Resummed propagators in multi-component cosmic fluids with the eikonal approximation
RResummed propagators in multi-component cosmic fluidswith the eikonal approximation
Francis Bernardeau, ∗ Nicolas Van de Rijt, † and Filippo Vernizzi ‡ Institut de Physique Th´eorique, CEA, IPhT, F-91191 Gif-sur-Yvette,France CNRS, URA 2306, F-91191 Gif-sur-Yvette, France (Dated: September 25, 2018)We introduce the eikonal approximation to study the effect of the large-scale motion of cosmicfluids on their small-scale evolution. This approach consists in collecting the impact of the long-wavelength displacement field into a single or finite number of random variables, whose statisticalproperties can be computed from the initial conditions. For a single dark matter fluid, we showthat we can recover the nonlinear propagators of renormalized perturbation theory. These areobtained with no need to assume that the displacement field follows the linear theory. Then weextend the eikonal approximation to many fluids. In particular, we study the case of two non-relativistic components and we derive their resummed propagators in the presence of isodensitymodes. Unlike the adiabatic case, where only the phase of small-scale modes is affected by thelarge-scale advection field, the isodensity modes change also the amplitude on small scales. Weexplicitly solve the case of cold dark matter-baryon mixing and find that the isodensity modesinduce only very small corrections to the resummed propagators.
I. INTRODUCTION
The development of wide-field surveys has triggeredrenewed interest in the implementation of perturbationtechniques for the computation of the statistical prop-erties of large-scale structures. Several approaches havebeen proposed to significantly extend the standard per-turbation theory (PT) methods (see [1]). A particularlyinteresting approach is the so-called renormalized pertur-bation theory (RPT), pioneered by Crocce and Scocci-marro [2–4]. This method relies on the use of the 2-pointpropagator as a measure of the memory of the initial con-ditions. This appears as the fundamental building blockfrom which perturbation theory can be re-constructedand allows to take into account nonlinearities from verysmall scales, reducing their impact in the neglected termsof the perturbative expansion. This idea was later ex-tended in [5, 6] with the introduction of multi-point prop-agators.The key result of RPT is that in the high- k limit thepropagators can be computed exactly, by summing up aninfinite subset of contributions in the standard perturba-tion theory expansion. However, this result has beenproved only for a single pressureless fluid – describingcold dark matter (CDM) – using a technique that seemsdifficult to extend to more complex scenarios. Thus,there is no systematic way to implement the RPT ap-proach when the content of the cosmic fluid is richer –e.g. when it includes various matter components, non-relativistic neutrinos, or even modification of gravity –and its application range has been so far limited to simplecosmological models (see however [7]). Note that other ∗ [email protected] † [email protected] ‡ fi[email protected] approaches, such as the so-called time renormalizationgroup approach [8], do not suffer from such limitations.On the other hand, the resummation of the propaga-tors can be obtained by a more direct technique than thatoriginally introduced in [2–4]. As mentioned in [9] andexplicitly used in [6, 10], in the high- k limit it is possibleto resum the same class of contributions making use ofa single or a finite number of random variables, whichdescribe the effect of the long-wavelength fluctuations onsmaller scales. This has been called the α -method [6].Here we will explicitly present how to compute nonlinearpropagators in this framework. Moreover, we will showthat this method can be employed to extend the RPTapproach to arbitrarily complicated cosmologies. Bor-rowing the terminology from quantum field theory, wheresimilar techniques are used (see e.g. [11, 12]), we proposeto dub this method the eikonal approximation .The fluid content of the universe is richer than a simplesingle-dark matter component. In practice we know thatthe properties of the large-scale structure of the universecan be significantly affected by the presence of a sub-dominant species. This is the case of baryons, which athigh enough redshift behave very differently from CDM.Indeed, the net result of this different behavior is theexistence of the baryonic oscillations.From a theoretical point of view, the CDM-baryon sys-tem is very appealing: After decoupling both componentsare pressureless fluids (at least above the baryonic Jeansscale) and thus follow geodesic motion [13]. However, thematter fluid as a whole cannot be described as a pres-sureless effective fluid. The reason is that the CDM andbaryon fluids are moving at different velocities – see [14]for a study of some of the consequences of this differentbehavior and their possible observational implications.Such a velocity dispersion induces an effective anisotropicpressure in the total fluid, modifying its equations of mo-tion. Thus, one is forced to study the system of coupledequations for the CDM and baryons, as previously done a r X i v : . [ a s t r o - ph . C O ] O c t in [7]. In particular, this is the system that we will ex-plore with the help of the eikonal approximation.The plan of the paper is the following. In Section IIwe review the basic concepts of the RPT approach for asingle CDM fluid and we discuss the eikonal approxima-tion in this context. In Section III we extend this dis-cussion to the multi-fluid case. In particular, we derivethe evolution equations for several gravitationally cou-pled pressureless fluids, we describe the various modesthat appear in this case, and we present how they canbe incorporated in the eikonal approximation. Finally,in Section IV we illustrate our concepts in the case of thestandard CDM-baryon mixing. II. SINGLE FLUID
In this section we review the basic concepts of the RPTapproach with a single perfect fluid, developed in [2, 3].Moreover, we rederive the procedure to resum the non-linear propagators for cosmic fluids using the eikonal ap-proximation.
