PHYSICAL REVIEW D VOLUME 52, NUMBER 10 15 NOVEMBER 1995

Probability for primordial black holes

R. Bousso* and S. W. Hawkingt
Department of Applied Mathematics and Theoretical Physics, University of Cambridge,
Silver Street, Cambridge CB8 9EW; United Kingdom
(Received 28 June 1995)
We consider two quantum cosmological models with a massive scalar field: an ordinary Friedmann
universe and a universe containing primordial black holes. For both models we discuss the complex
solutions to the Euclidean Einstein equations. Using the probability measure obtained from the
Hartle-Hawking no-boundary proposal we find that the only unsuppressed black holes start at the
Planck size but can grow with the horizon scale during the roll down of the scalar field to the
minimum.

PACS number(s): 98.80.Hw, 04.60.Kz, 04.70.Dy, 98.80.Bp

I. INTRODUCTION the Euclidean Einstein equations for the S case. They

In this paper we ask how likely it is for the universe to
have contained primordial black holes. We investigate
universes which undergo a period of inflation in their
earliest stage, driven by a scalar field P with a poten-
tial V(P) with a minimum V(0) = 0. The results do not
depend qualitatively on the exact form of the potential,
so for simplicity we consider a massive minimally cou-
pled scalar V(P) = zm2$2. The scalar field starts out at
a large initial value Po and acts as a cosmological con-
stant for some time until it reaches the minimum of its
potential and inflation ends. We consider two difFerent
types of spacetimes: in the first, the spacelike sections
are simply three-spheres and no black holes are present;
in the second, they have the topology S x S, which is
the topology of the spatial section of the Schwarzschild-
de Sitter solution. Thus these spaces can be interpreted
as inflationary universes with a pair of black holes. In
the inflationary period, the first type will be similar to
will be slightly complex due to the time dependence of
the effective cosmological constant (mP) 2. We obtain
the Euclidean action for those solutions. In Sec. VI we
go through a similar procedure for the S x S case. We
find that the black hole grows during the inflationary pe-
riod, a noteworthy difI'erence to the Nariai case with a
fixed cosmological constant. In Sec. VII we use the ac-
tion to estimate the relative probability of the two types
of universes. We find that black holes are suppressed for
all but very large initial values of Po.
II. THE WAVE FUNCTION OF THE UNIVERSE
The Hartle-Hawking no-boundary proposal states that
the wave function of the universe is given by
@o(h;, , o&M] =
f o(g„,o) exp( I(g„,4)], (2. 1)—

a de Sitter universe, the second to a Nariai universe [1].
To find the likelihood for primordial black holes, we as-
sign probabilities to both types of spacetimes using the
Hartle-Hawking no-boundary proposal (NBP) [2]. This
is the only proposal for the boundary conditions of the
universe that seems to give a well-defined answer in this
situation. It is not clear how to apply the so-called "tun-
neling proposal" in the S x S case. If one takes the
action to appear with the opposite sign as is done in the
S case, one would reach the conclusion that a universe
with a pair of black holes was more likely than a universe
without, and that the probability would increase with the
size of the black holes. This is clearly absurd.
The NBP framework is summarized in Sec. II. In
Secs. III and IV we review its implementation for cases
with a fixed cosmological constant. In Sec. V we intro-
where (h;s, CpsM) are the three-metric and matter field on
a spacelike boundary OM and the path integral is taken
over all compact Euclidean four geometries g„ that have
BM as their only boundary and matter field configura-
tions 4 that are regular on them; I(g„„,4) is their action.
The gravitational part of the action is given by
d x g'~'(R —2A)
16'
d'~ '~'K,
h. (2.2)
aM
where B is the Ricci-scalar, A is the cosmological con-
stant, and K is the trace of K,~, the second fundamental
form of the boundary BM in the metric g. For the origin
of the boundary term, see, e.g. , Ref. [3].
In the standard 3+1 decomposition [4], the metric is

