HUTP-99/A048

MIT-CTP-2903
hep-th/9909134

Modeling the fifth dimension

with scalars and gravity
O. DeWolfe,1∗ D.Z. Freedman,2∗ S.S. Gubser,3∗ and A. Karch1∗
arXiv:hep-th/9909134 v4 9 May 2000

1
Center for Theoretical Physics,
Massachusetts Institute of Technology, Cambridge, MA 02139-4307, USA
2
Department of Mathematics and Center for Theoretical Physics,
Massachusetts Institute of Technology, Cambridge, MA 02139-4307, USA
3
Lyman Laboratory of Physics, Harvard University, Cambridge, MA 02138, USA

Abstract
A method for obtaining solutions to the classical equations for scalars plus gravity in
five dimensions is applied to some recent suggestions for brane-world phenomenology.
The method involves only first order differential equations. It is inspired by gauged
supergravity but does not require supersymmetry. Our first application is a full non-
linear treatment of a recently studied stabilization mechanism for inter-brane spacing.
The spacing is uniquely determined after conventional fine-tuning to achieve zero four-
dimensional cosmological constant. If the fine-tuning is imperfect, there are solutions
in which the four-dimensional branes are de Sitter or anti-de Sitter spacetimes. Our
second application is a construction of smooth domain wall solutions which in a well-
defined limit approach any desired array of sharply localized positive-tension branes.
As an offshoot of the analysis we suggest a construction of a supergravity c-function
for non-supersymmetric four-dimensional renormalization group flows.
The equations for fluctuations about an arbitrary scalar-gravity background are also
studied. It is shown that all models in which the fifth dimension is effectively com-
pactified contain a massless graviton. The graviton is the constant mode in the fifth
dimension. The separated wave equation can be recast into the form of supersymmetric
quantum mechanics. The graviton wave-function is then the supersymmetric ground
state, and there are no tachyons.
September 1999
∗
[email protected], [email protected], [email protected], [email protected]
1 Introduction
Phenomenologists have recently studied higher dimensional gravitational models con-
taining one or more flat 3-branes embedded discontinuously in the ambient geometry.
Scenarios with two 3-branes provide an explanation of the large hierarchy between the
scales of weak and gravitational forces and contain a massless 2++ mode which repro-
duces Newtonian gravity at long range on the branes [1, 2]. In the following paper we
present results of our study of models of this type: specifically, results on the smoothing
of discontinuities and stabilization of inter-brane spacings in 5-dimensional models with
gravity and a scalar field. The issue of fine-tuning in such models is also addressed.
We also discuss the fluctuation equations in these models somewhat differently from
treatments in the recent literature.
The centerpiece of this work is a supergravity-inspired approach to obtain exact so-
lutions of the nonlinear classical field equations in gravity-scalar-brane models which
is valid even without supersymmetry. After a brief introduction to the technical issues
in section 2, this approach is presented in section 3 and applied to a class of models
containing one positive and one negative tension brane [1] with compact S1 /Z2 geom-
etry in the fifth dimension. Stabilization of the brane spacing is a generic feature of
these models, but it is not guaranteed that the branes will be flat. Indeed, obtain-
ing flat branes requires a fine-tuning of the model precisely equivalent to setting the
four-dimensional cosmological constant to zero, and if the fine-tuning is imperfect, the
induced metric on the branes will be de Sitter space or anti-de Sitter space. The stabi-
lization mechanism is a generalization of the work of [3]; however, our treatment also
includes back-reaction of the classical scalar profile. An explicit model is presented in
section 4.
In section 5 we obtain smooth solutions of gravity-scalar models which approach
discontinuous brane geometries in a certain “stiff limit.” Any array containing only
positive tension branes can be smoothed in this way. We also remark on the usefulness
of our first-order formalism for the description of supergravity duals to renormalization
group flows.
Our constructions have some parallels in earlier supergravity domain wall literature
(see [4] for a review). There are also similarities with more recent literature, for example
[5, 6].
In section 6 we discuss the equations for linear fluctuations about a gravity-scalar-
brane configuration. We use the axial gauge and a parameterization in which the
4-dimensional graviton appears universally as a constant mode in the fifth dimen-
sion. This mode is normalizable since that dimension is either manifestly or effectively
compact. The graviton equation can be transformed into the form of a Schrödinger
equation in supersymmetric quantum mechanics. The graviton is the supersymmetric

1
ground state, so there is no lower energy state which would be a tachyon in the present
context.

2 The issues
We start with the five-dimensional gravitational action
√ 1 3
Z  
5
S= d x g − R+ 2 , (1)
4 L
in +−−−− signature. The most general five-dimensional metric with four-dimensional
Poincaré symmetry is
ds2 = e2A(r) ηij dxi dxj − dr 2 , (2)
with ηij = diag{1, −1, −1, −1, −1}. Anti-de Sitter space is the solution of the field
equations of (1) with A(r) = −r/L. This metric describes a Poincaré coordinate patch
in AdS5 with boundary region r → −∞ and Killing horizon region r → +∞.
The basic positive tension brane considered in [1, 2] is given by A(r) = −|r|/L. This
can be thought of as the discontinuous (in first derivative) pasting of the horizon halves
of two Poincaré patches with the 3-brane at r = 0. One can obtain this as the solution
of the field equations for an action consisting of (1) plus a brane tension term:
XZ q
Sbrane = − d4 x dr | det gij | λα δ(r − rα ) . (3)
α

Here we have generalized to any number of branes; gij is the metric induced on each
brane by the ambient metric gµν . For a single brane at r1 = 0 with brane tension λ1 ,
the AdS scale must be related by 1/L = λ1 /3 to achieve a solution in which the induced
metric is flat. This constraint represents a fine-tuning which is precisely equivalent to
setting the four-dimensional cosmological constant equal to zero.
One can obtain a system of one positive and one negative tension brane [1] by
considering two branes in (3) with λ2 = −λ1 and r2 = r0 . This leads to the piece-wise
linear scale function A(r) shown in figure 1a. The fifth dimension is then periodic
with period 2r0 and there is a reflection symmetry under r → −r. This is the S 1 /Z2
situation originally considered in [7, 8].
Another possibility is to consider [9] a second positive tension brane, which admits a
solution for A(r) shown in figure 1b. In this case, the bulk action (1) must be changed
to admit different scales L1 , L2 , L3 in the three spatial regions. The scales are related
to the brane tensions by 1/L1 + 1/L2 = 2λ1 /3 and 1/L3 − 1/L2 = 2λ2 /3 > 0. Again
these relations must be regarded as fine-tunings absent a dynamical mechanism by
which they arise.

2
 A A

(2)
L2
(1)
(3)
r r
r0 2r 0 r0
L1
(1) (2) L3

a) b)

Figure 1: a) A as a function of r for the S 1 /Z2 geometry, with one positive and one
negative tension brane, each at a fixed point of Z2 . b) A as a function of r for two
positive tension branes in an infinite fifth dimension.

There are solutions of the equations of motion for any choice of the inter-brane
spacing r0 in both scenarios above, so it is important to ask whether there is any
principle which fixes or stabilizes the value of r0 . A first thought is that the total
action integral of the configuration might depend on r0 , reflecting an imbalance of
forces on the two 3-branes, and therefore could be minimized. However it will be
shown in the next section that the action vanishes for all r0 , which apparently reflects
the fact that the “output” value of the classical four-dimensional cosmological constant
vanishes, as is consistent with the “input” value assumed when we considered solutions
containing flat 3-branes. In later sections we discuss models in which a real scalar field
φ with potential V (φ) is coupled to gravity with brane tensions λα (φ) depending on φ.
For a given choice of V (φ) and λα (φ), it is generally the case that the brane spacing
r0 is uniquely determined.
Discontinuous solutions of field equations would be less artificial if they could be
obtained as a limit of smooth configurations. In section 5 we present coupled scalar-
gravity models with potential V (ξ) (and no branes initially present). In these models
the scalar ξ plays a different role, that of an auxiliary field, and hence is given a different
symbol. The models have smooth domain wall solutions which approach any desired
discontinuous configuration of positive tension branes as a scale parameter in V (ξ) is
varied. Other parameters in V (ξ) determine the inter-brane spacing (e.g. r0 ) and AdS
scales (e.g. Li ) of the limiting solution, and the solutions have zero total action at all
stages of the limiting procedure. The scalar ξ is effectively frozen in the “stiff” limit
of discontinuous branes.
Only positive tension brane configurations can be smoothed in this way. A negative
tension brane effectively has negative energy which cannot be modeled in a conventional
gravitational theory. Nevertheless a negative tension brane is consistent with micro-

3
physical requirements if it is located at the fixed point of a discrete group action. The
crucial point is that transverse fluctuations are then projected out; otherwise they
would have negative kinetic terms.