A. Equations of motion
We assume the universe to be filled by one pressure-less fluid. We denote its density by ρ , and the densitycontrast by δ ≡ ρ/ρ −
1, where ρ is the average energy density. The continuity equation then reads ∂∂t δ + 1 a (cid:0) (1 + δ ) u i (cid:1) ,i = 0 , (1)where u i is the i -component of the peculiar velocity fieldof the fluid and a comma denotes the partial derivative.The Euler equation is ∂∂t u i + Hu i + 1 a u j u i,j = − a φ ,i , (2)where H is the Hubble rate, H ≡ d ln a/ d t , and φ isthe gravitational potential. Since we are only interestedin the dynamics on sub-horizon scales, φ is the usualNewtonian potential, satisfying the Poisson equation∆ φ = 4 πG a ρδ . (3)We also ignore small-scale shell crossings in the fluid.Then, since the gravitational force is potential the fluidvelocity remains potential at all orders in the perturba-tions. Thus, it can be entirely described by the dimen-sionless velocity divergence, defined by θ ≡ u i,i aH . (4)By using the following convention for the Fouriermodes, f ( k ) ≡ (cid:90) d x (2 π ) f ( x ) e − i k · x , (5)the equations of motion can then be rewritten in Fourierspace as1 H ∂∂t δ ( k ) + θ ( k ) = − α ( k , k ) θ ( k ) δ ( k ) , (6)1 H ∂∂t θ ( k ) + 1 H d ln( a H )d t θ ( k ) + 32 Ω m δ ( k ) = − β ( k , k ) θ ( k ) θ ( k ) , (7)where Ω m is the reduced matter density and α ( k , k ) = ( k + k ) · k k , (8) β ( k , k ) = ( k + k ) k · k k k . (9)On the right-hand side of eqs. (6) and (7), integra-tion over repeated wave modes and a Dirac function δ D ( k − k − k ) is implied. Note that these equations arevalid irrespective of the dark energy equation of state orcurvature term. B. RPT formulation
In order to recast these equations in RPT form letus first discuss their linear solutions. At linear order,the coupling terms in the right-hand side of eqs. (6) and(7) are absent. We are then left with the usual linearsolutions of a pressureless fluid, i.e. δ ( x , t ) = D + ( t ) δ + ( x ) + D − ( t ) δ − ( x ) , (10)where D + ( t ) and D − ( t ) correspond to the growing anddecaying modes, respectively. The corresponding expres-sion for the dimensionless velocity divergence is θ ( x , t ) = − f + ( t ) D + ( t ) δ + ( x ) − f − ( t ) D − ( t ) δ − ( x ) , (11)where f + and f − are the growth rates, defined as f ± ≡ d ln D ± / d ln a .As introduced in [7, 15], it is convenient to define theduplet Ψ a = (cid:18) δ Θ (cid:19) , (12)where Θ is the reduced velocity contrast defined asΘ( x , t ) ≡ − θ ( x , t ) /f + ( t )= D + ( t ) δ + ( x ) + f − ( t ) f + ( t ) D − ( t ) δ − ( x ) , (13)in such a way that the linear growing mode of Θ is thesame as that of δ . It is then convenient to rewrite the evo-lution equation using η as time variable, defined through D + d η ≡ d D + . (14)With this definition the equations of motion (6) and (7)can be recast as ∂∂η Ψ a ( k ) + Ω ab Ψ b ( k ) = γ abc ( k , k , k )Ψ b ( k )Ψ c ( k ) , (15)where Ω ab ≡ (cid:32) − −
32 Ω m f
32 Ω m f − (cid:33) , (16)and the non-zero elements of the coupling matrix γ abc are γ ( k , k , k ) ≡ α ( k , k )2 ,γ ( k , k , k ) ≡ α ( k , k )2 ,γ ( k , k , k ) ≡ β ( k , k ) . (17)From eqs. (10) and (13), the growing and decayingsolutions are proportional to u (+) a ∝ (1 , T and u ( − ) a ∝ (1 , f − /f + ) T (18)respectively. It has been widely stressed that f − /f + isvery weakly dependent on the background. Indeed, itdeparts little from the value it takes in an Einstein-deSitter universe (EdS), i.e. Ω m = 1, where f − /f + = − / linear propagator g ab ( η, η (cid:48) ). This is such that g ab ( η, η ) = δ ab , (19)and ∂∂η g ab ( η, η ) + Ω ac ( η ) g cb ( η, η ) = 0 . (20) Ψ (1) a ( k ) =Ψ (2) a ( k ) = g ab ( η, η ) g ab ( η, η ) g ce ( η , η ) g df ( η , η ) Φ b ( k )Φ e ( k )Φ f ( k ) γ bcd FIG. 1. Diagrammatic representation of the series expansionof Ψ a ( k ) up to fourth order in the initial conditions denotedhere by Φ a ( k ). Time increases along each segment accordingto the arrow and each segment bears a factor g cd ( η f − η i )if η i is the initial time and η f is the final time. At eachinitial point and each vertex point there is a sum over thecomponent indices; a sum over the incoming wave modes isalso implicit and, finally, the time coordinate of the vertexpoints is integrated from η to the final time η according tothe time ordering of each diagram. It can be built from a complete set of independent solu-tions of the evolution equation. For a single fluid we canuse