PreprinttypesetinJHEPstyle-HYPERVERSION Ghosts in asymmetric brane gravity and the 9 decoupled stealth limit 0 0 2 n a J 6 Kazuya Koyama1, Antonio Padilla2 and Fabio P Silva1 ] h 1Institute of Cosmology & Gravitation, University of Portsmouth, t - p Portsmouth PO1 2EG, United Kingdom e 2School of Physics and Astronomy, University of Nottingham, h [ Nottingham NG7 2RD, United Kingdom 1 v 3 Abstract: We study the spectrum of gravitational perturbations around a vacuum 1 7 de Sitter brane in a 5D asymmetric braneworld model, with induced curvature on 0 . the brane. This generalises the stealth acceleration model proposed by Charmousis, 1 0 Gregory and Padilla (CGP) which realises the Cardassian cosmology in which power 9 law cosmic acceleration can be driven by ordinary matter. Whenever the bulk has 0 : infinite volume we find that there is always a perturbative ghost propagating on v i the de Sitter brane, in contrast to the Minkowski brane case analysed by CGP. We X discuss the implication of this ghost for the stealth acceleration model, and identify r a a limiting case where the ghost decouples as the de Sitter curvature vanishes. Keywords: braneworlds, cosmology, modified gravity. Contents 1. Introduction 1 2. The CGP model: set up and background solutions 3 3. Vacuum fluctuations 6 3.1 Spin 2 modes 8 3.2 Spin 1 modes 10 3.3 Spin 0 modes 10 4. Coupling to matter 12 5. The effective action 13 6. Discussion 16 1. Introduction There is now a wealth of evidence [1, 2, 3] indicating that the universe is currently in a period of accelerated expansion. One of the biggest challenges in cosmology today is understanding the origin of this late time acceleration. One possibility is that 70% of the energy content of the universe is dominated by an as yet unknown form of energy, so-called dark energy. The most popular dark energy candidate is the vacuum energy, which takes the form of a small and positive cosmological constant. In order to explain the current acceleration, the value of the cosmological constant must contribute a vacuum energy density of the order ρ 10 12(eV)4. This is 10120 Λ − ∼ times smaller thanwhat wemight expect, given ourcurrent understanding ofparticle physics. Given that particle physics is doing such a miserable job of explaining the accelerated expansion, it is important to look for alternative explanations. A popular alternative is to interpret this acceleration as a sign that our un- derstanding of gravity is breaking down, and that a large distance modification of Einstein’s General Relativity is required. Despite numerous attempts, it is fair to say that an established proposal has yet to emerge that is consistent on both a fun- damental and a phenomenological level. Arguably the most successful attempts have been inspired by the braneworld paradigm (for a review see [4]). In particular, the Dvali-Gabadadze-Porrati (DGP) model [5] was discovered to have two cosmological – 1 – branches, one of which gave rise to cosmic acceleration even when no matter was present on the brane [6]. This branch became known as the self-accelerating branch, for obvious reasons, but was later discovered to be haunted by ghost instabilities around the vacuum de Sitter brane [7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17] (for a review see [18]). In this context a ghost is a field whose kinetic term has the ”wrong” sign. This pathology leads to a choice: either the ghost state has negative norm and unitarity is violated, or the ghost can have arbitrarily negative energy. A ghost in the perturbation spectrum (specteroscopy!) indicates a catastrophic instability of the background, and therefore an unacceptably sick perturbative theory. In DGP, the other cosmological branch (the ”normal” branch), is ghost-free but cannot be an alternative to ΛCDM since it still needs the introduction of the cosmological con- stant Λ to explain the acceleration. Nevertheless it still has plenty of interesting phenomenological features [19, 20, 21]. More recently, Charmousis, Gregory and Padilla (CGP)[22] presented a general- isation of the DGP model in which they allowed for bulk curvature and introduced some asymmetry across the brane [23, 24, 25, 26, 27, 28]. This asymmetry could, in principle, apply tothe bulkcosmological constant or even thebulk Planck scales, giv- ing rise to a rich variety of cosmologies. The authors focussed on those