Cosmological matching conditions and galilean genesis in Horndeski's theory

We derive the cosmological matching conditions for the homogeneous and isotropic background and for linear perturbations in Horndeski's most general second-order scalar-tensor theory. In general relativity, the matching is done in such a way that the extrinsic curvature is continuous across the transition hypersurface. This procedure is generalized so as to incorporate the mixing of scalar and gravity kinetic terms in the field equations of Horndeski's theory. Our matching conditions have a wide range of applications including the galilean genesis and the bounce scenarios, in which stable, null energy condition violating solutions play a central role. We demonstrate how our matching conditions are used in the galilean genesis scenario. In doing so, we extend the previous genesis models and provide a unified description of the theory admitting the solution that starts expanding from the Minkowski spacetime.


I. INTRODUCTION
Scalar fields are ubiquitous in cosmology. Inflation [1] is considered to be driven by one or multiple scalar fields, which can seed the large-scale structure of the Universe as well. The current cosmic acceleration may also be caused by a scalar field dominating the energy content of the Universe as dark energy (see e.g. [2] for a review). A great variety of modified gravity models have been proposed as an alternative to dark energy (see e.g. [3] and references therein), many of which involve an additional scalar degree of freedom in the gravity sector. Earlyuniverse scenarios other than inflation have also been explored (see e.g. [4] for a recent review), such as bounce models, and they are often based on some scalar-field theory.
Almost forty years ago, Horndeski constructed the most general theory composed of the metric g µν and the scalar field φ with second-order field equations [5], which has long been ignored until recently [6]. In the course of generalizing the galileon scalar-field theory, Horndeski's theory was rediscovered in its modern form [7][8][9]. (The equivalence of the generalized galileon and Horndeski's theory was first shown in Ref. [10].) The action is given by (1) * Email: sakine˙n"at"rikkyo.ac.jp † Email: tsutomu"at"rikkyo.ac.jp ‡ Email: norihiro.tanahashi"at"ipmu.jp § Email: gucci"at"phys.titech.ac.jp with L 2 = G 2 (φ, X), L 3 = −G 3 (φ, X)✷φ, where X := −g µν ∂ µ φ∂ ν φ/2, R is the Ricci scalar, and G µν is the Einstein tensor, and G 2 , G 3 , G 4 , and G 5 are arbitrary functions of φ and X. (Here and hereafter we use the notation G iX := ∂G i /∂X, G iφ := ∂G i /∂φ, and so on.) Since this theory contains all the single-field inflation models and modified gravity models with one scalar degree of freedom as specific cases, it is of great importance in cosmology and hence considerable attention has been paid in recent years to various aspects of Horndeski's theory (see the nonexhaustive list of references [11]). In this paper, we will address the following issue: suppose that the Universe undergoes a sharp transition caused, for example, by sudden halt of the scalar field or by a discontinuous jump in matter pressure, and then what are the continuous quantities across the transition hypersurface in Horndeski's theory? In general relativity, it is known that the induced metric on the surface and its extrinsic curvature must be continuous. This implies that the Hubble parameter H is continuous. As for linear cosmological perturbations, the matching conditions in general relativity are clarified in Refs. [12,13]. In Horndeski's theory, however, scalar and gravity kinetic terms are mixed due to second derivatives on φ in the Lagrangian [14], and as a result the matching conditions would be nontrivial both for the homogeneous and isotropic background and for cosmological perturbations. This point was raised in the context of galilean genesis [15] and was studied based on specific Lagrangians [16,17]. In this paper, we start from the boundary terms in Horndeski's theory [18] and derive rigorously the cosmological matching conditions in their most general form.
The matching conditions obtained in this paper have a wide range of applications. In particular, Horndeski's theory allows for stable violation of the null energy condition (NEC), leading to interesting possibilities such as galilean genesis mentioned above and non-singular bounce models [19][20][21][22][23]. Our matching conditions provide a generic algorithm to follow the evolution of the cosmological background and perturbations, which is applicable to those scenarios. This paper is organized as follows. In the next section we summarize the boundary terms in Horndeski's theory [18], which are the basis of the present work. Then, in Sec. III, we derive the cosmological matching conditions both for the background and perturbations. To present an example, we develop a unified Lagrangian accommodating all the previous models of galilean genesis, and apply our matching conditions to this general model in Sec. IV. We draw our conclusions in Sec. V.