u (+) a and u ( − ) a defined in eq. (18). We then have g ab ( η, η ) = (cid:88) α u ( α ) a ( η ) c ( α ) b ( η ) , (21)where the coefficients c ( α ) b are chosen such that (cid:88) α u ( α ) a ( η ) c ( α ) b ( η ) = δ ab . (22)For an EdS background the explicit form of g ab reads g ab ( η, η ) = e η − η (cid:18) (cid:19) + e − ( η − η ) (cid:18) − − (cid:19) . (23)The linear propagator is useful to formally write thesolution for Ψ a in integral form. Indeed, using eqs. (12)and (13), the equations of motion (6) and (7) can bewritten as [15]Ψ a ( k , η ) = g ab ( η, η )Ψ b ( k , η )+ (cid:90) ηη d η (cid:48) g ab ( η, η (cid:48) ) γ bde ( k , k , k )Ψ d ( k , η (cid:48) )Ψ e ( k , η (cid:48) ) . (24)As illustrated in Fig. 1, this equation has a diagrammaticrepresentation in the RPT context [2].Another important quantity introduced in the RPTapproach is the nonlinear multi-point propagator. Moreprecisely, the ( n + 1)-point propagator Γ ( n ) ab ...b n is definedby (cid:28) ∂ n Ψ a ( k , η ) ∂ Ψ b ( k , η ) . . . ∂ Ψ b n ( k n , η ) (cid:29) ≡ δ D ( k − n (cid:88) i k i )Γ ( n ) ab ...b n ( k , . . . , k n ; η, η ) . (25)Propagators represent the way the Ψ a ’s respond to aninfinitesimal change of the modes at an earlier time andthey are important in the construction of multi-pointspectra [5, 6].In the large- k limit (to be better specified below) thesepropagators enjoy a remarkable property. Indeed, in [3] ithas been shown that in this limit and for Gaussian initialconditions, the nonlinear 2-point propagator G ab ≡ Γ (1) ab has a simple expression G ab ( k ; η, η ) = g ab ( η, η ) exp (cid:0) − k σ ( e η − e η ) / (cid:1) , (26)where σ is the variance of the initial displacement field.Note that the linear propagator g ab is simply the treelevel analog of G ab . This result has been generalized to( n + 1)-point propagators with n ≥ ( n ) ab ...b n = Γ ( n ) − tree ab ...b n exp (cid:0) − k σ ( e η − e η ) / (cid:1) , (27)where Γ ( n ) − tree is the corresponding propagator com-puted at tree level.The exponentiation in eqs. (26) and (27) has been ob-tained in [2, 3] by summing up an infinite number ofdiagrams thought to dominate in the large- k limit. Inorder to identify which diagrams dominate in this limit,the concept of principal line and its generalization for the( n + 1)-point propagators, the principal tree , have beenintroduced. In [3] it has been shown that each diagramcontributing to the nonlinear propagator G ab ( k ; η, η ) al-ways contains a unique line that goes from some time η (symbolized by the vertical dotted line) to a final time η . To this line may be attached loops containing powerspectra evaluated at an initial time η in . This is illus-trated in Fig. 2, upper panel. The principal line is theunique way to go from η to η without crossing an initialpower spectrum ⊗ , thus moving always in the directionof increasing time. Similarly, for each diagram contribut-ing to Γ ( n ) ab ...b n there always exists a unique tree with n branches, the principal tree , that joins η to η (see bot-tom diagram of Fig. 2) [5].We can now specify under which assumption the rela-tions (26) and (27) have been derived. These are: • The multi-point propagators are dominated bythose diagrams in which every loop is directly con-nected to the principal tree. • The diagrams are computed and summed up in thelimit where the incoming wave modes q i are soft,i.e. q i (cid:28) k .As we will show below, the eikonal approximation corre-sponds exactly to the last assumption. It can incorporatethe first one if necessary. k kk k k k G ab ( k ) =Γ (3) abcd ( k , k , k ) = FIG. 2. Example of diagrams contributing to G ab ( k ) (top)and Γ (3) abcd ( k , k , k , k ) (bottom). The dominant contributionafter resuming all possible configurations is expected to comefrom those diagrams where all loops are directly connectedto the principal line (top) or principal tree (bottom). Theprincipal line and tree are drawn with a thick solid line. Asymbol ⊗ denotes a power spectrum evaluated at initial time η in . The dominant loops are those drawn by dashed lines,while the sub-dominant loops are those in dotted lines. C. Resumming the 2-point propagator with theeikonal approximation