3 The Goldberger-Wise mechanism

It was proposed in [3] that the dynamics of a scalar field could stabilize the size of an
extra dimension in the brane-world scenario of [1]. The mechanism was to have a scalar
φ with some mass in the bulk of a five-dimensional spacetime and some potentials λ1 (φ)
and λ2 (φ) on two four-dimensional branes at the boundaries of this spacetime. Such a
situation might be realized in the context of type I′ string theory [10, 7], the Horava-
Witten version of the heterotic string [8], or some more ornate string theory realization
of the basic scenario of [1]: in all cases, spacetime has the topology R3,1 × S 1 /Z2 . The
claim of [3] is that stabilization of the length of the interval S 1 /Z2 can be achieved
without fine-tuning the parameters of the model (namely the mass of the scalar and
the potentials λ1 (φ) and λ2 (φ)).
The analysis presented in [3] neglected back-reaction of the scalar field on the metric
as well as the effect of different scalar VEV’s on the tensions of the branes. The aim of
this section is to include these effects exactly. To achieve a static solution with 3 + 1-
dimensional Poincaré invariance to the full gravity-plus-scalar-plus-branes equations,
one fine-tuning is necessary. This fine-tuning amounts to setting the four-dimensional
cosmological constant to zero.
The fine-tuning is somewhat different from the ones discussed in [11, 12]. In [11] it
was argued for a theory with only gravity in the bulk that a nonzero four-dimensional
cosmological constant must necessarily be accompanied by rolling moduli (correspond-
ing to changing brane separations). In [12] it was conjectured that a state with nonzero
cosmological constant might relax to zero cosmological constant, again through evolu-
tion of some moduli specifying a brane configuration: in short, it was suggested that
an appropriate brane dynamics might fine-tune itself to zero cosmological constant.
We will find a more conventional alternative: there is generically a solution which is a
warped product of a maximally symmetric four-dimensional spacetime and an interval.
The four-dimensional spacetime can be flat Minkowski spacetime, de Sitter spacetime,
or anti-de Sitter spacetime, and which is chosen depends on the details of the scalar
potentials in the bulk and on the branes. Roughly speaking, one can construct a four-
dimensional effective potential Veff whose extremal value determines the cosmological
constant. There is no obvious dynamical principle in the absence of supersymmetry
which seems capable of forcing Veff = 0. In particular, the presence of a fifth dimension
simply does not constrain the extremal value of Veff . From a certain viewpoint this
should not come as a surprise: brane-world scenarios must reduce at low energies to a

4
four-dimensional gravity-plus-matter theory, including some brane moduli with some
potential, and it would seem rather accidental than otherwise for this potential to enjoy
a fantastic property like zero extrema.

3.1 A solution generating technique

We generalize the action (1) + (3) to include a scalar field φ(xi , r):
1 1
Z q   Z q
4
| det gµν | − R + (∂φ)2 − V (φ) − 4
X
S= d xdr d x | det gij |λα (φ) , (4)
M 4 2 α Bα

where M is the full five-dimensional spacetime and Bα is the codimension one hyper-
surface where each brane is located. It will always be assumed that the branes are at
definite values of r, so that the xi are perpendicular to the brane hypersurfaces.
The solution generating method described in this section could be applied to a fairly
general setup with many codimension one branes on a finite or infinite interval. In this
section our focus will be the case of a finite interval S 1 /Z2 where the only branes are the
ones at the ends of the interval. We will work in the “upstairs” picture: Z2 -symmetric
configurations on the circle S 1 . The bulk integration will extend over the entire S 1 .
Properly speaking, the action should be cut in half after this integration. This can be
achieved simply by setting G5 = 1/8π rather than 1/4π.
We will initially assume a five-dimensional metric of the form (2). We also assume
that the scalar depends only on r. These assumptions follow if one demands a solution
with 3+1-dimensional Poincaré invariance. We will later generalize slightly by replacing
ηij with a de Sitter or anti-de Sitter metric. It is straightforward to obtain the Ricci
tensor:
Rij = e2A (4A′2 + A′′ ) ηij R55 = −4A′2 − 4A′′ , (5)
and to show that the equations of motion are

∂V (φ) X ∂λα (φ)

φ′′ + 4A′ φ′ = + δ(r − rα ) ,
∂φ α ∂φ
2 2X
A′′ = − φ′2 − λα (φ)δ(r − rα ) , (6)
3 3 α
1 1
A′2 = − V (φ) + φ′2 .
3 6
We generally use primes to denote d/dr. The last of the equations in (6) is the usual
zero-energy condition that follows from diffeomorphism invariance. If one differentiates
it with respect to r, the result can be shown to vanish identically if the first two
equations are satisfied.

5
 By integrating the first two equations on a small interval (rα − ǫ, rα + ǫ) one can
derive the jump conditions
rα +ǫ 2 rα +ǫ ∂λα
A′  = − λα (φ(rα )) , φ′  = (φ(rα )) . (7)
 
rα −ǫ 3 rα −ǫ ∂φ
If these conditions are satisfied at each brane, and if the first and third equations of (6)
are satisfied away from the branes, then we have a consistent solution of the equations
of motion everywhere.
Unfortunately we are still left with a difficult non-linear set of equations. We have
been able to take advantage of one integral of the motion (namely the zero-energy
condition) to eliminate A′′ , and if we wished we could eliminate A′ algebraically in the
φ equation by using the zero-energy condition, but we would still have a difficult second
order equation for φ with no further obvious conserved quantities. The purpose of this
section is to exhibit a general method of reducing the system (6) to three decoupled
first order ordinary differential equations, two of which are separable. The method
is inspired by supersymmetry but can be carried out independent of it. We should
remark at the outset that our method is only simple in the case of a single scalar φ:
one of our differential equations has φ as the independent variable, and if there were
several scalars it would become a difficult partial differential equation.
Suppose V (φ) has the special form
!2
1 ∂W (φ) 1
V (φ) = − W (φ)2 , (8)
8 ∂φ 3
for some W (φ). Then it is straightforward to verify that a solution to
1 ∂W (φ) 1
φ′ = , A′ = − W (φ) , (9)
2 ∂φ 3
is also a solution to (6), provided we have
r +ǫ
1 rα +ǫ 1 ∂W (φ)  α ∂λα
W (φ) = λα (φ(rα )) , = (φ(rα )) . (10)
2 2 ∂φ rα −ǫ ∂φ

rα −ǫ 

(It was previously noted in [13] that the jump conditions could be satisfied in a specific
model if the brane tension was given identically by W (φ), which is a much stronger
constraint on the model than we assume.) Potentials of the form (8) occur in five-
dimensional gauged supergravity [14], and the conditions (9) arise as conditions for
unbroken supersymmetry: the vanishing of the dilatino variation leads to the first
equation in (9) and gravitino variation leads to the second.
For us, the key observation is that, given V (φ), (8) can be solved for W (φ), and there
is one integration constant in the solution. Whether a gauged supergravity theory can

6
be constructed so that the supersymmetry conditions lead to any desired W (φ) is an
interesting question which we will not address in this paper. (It would also be amusing
to ask whether one could come up with interesting supersymmetry-breaking scenarios
by starting with a five-dimensional gauged supergravity and constructing a solution
using (9) with the “wrong” W (φ).) The relevant point for the analysis at hand is
that (8) and (9) together have solutions specified by three integration constants, one
of which is the trivial additive constant on A. There are likewise three integration
constants for the solutions of (6), and again one is the trivial additive constant on A.
From this simple parameter count we may expect that the space of solutions includes
all possible solutions to (6).∗ Issues of global existence and discrete ambiguities seem
to be the only obstacles to realizing this expectation. These are best seen in a more
definite framework, so we will now proceed to our main example.
The rest of this section is devoted to the case where the only branes are the ones
at the ends of the interval S 1 /Z2 . Again, we work in the “upstairs” picture where
these branes are realized as kinks in A(r) at the fixed points of Z2 . If the Z2 reflection
includes an orientifolding, then string theory allows one of these two branes to have
negative tension. The negative tension brane must be located at a fixed point of the
discrete group action: it does not introduce difficulties with negative kinetic terms or
unboundedness of energy because it is just part of a background, not something which
can be dynamically created anywhere in space. We fix the additive ambiguity on the
variable r by taking the positive tension brane to be at r = 0. The negative tension
brane then lives at some r0 (see figure 1) which is the modulus of the theory that the
mechanism of [3] purports to stabilize. The physical parameters that go into the theory
are the scalar potential V (φ) and the tensions λ1 (φ) and λ2 (φ). These are assumed to
emerge from the microscopic physics (for instance string theory) which leads to this
five-dimensional picture in a low-energy limit (that is, low-energy compared to string
scale and ten-dimensional Planck scale as well as any further compactification scales).
A moduli stabilization mechanism would be regarded as fine-tuned if one has to impose
some relationship among V (φ), λ1 (φ), and λ2 (φ) to achieve a static solution.
Before explaining how the solutions to (6) can be generated using (8) and (9), let us
do a quick count of parameters and constraints to show that a fine-tuning is necessary
to obtain a static solution with flat branes. There are three integration constants for
the φ equation plus the zero-energy equation in (6): they are φ(0), φ′ (0), and A(0).
There is one additional parameter, namely r0 , so four parameters in all. There are
four constraints coming from the two jump conditions at the two branes. Naively one
would conclude that there is no fine-tuning: four contraints on four parameters can
∗
R. Myers [15] has also noted that (8) and (9) can be used to generate kink solutions, independent
of supersymmetry. In the study of RG flows in AdS/CFT he has considered an example with cubic
W (φ) which is similar to the single-brane solution which we will discuss in section 5.