solutions that possessed asymptotically Minkowski branes, despite the presence of self-accelerating solutions that they (correctly) assumed to be haunted by ghosts. A subset of these solutions were shown to contain vacuum branes that were perturbatively stable, free from the ghoulish instabitities that terrorized the self accelerating DGP brane. The cosmological evolution of this subset was then analysed, and in some cases yielded extremely interesting results. Two limiting models in particular (the ”decoupled” limit and the ”conformal” limit) were found to exhibit power law acceleration but only when matter is present on the brane. They dubbed this ’stealth acceleration’. The cosmology is reminiscent of the Cardassian cosmology proposed by Freese and Lewis [29]. Here the standard Friedmann equation is modified so that ρ → ρ + cρn, where n < 2/3, and one also finds that cosmic acceleration is driven by the presence of ordinary matter. The Cardassian model is an interesting empirical model, but didnot have aconcrete theoreticalbasis. The stealth model provides that by realising an effective Cardassian cosmology (with n 0.5) within the braneworld ≈ paradigm. In this paper we will consider vacuum de Sitter branes within the CGP set-up. This will include self-accelerating solutions, as well as the stealth models with some additional vacuum energy on the brane. We will study the spectrum of linearised perturbations about these solutions, closely following the corresponding analysis in the DGP model [9, 10, 13]. For an infinite volume bulk, we will find, without ex- ception, that the vacuum is unstable because of the presence of ghosts. Just as for the self-accelerating branch of DGP, a ghost will manifest itself either through the radion mode, or through the helicity 0 mode of the lightest graviton. In some cases – 2 – a ghost will also appear in the spin 1 sector. The ”decoupled” version of the stealth model is now of particular interest. We will find a class of de Sitter solutions that approach the ”decoupled” model as the Hubble scale H 0. As the limit is approached the ghost becomes more and → more weakly coupled, until eventually it decouples completely. We will infer some conclusions regarding the stability of the stealth models when matter is present. For small H it seems that we can carry our analysis of de Sitter branes over to the general Friedmann-Robertson-Walker case and conclude that the decoupled stealth model develops an instability albeit a very mild one softened by the weakness of the ghost coupling. For larger H the instability for de Sitter branes would be more severe, but itisnot clear whether ornotwe cantransfer this conclusion tothegeneral FRW case. The rest of this paper is organised as follows: in section 2 we describe the CGP model in detail, our generalisation, and the background solutions. In section 3 we analyse the spectrum of linearised perturbations and derive conditions for the presence ofanhelicity0ghostinthespin2sector. Westudythecouplingtomatterin section 4 and calculate the effective action in section 5. The effective action helps to reveal any further ghosts, including the radion ghost, which seems to take it in turns with the helicity 0 mode to haunt the background. We end with some concluding remarks in section 6. 2. The CGP model: set up and background solutions The CGP model is an asymmetric generalisation of its celebrated cousin, the DGP model. In both models, our Universe is taken to be a 3-brane, Σ, embedded in between two five dimensional spacetimes, , where i = L,R. In the original DGP i M Z scenario, we impose symmetry across the brane, identifying with and 2 L R M M having vanishing vacuum energy in the bulk. In the CGP model, however, we relax both of these assumptions. The key new ingredient is the introduction of asymmetry. Each spacetime generically has a five dimensional Planck scale given by M , and i i M a negative (or zero) cosmological constant given by Λ = 6k2,. However, since i − i we are no longer assuming Z symmetry across the brane, we can have M = M 2 L R 6 and Λ = Λ . Allowing for Λ = Λ is familar enough in domain wall scenarios L R L R 6 6 [31]. The Planck scale asymmetry is less familiar, but could arise in a number of ways. Suppose, for example, that this scenario is derived from a fundamental higher dimensional theory. This theory could contain a dilaton field that is