In [6] it has been shown that eqs. (26) and (27) canbe obtained irrespective of the diagrammatic represen-tations and of the nature of the initial conditions. In-deed, the nonlinear fluid equations contain nonlinearterms that couple short and long-wavelength modes. Theeikonal approximation corresponds to study the effect ofvery long-wavelength modes q on the dynamics of a givenshort-wavelength mode k , in the limit of q (cid:28) k . In thislimit, space variations of the long-wavelength modes aretiny with respect to the mode k , and the long modescan be treated as an external random background. If weneglect the mode couplings between short scales, the non-linear fluid equations can be rewritten as linear equationsembedded in an external random medium.Let us be more explicit here. Coupling terms are givenby a convolution of fields taken at wave modes k and k such that k = k + k . These nonlinear terms canbe split into two different contributions: the one comingfrom coupling two modes of very different amplitudes, k (cid:28) k or k (cid:28) k , and the one coming from couplingtwo modes of comparable amplitudes. In the first case,the small wave modes ought to be much smaller than k itself. Let us denote these small modes by q . In the limitof q (cid:28) k , the equations of motion (15) can be rewrittenas ∂∂η Ψ a ( k ) + Ω ab Ψ b ( k ) = Ξ ab ( k )Ψ b ( k )+ [ γ abc ( k , k , k )Ψ b ( k )Ψ c ( k )] H , (28)with Ξ ab ( k , η ) ≡ (cid:90) S d q γ abc ( k , k , q )Ψ c ( q , η ) . (29)The key point is that in eq. (29) the domain of integra-tion is restricted to the soft momenta, for which q (cid:28) k .Conversely, on the right-hand side of eq. (28) the convo-lution is done excluding the soft domain, i.e. it is overhard modes or modes of comparable size.In the limit of separation of scales, Ξ ab is a randomquantity which depends on the initial conditions. Us-ing eqs. (8) and (9), for q (cid:28) k the leading expres-sion of the coupling matrix is obtained with the fol-lowing limit values α ( q , k ) ≈ ( q · k ) /q , α ( k , q ) ≈ β ( q , k ) = β ( k , q ) ≈ ( q · k ) / (2 q ). Thus, γ abc in eq. (29)simplifies and Ξ ab becomes proportional to the identity,with Ξ ab ( k , η ) = Ξ( k , η ) δ ab , Ξ( k , η ) ≡ (cid:90) S d q k · q q Θ( q , η ) . (30)Note that only the velocity field Θ (and not the densityfield δ ) contributes to Ξ ab . Furthermore, as Θ( x , η ) isreal Θ( − q ) = Θ ∗ ( q ) and thus Ξ is purely imaginary.In eq. (28) we have reabsorbed the effect of the non-linear coupling with long-wavelength modes in the linearterm Ξ ab Ψ b . The solution to this equation can be givenin terms of the resummed propagator ξ ab ( k , η, η (cid:48) ) [17]satisfying the equation (cid:18) ∂∂η − Ξ( k , η ) (cid:19) ξ ab ( k , η, η (cid:48) ) + Ω ac ( η ) ξ cb ( k , η, η (cid:48) ) = 0 , (31)and readsΨ a ( k , η ) = ξ ab ( η, η )Ψ b ( k , η )+ (cid:90) ηη d η (cid:48) ξ ab ( η, η (cid:48) ) [ γ bde ( k , k , k )Ψ d ( k , η (cid:48) )Ψ e ( k , η (cid:48) )] H , (32)where in the last line the convolution is done on the harddomain H .In the case of a single fluid, as discussed here, eq. (31)can be easily solved. Taking into account the boundarycondition ξ ab ( k , η, η ) = δ ab , one obtains ξ ab ( k , η, η ) = g ab ( η, η ) exp (cid:18)(cid:90) ηη d η (cid:48) Ξ( k , η (cid:48) ) (cid:19) . (33)The argument of the exponential is the time integral ofthe velocity projected along the direction k , i.e. the dis-placement component along k . Note that in their originalcalculation, Crocce and Scoccimarro assumed that the in-coming modes in the soft (i.e. large-scale) lines were inthe linear and growing regime. Here we need not makethis assumption. Equation (33) is valid irrespective ofthe fact that the incoming modes in Ξ are in the growingmode or not. There is another important aspect of eq. (33). SinceΞ( k , η ) is a purely imaginary number, the soft modeschange only the phase of the small-scale modes but nottheir amplitude. Such an effect will then have no impacton the equal-time power spectra. However, it has someimpact on the amplitude of the propagators. Indeed, thephase change inevitably damps the correlation betweenmodes at different times. This effect is at the heart of theregularization scheme used by approaches such as RPT.To illustrate this last point, let us see how one canrecover eq. (26) using the solution (32) and the resummedpropagator (33) derived with the eikonal approximation.Deriving eq. (32) with respect to an initial field Ψ b ( k , η )as in eq. (25), and taking the ensemble average one finds G ab ( k, η, η ) = (cid:104) ξ ab ( k , η, η ) (cid:105) Ξ . (34)The nonlinear 2-point propagator G ab is given by theensemble average of ξ ab ( k , η, η ) over the realizations