7
generically be solved. But A(0) is completely irrelevant because A(r) enters into the
equations of motion and the jump conditions only through its derivatives. That leaves
three parameters subject to four constraints: indeed fine-tuned. This fine-tuning is
equivalent to the fine-tuning required in a theory without scalars between the brane
tensions and the bulk cosmological constant.
We will now argue in detail that any solution of (6) can be written as a solution to
(8) and (9) with an appropriately chosen W (φ). It is necessary to choose W odd under
the Z2 symmetry, just because A′ is equal and opposite at the two points on any given
Z2 orbit away from the fixed points. With this in mind we can restrict our attention
to region a in figure 1. The jump conditions become
1 1 ∂λ1
A′ (ǫ) = − λ1 (φ(0)) , φ′ (ǫ) = (φ(0)) ,
3 2 ∂φ
(11)
1 1 ∂λ2
A′ (r0 − ǫ) = λ2 (φ(r0 )) , φ′ (r0 − ǫ) = − (φ(r0 )) .
3 2 ∂φ
Plugging these relations into the zero energy condition, we learn that
!2
1 ∂λ1 1
− λ21 = V at φ = φ1 ,
8 ∂φ 3
!2 (12)
1 ∂λ2 1
− λ22 = V at φ = φ2 ,
8 ∂φ 3
where φ1 and φ2 are the values attained by φ(r) at r = 0 and r = r0 , respectively.
Notice these constraints have the same form as (8) , with the λα playing the role of W .
For generic V and λα , the equations (12) admit only a discrete set of solutions for φ1
and φ2 . Given the physical input into the model, namely V (φ), λ1 (φ), and λ2 (φ), the
discrete values φ1 , φ2 are the points in field space where flat branes can be consistently
inserted.
Let us now integrate the equation (8) and fix the single integration constant by
requiring W (φ1 ) = λ1 (φ1 ). Because of (8) we have ∂W ∂φ
(φ1 ) = ± ∂λ
∂φ
1
(φ1 ), and the plus
sign is guaranteed if we assume that ∂W∂φ(φ) has the same sign as ∂λ∂φ
1
(φ1 ) in the vicinity
′
of φ = φ1 . The solution (A (r), φ(r)) of (9) subject to φ(0) = φ1 must coincide with the
solution (A′ (r), φ(r)) of (6) subject to φ(0) = φ1 and φ′ (0) = 21 ∂λ
∂φ
1
(φ1 ), because both
of them satisfy the same boundary data. This is enough to conclude that locally every
solution of (6) can be generated by solving (8) and (9). Global issues of the existence
and uniqueness of solutions to (8) and (9) are best addressed with a specific model in
hand. We will return to these points in section 4.
Besides providing an efficient method for generating solutions to (6), the use of (8)
and (9) also allows us to characterize in a simple way how λ1 (φ), λ2 (φ), and V (φ)
have to be fine-tuned. Having first fixed W (φ) in the manner described in the previous

8
 Λ1 HΦL

VHΦL

Φ
Φ2 Φ1

WHΦL

-Λ2 HΦL

Figure 2: Sample W (solid line), V (dotted line), λ1 and λ2 (grey lines) as functions of
φ. By adjusting the integration constant of (8) one can arrange for λ1 to be tangent
to W , but then for λ2 also to be tangent amounts to a fine-tuning.

paragraph, and then integrated (9) to obtain φ(r), we can determine the position
of the second brane by φ(r0 ) = φ2 . There are no parameters left to fix (except for
the trivial additive constant on A), but we must still demand W (φ2 ) = −λ2 (φ2 ) and
∂W
∂φ
(φ2 ) = − ∂λ
∂φ
2
(φ2 ) in order that the jump conditions at the second brane be satisfied.
Because of the defining property (12) of φ2 , either one of these last two equations
implies the other up to a sign. Thus there is precisely one fine-tuning, as expected from
the earlier parameter count. The advantage of introducing W is that the fine-tuning
condition can be expressed in terms of the solutions of the single ordinary differential
equation (8) (see figure 2). It should be kept in mind that we are working strictly at
the classical level. If we tune parameters so that W (φ) and −λ2 (φ) are tangent, then
loop corrections to λα (φ) and V (φ) must be expected to spoil the relation.
It is true that if this fine-tuning can be achieved, there is no cosmological constant
allowed in the four-dimensional action. A quick way to see this is to show that the
lagrangian is a total derivative with respect to r when (8), (9), and (10) are satisfied:
then the four-dimensional lagrangian must vanish.† Let us define

W (φ) for 0 < r < r0
Ŵ (φ, r) = (13)
 − W (φ) for r0 < r < 2r0 ,
†
Since we assumed 3 + 1-dimensional Poincaré invariance in our ansatz from the start, zero four-
dimensional cosmological constant was guaranteed. The following computation is therefore only a
consistency check.

9
which is appropriately Z2 odd. Then it is straightforward to show that
1 1
q  Xq 
L= | det gµν | − R + (∂φ)2 − V (φ) − | det gij |λα (φ)δ(r − rα )
4 2 α

2 !2 
1 1 1 ∂ Ŵ 1 ∂ Ŵ

= e4A 3 A′ + Ŵ φ′ − (14)
X
− + − λα δ(r − rα )
3 2 2 ∂φ 2 ∂r α

d 1
  
− e4A 2A′ + Ŵ .
dr 2

In (14) we have used (8) (with W replaced by Ŵ ) but not (9). If the perfect squares
in (14) vanish, then we have

1 ∂ Ŵ X
= W (φ1)δ(r − r1 ) − W (φ2 )δ(r − r2 ) = λα δ(r − rα ) , (15)
2 ∂r α

where in the second equality we have used the jump conditions, (10). In comparing
with (13), recall that by convention r1 = 0 and r2 = r0 .
The form of (14) makes it clear that (9) are indeed a sort of BPS condition for solu-
tions of (6). However, because the perfect squares in (14) come in with opposite signs,
there is no obvious analog of a Bogomolnyi bound. Another important implication of
(14) is that the total action of any configuration of flat branes vanishes. This is even
true of non-periodic arrays provided A → −∞ as r → ±∞.

3.2 Non-zero cosmological constant

The fine-tuning to achieve zero cosmological constant was already commented on in
[3]. The purpose of this section is show that if the fine-tuning is imperfect, then there
are solutions without rolling moduli but where the metric on the branes is de Sitter
space or anti-de Sitter space.
Most of the analysis is similar to section 3.1, so we will be brief. The metric ansatz
is
ds2 = e2A(r) ḡij dxi dxj − dr 2 , (16)
where ḡij is the metric of four-dimensional de Sitter or anti-de Sitter spacetime: R̄ij =
−3Λ̄ḡij , where Λ̄ is the four-dimensional cosmological constant (positive for de Sitter
spacetime and negative for anti-de Sitter spacetime). Explicitly, we may write the
four-dimensional metrics as
√
dS4 : ḡij dxi dxj = dt2 − e2 Λ̄t (dx21 + dx22 + dx23 )
√ (17)
AdS4 : ḡij dxi dxj = e−2 −Λ̄x3 (dt2 − dx21 − dx22 ) − dx23 .

10
The five-dimensional Ricci tensor and the equations of motion are
 
Rij = e2A 4A′2 + A′′ − 3Λ̄e−2A ḡij , R55 = −4A′2 − 4A′′ ,
∂V (φ) X ∂λα (φ)
φ′′ + 4A′ φ′ = + δ(r − rα ) ,
∂φ α ∂φ
2 2X (18)
A′′ + Λ̄e−2A = − φ′2 − λα (φ)δ(r − rα ) ,
3 3 α
1 1
A′2 − Λ̄e−2A = − V (φ) + φ′2 .
3 6
The jump conditions (7) are unchanged. Neither A(r) nor |Λ̄| can be determined
unambiguously from the equations of motion because they enter only in the com-
bination A(r) − 21 ln |Λ̄|. We will see that this combination is what determines the
four-dimensional cosmological constant in four-dimensional Planck units. We could
adjust the additive constant on A, if we so desired, to set Λ̄ = 1 for de Sitter space-
time or Λ̄ = −1 for anti-de Sitter spactimee. The important point is not to count the
magnitude of Λ̄ as an adjustable parameter separate from the additive constant on A.
Already from (18) we can see why there should be a solution with no fine-tuning
of parameters. The φ equation and the zero-energy condition together have three
integration constants, and there is also the brane separation r0 . Because A itself rather
than just its derivatives enters into the equations (18), the additive constant on A is no
longer trivial. As before there are four boundary conditions (two jump conditions at
each brane), so generically one expects a (locally) unique solution for any given V (φ),
λ1 (φ) and λ2 (φ).
The solution in the bulk (more precisely, in region A of figure 1) can still be obtained
as a solution of a slightly modified system of first order equations,‡
1
A′ = − W γ(r) , (19)
3
1 ∂W
φ′ = ,
2γ(r) ∂φ
!2
1 ∂W 1 2
V = − W ,
8γ(r)2 ∂φ 3

which differ from (9) just by inclusion of the factor

v
9Λ̄ −2A(r)
u
u
γ(r) ≡ t1 + e . (20)
W (r)2
‡
We are grateful to Martin Gremm and Lisa Randall for pointing out to us an error in an earlier
version concerning these first order equations.