stabilised in different fundamental vacua on either side of Σ. From the point of view of a 5D effective description, the 5D Planck scales would then differ accordingly. Indeed naive expectations from string theory point towards this asymmetric scenario as opposed to a symmetric one. Different effective Planck scales can also appear on either side of a domain wall that is bound to a five-dimensional braneworld [32]. – 3 – In keeping with the braneworld paradigm, all matter and standard model in- teractions are confined to the brane, although gravity can propagate into the fifth dimension. As in the DGP scenario, we include some intrinsic curvature induced on the brane. This term is rather natural and can be induced by matter loop cor- rections [33], finite width effects [34] or even classically from higher dimensional modifications of General relativity [35]. We will also include some vacuum energy on the brane in the form of some brane tension, σ. At this point we introduce an important new development. In the original CGP paper, the brane tension was fine-tuned against the bulk cosmological constants in order to admit a Minkowski vacuum solution. This choice corresponds to having vanishing effective cosmological constant on the brane and was the analogue of the Randall-Sundrum fine-tuning. In this paper we will introduce some additional tension so that the vacuum brane is de Sitter. Such detuning of brane tensions helped conjure up the ghost in the DGP model, and we will ultimately find that the same is true here. This set-up is described by the following action, S = M3 √ g(R 2Λ )+2M3 √ γK(i)+ √ γ(M2 σ+ ), i − − i i − − 4R− Lmatter i=XL,R ZMi Z∂Mi ZΣ (2.1) where g is the bulk metric with corresponding Ricci tensor, R. The metric induced ab on the brane is given by γ = g n n where na is the unit normal to ∂ in ab ab a b i i − M M pointing out of . Of course, continuity of the metric at the brane requires that i M γ is the same, whether it is calculated from the left, or from the right of the brane. ab In contrast, the extrinsic curvature of the brane can jump from right to left. In , i M it is defined as K(i) = γcγd n , (2.2) ab a b∇(c d) with its trace appearing in the Gibbons-Hawking boundary term in (2.1). In the brane part of the action we have included the brane tension, σ, and the induced intrinsic curvature term, , weighted by a 4D mass scale, M . includes any 4 matter R L additional matter excitations. The equations of motion in the bulk region, , are just the Einstein equations, i M with the appropriate cosmological constant, Λ . i 1 E = R Rg +Λ g = 0. (2.3) ab ab ab i ab − 2 The equations of motion on the brane are described by the Israel junction conditions, and can be obtained by varying the action (2.1), with respect to the brane metric, – 4 – γ . This gives1 ab 1 σ 1 Θ = 2 M3(K Kγ ) +M2 γ + γ = T , (2.4) ab ab − ab 4 Rab − 2R ab 2 ab 2 ab (cid:18) (cid:19) (cid:10) (cid:11) where Tab = −√2γ∂√−∂γγLambatter. Note that the Israel equations here do not use the − familiar “difference”, because we have defined the unit normal as pointing out of on each side. We adopt this (slightly) unconventional approach since it is more i M convenient in the asymmetric scenario where the brane is best thought of as the common boundary Σ = ∂ = ∂ . L R M M We will now derive the vacuum solutions to the equations of motion (2.3) and (2.4). This corresponds to the case where there are no matter excitations, and so, T = 0. In each region of the bulk, we introduce coordinates xa = (xµ,y), with the ab brane located at y = 0. We are interested in de Sitter brane solutions of the form ds2 = g¯ dxadxb = dy2+N(y)2γ¯ dxµdxν. (2.5) ab µν where γ¯ is the four dimensional de Sitter metric with curvature, H. Inserting this µν into the bulk equations of motion (2.3) gives N 2 H2 N ′ = +k2, ′′ = k2, (2.6) N N2 N (cid:18) (cid:19) where ”prime” denotes differentiation with respect to y, and we have dropped the index i for brevity. One can easily show that H 1 N(y) = sinhk(y +θy), y sinh 1k/H, (2.7) h h − k ≡ k where θ = 1. Each region of the bulk corresponds to 0 < y < y where max ± for θ = 1, y = ∞ (2.8) max (y for θ = 1. h − If we transformed to global coordinates in the bulk, θ = 1 would correspond to retaining the asymptotic region (large radius), whereas θ = 1 would correspond to − retaining the central region (small radius). For k = 0, this means that