ofΞ( k ). In general, the expression of the nonlinear propa-gator introduces the cumulant generating functions of Ξ.Indeed, using (33) eq. (34) yields G ab ( k, η, η ) = g ab ( η, η ) exp (cid:32) ∞ (cid:88) p =2 c p p ! (cid:33) , (35)where c p is the p -order cumulant of the field (cid:82) ηη d η (cid:48) Ξ( k , η (cid:48) ) and for symmetry reasons the sum is re-stricted to even values of p , thus ensuring that the non-linear propagators are real.For Gaussian initial conditions and assuming that atlate time the long-wavelength Ξ is in the linear growingmode, cumulants with p > c . This is given by c ( k ) = (cid:90) ηη d η (cid:48) d η (cid:48)(cid:48) (cid:104) Ξ( k , η (cid:48) )Ξ( k , η (cid:48)(cid:48) ) (cid:105) . (36)Then, exploiting the time dependence of the linear grow-ing mode, Θ ∝ D + = e η − η in , and using eq. (30), we haveΞ( k , η ) = D + ( η ) (cid:90) S d q k · q q Θ( q , η in ) . (37)Plugging this expression in eq. (36), we can express c interms of the initial power spectrum P in ( q ), defined by (cid:104) Θ( q , η in )Θ( q (cid:48) , η in ) (cid:105) ≡ δ D ( q + q (cid:48) ) P in ( q ) . (38)Indeed, we have c ( k ) = − k σ (cid:0) e η − η in − e η − η in (cid:1) , (39)where σ gives the variance of the displacement field de-fined as [2, 3] σ ≡ (cid:90) S d q P in ( q ) q . (40)At this stage σ depends on the domain of integrationand hence on k . The standard RPT results are obtainedby taking the value of σ in the large- k limit. We willcomment on this assumption in the conclusion. Then,setting here and in the following η in = 0 for convenience,from eq. (35) we recover eq. (26), G ab ( k, η, η ) = g ab ( η, η ) exp (cid:0) − k σ ( e η − e η ) / (cid:1) . (41) D. Higher-order propagators
Although the focus of this paper is on the 2-point prop-agator, let us comment on the use of the eikonal approx-imation, in particular of eq. (32), in investigating theresummation of higher-order propagators.The computation of the nonlinear 3-point propagatorproceeds by replacing Ψ d and Ψ e in the second line ofeq. (32) by the linear solution given by the first line ofthis equation. Deriving twice with respect to the initialfield yields ∂ Ψ a ( k , η ) ∂ Ψ b ( k , η ) ∂ Ψ c ( k , η ) = (cid:90) ηη d η (cid:48) ξ ad ( k ; η, η (cid:48) ) × [ γ def ( k , k , k ) ξ eb ( k ; η (cid:48) , η ) ξ fc ( k ; η (cid:48) , η )] H . (42)This is the same formal expression as for the naked the-ory except that here the convolution is restricted to thehard-mode domain H . Note that the coupling vertex be-tween modes with hard momenta in the second line isnot affected by the use of the eikonal approximation: Itis identical to the one of the naked theory. Moreover, it isremarkable to see that, using the form given by eq. (33),the exponential terms factor out of the time integral andtheir arguments sum up to give ∂ Ψ a ( k , η ) ∂ Ψ b ( k , η ) ∂ Ψ c ( k , η ) = exp (cid:18)(cid:90) ηη d η (cid:48) Ξ( k , η (cid:48) ) (cid:19) × (cid:90) ηη d η (cid:48) g ad ( η, η (cid:48) ) [ γ def ( k , k , k ) g eb ( η (cid:48) , η ) g fc ( η (cid:48) , η )] H . (43)Finally, taking the ensemble average and using the def-inition of multi-point propagators, eq. (25), one obtainsin the Gaussian caseΓ (2) abc ( η, η ) = Γ (2) − tree abc ( η, η ) exp (cid:0) − k σ ( e η − e η ) (cid:14) . (44)This result can be generalized to propagators of anyhigher order. The formal expressions of the resummed trees computed in the eikonal approximation are ob-tained from those computed in the naked theory by sim-ply changing the propagators from g ab to ξ ab . Then, foreach pair of merging branches with equal initial time,one can factor out the phase similarly to what is donewhen going from eq. (42) to (43). Finally, this leaves anoverall factor exp (cid:16)(cid:82) ηη d η (cid:48) Ξ( k , η (cid:48) ) (cid:17) , which can be factor-ized out, recovering eq. (27). The eikonal approximationexplicitly shows how the results [5, 6] can be recoveredand generalized to any time-dependent large-scale wavemode. III. MULTI-FLUIDS
In this section we explore the case where the universeis filled with several non-interacting pressureless fluidsand show how the eikonal approximation can be imple-mented in this case. Due to the gravitational couplingand the expansion, at late time such a system becomesindistinguishable from a single-fluid component. How-ever, during its evolution it can behave very differentlyfrom a single perfect fluid depending on the initial con-ditions.