11
Note that this completely changes the character of the problem. In the case of zero
cosmological constant, the first order equations (8), (9) allowed us to find solutions for
given V directly by first integrating (8) to solve for W (φ), then using the equations (9)
to solve consecutively for φ(r) and A(r). We now see that if we do not fine-tune the
cosmological constant to zero, we obtain a complicated non-linear first order system of
differential equations for 3 functions W (r), φ(r) and A(r), now viewed as functions of
a single independent variable r, which we cannot simply solve for in sequence. V (φ)
is still to be considered the information that is put in from the Lagrangian, but its
relationship with W can no longer be isolated from the rest of the system. Derivatives
with respect to φ should now be thought of as
∂ 1 ∂
= ′ . (21)
∂φ φ (r) ∂r

To make this point more transparent, it is useful to rewrite the system (19) as an
autonomous system, that is in the form

A′ = f (A, W, φ), φ′ = g(A, W, φ), W ′ = h(A, W, φ) , (22)

where
1
f (A, W, φ) = − γ(r) W (r) , (23)
s3
2
g(A, W, φ) = 2 V (φ(r)) + W (r)2 ,
3
2
 
2
h(A, W, φ) = 2γ(r) 2 V (φ(r) + W (r) .
3
While for a generic V (φ) this system will still be hard to solve, it is very well suited
for generating examples where V (φ) is determined at the end. For any given shape of
the warp factor A(r) one desires, one can find a potential that supports such a solution
by the following procedure:
q ′
pick A(r), calculate A′ (r), solve A′ = − W3 γ algebraically
for W (r), and use φ′ = W 2γ
to obtain φ(r). V (r) can now simply be determined by
plugging in W and φ, and after inverting φ(r) to r(φ) one obtains the desired V (φ).
This procedure for example can be used to generate fat branes (as we will discuss them
in later chapters for the Minkowski case) with an AdS or dS worldvolume. Note that
this simple technique for generating examples is not possible in the obvious first order
system one could write down simply by introducing one new variable y with the one
new defining equation y = φ′ , as it is a standard technique for converting a system of
higher order equations into a first order system.

12
 Assuming (19), the jump conditions reduce to

∂λ1 1 ∂W 
λ1 (φ(0)) = W (r = 0)γ(0), (φ(0)) = ,
∂φ γ(0) ∂φ φ(0)

 (24)
∂λ2 1 ∂W 
λ2 (φ(r0 )) = W (r = r0 )γ(r0 ), (φ(r0 )) = .
∂φ γ(r0 ) ∂φ φ(r0 )


If for a given V (φ) we fix A(0) arbitrarily, then the 5 other initial conditions, φ(0),
A (0), W (0), W ′ (0), φ(0), φ′ (0) can be determined up to discrete choices, using the
′

3 equations from (19) evaluated at r = 0 and the 2 from the first line of (24). Then
(22) can be solved unambiguously for φ(r), W (r) and A(r). r0 is fixed by the last
equality in (19). One is left with one condition, namely the third equality in (24). It
is a (very complicated) constraint on A(0), which generically will have only discretely
many solutions. The point is that we wind up with exactly as many parameters as
constraints, so it doesn’t take any fine-tuning to get a solution.
There does not seem to be a simple way to express the action as a sum (or difference)
of squares plus total derivatives, in analogy to (14). However it is straightforward to
use the equations of motion to show that
1 1
q  Xq 
L = | det gµν | − R + (∂φ)2 − V (φ) − | det gij |λα (φ)δ(r − rα )
4 2 α
" # (25)
3 2A d 1 4A ′
q 
= | det ḡij | e Λ̄ − e A .
2 dr 2
When L is integrated over the S 1 parametrized by r, it must for consistency reduce to
the four-dimensional lagrangian,
1 q 1 3
 
L̄ = | det ḡij | − R̄ − Λ̄ , (26)
4πG4 4 2
evaluated on de Sitter or anti-de Sitter spacetime, where R̄ij = −3Λ̄ḡij , with Λ̄ positive
or negative, respectively. Comparison yields the relation
1
Z
= 4π dr e2A , (27)
G4
where as usual the r integration is over the whole of S 1 . For consistency with observa-
tion we must demand the bound
!2
1 > ℓHubble
≈ 10120 . (28)
G4 |Λ̄| ∼ ℓ4d Planck

In view of (27) this translates to

Z
dr e2(A(r)− 2 ln |Λ̄|) >
1
120
4π ∼ 10 . (29)

13
The function A(r) − 21 ln |Λ̄| is fixed by (19) and (24) once V (φ), λ1 (φ), and λ2 (φ) are
specified. A dramatic fine-tuning in these quantities is required to achieve (29).
In general it is difficult to obtain solutions to (18) or (19) in closed form. We can
however give a complete treatment of the case where there is no scalar and W is just a
constant (namely the square root of the bulk cosmological constant); see also [16, 17].
In this case the only equations we have to solve are the first equations in each line of
(19) and (24). The solutions can be expressed as follows:

dS4 : Λ̄ > 0
√ r1 − r 3 r1 3 r1 − r0
eA = Λ̄L sinh , λ1 = coth , λ2 = − coth
L L L L L (30)
AdS4 : Λ̄ < 0
r1 − r 3 r1 3 r1 − r0
q
eA = −Λ̄L cosh , λ1 = tanh , λ2 = − tanh
L L L L L
In the dS4 case it is necessary to restrict r0 < r1 . The main point which (30) demon-
strates is the following. Suppose one starts with any fixed negative bulk cosmological
constant, −4/L2 , and arbitrary but specified λ1 and λ2 , subject only to the constraint
that if one of the λα exceeds 3/L in magnitude, then the other must also exceed 3/L
in magnitude and be of the opposite sign. Then there is a unique solution to (30) up
to the usual ambiguity between the additive constant on A and the magnitude of Λ̄.
Both r1 and r0 will be fixed in this solution, and so will the combination A − 12 ln |Λ̄|
which determines the four-dimensional cosmological constant in Planck units. The only
exception is when λ1 = −λ2 = L3 : in this case the branes are flat, r1 is a meaningless
additive constant on A, and the brane separation r0 is not fixed.
The bulk solutions in (30) have vanishing Weyl tensor, hence they are locally AdS5 .
All we have found, then, is an embedding of AdS4 and dS4 as codimension one hy-
persurfaces in AdS5 . To verify this one can find an explicit change of variables which
brings the bulk metric into the standard form

ds2 = e−2r̃/L (dt̃2 − dx̃21 − dx̃22 − dx̃23 ) − dr̃ 2 . (31)

If we demand that the map from untilded to tilded coordinates be orientation preserv-
ing, then the natural choice is
√ r1 − r −√Λ̄t √ r1 − r
dS4 : t̃ = − Λ̄ coth e , r̃ = − Λ̄Lt − L log sinh ,
L L
x̃1 = x1 , x̃2 = x2 , x̃3 = x3
(32)
q
r1 − r √−Λ̄x3 q
r1 − r
AdS4 : x̃3 = −Λ̄ tanh e , r̃ = −Λ̄Lx3 − L log cosh ,
L L
t̃ = t , x̃1 = x1 , x̃2 = x2 .