II. BOUNDARY TERMS FOR HORNDESKI'S THEORY
We begin with summarizing the result of Ref. [18]. The action we are going to study is given by where S Hor is Horndeski's action (1), S m is the action for usual matter, and S B is the boundary term. This last term is necessary when one considers a spacetime M divided into two domains, M ± , by a surface Σ. In what follows Σ is supposed to be spacelike. Let us take a look at the case of general relativity. The variation of the Einstein-Hilbert term with respect to the metric involves a normal derivative of the metric variation, where γ µν = g µν + n µ n ν is the induced metric on Σ ± and n µ is the future directed unit normal. Here and hereafter we write with Σ ± denoting the two sides of Σ. The presence of the normal derivative of the metric variation is problematic; to obtain a well-defined variational problem, one has to add a boundary term that cancels the contribution (4). By noticing that the variation of the trace of the extrinsic curvature, K µν := γ a µ γ b ν ∇ (a n b) , gives rise to the same contribution, we are lead to add the well-known Gibbons-Hawking term on the boundary [24].
Since the most general scalar-tensor Lagrangian having second-order field equations contains second derivatives of the scalar field which is nonminimally coupled to gravity, as well as the second derivatives of the metric, the corresponding boundary action is not simply given by the Gibbons-Hawking term. For example, G 3 ✷φ produces the following problematic normal derivative: This can be canceled by adding where with X 0 := (n µ ∇ µ φ) 2 /2 and X := −γ µν ∂ µ φ∂ ν φ/2. (Note that X = X 0 + X.) Similarly, one can obtain the boundary contributions corresponding to L 4 and L 5 . The boundary term for the galileon Lagrangian was considered in Ref. [25], and then the complete boundary term in Horndeski's theory, which is composed of three different parts, S B = B 3 + B 4 + B 5 , was derived for the first time in Ref. [18]. The latter two terms are given by where each F i (i = 4, 5) is defined similarly to F 3 as D µ is the covariant derivative on the boundary, and R (3) is the boundary Ricci scalar.
Having found the boundary term, one can obtain the junction conditions that describe discontinuity across the hypersurface Σ, as a generalization of Israel's conditions [26]. The variational principle for (3) yields the equations of motion and Here, J µν = J µν and lengthy expressions for J µν 4 and J µν 5 , for which we refer the reader to Ref. [18]. 2 In the case of general relativity (G 4 = const, G 3 = 0 = G 5 ), one finds J µν = −G 4 (K µν − γ µν K). A concrete expression for J φ is also found in Ref. [18].
We allow for a localized source on Σ whose action is denoted by S Σ . Variation of the action S Σ will take the form where τ µν is the surface stress tensor from the localized source, giving the jump in J µν . The surface action also gives rise to the source ∆ φ for the jump in J φ . From Eqs. (13) and (15), we obtain [18] [ and where [(· · · )] + − := (· · · )| Σ+ − (· · · )| Σ− . The above junction conditions together with the continuity [γ µν ] + − = 0 and [φ] + − = 0 determine how the metric and the scalar field are matched across the surface Σ. It is now clear from those conditions that the first time derivatives of the metric and φ can be discontinuous, and hence the second time derivatives can be singular at Σ.