A. The equations of motion
Denoting each fluid by a subscript α , the continuityequation reads, for each fluid, ∂∂t δ α + 1 a (cid:0) (1 + δ α ) u iα (cid:1) ,i = 0 , (45)while the Euler equation reads ∂∂t u iα + Hu iα + 1 a u jα u iα,j = − a φ ,i . (46)The Poisson equation (3), where now δ is the densitycontrast of the total fluid energy density, i.e. ρ m ≡ (cid:88) α ρ α ≡ (1 + δ m ) ρ m , (47)allows to close the system. This introduces couplingsbetween the fluids.In Fourier space, the equations of motion become now1 H ∂∂t δ α ( k ) + θ α ( k ) = − α ( k , k ) θ α ( k ) δ α ( k ) , (48)1 H ∂∂t θ α ( k ) + 1 H d ln( a H )d t θ α ( k ) + 32 Ω m δ m ( k ) = − β ( k , k ) θ α ( k ) θ α ( k ) , (49)where θ α is the dimensionless divergence of the velocityfield of the fluid α and Ω m is the reduced total density ofthe pressureless fluids. The coupling between the fluids isonly due to the term δ m appearing in the Euler equation.Before studying these equations let us discuss the equa-tions for the total fluid. As we are describing here a col-lection of pressureless particles, it is tempting to writedown the equations of motion for the total fluid. Thecontinuity equation is simply identical to eq. (1). Thetotal fluid velocity u i is defined by u i ≡ (cid:88) α f α u iα , (50)where f α ≡ ρ α /ρ m , and its evolution equation reads ∂∂t u i + Hu i + 1 a u j u i,j = − a φ ,i − aρ m (cid:0) ρ m σ ij (cid:1) ,j , (51)where and σ ij is the velocity dispersion of the mean fluid,given by σ ij ≡ (cid:88) α f α u iα u jα − u i u j . (52)Thus, due to the multi-fluid nature of the system, the Eu-ler equation contains an anisotropic stress term. One canwrite down an equation of motion for this term, but thiswill involve higher moments of the fluid distribution andso on. Thus, the complete description of the total fluidat nonlinear order requires an infinite hierarchy of equa-tions in the moments of the fluid. Another consequenceof this expression is that, even though the velocity fieldof each fluid remains potential, the total velocity field isno longer potential. Indeed, we expect that it developsa rotational part due to the presence of the dissipativeterm σ ij [18]. B. Adiabatic and isodensity modes
As we did for the single-fluid case, let us study the lin-ear evolution of the multi-fluid system by dropping theright-hand side of eqs. (48) and (49). Since at linear orderthere is no anisotropic stress σ ij , which is second orderin the velocities, the two linear solutions (10) and (13)found in the single-fluid case are expected to be also solu-tions of the linear multi-fluid system. This correspondsto the case where all the fluids start comoving and, asthey all follow geodesic motion, remain comoving duringtheir entire evolution. Analogously to the jargon adoptedin the physics of the early universe, these solutions cor-respond to the so-called growing and decaying adiabatic modes. Note that if only these two modes are initiallyexcited, the right-hand side of eq. (52) vanishes and thetotal fluid is indistinguishable from a pure dark matterfluid.However, the presence of multiple components givesbirth also to isocurvature or rather, given our scales of in-terest, isodensity modes. To examine their properties, let us turn to the equations describing the multi-componentsystem, eqs. (48) and (49). In this case it is convenientto introduce a multiplet Ψ a which generalizes the dupletdefined in eq. (12), i.e. [15]Ψ a = ( δ , Θ , δ , Θ , . . . ) T , (53)where Θ α ≡ − θ α /f + ( t ). Thus, for N components Ψ a has 2 N elements. Equations (48) and (49) can then berewritten as eq. (15) where in this case the matrix ele-ments of Ω ab are given byΩ (2 p −
1) (2 p ) = − , Ω (2 p ) (2 p ) = 32 Ω m f − , Ω (2 p ) (2 q − = −
32 Ω m f f q , (54)for any integers p and q running from 1 to N and where allthe other elements of Ω ab vanish. The non-zero elementsof the coupling matrix γ abc are γ (2 p −
1) (2 p −
1) (2 p ) ( k , k , k ) = α ( k , k )2 ,γ (2 p −