14
Let us now focus on the dS4 case with one positive and one negative tension brane at
the ends of the bulk. A solution of the form (30) maps to a strip of the t̃ − r̃ plane
between two curves of the form t̃ = −cα er̃/L . Here c1 and c2 are positive constants.
Because ∂/∂ t̃ is a Killing vector of the bulk geometry, we can trivially obtain a broader
class of solutions which have as their boundaries curves of the form t̃ − t̃α = −cα er̃/L ,
where now t̃1 and t̃2 are additional constants, only one of which can be set to 0 through
diffeomorphism freedom. In these solutions the proper distance between the branes is
not constant. In fact, generically the branes intersect at some point, or they intersect
the boundary of AdS5 at different points—or both. In the latter case the graviton
bound state ceases to exist at some finite time as measured on the negative tension
brane. This reinforces the intuition that brane-world cosmology can encounter some
curious pathologies.
The strategy of displacing one boundary by some distance along the flow of a Killing
vector of AdS5 can also be applied to flat branes. For instance, one could shift the
negative tension brane forward along the global time of AdS5 to obtain a new solution
where the proper distance between the branes is non-constant. The positive and neg-
ative tension branes would then intersect at some time in the distant past, and the
positive tension brane would again retreat to the true boundary of AdS5 at a finite
time as measured on the negative tension brane. This is a catastrophe since it means
that gravity would cease altogether in four dimensions: the four-dimensional Planck
length would vanish.

4 An explicit model
It is useful now to turn to an explicit example with non-trivial dynamics for a single
scalar. For simplicity, we choose quadratic W (φ), λ1 (φ), and λ2 (φ) which are tangent
to one another in the manner illustrated in figure 2. Explicitly,
3
W (φ) = − bφ2 ,
L
b2 2b b2 φ4
!
3
V (φ) = − 2 + + φ2 − , (33)
L 2 L 3
λ1 (φ) = W (φ1 ) + W ′ (φ1 )(φ − φ1 ) + γ1 (φ − φ1 )2 ,
λ2 (φ) = −W (φ2 ) − W ′ (φ2 )(φ − φ2 ) + γ2 (φ − φ2 )2 .
We stress that the physical properties of the model are summarized by V (φ) and the
λα (φ): in the absence of supersymmetry, there is no preferred choice of W (φ). In
section 4.2 we will analyze the different possible W (φ) that lead to the particular
quartic V (φ) exhibited in (33). Until then we will just assume that the particular
W (φ) that is tangent to λ1 (φ) happens to be the quadratic one shown in (33). We

15
make this assumption in order to obtain solutions in closed form. The only physical
fine-tuning is the requirement that −λ2 (φ) is also tangent to W (φ). The quantities
L, b, φ1 , φ2 , γ1 , and γ2 are parameters of the various potentials, and no dimensionless
ratio of them should be large if we want to preserve naturalness.
We will always assume that γ1 and γ2 are positive so that the energetics of λ1 and
λ2 tend to stabilize the positions of the branes in field space. We will usually assume
b > 0 as well. It should be noted that V (φ) is unbounded below, as is common and
without pathology in AdS supergravity.

4.1 Analytical calculations

It is trivial to solve the first order equations (9) in the model (33) to obtain

φ(r) = φ1 e−br ,
r 1 (34)
A(r) = a0 − − φ21 e−2br .
L 6
The brane spacing is determined by the condition br0 = ln(φ1 /φ2 ). The difference
A(0) − A(r0 ) gives the number of e-foldings in discussions [1, 9] of the gauge hierarchy
problem,§ and one easily obtains
1 φ1 1 2
A(0) − A(r0 ) = ln − (φ1 − φ22 ) . (35)
bL φ2 6
Phenomenologically one wants
MPlanck
A(0) − A(r0 ) ≈ ln ≈ 37 . (36)
Melectroweak

If b > 0, then φ21 − φ22 > 0, and only first term can contribute to the hierarchy. This is
conceivable if bL is fairly small: for instance, if φ1 /φ2 = e then one needs bL ≈ 1/37.
If b < 0, then both terms in (35) could contribute to the hierarchy. One could for
instance obtain an acceptable hierarchy by taking bL = 1, φ1 = 1, and φ2 = 15.
The treatment of [3] ignored back-reaction of the scalar profile on the geometry.
Crudely speaking this means one should drop the second term in (35) since it came
from a term proportional to the square of the scalar field in (34). More precisely, (14)
of [3] can be reproduced exactly by dropping the second term in (35) and identifying
their m2 L2 with our bL in the limit of small bL. Thus the analysis of [3] was essentially
adequate for the case b > 0, where to obtain a large hierarchy one wants a bulk
geometry which is not so far from AdS5 that the second term of (35) is large. However
§
We assume that the four-dimensional and five-dimensional Planck scales are comparable. It is
possible to relax this assumption [18] since the additive constant on A(r) is a free parameter.

16
the inclusion of back-reaction becomes quite important in the b < 0 case, where a large
hierarchy can be most easily obtained via a geometry which deviates strongly from
AdS5 .
Any mechanism for generating large numbers must be probed for robustness. We
may ask, once the hierarchy (36) is obtained, how much can the parameters change
and still give the same weak scale to within errors? For definiteness, let us ask what
change of parameters shifts A(0) − A(r0 ) by no more than 0.02: this would amount to
a shift of the weak scale by two percent, which is about the ratio of the Z width to its
mass. In the b > 0 scenario we described above, a change of φ1 /φ2 by about one part
in 2000 changes the weak scale by two percent: multiplicative shifts in this ratio are
magnified by the factor 1/bL. In the b < 0 scenario, changing φ2 by about one percent
changes the weak scale by two percent. Thus (superficially at least) the b < 0 scenario
is more robust.

4.2 Numerics
We now change gears and refocus on (8). The purpose is to illustrate the problem
of selecting a superpotential W (φ) which reproduces a given potential function V (φ).
However, we shall be content to explore this question only in the model of this section,
where V (φ) is given in (33). It is convenient to rescale variables, partly to prepare for
use of the MATLAB linked program DFIELD5 [19]. We therefore define
s
3
φ= t,
8
1 4 t2 t4
 
2
V (φ) = 3 b U(t) , U(t) = − 2 2 + 1 + − , (37)
bL bL 16 64
1 t2
W (φ) = 3 b x(t) , x0 (t) = − .
bL 8
We denote the rescaled preferred superpotential by x0 (t) since we will consider other
superpotentials corresponding to the potential U(t).
In this notation (8) takes the form
!2
dx
= x(t)2 + U(t) . (38)
dt
There is a sign ambiguity in taking the square root which must be kept in mind, but
we will discuss only the features of the differential equation which results from the
positive root, namely
s
dx 1 4 t2 t4
 
= x2 − 2 2 + 1+ − . (39)
dt bL bL 16 64

17
 8

0
x

8 6 4 2 0 2 4 6 8
t

Figure 3: The t–x plane, showing forbidden regions.

The equation is roughly like the energy equation in the mechanics problem of a particle
in an inverted harmonic potential. As in mechanics there are forbidden regions of the
t−x plane where x2 +U(t) < 0. At a boundary of this region, which would be a turning
point in a mechanics problem, the slope dx dt
vanishes. According to the general theory
of first order differential equations there is a unique solution curve through every point
not in a forbidden region. The inequality
 
 dx  q
≤ |x| + |U(t)| , (40)
 
 dt 
 

shows that no solution reaches |x| = ∞ at a finite field value.

The DFIELD5 program quite rapidly provides a reasonable global and quantitative
picture of the space of solutions. The quantities of our problem depend only on the
single dimensionless parameter bL, and we set bL = 1 in our numerical work.
A large-scale plot of the t–x plane is shown in figure 3, and we see two large forbidden
regions on the left and right and a small one in the center. The inclined lines at a grid
of points are the slopes, obtained from (39), of the solution curves through each point.
The solution through (t, x) = (0, 1) is shown, and it is easy to see that it gives the
preferred superpotential x0 (t) = 1 − t2 /8 only for t < 0. This is related to the sign
ambiguity of the square root in (39), and it is not a difficulty for us because we are

18
primarily concerned with the region t < 0 which includes the full range of the geometry
containing two branes which was discussed in the first part of this section.
Some other representative solutions are also plotted in figure 3. It is not proven, but
it appears to be the case that the only solutions which give a superpotential defined
on the full field space −∞ < t < ∞ are the curve through (t, x) = (0, 1) and its mirror
image through (t, x) = (0, −1), which is also shown in figure 3. Other solution curves
reach the boundary of the allowed region at a finite value of t in one direction, and
one can see that x′ (t) vanishes but x′′ (t) diverges as one approaches the boundary.
By examining an approximate form of (39) and (9) , one can show that these curves
approach the boundary at a finite value of the coordinate r. It then appears that the
solution curve reflects, and one must consider solutions of (39) with the other sign
of the square root. The scale factor A(r) is smooth at the turning point. This issue
does not affect our application, since the full brane geometry is contained in a region
without turning points.
Let’s recall the logic of our construction. The potential V (φ) and left-hand brane
tension λ1 (φ) are matched at a chosen value φ = φ1 . We then choose the unique
superpotential W (φ) which satisfies W (φ1) = λ1 (φ1 ) and agrees in sign of slope with
λ′1 (φ1 ). Agreement in the magnitude of the slope is guaranteed by (8) and (12). We
then integrate the first order equations (9) which gives the unique solution of the
second order problem (6) with the initial conditions φ(0) = φ1 , φ′ (0) = λ′1 (φ1 ), the
latter from the jump condition (7). For consistency, it is useful to know that any other
choice of W (φ) leads to a different solution of (6), one which does not satisfy the jump
conditions. This is quite clear from figure 3, since the jump conditons. e.g. (10),
q are no
longer satisfied if we change solution chosen at the relevant fixed value t1 = 8/3 φ1 .
We have explored our suggested solution generating technique in only one model.
Global issues associated with the turning points do not spoil the applicability of our
method, and the method is certainly easy to use in the reverse mode where we start
with a conveniently chosen W (φ). We believe that this favorable situation is generic.

