiclassical approxi- corresponding classical histories. The critical value φc0
mation (2) by numerically solving the Friedman-Lemaı̂tre increases slightly with µ and tends to 1.27 when Λ → 0,
equations for each value of b and χ along a suitable com- for fixed m.
plex contour C(0, υ). This gives complex analytic func- The classicality constraint is closely related to the slow
tions (a(τ ), φ(τ )) which are an extemum of the action. roll condition of scalar field inflation. We make this pre-
The value of the action at an extremum gives IR (b, χ) cise in Figure 2 where we plot the trajectories in (H, φ)
and S(b, χ). variables where H is the instanteous Hubble constant
The integral curves of S are the classical solutions when H = ḃ/b. We show five members of the ensemble of clas-
the classicality constraint (3) is satisfied. The relation be- sical histories in the µ = 3 model for φ0 between 1.3 and
tween position and momenta that follows from S means 2. When we follow the histories back in time to higher
that, in the semiclassical approximation, the NBWF pre- values of H, they all lie within a very narrow band around
dicts non-zero probabilities only for a one-parameter en- H = µφ. But this is precisely the regime that corre-
semble of the two-parameter family of classical histories. sponds to slow roll inflationary solutions, as emphasized
Classical histories not in the ensemble have zero probabil- recently in [9]. Hence the NBWF plus classicality at late
ity. The relative probabilities for histories in this classical times implies inflation at early times.
ensemble are given by exp[−2IR (b, χ)/h̄] in the leading For φ0 smaller than a critical value φs0 > φc0 the al-
 3

H ferent from the causality in pre-big bang universes where

the arrow of time points in one direction throughout the
0.5 spacetime.
Top Down Cosmology
0.4 The NBWF gives the probabilities of entire classical his-
0.3 tories. But we are interested in probabilities that refer
to our data which are limited to a part of our past light
0.2 cone. Among these are the top-down probabilities [2]
0.1 for our past conditioned on our present data. These are
obtained by summing over the probabilities for classical
Φ spacetimes that contain our data at least once, and over
-0.5 0.5 1 1.5 2
the possible locations of our light cone in them.
FIG. 2: The no-boundary wave function predicts that all his- These sums can be implemented concretely in our
tories that behave classically at late times undergo a period closed, homogeneous, isotropic minisuperspace models as
of inflation at early times as shown here by the linear growth follows: Approximate the probability for our data on the
of the instantaneous Hubble constant H in five representative
past light cone by the probability of data in a Hubble vol-
µ = 3 classical histories.
ume on an appropriate surface of homogeneity. Assume
that our data are otherwise detailed enough that they
occur only once on this surface [8]. The sum is then over
lowed classical histories of the universe are singular in the spatial locations of our Hubble volume in that sur-
the sense that their matter densities exceed the Planck face of homogeneity in all classical spacetimes that last
density. But for φ0 > φs0 they bounce at a finite ra- sufficiently long.
dius in the past. This is possible despite the singular-
The classicality constraint φ0 > φc0 implies that all his-
ity theorems because a scalar field and the cosmological
tories in the classical ensemble inflate (Figure 2). The
constant violate the strong energy condition. Near the
condition that the universe lasts ∼14 Gyr further re-
bounce the universe approaches a de Sitter state with ra-
stricts the ensemble, requiring φ0 to be larger than a
dius ∼ (µφ0 )−1 . Such non-singular solutions form only a
critical value φg0 > φc0 . On average each history has the
small subset of all scalar field gravity solutions but have
same behavior shortly after inflation ends and thus pre-
significant probability in the no-boundary state.
dicts the same observable physics for every Hubble vol-
Even for the histories in the ensemble that are clas- ume at the present time. But the classical histories differ
sically singular at an early time the NBWF unambigu- in the value φi ≈ φ0 of the inflaton at the start of in-
ously predicts probabilities for late time observables such flation, and consequently in the volume of the present
as CMB fluctuations, because it predicts probabilities for surface of homogeneity. None of these properties is di-
histories rather than their initial data. The existence of rectly observable and should be summed over. The sum
singularities in the extrapolation of some classical ap- over our location therefore multiplies the NBWF proba-
proximation in quantum mechanics is not an obstacle bility for each classical spacetime in the ensemble by the
to prediction but merely a limitation of the validity of number of Hubble volumes in the total present volume
the approximation. Indeed, there could be quantum me- — a factor proportional to exp(3N ). This favors larger
chanical transitions rather than classical ones across cos- universes and more inflation. In a larger universe there
mological singularities that connect two classical regimes are more places for our Hubble volume to be.
[10]. Volume weighting increases the probability of a large
Individual classical bouncing histories are not gener- number of efoldings. For quadratic potentials with re-
ally time-symmetric about the bounce, although the time alistic values of m and Λ the constraints of classicality
asymmetry is small for large φ0 . However, the reality and minumum age yield a restricted ensemble of histories
of the NBWF implies the ensemble of allowed classical whose volume weighted probabilities slightly favor a large
histories is time symmetric. For every history in this number of efoldings [5] that are necessary for explaining
ensemble, its time reverse is also a member. the observed spatial flatness. An important feature of the
For the universes that bounce at a minimum radius it volume weighted probability distribution is that there is
seems likely that the NBWF will predict that fluctua- a wide region where the probability is strongly increas-
tions away from homogeneity and isotropy will be at a ing with N . The gradient of the probability distribution
minimum at the bounce and grow away from the bounce ∼ exp(3N − 2IR ) with respect to φi is positive if
for at least a while on either side (cf. [7]). This means
that the thermodynamic arrow of time is likely to point V 3 ≥ |V,φ |2 (6)
away from the bounce on either side of it. Events on one
side are therefore unlikely to have a causal impact on the which, intriguingly, is the same as the condition for eter-
other and have much explanatory value. This is very dif- nal inflation [3, 11].
 4