1) (2 p ) (2 p − ( k , k , k ) = α ( k , k )2 ,γ (2 p ) (2 p ) (2 p ) ( k , k , k ) = β ( k , k ) , (55)for any integer p . Note that there are no explicit cou-plings between different species in the γ abc -matrices.The isodensity modes are obtained under the con-straint that the total density contrast vanishes, i.e. δ = 0.Since the evolution equations decouple under this con-straint, the time dependence of these modes can be easilyinferred. One solution is given byΘ (is) α ( η ) ∝ exp (cid:20) − (cid:90) η d η (cid:48) (cid:18)
32 Ω m f − (cid:19)(cid:21) ,δ (is) α ( η ) = (cid:90) η d η (cid:48) Θ (is) α ( η (cid:48) ) , (56)with (cid:88) α f α Θ (is) α = 0 , (57)which automatically ensures that (cid:80) α f α δ (is) α = 0. Notethat as Ω m /f departs little from the value taken in anEdS cosmology, i.e. Ω m /f = 1, the isodensity modesare expected to depart very weakly fromΘ (is) α ( η ) ∝ exp( − η/ , δ (is) α ( η ) = − (is) α ( η ) . (58)A second set of isodensity modes is given byΘ (ci) α ( η ) = 0 , δ (ci) α ( η ) = Constant , (59)again with (cid:88) α f α δ (ci) α = 0 . (60)To be specific, let us concentrate now on the case oftwo fluids and assume an EdS background. In this casethe growing and decaying solutions are proportional, re-spectively, to u (+) a ∝ (1 , , , T ,u ( − ) a ∝ (1 , − / , , − / T . (61) Moreover, the isodensity modes are proportional to u (is) a ∝ ( − f , f , f , − f ) T ,u (ci) a ∝ ( f , , − f , T . (62)We are then in position to write down the linear prop-agator g ab ( η, η ) satisfying eqs. (21) with (22). For twofluids and an EdS background it reads [7] g ab ( η, η ) = e η − η f f f f f f f f f f f f f f f f + e − ( η − η ) f − f f − f − f f − f f f − f f − f − f f − f f + e − ( η − η ) − f f f − f f − f − f f + f f − f − f − f − f f f . (63)In the following we explore how this propagator ischanged by the coupling with the long-wavelength modesin the eikonal approximation. C. Resummation of the propagator with theeikonal approximation
Let us study the resummed propagator in the presenceof more than one fluid. For simplicity, we will restrictthe study to the two-fluid case and an EdS background.The eikonal equation, eq. (28) with (29), also holds inthe multi-fluid case. However, in this case Ξ ab is givenby a sum of adiabatic contributions, for which the fluiddisplacements are the same, and isodensity contributions,for which their weighted sum vanishes, i.e.,Ξ ab ( k , η ) = Ξ (ad) ( k , η ) δ ab + Ξ (is) ab ( k , η ) , (64)where Ξ (is) ab takes the formΞ (is) ab = Ξ (is) h ab , h ab ≡ f f − f
00 0 0 − f . (65)If we assume Ξ ab to be in the linear regime, thenΞ (ad) ( k , η ) ≡ (cid:90) S d q k · q q (cid:0) Θ (+) ( q , η ) + Θ ( − ) ( q , η ) (cid:1) , (66)where Θ (+) and Θ ( − ) are, respectively, the growing anddecaying adiabatic modes of the long-wavelength dis-placement field. The isodensity contribution Ξ (is) ab con-tains the decaying isodensity mode given in eq. (58), so that it readsΞ (is) ≡ f (cid:90) S d q k · q q Θ (is)1 = − f (cid:90) S d q k · q q Θ (is)2 . (67)Note that, because of eq. (59), the constant isodensitymode does not contribute to Ξ ab .We are now interested in computing the resummedpropagator in the eikonal approximation under the mod-ulation of the long-wavelength modes in eq. (64). As theadiabatic modes in Ξ ab are proportional to the identity,their effect can be incorporated in exactly the same man-ner as in the single-fluid case. The adiabatic modes willcontribute to the resummed propagator by a multiplica-tive factor of the exponential of the adiabatic displace-ment field, as in eq. (33), ξ ab ( k ; η, η ) = ξ ab ( k ; η, η ; Ξ (ad) = 0) × exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) . (68)Note again that, as in the single-fluid case, the soft adia-batic modes induce a phase change but do not affect theamplitude of the small-scale modes.Including the isodensity mode in the resummed prop-agator proved difficult. We have not been able to finda closed analytic form for it. Thus, we have to rely ei-ther on numerical studies or on perturbative calculations.As