5 Smooth solutions modeling branes

So far we have been considering solutions to an action that contains explicit δ-functions
at the positions of the branes. One might wonder to what extent this approach has
already built in the answers one wants to obtain. The purpose of this section is to
present a one-parameter family of purely 5d Lagrangians for gravity coupled to a scalar,
labelled by the parameter β, whose solutions are generically smooth and asymptote to
a specific δ-brane solution of the type considered so far. For generic β, the smoothed
branes appear as domain walls interpolating between various scalar vacuua. In the

19
“stiff” limit (β → ∞) the second derivative of the scalar potential goes to infinity, so
the scalar becomes very heavy and can be integrated out. The parameters entering the
scalar potential become the brane tensions and positions associated with δ-function
terms in an action of the type (1) + (3) after integrating out the scalar.
Several comments are in order. First, as mentioned before, we will not be able to
treat negative tension branes in this framework. Second, the solutions presented in this
section do not have any fields living on the brane, since the smooth solitons that in the
stiff limit become the branes do not have any zero modes. Both these obstacles can be
avoided by introducing “by hand” the δ-functions in the action, but this is precisely
what we want to avoid with the smooth formalism. In principle, the second limitation
above could be overcome by studying a more complicated smooth model which allows
for non-trivial zero modes on the brane.
Last but not least we should emphasize that even though we are considering once
more 5d gravity coupled to a scalar, this time the scalar should not be thought of as
the bulk scalar φ we studied so far, which plays the role of a modulus for the fifth
dimension. Instead it is the scalar that the branes are made of! In order to avoid
confusion we will call this auxiliary scalar ξ and reserve the symbol φ for the modulus
scalar. In the stiff limit, where the soliton approaches the array of localized δ-like
branes the fluctuations of ξ are frozen out. The bulk scalar φ has to be introduced as a
second scalar. Interactions localized on the brane, like the λ(φ) we introduced earlier,
can be mimicked by coupling the bulk scalar φ only to derivatives of ξ.
We study a five-dimensional action of the form
1 1
Z q  
4
S= d x dr |detgµν − R + (∂ξ)2 − V (ξ) . (41)
M 4 2
We will work in the first order framework and hence take V (ξ) to be given in terms of a
“superpotential” as in (8) and study solutions to the first order equations (9). We will
show that once we specify the potential appropriately, the resulting solitonic solution
describing an array of branes with tension λα at positions rα in the fifth dimension is
specified uniquely.
We are interested in the case where the scalar profile is given as a solitonic domain
wall configuration interpolating between various vacua for the scalar field, e.g. written
as
X 1
ξ(r) = √ κα tanh(β(r − rα )), (42)
α β
or a similar function that has the properties that

• in the “stiff” limit (β → ∞) it reduces to an array of step functions of height

κα
∼√ , and that
β

20
 • its first derivative is always negative and approaches a collection of δ-functions
κα
at position rα of strength ∼ √ .
β

Note that latter property requires all κα to be positive, ensuring that the function ξ(r)
is invertible. This solution in the stiff limit becomes an array of branes of tension
4
λα = κ2α (43)
3
and only positive tensions appear.
Can we find a W (ξ) in such a way that it allows a solution of the form specified in
(42)? In order to do so, we just rewrite the first order equation for the scalar flow in
(9) as
∂W (ξ) ∂W (ξ(r)) ∂r(ξ) W′
2ξ ′ = = = ′ (44)
∂ξ ∂r ∂ξ ξ
Z r
W (r) = 2 (ξ(r ′ )′ )2 dr ′. (45)

Using invertibility of ξ(r) we can re-express W (r) as W (ξ) and hence obtain a potential
V (ξ) which leads to a solution of the desired form. The one integration constant in W
corresponds to an “overall” bulk cosmological constant. It should be chosen in such
a way that A′ (r) = − 31 W (r) is positive (negative) to the left (right) of all branes.
Since A′′ is always negative, it is always possible to choose the integration constant
this way. As we will see in the next section this property is enough to ensure that there
exists a 4-dimensional graviton. Now we can turn the philosophy around and say that
once we have specified V and hence specified the action, or more precisely the bulk
cosmological constant and the cosmological constants between the various branes given
in terms of the value of V (ξ) at its minima, the first order equations then provide us
with a solution of the form (42) for ξ(r) together with the A(r). In the stiff limit this
solution approaches an array of sharply localized branes at positions rα and tensions
λα .
One should think of V (ξ) as being obtained from integrating out the microscopic
physics. One then can ask again whether there is some dynamical principle that de-
termines the parameters in V . Since we expressed W as an integral over (ξ ′)2 those
parameters are the κα and the rα . Calculating the action integral of the solution as a
function of κα and rα one finds once more that it is always zero. We remain with a
serious fine-tuning problem: the underlying theory has to be arranged in such a way,
that for given λα and rα the potential has precisely the form specified by (45). In the
stiff limit all that remains of V are its values at the minima – the inter-brane cosmo-
logical constants¶ – and the fine-tuning problem reduces to the standard fine-tuning
¶
The normalization in (42) was chosen in such a way, that those inter-brane cosmological constants
2
remain finite in the stiff limit, L1 jumps by 8κ9 when crossing a brane.

21
of the bulk cosmological constants against the brane tensions.
For example, in the case of a single brane we start with
κ
ξ(r) = √ tanh(βr) , (46)
β
leading to √
′ β
ξ (r) = κ 2 , (47)
cosh (βr)
and hence
1 β
Z  q 
W = 2κ2 (ξ ′)2 dr = 2κ2 tanh(βr) − tanh3 (βr) = 2κ β(ξ − 2 ξ 3 ) . (48)
3 3κ
A is simply obtained by integrating W . In the multi-brane arrays the solution becomes
slightly more complicated due to the cross-terms in (ξ ′)2 but it is still analytical. One
can show that in the stiff limit all possible smoothings lead to the same brane array.
Before we end our discussion on smoothing of the singular solutions, let us comment
on how the coupling to the additional bulk scalar looks in this framework. In order to
mimic the localized interactions for the bulk scalar φ we couple it to the derivatives of
the auxiliary scalar ξ. Basically, this means that we couple a σ-model for the scalars to
gravity, where the kinetic terms of the auxiliary scalar ξ depend on the bulk scalar φ.
In the stiff limit this once more will reduce to the solutions discussed in the previous
sections.
Similar to (8) and (9) we can find a first order formalism for the general action
1 1
Z q  
S= d xdr |detgµν | − R + GIJ ∂ µ φJ ∂µ φI − V (φ) ,
4
(49)
M 4 2
where GIJ is a metric on the scalar target space. Any solution to
1 IJ ∂W (φ) 1
(φI )′ = G , A′ = − W (φ) , (50)
2 ∂φJ 3
is also a solution to the full second order equations provided V is of the special form
1 IJ ∂W (φ) ∂W (φ) 1
V (φ) = G − W (φ)2 . (51)
8 ∂φI ∂φJ 3
Choosing a two scalar model with φ1 = φ and φ2 = ξ and choosing G12 = G21 = 0,
G11 = 1 and G22 (φ) to be an arbitrary function of φ we should once more be able (50)
to engineer a smooth model, this time limiting to multi-brane-array in the presence of
the bulk scalar φ with localized interactions.
A count of parameters similar to the ones in section 3.1 and 3.2 allows us to conclude
that—at least locally—any solution of the equations of motion following from (49)