V Sitter like state near a broad saddle-point of V . Because

3N-2IR
the dominant saddle-points are well below the Planck
density we expect the most probable histories are bounc-
ing solutions of the field equations which lie entirely in
the semi-classical regime. They have a large amount of
slow roll inflation. During this the scalar field evolves
Φ0c
Φ
Φ0c
Φ0 from the saddle-point to the neighbouring minima of V ,
populating only a few of the possible vacua in the land-
FIG. 3: To account for the different possible locations in the scape.
universe of the Hubble volume that contains our data one Inhomogeneities
ought to multiply the no-boundary amplitudes by a volume In this paper we have discussed homogeneous universes
factor. In regions of the landscape around a maximum of the only. However, one can also consider inhomogeneous per-
potential (left), we expect this to have a significant effect on turbations. It appears that the volume weighting can
the probability distribution over φ0 and hence over N (right).
The effect of a classicality constraint is also shown.
overcome the gradient action for very long wavelength
perturbations that leave the horizon while (6) is satisfied.
This suggests the NBWF with volume weighting will pre-
Hence, there is a striking contrast between the un- dict a universe that is very inhomogeneous on very large
conditioned bottom-up probabilities that favor small scales. Eternal inflation [11] also predicts large scale in-
amounts of inflation and the top-down probabilities con- homogeneities but the connection, if any, with this pic-
ditioned on our data that favor larger amounts. ture is not yet clear to us. In any event no additional
Landscape Potentials ‘measure’ would be needed to derive the probabilities for
A typical landscape potential will have several saddle- this structure. The NBWF in principle provides that.
points besides the quadratic directions discussed above.
For saddle points with more than one descent direction,
there will generally be a lower saddle-point with only one Acknowledgments We thank Neil Turok for stimulat-
descent direction, and with lower action. If this descent ing discussions. The work of JH was supported in part
direction is sharply curved we expect the classicality con- by the National Science Foundation under grant PHY05-
straint (3) not to be satisfied in analogy with the case of 55669.
quadratic potentials. Hence the no-boundary amplitude
for universes that emerge from around such saddle-points
will be approximately zero. Thus only broad saddle-
points with a single descent direction will give rise to [1] J. B. Hartle, S. W. Hawking (1983): The Wave Function
significant no-boundary amplitudes for universes that be- of the Universe. Phys. Rev. D 28, 2960-2975.
have classically at late times. Only a few of the saddle- [2] S.W. Hawking, T. Hertog (2006): Populating the Land-
points will satisfy the demanding condition that they be scape: A Top Down Approach, Phys. Rev. D 73, 123527,
broad, because it requires that the scalar field varies by hep-th/0602091.
[3] D. Page (1997): Space for both No-Boundary and Tun-
order the Planck value across them. The classicality con-
neling Quantum States of the Universe, Phys. Rev. D,
straint, therefore, acts as a vacuum selection principle. 56, 2065.
By analogy with quadratic potentials we expect the [4] S.W. Hawking (2007): Volume Weighting in the No
classical histories predicted by the NBWF for φ0 near a Boundary Proposal, arXiv:0710.2029.
broad maximum of V to have an early period of inflation, [5] J. B. Hartle, S. W. Hawking, T. Hertog (2007): The
during which the scalar field rolls down to a nearby min- Classical Histories of the No Boundary Quantum State,
imum of V . (We assume for simplicity that all vacua are to be published in Phys. Rev. D, arXiv:0803:1663.
[6] L.P. Grischuk and L.V. Rozhansky (1990): Does the
consistent with the Standard Model.) As before, the no-
Hartle-Hawking Wave Function Predict the Universe We
boundary proposal favors a small number of efoldings, Live in?, Phys. Lett. B 234, 9.
i.e. histories where φ0 ≈ φc0 (see Fig 3a). However in [7] S. W. Hawking, R. Laflamme, and G. W. Lyons (1993):
contrast with quadratic potentials, near a broad maxi- Origin of time asymmetry. Phys. Rev. D 47, 5342-5356.
mum the volume factor more than compensates for the [8] J.B. Hartle and M. Srednicki (2007): Are We Typical?,
reduction in amplitude due to the higher value of the Phys. Rev. D 75, 123523, arXiv:0704.2630.
potential. The resulting probabilities of past histories [9] G. W. Gibbons, N. Turok (2006): The Measure Problem
in Cosmology, hep-th/0609095.
consistent with present data significantly favor a large
[10] N.G. Turok, B. Craps, T. Hertog (2007): From Big
number of efoldings. This is illustrated in Figure 3b and Crunch to Big Bang with AdS/CFT, arXiv:0711.1824.
discussed in [5]. [11] A. Vilenkin (1983): Birth of Inflationary Universes, Phys.
This leads us to predict that in a landscape potential, Rev. D 27, 2848; A.D. Linde (1986): Eternally Exist-
the most probable homogeneous history of the universe ing Selfreproducing Chaotic Inflationary Universe, Phys.
that is consistent with our data started in an unstable de Lett. B 175, 395