3).
duce a massive scalar field and discuss the solutions of
written as
dsz =N dr + h;;(dx'+ N'dr)(dz~ + Nsdr) (2.
'Electronic address: R.Boussoodamtp. cam. ac.uk
t Electronic address: S.W. Hawkingodamtp. cam. ac.uk Assuming that the NBP is satisfied at w = 0, the Eu-

0556-2821/95/52{10)/5659{6)/$06. 00 52 5659 1995 The American Physical Society

5660 R. BOUSSO AND S. W. HAWKING 52

clidean action then takes the form This allows us to reintroduce a concept of Lorentzian
TBM time in such regions: We find the integral curves of
Sd d h~ V'I in minisuperspace and define the Lorentzian time
16m 0 t as the parameter naturally associated with them. Re-
x ( —K . K'' + K + R —2A) versely, if we demand that the NBP should predict clas-
sical Lorentzian universes at suFiciently late Lorentzian
d~g/ ~ (2 4) time, condition (2.10) must be satisfied. This means that
8m 0
there must be saddle-point solutions for which the path
Here 3B is the scalar curvature of the surface, and tensor in the v plane can be deformed such that it is eventually
operations are carried out with respect to the surface almost parallel to the imaginary w axis and that all the
metric 6;~. In the first term the boundary terms are q should be virtually real at late Lorentzian times. In
implicitly subtracted out at v = 0 and 7 = 7gM. But summary, the following conditions must be met.
it is an essential prescription of the NBP that there i8 (i) The NBP must be satisfied at w = 0.
no boundary at 7 = 0. So the second term explicitly (ii) At the end point 7sM of the path, the q must
adds the contribution f'rom v = 0 back in. It vanishes for take on the real values qM of the arguments of the wave
universes with spacelike sections of topology but can S, function:
be nonzero for the topology S x S2.
There are unresolved questions on how to choose the q (~sM) =qsM. (2.11)
integration contour and make the integral converge [5],
but we shall not discuss them here. Instead, we will use (iii) The q must remain nearly real in the Lorentzian
the semiclassical approximation vicinity of the end point:

CJO[h;, , @sM] = ) A„e '", (2 5) Re

&

~de
dq-
)
—0. (2.12)

where the sum is over the saddle points of the path in-
tegral, i.e. , the solutions of the Euclidean Einstein equa-
tions. In this paper, we neglect the prefactors A and
take only one saddle point into account for a given argu-
ment of the wave function. So the probability measure
will be
III. THE de SITTER SPACETIME
In this and the next section we review vacuum solu-
tions of the Euclidean Einstein equations with a cosmo-
logical constant A. First we look for a solution with

)@0[h;„C'aM](' = e ' (2 6)
spacelike sections S . Therefore we choose the metric
ansatz
where I ' is the
real part of the Euclidean saddle-point ds = N(7)d~ + a(~) . dOs. (3.1)
action.
By considering only spaces of high symmetry (homo- The Euclidean action is
geneous Ss or S x S spacelike sections) we restrict the
degrees of &eedom in the metric to a finite number q
The Euclidean action for such a minisuperspace model
I = ——
3~
4
Nd~a~
f a' +1 ——
qN'
A ,)
3
a (3.2)
with bosonic matter will typically have the form
An overdot denotes difFerentiation with respect to v. . We
define
I= — N& —
2 p d7-
dq dq~
d7-
+U q (2.7)
(3.3)
The saddle points will in general be complex solutions
q (w) in the v plane. In the semiclassical approximation
Variation of a and N yields the equation of motion
the following relations for the real and imaginary parts
of the saddle-point actions hold: a
~~