when θ = 1 6 we keep the adS boundary (growing warp factor) whereas when θ = 1 we keep − the adS horizon (decaying warp factor). Since we are interested in a modification of gravitational physics in the infra-red, we will assume that the bulk volume is infinite, and retain the asymptotic region on at least one side of the bulk. In other words, we do not consider the case θ = θ = 1. L R − 1Theangledbracketsdenoteanaveragedquantityatthebrane. Moreprecisely,forsomequantity Q defined on the brane in ∂ , we define the average Q = QL+QR. Later on we will also make i Mi h i 2 use of the difference, ∆Q=Q Q . L R − – 5 – The boundary conditions at the brane (2.4) yield σ 6 M3N (0) + 3H2M2 = 0, (2.9) h ′ i 2 − 4 so that the curvature H is given by the real roots of σ = 6M2H2 12 M3θ√H2 +k2 . (2.10) 4 − D E In [22], the brane tension was fine tuned to a critical value, σ = 6 M3k , so that c − h i the effective cosmological constant on the brane vanished. We now introduce some additional tension ǫ > 0 so that σ = σ +ǫ. This introduces some positive curvature c given by the roots of ǫ = F(H2) where, as in [22], we have F(H2) = 6M2H2 12 M3θ √H2 +k2 k . (2.11) 4 − − D (cid:16) (cid:17)E As in DGP, we have two classes of solution. There are those that vanish as ǫ 0, → so that we recover the Minkowksi brane studied in [22], and there are those that approach a finite positive value, so that we have a de Sitter brane, even in the absence of an effective cosmological constant. The former are the analogue of the normal branch in DGP, whereas the latter are the analogue of the self-accelerating branch. Of course, the class of solution depends on the form of the function F(H2), discussed in some detail in section 4 of [22]. For example, the following represent necessary and sufficient conditions for the existence of a normal branch solution: M2 > M3θ/k , (2.12) 4 h i or M2 = M3θ/k , M3θ/k3 > 0, (2.13) 4 h i h i or M2 = M3θ/k , M3θ/k3 = 0, M3θ/k5 < 0. (2.14) 4 h i h i h i Although we will study both classes of solution, we will be particularly interested in the normal branch since these will include small fluctuations about the finely tuned ”stealth” scenarios discussed in [22]. 3. Vacuum fluctuations We shall now consider metric perturbations in the vacuum so that g = g¯ + δg ab ab ab and T = 0. In the unperturbed spacetime, given by (2.5) and (2.9), the gauge µν was fixed in both and so that the brane was at y = 0. However, a general 1 2 M M perturbation of the system must also allow the brane position to flutter. In , the i M brane will be located at y = ζ (xµ). (3.1) i It is convenient to work in a Gaussian Normal (GN) gauge, so that in we have i M δg = δg = 0, δg = h (x,y). (3.2) yy µy µν iµν – 6 – In most of this discussion, we will drop the index i although its should be understood thatitisreallythere. Now,itiswellknown(see,forexample, [36])thatintheabsence of anybulkmatter, wemay takeh tobetransverse-tracefree Dµh = hµ = 0. This µν µν µ is known as Randall-Sundrum gauge. It follows that the bulk equations of motion, δE = 0 give ab 1 ∂2 + (D2 4H2) 4k2 h (x,y) = 0, (3.3) y N2 − − µν (cid:20) (cid:21) where D is the covariant derivative on the 4D de Sitter slicings, and indices are µ raised/lowered using the 4D metric γ¯ . To impose the boundary conditions at the µν brane, we need to apply a GN to GN gauge transformation that shifts the brane position back to y = 0. The most general such transformation is given by y dz y y ζ(x), xµ xµ ξµ(x)+Dµζ , (3.4) → − → − N2(z) Z0 so that h h¯ = h +h(ζ) +2N2D ξ . (3.5) µν → µν µν µν (µ ν) We call this new gauge ”brane-GN” gauge. Although the brane position is fixed in this gauge, the originalposition ζ(x) still enters the dynamics througha bookkeeping term y dz h(ζ) = 2 N2 D D ζ +2NN γ¯ ζ. (3.6) µν − N2 µ ν ′ µν (cid:18) Z0 (cid:19) The metric perturbation in the new gauge is no longer transverse-tracefree, although it is now straightforward to apply continuity of the metric at the brane ¯ ∆h (x,0) = 0, (3.7) µν and the vacuum Israel equations (2.4) ¯ ¯ h hγ¯ ′ δΘ = M3 µν − µν +M2X (h¯) = 0, (3.8) µν −* N2 y=0+ 4 µν (cid:18) (cid:19) (cid:12) (cid:12) where (cid:12) X (h¯) = δG (h¯)+3H2h¯ µν µν µν 1 1 = (D2 2H2)h¯ +D Dαh¯ D D h¯ µν (µ ν)α µ ν −2 − − 2 1 γ¯ DαDβh¯ (D2 +H2)h¯ . (3.9) µν αβ −2 − (cid:2) (cid:3) If we substitute the expression (3.5) into equation (3.8) we find M3 hµν ′ + M42(D2 2H2)h (x,0) = N2 y=0 2 − µν (cid:28) (cid:18) (cid:19) (cid:29) (cid:12) 2(D D (D(cid:12)2 +3H2)γ¯ ) (M3 M2N (0))ζ . (3.10) µ ν − (cid:12) µν − 4 ′ (cid:10) (cid:11) – 7 – Note that this expression is independent of ξµ(x), as expected, since this just corre- sponds to diffeomorphism invariance along the brane. It is convenient to decompose h in terms of the irreducible representations of the 4D de Sitter diffeomorphism µν group h = h(2) +h(1) +h(0), (3.11) µν µν µν µν (n) where h corresponds to the spin n contribution. We can treat these modes inde- µν pendently of one another provided they have different masses2. Let us now assume thatthisisindeedthecaseandanalyseeachspinseparately. Itwillalsobeconvenient (1) (0) to decompose the field ξ (x) into its spin 1 and spin 0 components ξ = ξ +ξ . µ µ µ µ The field ζ(x) is just spin 0. 3.1 Spin 2 modes We begin by analysing the spin 2 modes. Since neither ζ nor ξ have a spin 2 µ contribution, we can set them zero here, and can further decompose the spin 2 piece of the metric by separating variables h(2)(x,y) = u (y)χ(m)(x), (3.12) µν m µν Zm whereχ(m) isa4D tensorfieldofmassmsatisfying(D2 2H2)χ(m)(x) = m2χ(m)(x), µν µν µν − and denotes a generalised sum, summing over discrete modes and integrating over m continuum modes. The bulk equations of motion (3.3) now give R m2 2H2 u (y)+ − 4k2 u (y) = 0, (3.13) ′m′ N2 − m (cid:18) (cid:19) This is easily solved in terms of the associated Legendre functions: 2 2 k k u (y) = C 2 (cothk(y +θy))+C 2 (cothk(y +θy)), m 1 H P±1/2 ν h 2 H Q±1/2 ν h − ± − ± (cid:18) (cid:19) (cid:18) (cid:19) (3.14) where ν = 9/4 m2/H2. m(z) and m(z) are the associated Legendre functions − Pν Qν of the first and second kind, respectively. Of course, the expression (3.14) is only p well defined for m2 9H2, We could, in principle analytically continue our solution ≤ 4 to m2 > 9H2, although this will not be necessary since our ultimate goal is to 4 establish the existence of an helicity-0 ghost which is found in spin 2 modes of mass 0 < m2 < 2H2 [30]. Normalisability requires that [27] ymax u2 dy m < , (3.15) N2 ∞ Z0 so that for θ = 1 we only keep the part proportional to 2 (z), whereas for − P 1/2+ν θ = 1 we only keep the part proportional to 2 (z). −Since we may assume − Q 1/2+ν − 2In4D deSitter,atransverse-tracefreetensorofmassmsatisfies(D2 2H2)q(m) =m2q(m) [37] µν µν − – 8 – that u (0) = 1, without loss of generality, we get that the normalizable modes are m given by P−−12/2+ν(cothk(yh+y)) for θ = +1, um(y) = Q2−P1−−/212+/2ν+(νco(cthotkh(kyhyh−)y)) for θ = 1. (3.16)  Q2−1/2+ν(cothkyh) − It will be instructive to take a closer look at two special cases. For massless modes,   this expression simplifies to give e 2ky 2+cothk(yh+y) = N2Ryymaxdz/N4 for θ = +1, u0(y) = − 2+cothkyh R0ymaxdz/N4 (3.17) N2(y)(cid:16) (cid:17) for θ = 1.  − whereas for ”partially massless” modes of mass m2 = 2H2 we have e 2ky for θ = +1, − u (y) = (3.18) √2H (NN′ for θ = 1. N′(0) − Of course, neither the massless modes, nor the partially massless modes get excited in general. This is determined by the boundary conditions at the brane. The spin (m) 2 part of the continuity equation (3.7) now implies that ∆χ (x) = 0 for each m, µν so that the spin 2 part of Israel equations (3.10) yield the following quantization condition u M2 f(m2) = M3 m ′ + 4m2 = 0. (3.19) N2 y=0 2 (cid:28) (cid:29) (cid:16) (cid:17) (cid:12) Let us consider the lightest mode. For a fi(cid:12)nite volume bulk (θ = θ = 1), it is (cid:12) L R − well known that this mode is massless so that gravity looks four dimensional out to arbitrarily large distances. We do not consider this case here, and assume, without further loss of generality, that θ = +1. The lightest mode is now guaranteed to R be massive. If the mass lies in the forbidden region 0 < m2 < 2H2, then this mode contains an helicity-0 ghost [30]. We can now check if such a mode exists, by application of Bolzano’s theorem: f(0)f(2H2) < 0, (3.20) since f(m2) is continuous over the forbidden region. Although not necessary for the existence of a ghost, this condition is certainly sufficient. For an infinite bulk ((θ ,θ ) = ( 1, 1)), it is easy enough to see that L R 6 − − 1 ymax dz −1 f(0) = M3(1+θ) < 0. (3.21) −2 N4 * (cid:20)Z0 (cid:21) + This means we have an helicity-0 ghost whenever M3 (1 θ)H2 f(2H2) = − 2(1+θ)(k +√H2 +k2) +M2H2 > 0. (3.22) 2 √H2 +k2 − 4 (cid:28) (cid:18) (cid:19)(cid:29) – 9 –