III. COSMOLOGICAL MATCHING CONDITIONS
We consider a slightly perturbed universe whose metric is given by 1 In deriving Eq. (14) we used to rearrange the original expression of Ref. [18]. 2 In arXiv:1206.1258v1 [18] there is a typo in the expression for J µν 5 , so the reader should refer to the updated version of Ref. [18].
where A and ψ are scalar perturbations, h ij is a traceless and transverse tensor perturbation, and B i and E ij are decomposed into scalar and transverse vector parts as The scalar field also has a homogeneous part and a small inhomogeneous perturbation as φ(t, x) =φ(t) + δφ(t, x).
We will omit the bar on the homogeneous part when there is no worry about confusion. Let the matching surface be specified by q(t, x) = 0. This equation can be decomposed asq(t) + δq(t, x) = 0. The cosmological matching conditions on this hypersurface are derived by calculating . By using the temporal gauge transformation t →t = t + ξ 0 , one can move to the uniform q gauge, i.e., the coordinate system satisfying Then, the matching surface is determined simply by the equationq(t) = 0, or, equivalently,t = const =: t * . Although the choice of the temporal gauge has no relevance to the matching conditions for the homogeneous background and tensor and vector perturbations, this coordinate system is convenient for the computation of δJ φ and the scalar part of

A. Matching conditions for the homogeneous background
Let us first consider the matching conditions for a homogeneous and isotropic background. The homogeneous part of J ij is of the form J with Assuming that there are no localized sources on Σ, the matching conditions for the background are given by [a] + − = 0 and In general relativity Eq. (23) reduces to the standard matching condition [H] + − = 0. The same condition can be derived by integrating the background equation P = −p (see Appendix) from t = t * − ǫ to t = t * + ǫ. Isolating the second time derivatives and denoting them with the subscript ••, one gets which implies Eq. (23). It is worth emphasizing that even if G 4 = const and G 5 = 0, i.e., even if φ is minimally coupled to gravity, G 3X gives rise to a nonstandard term f 3 in the junction condition (21). This is because the gravitational field equations contain second derivatives of φ in the presence of G 3X .
Similarly, it is straightforward to get where In the absence of a localized source, we obtain the scalarfield matching condition with the continuity [φ] + − = 0. In general relativity with a scalar field whose kinetic term is canonical, we have J φ = −φ. The same equation as Eq. (27) can be derived as well by integrating the scalar-field equation of motion (see Appendix) from t = t * − ǫ to t = t * + ǫ, noting that the second derivatives in the scalar-field equation are given by The matching conditions (23), (27), and [φ] + − = 0 admit the solution satisfying the same conditions as in general relativity: The second derivatives,Ḣ andφ, can however be discontinuous (but not singular) across Σ. Obviously, H + andφ + determined from Eq. (29) satisfy the Hamiltonian constraint, E(φ + ,φ + , H + ) = −ρ + . There could be other nontrivial solutions, H + = H − ,φ + =φ − , to the matching conditions (23) and (27). However, in contrast to the trivial solution (29)  continuous across Σ, i.e., there is no essential modification compared with the result of general relativity. This is indeed the case if the matter equation of state undergoes a sudden transition, p = p − (ρ) → p + (ρ), at some ρ = ρ * = const hypersurface. Another example is the model where the nonsingular bounce is caused by some scalar-field dynamics: in the scenario of [23], H andφ are continuous whileḢ andφ can be approximated to be discontinuous at the beginning and end of the bounce phase.
To see the situation where the matching conditions do not reduce simply to Eq. (29), let us investigate the model with a step-like potential for the scalar field, In deriving the above scalar matching condition, we have implicitly assumed that the singular part in the scalar-field equation of motion comes only from the second derivativesḢ andφ. However, variation with respect to φ now gives δ φ G 2 ⊃ −V 0 δ(φ − φ * )δφ, leading to a non-vanishing localized source of the jump in the right hand side of Eq. (27). Equivalently, one can collect the singular part of the scalar-field equation of motion, to see that the scalar matching condition in the form (27) cannot be used due to the extra singular contribution V 0 δ(φ−φ * ). In this case, both H andφ are discontinuous in general, but J is continuous. In the genesis scenario which will be discussed in the next section [15][16][17]27], such a step in the potential will cause an instantaneous change inφ to end the genesis phase. We present our numerical result in Fig. 1 corresponding to the situation in which φ suddenly slows down. As a simple example containing only G 2 and G 3 plus the Einstein-Hilbert term [14,28], the Lagrangian with c, V 0 = const, and ξ ≫ 1 is employed for the numerical calculation to mimic the case of the step-like potential. It can be seen that H andφ experience a sharp jump, but the matching condition (23) still holds.