22
which preserves 3 + 1-dimensional Poincaré invariance can be written as a solution of
(50) for an appropriately chosen W (φ) satisfying (51). Suppose there are n scalars
involved in the action (49). Each of them satisfies a second order equation of motion.
The scale factor A satisfies a first-order zero-energy constraint analogous to the last
line of (6). So there are 2n + 1 integration constants. One of them can be absorbed
into an additive shift on r. Now, (50) leads to only n+1 integration constants since the
scalar equations are now first order. But there are also n integration constants in (51)
regarded as a partial differential equation for W (φ). Again one integration constant
can be absorbed into an additive shift on r. The point is that either way we have the
same number of integration constants, so barring non-generic phenomena and global
obstructions, the solution spaces are the same.
This is quite an interesting result in view of the AdS/CFT correspondence [20,
21, 22]. One of the main puzzles in the correspondence is how one might translate
the renormalization group (RG) equations, which are first order, into supergravity
equations, which are second order. In [14] first order equations were extracted from
the conditions for unbroken supersymmetry. These equations are suggestive of an RG
flow based on the gradient of a c-function. The c-function is W (φ), and its relation
to the conformal anomaly arises because of the equation A′ = − 31 W : in regions where
the scalars are nearly constant and the geometry is nearly AdS5 , an application of the
analysis of [23] shows that the Weyl anomaly coefficients in the conformal field theory
are proportional to the third power of the radius of AdS5 , or equivalently to |W |−3.
(Thus in a sense it would be more appropriate to speak of |W |−3 as the c-function.)
In a non-supersymmetric “flow,” the c-function can still be defined [24, 14] as −3A′ ,
and it is possible to demonstrate A′′ ≤ 0 using only the weakest of positive energy
conditions [14]. But then the spirit of RG is lost: one wants to have a notion of
a first order flow through the space of possible theories labelled by different values
of parameters, and whatever c-function one constructs should be defined in terms of
those parameters. The construction of W indicated in (51) seems to realize this idea
explicitly.
However there are some caveats. First, W depends on n integration constants,
where as before n is the number of scalars. It seems reasonable that these integration
constants can be interpreted as specifying the state of the dual field theory, which does
not change under RG—only the Hamiltonian evolves. Second, the same phenomena of
forbidden regions and turning points that we discussed in section 4.2 occur also in the
case of several scalars. A forbidden region is a region of (W, φ) space where V (φ)+ 13 W 2
is negative. Barring singular behavior in GIJ , one finds that the gradient of W vanishes
at the border of these regions, so no flow can cross over. Rather, flows reflect from the
border and the subsequent flow is controlled by a different branch of W . Because of
the multi-valued nature of W , we do not regard (50) as a wholly satisfactory starting

23
point for the transcription of supergravity equations into RG equations. However it is
perhaps a step in the right direction.

6 Fluctuations around the solution

Finally we examine the equations governing fluctuations of the metric and scalar around
the classical background solutions of the equations of motion of the action (4). Our
methods are somewhat different from those in the literature. We choose an axial-
type gauge, and the resulting form of the four-dimensional graviton is particularly
simple. Transverse traceless modes in general obey the equation of a massless scalar
in the curved background, and by recasting this as the Schrödinger equation for a
supersymmetric quantum mechanics problem, we argue that there are no space-like
modes threatening stability.
We impose the “axial gauge” constraint, so named for its resemblance to A3 = 0 in
electrodynamics:
hµ5 = 0 , (52)
where µ = 0, 1, 2, 3, 5. We can then write the total metric in the form

ds2 = e2A(r) (ηij + hij ) dxidxj − dr 2 , (53)

where we extracted a factor e2A from the fluctuation term to simplify future equations.
Axial gauge is not a total gauge fix, as diffeomorphisms generated by a vector field
ǫi = e2A(r) ωi (xj ), ǫ5 = 0 preserve the condition (52) while transforming the fluctuations
hij as
hij (xk , r) → hij (xk , r) + ∂i ωj (xk ) + ∂j ωi (xk ) . (54)
Note the resemblance to four-dimensional diffeomorphisms. k
The Ricci tensor can be computed from the metric (53). To zeroth order in the
fluctuations we continue to have (5), while to first order we calculate (using Maple):
1 2 1 1
 
(1) 2A
Rij =e ∂r + 2A′ ∂r + A′′ + 4A′2 hij + ηij e2A A′ ∂r (η kl hkl ) − hij
2 2 2
1 kl
− η (∂i ∂j hkl − ∂i ∂k hjl − ∂j ∂k hil ) , (55)
2
(1) 1 (1) 1
R55 = − (∂r2 + 2A′ ∂r ) η kl hkl , Rj5 = η kl ∂r (∂k hjl − ∂j hkl ) ,
2 2
where = η ij ∂i ∂j is the flat four-dimensional Laplacian. Einstein’s equations in Ricci
k
There is a more general residual gauge invariance involving a non-vanishing ǫ5 (xk ). See [25].

24
form require that Rµν = T̄µν ≡ Tµν − 31 gµν T αα , and we find
" !
(1) 2 ∂V (φ) X ∂λα (φ)
T̄ij = − e2A 2 + δ(r − rα ) φ̃ ηij
3 ∂φ α ∂φ
! #
X
+ 2V (φ) + λα (φ)δ(r − rα ) hij (56)
α
!
(1) 4 ∂V (φ) X ∂λα (1)
T̄55 = 4φ′ φ̃′ + +2 δ(r − rα ) φ̃ , T̄j5 = 2φ′ ∂j φ̃ .
3 ∂φ α ∂φ

Additionally, the equation of motion for the scalar fluctuation φ̃ is

∂ 2 V (φ) X ∂ 2 λ(φ)
!
−2A ′′ ′ ′ 1
e φ̃ − φ̃ − 4A φ̃ + 2
+ 2
δ(r − rα ) φ̃ = φ′ η ij h′ij . (57)
∂φ α ∂ φ 2
(1) (1)
The equation Rij = T̄ij further simplifies as a consequence of the zeroth-order equa-
tion of motion (6) :
4 2X
A′′ + 4A′2 = − V (φ) − λα (φ)δ(r − rα ) , (58)
3 3 α
to
1 2 1
 
2A
e ∂r + 2A′ ∂r hij − hij
2 2
1 1
+ ηij e2A A′ ∂r (η kl hkl ) − η kl (∂i ∂j hkl − ∂i ∂k hjl − ∂j ∂k hil ) = (59)
2 2 !
2 2A ∂V (φ) X ∂λα (φ)
− e 2 + δ(r − rα ) φ̃ ηij .
3 ∂φ α ∂φ
Let us now consider the transverse traceless components of hij , defined by the non-local
projection [26]:
1 1
 
hij = (πik πjl + πil πjk ) − πij πkl hkl = hij + . . . , (60)
2 3
where πij ≡ (ηij − ∂i ∂j / ) and . . . indicates nonlocal terms. The hij satisfy

∂ j hij = η ij hij = 0 . (61)

We emphasize that (61) applies only to the components defined in (60) and is not a
gauge choice; it would be incompatible with (52) and the residual gauge freedom (54).
For the hij , (59) simplifies enormously. The transverse traceless projection removes
the right-hand side, and we are left with
 
∂r2 + 4A′ ∂r − e−2A hij = 0 . (62)

25
Notice that all δ-function jumps have canceled out; this is nothing but the equation
of motion for a free massless scalar in our curved background. In an AdS5 black hole
background, the spin-2 components of the graviton were also found to obey a free scalar
wave equation [27, 28].
We expect one solution of our equations to be the four-dimensional graviton. Since
it is massless in the four-dimensional sense, it must obey hij = 0. We can easily see
that such a solution to (62) is simply the r-independent plane wave

hij = Cij eipx , (63)

where p2 = 0 and Cij is a constant. Thus in this presentation the phenomenological

graviton has a very simple form.
As we will argue below, the norm of metric fluctuations is
Z
ij
|| h ||2 = dr e2A(r) hij h , (64)

where indices are raised with η ij . We see that the graviton mode (63) is normalizable
because the r-direction is effectively compactified in these models. The S 1 /Z 2 geome-
tries are manifestly compact. For arrays of positive-tension branes only, the range of r
is − ∞ < r < ∞, but the norm converges if we restrict to cases where

A′ → 1/L− > 0 as r → −∞ ,
(65)
A′ → −1/L+ < 0 as r → ∞ ,

which are asymptotically anti-de Sitter geometries. In all such models, which include
the smooth configurations of section 5, there is a naturally massless four-dimensional
graviton as described above.
Having identified the four-dimensional graviton, we next turn to the question of
stability. If the equations of motion were to admit fluctuations with a space-like mo-
mentum, it would be evident that the zeroth-order solution — our classical background
— is not stable. For the transverse traceless components, we can cast the expression
(62) in the form of a supersymmetric quantum mechanics problem, where p2 plays the
role of the energy, and thus argue that p2 ≥ 0.
To accomplish this, we first need to eliminate the factor e2A multiplying the momen-
tum. We can do this by changing variables to coordinates in which the background is
conformally flat:  
ds2 = e2A(z) (ηij + hij ) dxi dxj − dz 2 . (66)
Now (62) takes the form

(−∂z2 − 3 A′(z) ∂z + ) hij = 0 . (67)

26
In terms of Hij (z) = e−ipx e3A/2 hij , this becomes

9 3
 
−∂z2 + A′ (z)2 + A′′ (z) Hij (z) = p2 Hij (z) . (68)
4 2
This differential operator has the same form as a Hamiltonian in quantum mechanics,
with a potential V (z) = 49 A′ (z)2 + 32 A′′ (z) and p2 as the energy eigenvalue. One can
easily check that it factorizes
3 3
 