—
a
+H =0 (3.4)
——(V'I ') + —(V'I ) + U(q ) = 0, (2.8)
and the Hamiltonian constraint
a ——
— 1
~g Re ~yIm 0 (2.9) +H2 =0 (3.5)
a2 a2
where the gradient and the dot product are both with = 1. A solution
in the gauge N of Eqs. (3.4) and (3.5) is
respect to f~ Therefore Ii. will be a solution of given by
the Lorentzian Hamilton-Jacobi equation in regions of
minisuperspace where 4' has the property that a(~) = II 'sinII~ (3.6)
(VI ) «(VI'-) . (2.10) It is called the de Sitter spacetime. The NBP is satisfied
52 PROBABILITY FOR PRIMORDIAL BLACK HOLES 5661

at w = 0, = 0 and & —1. If we choose a path

where a w = 0, where
along the v '
axis to v. = z~, the solution will describe
half of the Euclidean de Sitter instanton S . Choosing
a=0, a=1, b=bo, andb=0. (4.8)
the path to continue parallel to the r axis, a(r) remains (There is a second way of satisfying the NBP for the
real and the conditions (i) to (iii) of the previous section Kantowski-Sachs metric [6], but it will not lead to a
will be satisfied: universe containing black holes. ) The path along the
axis describes half of the Euclidean Nariai instan-
a(r ) „. 2H
=H coshHr (3.7) ton S x S . Both two-spheres have the radius H
Continuing parallel to the w axis, the solution remains
This describes half of an ordinary Lorentzian de Sitter real:
universe.
So with the above choice of path, Eq. (3.6) corre-
a(r' „.) .=
2H
H ' cosh Hr' (4.9)
sponds to half of a real Euclidean 4-sphere joined to a b(r™)„. =H (4.10)
real Lorentzian hyperboloid of topology R x S . It can
be matched to any agM
appropriately, and for ay~
))
0 by choosing the end point
H the wave function os-
This describes half of a Lorentzian Nariai universe. Its
spacelike sections can be visualized as three-spheres of
cillates and a classical Lorentzian universe is predicted. radius a with a "hole" of radius b punched through the
The real part of the action for this saddle point is North and South pole. This gives them the topology of
S x S . Their physical interpretation is that of three-
I~,'»«« —— — 0
dr 'a (H a —1) = — . (3.8) spheres containing two black holes at opposite ends. The
black holes have the radius b and accelerate away Rom
each other as a grows. The Nariai universe is a degener-
The Lorentzian segment of the path only contributes to ate case of the Schwarzschild —de Sitter spacetime, with
IIm
the black hole horizon and the cosmological horizon hav-
ing equal radius [7].
The above path corresponds to half of a two-sphere
IV. THE NARIAI SPACETIME joined to a two-dimensional hyperboloid at its minimum

We still consider vacuum solutions of the Euclidean

radius H, cross a two-sphere of constant radius H
It can be matched to any agM )
0 but only to bg~ ——
Einstein equations with a cosmological constant, but we H so the wave function will be highly peaked around
now look for solutions with spacelike sections S x S . that value of b.
The corresponding ansatz is the Kantowski-Sachs metric The first term of Eq. (4.2) vanishes and so the real part
of the action for the Nariai solution comes entirely from
ds = N(r) dr ~ a(r) dx + b(r) dO, '. (4. 1) the second term:

(¹
The Euclidean action is (4.11)

(b~ Now we compare the probability measures correspond-

b ab
I=— 7r Nd~a ~
—
aN~~ +1 Ab
+2 — ing to the de Sitter and Nariai solutions. We find that
in these models with a fixed cosmological constant pri-
—ab —2abb mordial black holes are strongly suppressed, unless A is
+7r (4.2)
7=0 at least of order 1 in Planck units:
where the second term is the surface term of equation R.
exp ( —2IN'„;) = exp i

(2.4). We define

(4.3) (( exp f 37ri

&~
= exp ( —2I&,'s;«„j .
~

Variation of a, b, and N gives the equations of motion (4.12)

and the Hamiltonian constraint:
V. AN INFLATIONARY MODEL %WITHOUT
b
——
ab
=0 (4.4)
BLACK HOLES
b ab Of course, we know that A —0, and therefore the mod-
b ab a +H' =0, els of the previous section are rather unrealistic. How-
(4.5) ever, in inQationary cosmology it is assumed that the very
b ab a
early universe underwent a period of exponential expan-
2 —
ab
ab
+ ———
b
b~
1
b~
= 0. (4.6) sion. It has proven very successful to model this behavior
by introducing a massive scalar field 4 with a potential
A solution is given by &m 4 . If this field is suKciently far from equilibrium at
a(r) = H sinHr, b(7) = H = const. (4.7) the beginning of the universe, the corresponding energy
density acts like a cosmological constant until the field
It is called the Nariai spacetime. The NBP is satisfied at has reached its minimum and starts oscillating. During
5662 R. BOUSSO AND S. W. HAWKING 52