B. Matching conditions for cosmological perturbations
We are now in position to consider matching of cosmological perturbations. The matching conditions for scalar, vector, and tensor modes can be studied separately.

(i) Tensor perturbations
The transverse and traceless part of δJ j i is where and f 5 is defined similarly to f 3 as Thus, the matching conditions for the tensor perturbations are given by Since the tensor perturbations are subject to a secondorder differential equation, the above two conditions are enough to determine their evolution after the transition. The matching conditions (35) can further be simplified if H andφ are continuous across Σ. For such a background, G T and f 5 are continuous and hence the matching conditions (35) reduce to the same ones as in general relativity: The vector part of δJ j i is of the form The matching condition for the vector perturbations is therefore given in any coordinates by The continuity of the induced metric, [E V i ] + − = 0, can always be satisfied by choosing the spatial coordinates appropriately. Since the vector perturbations are governed by a first-order differential equation, Eq. (38) completely fixes the integration constant at the transition. If H anḋ φ (and hence G T ) are continuous, Eq. (38) reduces to the same matching condition as in general relativity.
(iii) Scalar perturbations As mentioned above, we will work in the uniform q gauge, δq = q(t, x) −q(t) = 0. In this gauge, the continuity implies that where the second condition can always be satisfied by choosing appropriately the spatial coordinates. The junction conditions are derived from where we defined the shear as and The above equations yield the uniform q gauge expression of the matching conditions. However, using the following formulas,Ã one can undo the gauge fixing to move from one gauge to the other. Let us first consider the case where the equation of state of matter experiences a sudden jump, p = p − (ρ) → p + (ρ), at the time when ρ = ρ * = const, so that Σ is determined by the equation We assume that the localized source is absent on Σ, and so can use the matching conditions for the background in the form [J ] + − = [J φ ] + − = 0. From the discussion in the previous subsection, it turns out in the end that H andφ are continuous. From Eq. (45) we see that the continuity (39) can be written in an arbitrary gauge as The trace and traceless parts of the equations [ δJ The two conditions (49) and (51) can be rearranged to give Interestingly, these matching conditions are independent of the concrete form of G i (φ, X), and hence are the same as those in general relativity with a conventional scalar field. Note, however, that the matching procedure requires the use of the constraint equations presented in Appendix, which depend on the concrete form of G i (φ, X).
Having thus obtained the matching conditions in an arbitrary gauge, let us see how one can consistently determine the perturbation variables at t = t * + ǫ in the unitary gauge (δφ = 0). In this gauge, Eq. (48) reads [δρ u /ρ] + − = 0, and then Eq. (47) implies that where R is the curvature perturbation in the unitary gauge, Here and hereafter the subscript u refers to the unitary gauge variable. Equation (50) simply becomes [σ u ] + − = 0. Equations (52) and (53) can be used to determineṘ and A u at t = t * + ǫ. The Hamiltonian constraint is consistent with Eq. (51), while the momentum constraint is used to fix the velocity perturbation δu u . Thus, all the perturbation variables at t = t * + ǫ can be determined.
The matching procedure in the Newtonian gauge (σ = 0) is slightly different from that in the unitary gauge. In the Newtonian gauge, Eq. (50) reads [δρ N /ρ] + − = 0, where the subscript N stands for the Newtonian gauge variable. In terms of the metric potentials in the Newtonian gauge, Eqs. (47) and (48) are rewritten as while Eqs. (52) and (53) yields the two relations amonġ Ψ, Φ, andδ φ N . We then invoke the traceless part of the (i, j) components of the field equations, to remove Φ, and thus determineΨ andδ φ N at t = t * +ǫ. The next example we would like to study is the transition that occurs when φ reaches some value φ * : In the previous example of q = ρ(t, x) − ρ * , we considered the case where H andφ are continuous. In present example, however, we allow for discontinuous H andφ, because such a situation can easily be realized at the moment when φ(t, x) passes a step in the potential at φ * , as already demonstrated. In this case, it is convenient to stay in the uniform q gauge since it coincides with the uniform φ gauge. Then, the curvature perturbation on uniform φ hypersurfaces is given by R = ψ+Hδφ/φ =ψ, and the matching conditions [ψ] + − = 0 and [ δJ Let us first assume for simplicity that usual matter is absent. Equation (61) automatically holds thanks to the momentum constraints. Combining the Hamiltonian and momentum constraints, we find The matching condition (62) then reads Using Eqs. (60) and (64) one can do the matching of R andṘ. In the case whereφ and H are continuous, the latter condition is simplified to [Ṙ] + − = 0. However, if the second derivatives diverge and henceφ and H are discontinuous, one must employ the full equation (64).
In the presence of usual matter, the matching condition (64) is modified as while Eq. (60) remains unchanged. Since the continuity and Euler equations for matter do not contain second derivatives of the metric, all the matter-related quantities are continuous across the matching surface specified by q = φ(t, x) − φ * = 0. If the transition is such thatφ and H are continuous, then Eq. (65) implies that we are still allowed to use the condition [Ṙ] + − = 0.