(∂z + A′ (z))(−∂z + A′ (z)) Hij (z) = p2 Hij (z) . (69)
2 2
In flat space, these terms are one another’s adjoint, and (69) can be regarded as a
factorization of the Hamiltonian into QQ. This is supersymmetric quantum mechan-
ics, and the transformed graviton wave-function is the supersymmetric ground state.
However, to complete the argument we must show that a flat-space norm is correct for
Hij (z) in our curved background.
In Lorentzian signature field theory, the norm of fluctuations is determined by the
requirement that formally conserved quantities such as the contraction T µν Kν of the
stress tensor and a Killing vector of the background have convergent integral
√
Z
dz d3 x g T 0ν Kν , (70)

over a constant time 4-surface and vanishing flux through its boundary 3-surface. Stress
tensors for metric fluctuations are complicated, but in this linearized situation the stress
tensor must be covariantly conserved for all solutions of the equation of motion (62)
or (67) - (68) . Thus for the Killing vector (K 0 = 0, K i =constant, K 5 = 0) of spatial
translations parallel to the domain wall, one can take the form
kl
Ti0 = e−2A ∂0 hkl ∂i h , (71)

obtained by specializing the obvious covariant expression for Ti0 to our description of
the background. (The index i takes values 1,2,3 in (71) while k, l are raised with
η kl .) The requirement of a convergent integral for the spatial momentum carried by
the fluctuation then constrains the radial eigenfunctions Hij (z) to satisfy∗∗
Z
dz Hij H ij = finite. (72)

which is the usual Schrödinger norm for (68) (and equivalent to (64) when rephrased
in terms of hij and the radial coordinate r). Supersymmetric quantum mechanics thus
∗∗
We thank the authors of [29] for pointing out that our initial discussion of the norm was incorrect.
The correct norm appears in [29] and elsewhere; see, for example [13, 30].

27
ensures that there are no normalizable modes with p2 < 0. Thus we can state that there
are no transverse traceless modes with space-like momentum that might destabilize the
backgound solution.
Before concluding this section, we briefly remark on the non-transverse traceless
components of the metric fluctuation, which are coupled to the scalar by the equations
(55), (56), and (57). These coupled equations are not easy to solve, and we have not
attempted to rule out tachyonic modes of these fluctuations here.
However, it seems likely that the Boucher non-supersymmetric positive-energy the-
orem [31, 32] can be extended to include actions such as ours with potentials localized
on hypersurfaces, in which case stability would be guaranteed for our solutions, by
virtue of their satisfying the first-order equations.

Note Added
As this manuscript was nearing completion, several papers appeared [33, 34, 35, 36]
which overlap somewhat with our results. For instance, (14) was also derived in [36],
and the dS4 solution in (30) was also obtained in [33]. In [35], solutions similar to the
single domain wall of section 5 were shown to emerge from a U(1) gauged supergravity
theory.
The coupled equations relating scalar and non-transverse metric fluctuations have
recently been studied in [37]. The equations can again be reduced to the form of super-
symmetric quantum mechanics, and consequently there are no normalizable spacelike
modes. Thus our backgrounds have been shown to be entirely free from tachyonic
fluctuations.

Acknowledgements
We would like to thank D. Gross, S. Kachru, R. Myers, V. Periwal, and M. Perry
for useful discussions. The research of D.Z.F. and was supported in part by the NSF
under grant number PHY-97-22072. The research of O.D. and A.K. was supported by
the U.S. Department of Energy under contract #DE-FC02-94ER40818. The research
of S.S.G. was supported by the Harvard Society of Fellows, and also in part by the
NSF under grant number PHY-98-02709, and by DOE grant DE-FGO2-91ER40654.
D.Z.F., S.S.G., and A.K. thank the Aspen Center for Physics for hospitality.

28
References
[1] L. Randall and R. Sundrum, “A Large mass hierarchy from a small extra
dimension,” hep-ph/9905221.

[2] L. Randall and R. Sundrum, “An Alternative to compactification,”

hep-th/9906064.

[3] W. D. Goldberger and M. B. Wise, “Modulus stabilization with bulk fields,”

hep-ph/9907447.

[4] M. Cvetic and H. H. Soleng, “Supergravity domain walls,” Phys. Rept. 282
(1997) 159, hep-th/9604090.

[5] A. Lukas, B. A. Ovrut, K. S. Stelle, and D. Waldram, “The universe as a domain

wall,” Phys. Rev. D59 (1999) 086001, hep-th/9803235.

[6] A. Kehagias, “Exponential and power-law hierarchies from supergravity,”

hep-th/9906204.

[7] J. Polchinski and E. Witten, “Evidence for heterotic - type I string duality,”
Nucl. Phys. B460 (1996) 525–540, hep-th/9510169.

[8] P. Horava and E. Witten, “Heterotic and type I string dynamics from eleven-
dimensions,” Nucl. Phys. B460 (1996) 506–524, hep-th/9510209.

[9] J. Lykken and L. Randall, “The Shape of gravity,” hep-th/9908076.

[10] J. Dai, R. G. Leigh, and J. Polchinski, “New connections between string

theories,” Mod. Phys. Lett. A4 (1989) 2073–2083.

[11] P. J. Steinhardt, “General considerations of the cosmological constant and the

stabilization of moduli in the brane world picture,” hep-th/9907080.

[12] C. Csaki and Y. Shirman, “Brane junctions in the Randall-Sundrum scenario,”

hep-th/9908186.

[13] A. Brandhuber and K. Sfetsos, “Nonstandard compactifications with mass gaps

and Newton’s law,” hep-th/9908116.

[14] D. Z. Freedman, S. S. Gubser, K. Pilch, and N. P. Warner, “Renormalization

group flows from holography supersymmetry and a c theorem,” hep-th/9904017.

[15] R.C. Myers, unpublished notes, April 1999.

29
[16] T. Nihei, “Inflation in the five-dimensional universe with an orbifold extra
dimension,” hep-ph/9905487.

[17] N. Kaloper, “Bent domain walls as brane-worlds,” hep-th/9905210.

[18] T. jun Li, “Gauge hierarchy from AdS(5) universe with three-branes,”
hep-th/9908174.

[19] D. Arnold and J. Polking, Ordinary Differential Equations using MATLAB.

Prentice Hall, Upper Saddle River, 1999. DFIELD5 is available at
http://math.rice.edu/~polking/.

[20] J. Maldacena, “The Large N limit of superconformal field theories and

supergravity,” Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.

[21] S. S. Gubser, I. R. Klebanov, and A. M. Polyakov, “Gauge theory correlators

from noncritical string theory,” Phys. Lett. B428 (1998) 105, hep-th/9802109.

[22] E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. 2
(1998) 253, hep-th/9802150.

[23] M. Henningson and K. Skenderis, “The Holographic Weyl anomaly,” JHEP 07

(1998) 023, hep-th/9806087.

[24] L. Girardello, M. Petrini, M. Porrati, and A. Zaffaroni, “Novel local CFT and
exact results on perturbations of N=4 superYang Mills from AdS dynamics,”
JHEP 12 (1998) 022, hep-th/9810126.

[25] J. Garriga and T. Tanaka, “Gravity in the brane-world,” hep-th/9911055.

[26] D. Anselmi, “Central functions and their physical implications,” JHEP 05

(1998) 005, hep-th/9702056.

[27] N. R. Constable and R. C. Myers, “Spin two glueballs, positive energy theorems
and the AdS/CFT correspondence,” hep-th/9908175.

[28] R. C. Brower, S. D. Mathur, and C.-I. Tan, “Discrete spectrum of the graviton
in the AdS(5) black hole background,” hep-th/9908196.

[29] C. Csaki, J. Erlich, T. J. Hollowood, and Y. Shirman, “Universal aspects of

gravity localized on thick branes,” hep-th/0001033.

[30] A. G. Cohen and D. B. Kaplan, “Solving the hierarchy problem with

noncompact extra dimensions,” Phys. Lett. B470 (1999) 52, hep-th/9910132.

30
[31] W. Boucher, “Positive energy without supersymmetry,” Nucl. Phys. B242
(1984) 282.

[32] P. K. Townsend, “Positive energy and the scalar potential in higher dimensional
(super)gravity theories,” Phys. Lett. 148B (1984) 55.

[33] H. B. Kim and H. D. Kim, “Inflation and gauge hierarchy in Randall-Sundrum

compactification,” hep-th/9909053.

[34] H. Hatanaka, M. Sakamoto, M. Tachibana, and K. Takenaga, “Many brane

extension of the Randall-Sundrum solution,” hep-th/9909076.

[35] K. Behrndt and M. Cvetic, “Supersymmetric domain wall world from D = 5

simple gauged supergravity,” hep-th/9909058.

[36] K. Skenderis and P. K. Townsend, “Gravitational stability and renormalization

group flow,” hep-th/9909070.

[37] O. DeWolfe and D. Z. Freedman, “Notes on fluctuations and correlation

functions in holographic renormalization group flows,” hep-th/0002226.

31