this time the universe behaves much like the I orentzian imation we must Gnd solutions in the complex v plane
de Sitter or Nariai universes described above. that meet conditions (i) to (iii) of Sec. II. In particular,
But there are two important differences due to the time the NBP must be satisfied:
dependence of the effective cosmological constant A g.
Firstly, for the solutions of the Euclidean Einstein equa- a=O, a=1, P=Pp, and/=0 for7. =0. (5.10)
tions in the complex w plane one can no longer find a
path on which the minisuperspace variables are always Assume that the initial value of the scalar field is large
real. However, we shall see that it is possible to satisfy and nearly real:
conditions (i) to (iii) of Sec. II by choosing appropriate
complex initial values. Secondly, it will be found in the Pp' )) 1 )) Qp (5.11)
next section that the black hole radius 6 is no longer con-
An approximate solution near the origin is given by
stant during inQation.
In this section, we introduce the massive scalar field for 1
az(r) = Re
~

the model corresponding to de Sitter spacetime, where R sin Hp (5.12)

)..
the spacelike slices are three-spheres containing no black
holes. This model was first put forward by Hawking
4r(r) = 0o+
1
,V
r"— (5.13)
[8]. From the Buctuations in the cosmic microwave back-
ground as measured by the Cosmic Background Explorer
(COBE) [9] it follows that m is small compared to the
Planck mass [10]:
for ~r~(0 (1/Hp '), where the Taylor series is obtained
by solving Eq. (5.8) iteratively for P, using the NBP con-
m 10 (5.1) ditions (5.10) and the approximation (5.12) for a. It has
the property that
We will find complex solutions and the complex initial
value of the scalar Geld, and we calculate the real part of p2„+j —0 for all n. (5.14)
the action. This has been done before by Lyons [11], but
his paper contains a logical error to which we will come We call Eqs. (5.12) and (5. 13) the inner approximation.
back later. Writing down the Taylor expansion explicitly to lowest
The ansatz for the Euclidean metric is again nontrivial order,

ds = N(r) dr + a(r) dOs2. (5.2) = &p 1+ +O(r')

&(r) 8g2p
(H«) (5.15)
Using the rescaled Geld
shows that P is almost constant near the origin.
= 4vr4, (5.3) As an outer approximation we use

(¹
we obtain the Euclidean action zm
P~ (r) = @p + r + yp exp(3iHpr), (5.16)
3
I= —
—3~
4
Ndra~ +1 ——
3
a
¹
——a
3
m P
aci (r) = ap exp —zm P(r') dr'
3 p

(5 4) zm
+cp exp P(r')dr' (5.17)
0
so that the effective cosmological constant is

A, g( )=rm P(r)2. (5 5)
for 0 ( ri ((
~3@'/m. While this solution does not
satisfy the NBP, it will be good outside the validity of
In analogy to Eq. (3.3) we define the inner approximation. Both the yp term and the cp
term can be neglected for ri 1/Ho ', but they are ))
useful for matching aci and Pci to a~ and P~ at some
A, rr(r) mP(r)
(5.6) 1/Hp
'
Compariso. n with Eq. (5.15) shows that
3
Variation with respect to a, P, and N gives the Euclidean ~3
equations of motion and the Hamiltonian constraint:
Xo yRe ' ™(Xo)= 0. (5.18)

a 2.2 + —
1
mP =0, In the region of the inner approximation, a will be
a 3 3
(5 7)
nearly real on the Lorentzian line ~ = 2 H„. . Matching '
0
P+3 — —m'P = 0, (5.8) a~ to ag fixes
aP
—
a ——
a2
1 —1
a2
-y
3
2 1
+-m
3 y =0. (5.9) ap
z
Re ') Cp
2~Re (5.19)
0 0

To evaluate 4'p(aoM, Ps~) using a semiclassical approx- and ensures that a will remain nearly real on this line. To
52 PROBABILITY FOR PRIMORDIAL BLACK HOLES 5663

make P(r) roughly real on the same line, by Eqs. (5.16) proximate solutions for this regime have been given by
and (5.18) we have to choose Hawking and Page [12].