IV. GENESIS MODELS FROM HORNDESKI'S THEORY
In this section, we demonstrate how the matching conditions are used at the transition from the galilean genesis phase to the standard radiation-dominated Universe. This is probably the most illustrative example because stable galilean genesis is realized thanks to the terms L i with i ≥ 3, which give rise to the new boundary terms. The matching procedure has been carried out in specific examples of galilean genesis in Refs. [16,17]. We will extend those previous models and present a unified analysis of the theory admitting galilean genesis. To do so, we generalize the Lagrangian of Ref. [29] and study a subclass of Horndeski's theory defined by where each g i (i = 2, 3, 4, 5) is a function of and λ and M Pl are constants. We assume that g 4 (0) = 0. Let us look for a solution of the form where Y 0 and h 0 are positive constants. Note that Y ≃ Y 0 for this background. Equation (68) should be regarded as an approximate solution valid for |t| ≫ √ h 0 , and in this section we only consider the case where this approximation is good. The spacetime is close to Minkowski for |t| ≫ √ h 0 and expands as a ≃ 1 + h 0 (−t) −2 /2. Sincė H = 3h 0 (−t) −4 > 0, one can interpret this solution to be NEC violating. The above solution is essential for the galilean genesis scenario [15][16][17]27]. The Lagrangian defined by Eq. (66) contains different models of galilean genesis as specific cases, and allows us to study the genesis scenario in a unified manner. 3 The background equations read whereρ and a prime stands for differentiation with respect to Y . The constant Y 0 is determined as a positive root of and then h 0 is determined from Eq. (70) as As will be seen shortly, this background is stable for G(Y 0 ) > 0. Hence, the above NEC violating solution is possible provided that For tensor perturbations, it is straightforward to compute and therefore the background is stable against tensor perturbations if For scalar perturbations, we find where From Eq. (82) it is easy to show so that the background is stable against scalar perturbations ifρ Since G S ∝ (−t) 2 and F S ∝ (−t) 2 , the sound speed, c s = F S /G S , stays constant during the genesis phase. The genesis phase is supposed to be followed by the standard radiation-dominated phase. As in [16], we consider the model in which the transition is caused by sudden halt of the scalar field due to some upward lift of its potential, G 2 ⊃ −V 0 θ(φ − φ * ). Then, the second derivativesφ andḢ diverge at t = t * . We neglect the contribution from the scalar field to the expansion rate in the radiation-dominated phase, assumingφ rad ≃ 0. It is then found that in the genesis phase and (1/3)J rad ≃ M 2 Pl H rad in the radiation-dominated phase. The radiation-dominated universe is required to be expanding, H rad > 0. The matching condition J gen − J rad = 0 therefore reads J gen > 0. Using Eq. (75), this condition can be written as It is easy to fulfill all of these conditions simultaneously even in the simple Lagrangian with [15,16] where c 2 and c 3 are some constants. Indeed, all the requirements are satisfied for 4λc 3 > c 2 > 0. Let us then investigate the matching of the perturbation variables. On superhorizon scales, the general solution to the tensor perturbation equation in Fourier space is given by where C g± k and C g± k are integration constants that depend on the wavenumber k. From the matching conditions (35), the integration constants in the post-genesis phase are determined as Note, however, that tensor perturbations generated during the genesis phase are observationally irrelevant, because a ∼ 1, G T , F T ∼ const, so that the vacuum fluctuations of h ij are not amplified. As for the scalar perturbations, it is most convenient to use the curvature perturbation on uniform φ slices, R. The central quantity for the matching of R is G S , because from the matching conditions we see that R and G SṘ are continuous on superhorizon scales. In the present case, G S is of the form G S = A(Y 0 )(−t) 2 for t < t * , and G S = G + S = const for t > t * , where the concrete expression for A(Y 0 ) is not so illuminating. The superhorizon solution for R is given by where it follows from the matching conditions that The second term in Eq. (92) is matched to the second one in Eq. (93), i.e., the decaying mode in the post-genesis Universe. Thus, we have R ≃ C − k at sufficiently late times. Following the usual quantization procedure we determine |C − k | = (2 c s kA(Y 0 )|t * |) −1 ∼ k −1/2 , and hence we cannot get the scale-invariant fluctuations. See also Refs. [30,31] for discussions about the spectrum of fluctuations from galilean genesis.