„],Irn (5.20)
gyRe VI. AN INFLATIONARY MODEL WITH BLACK
HOLES
in the outer approximation.
in turn is fixed by matching P~ to Pci. Since it
is very small, this requires evaluation of Eq. (5.13) to a We now introduce a massive scalar field on a universe
very high order n. However, we need not calculate any with spacelike sections S x S . Thus we will obtain
coefficients since, by Eq. (5.14), P& is constant along the a cosmological model similar to the Nariai universe of
imaginary axis to any order n: Sec. IV. We find the complex solutions, initial conditions
and the action in analogy to the previous section, but
)Im( Im) )Im (5.21) point out a few differences.
Again we use the Kantowski-Sachs metric
Therefore it is convenient to choose a matching point 'TM
on the imaginary axis: ds = N(r) d~ + a(r) dz + b(~) d02 (6.1)

=0, ~'-=o(1r'H, ). and the rescaled field

By Eqs. (5.16) and (5.18) P~g is also constant along this

= 4vrcr . (6.2)
axis,
The Euclidean action is
)Im( Im) @Im (5.'-3) (b2
so the result of the matching analysis will be independent
I=— vr Nd7. a
(N2
+2 —
a N2
+1 —b N2
—b m
r
of the precise choice of 7M on the axis, as it should be.
The matching condition is +sr —ab —2abb (6.3)
7=p

4'z (~~) = Pc (rM) (5.24) and like in the previous section the effective cosmological
constant is given by
and by Eqs. (5.20), (5.21), and (5.23) we obtain
A, II(~) = m'P(~)'. (6.4)
yIm yIm
2/0 ' (5.25)
In analogy to Eq. (4.3) we define
This result is nontrivial (e.g. , PRQ' g vj'). We now see H(v) = QA, R(~) = mP(~). (6.5)
why the correct value for Po is obtained in Ref. [11],
although actually only @Oi is calculated there. Variation with respect to a, b, P, and N gives the Eu-
We have thus satisfied condition (ii) of Sec. II. By the clidean equations of motion and the Hamiltonian con-
continuity of the outer approximation, condition (iii) can straint:
be satisfied by fine-tuning $0 . Condition (i) is satis-
fied by the construction of the inner approximation. The b ab
P2 0 (6.6)

a'
only freedom left is the choice of $'. This variable ~~
b ab
parametrizes the set of solutions. b ab a = 0,
To calculate the Euclidean action for the solutions a
q P2+ m2$2 (6.7)
b ab
given above, we consider a path going along the real w
axis &om the origin to w =
0
and then parallel to. —+2- P —m2$=0,
b)
(6 8)
the imaginary w axis to 7sM. Both a and P are nearly real
on the Lorentzian segment of this path, so the real part of b2
the action can be approximated by an integral only over (6.9)
the first segment, using the inner approximation [11]:
The NBP conditions corresponding to an instanton of
' t1—a&m topology S x S2 are
R.
Ise = &~ d7 R. az 2
Pr —1
2 p
I

a=O, a=1, b=bp, b=O,

37r
(5.26) and P =0 for ~ = 0. (6.10)
2ni2 (QRe) 2
With the new definition (6.5) of H the inner approzi
The outer approximation is not valid after inflation mation is given by
ends, when P 0. However, at this point we are already
well inside the classical regime. A dust phase will ensue = Re
a~(r) R Slil Hp '7) (6.11)
where P oscillates; a and P will both remain real. Ap- p
5664 R. BOUSSO AND S. W. HAWKING 52