V. CONCLUSIONS
In this paper, we have obtained the matching conditions for the homogeneous and isotropic Universe and for cosmological perturbations in Horndeski's most general second-order scalar-tensor theory, starting from the generalization of Israel's conditions [18]. In the absence of any localized sources at the transition hypersurface, we have shown that the first derivatives of the metric and the scalar field, H(t) andφ(t), must be continuous as in the case of general relativity. This is the case where the equation of state of matter undergoes a sharp change. In the case where φ suddenly lose its velocity due to a step in the potential, some combination J of H andφ, defined in Eq. (21), is continuous across the transition hypersurface. For cosmological perturbations we have obtained the junction equations that can be used in any gauge.
Horndeski's theory can accommodate exotic but stable cosmologies such as galilean genesis [15]. The cosmological matching conditions we have presented in this paper can be applied to such interesting scenarios. To demonstrate this, we have developed a generic Lagrangian admitting the genesis solution that starts expanding from the Minkowski spacetime in the asymptotic past, and presented the conditions under which a stable genesis background is consistently joined to an expanding universe.

Background equations
The evolution of the homogeneous and isotropic background is determined from where −4H 2 X 2φ G 5XX + 4HX Ẋ − HX G 5φX +2 2 (HX)˙+ 3H 2 X G 5φ + 4HXφG 5φφ , and ρ and p are the energy density and pressure of usual matter, respectively. The first equation corresponds to the Friedmann equation (the Hamiltonian constraint), and the second one to the evolution equation containing the second derivatives of the metric and the scalar field. The equation of motion for φ is given bẏ and

Linear perturbations
In the main test we use the following equations for scalar cosmological perturbations: (i) the Hamiltonian constraint, and (iii) the traceless part of the (i, j) equations, where we have included the perturbations of the matter energy-momentum tensor: δT 0 0 = −δρ, δT 0 i = (ρ + p)∂ i δu, and δT j i = δpδ j i . Energy-momentum conservation implies ∂ t δρ + 3H(δρ + δp) − 3(ρ + p)ψ + ρ + p a 2 ∂ 2 (δu + σ) = 0, (A8) ∂ t [(ρ + p)δu] + 3H(ρ + p)δu + (ρ + p)A + δp = 0. (A9) In the above we defined The unitary gauge δφ = 0 is convenient in the absence of usual matter. The evolution equation for the curvature perturbation in the unitary gauge, R, follows from the quadratic action S (2) where Similarly, the quadratic action for the tensor perturbation is given by From those actions we see that the scalar and tensor perturbations are stable if