~&(r) = A+ ). 1
(6.12)
VII. THE PROBABILITY FOR PRIMORDIAL
BLACK HOLES
In the previous two sections we have calculated the
1 action for two infIationary universes. We now compare
b~(r) = (6.i3) the corresponding probability measures

for [r[ ( 0 (1/Ho '). The outer approximation is Pss ') = exp 3~
(7.1)
Pci (r) = 'leap + im7 + yp exp(iHpr), (6.i4)
(q o
i-'(~')')
and

aci(r) = im
T
—P(r')dr ( 2K
ao exp Ps s (Pp') = exp (7.2)
0
i-'(C.)'i
T The universe containing black holes is heavily suppressed,
+co exp im P(r') dr' (6.15) if P' is not large enough to make the initial effective cos-
0
mological constant equal to the Planck value. Thus the
1 formation of black holes with initial sizes significantly
hei(r) = (6.16) larger than the Planck scale is very unlikely. The semi-
classical approximation should be good in these situa-
for 0 (7™
((
Re tions, so one can have confidence in this conclusion.
The semiclassical approximation will break down for
A matching analysis completely analogous to that of
solutions with initial cosmological constants of the Planck
the previous section shows that a, b, and P will be nearly
value in a region where the curvature is on the Planck
real on the Lorentzian line w = 0 , if we choose the „. scale. However, this region contributes an action less
initial values than one in Planck units and one would not expect quan-
tum eKects to change this. Thus it seems clear that the
2pn' ' (6.i7) only primordial black holes with any significant proba-
mp
bility start with no more than the Planck size:
Po
' is a free parameter. )Re (1 (7.3)
An interesting feature of the outer approximation is
that the black hole radius grows with the horizon scale This corresponds to a large initial value of the scalar field
during inflation. On the Lorentzian line w =
0
the „. yRe ) 105 (7.4)
field decreases linearly with time until it reaches zero and
inflation ends. By Eqs. (6.14) and (6.16) 6 becomes very The Nariai solution is unstable to quantum fIuctua-
tions [7]. At the beginning of inflation it becomes a non-
large on the time scale
degenerate Schwarzschild —de Sitter spacetime. Once the
yRe black hole horizon is inside the cosmological horizon the
&&growth = (6.18) black hole will start to lose mass due to Hawking radia-
m
tion. If the black hole horizon is somewhat smaller than
Again the inner approximation is used to calculate the the cosmological horizon, the black hole will evaporate
real part of the Euclidean action. As in Sec. IV it comes and disappear. However, there is a significant probability
entirely &om the 7. = 0 term: that the areas of the two horizons will be nearly enough
equal for them to increase together. The consequences of
Is:.s =- Re
(bo ) =—
m
(6.19) this result for the global structure of the universe will be
presented in a forthcoming paper.

[1] H. Nariai, Sci. Rep. Tohoku Univ. 35, 62 (1951). (1990).

[2] J. B. Hartle and S. W. Hawking, Phys. Rev. D 28, 2960 [6] R. LaFlamme, Ph. D. thesis, 1988.
(1983). [7] P. Ginsparg and M. J. Perry, Nucl. Phys. B222, 245
[3] S. W. Hawking, in General Relativity: An Einstein Cen (1983).
tenary Survey, edited by S. W. Hawking and W. Is- [8] S. W. Hawking, Nucl. Phys. B239, 257 (1984).
rael (Cambridge University Press, Cambridge, England, [9] G. Smoot et al. , Astrophys. J. Lett. 396, Ll (1992).
1979). [10] J. J. Halliwell and S. W. Hawking, Phys. Rev. D 31, 1777
[4] A. Arnowitt, S. Deser, and C. Misner, in Gravitation, (1985).
edited by M. J. Bowick and F. Giirsey (World Scientific, [11] G. W. Lyons, Phys. Rev. D 46, 1546 (1992).
Singapore, 1986). [12] S. W. Hawking and D. N. Page, Nucl. Phys. B2$8, 789
[5] J. J. Halliwell and J. Louko, Phys. Rev. D 42, 3997 (1988).