an example, in Fig. 3 we show the effect of the softisodensity mode on the small-scale modes, by plottingthe evolution of the resummed CDM and baryon densitymodes with Ξ (ad) = 0, i.e., δ c ( η, Ξ (is) ) ≡ ξ a ( k ; η, η in ; Ξ (ad) = 0)Ψ a ( η in ) ,δ b ( η, Ξ (is) ) ≡ ξ a ( k ; η, η in ; Ξ (ad) = 0)Ψ a ( η in ) , (69)normalized to the growing mode e η . Initial conditionsare chosen such that Ψ a ( η in ) = (1 , , . , . T and we (cid:45) (cid:45) .50.51 Η e (cid:45) Η ∆ a (cid:72) Η , (cid:88) (cid:72) i s (cid:76) (cid:76) (cid:200) ∆ c (cid:200) (cid:200) ∆ b (cid:200) Exp (cid:72) Ν c Η (cid:76) Exp (cid:72) Ν b Η (cid:76) (cid:45) (cid:45) (cid:45) Η A r g (cid:64) ∆ a (cid:72) Η , (cid:88) (cid:72) i s (cid:76) (cid:76) (cid:68) Arg (cid:72) ∆ c (cid:76) Arg (cid:72) ∆ b (cid:76)(cid:217) d Η (cid:88) c (cid:72) is (cid:76) (cid:217) d Η (cid:88) b (cid:72) is (cid:76) FIG. 3. Evolution of the amplitudes (upper panel) and phases(lower panel) of the resummed density modes with Ξ (ad) = 0,i.e. δ c ( η, Ξ (is) ) and δ b ( η, Ξ (is) ) as defined in eq. (69) for CDM(thick solid line) and baryons (thick dashed line). The initialconditions are chosen such that Ψ a ( η in ) = (1 , , . , . T and | Ξ (is) ( η in ) | = 25. Thin lines represent the analytic solu-tions at early times of eq. (75). have taken | Ξ (is) ( k, η in ) | = 25. At early time the CDMand baryon density mode grow slower than e η (upperpanel) and since | Ξ (is) | (cid:29)
1. Early-time behavior
As Ξ (is) is a decaying mode, it can become arbitrarilylarge at early time. Let us consider a mode k for whichinitially | Ξ (is) | (cid:29)
1. This means that k · v S , i.e. the dis-placement field of the soft modes along k , is much largerthan the Hubble flow. In other words, the time scale ofthe motion of the large-scale modes is much shorter thanthe time-scale of growth of the small-scale ones, set bythe Hubble time.We can grasp the nature of the early-time evolution by making the following change of variable,˜Ψ a ( η ) = Ψ a ( η ) exp (cid:18) − (cid:90) ηη d η (cid:48) Ξ aa ( η (cid:48) ) (cid:19) , (70)where there is no summation over a in Ξ aa . In this casethe first line (i.e. the large- k part) of eq. (28) can berewritten in terms of ˜Ψ a as ∂∂η ˜Ψ a ( η ) + ˜Ω ab ˜Ψ b ( η ) = ˜Ξ ab ( η ) ˜Ψ b ( η ) , (71)where ˜Ω ab ≡ − − f / / −
10 0 − f / / , (72)and ˜Ξ ab ≡ f e − iϕ / f e iϕ / , (73)with ϕ ( η ) ≡ − i (cid:90) ηη d η (cid:48) Ξ (is) ( η (cid:48) ) . (74)The two fluids are only coupled through ˜Ξ ab . However,Ξ (is) in eq. (74) is purely imaginary: At early time thecoupling term contributes to a rapidly changing phase ϕ . When the time scale of these oscillations is muchshorter than that of structure growth, this force termcan effectively be neglected and the different species de-couple. Indeed, the velocity difference (in the directionalong k ) between the coherent flows of the two speciesis large enough that the short modes of one fluid do notgravitationally see those of the other fluid.The system we are left with is given by eq. (71) withvanishing right-hand side. For an EdS background thesolution of this equation is given by [19]˜ δ α ∝ exp( ν ( ± ) α η ) , (75)with ν ( ± ) α = 14 (cid:16) − ± (cid:112) f α (cid:17) . (76)The growing solutions in eq. (75) explain the early-timeevolution shown in Fig. 3. At early time, both CDMand baryons grow slower than e η and their phases aredominated by their respective large-scale isocurvaturedisplacement fields, as accounted for by the change ofvariable (70).0
2. Late-time behavior
As the isodensity mode decays, one expects torecover at late time the single-fluid propagator g ab ( η, η ) exp (cid:16)(cid:82) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:17) . More precisely, we can compute how the propagator ξ ab deviates from the adia-batic one with a perturbative analysis. Indeed, since Ξ (is) ab becomes small at late time, one can compute ξ ab ( k ; η, η )perturbatively in Ξ (is) .Solving the first line of eq. (28) at first order in Ξ (is) yields, ξ ad ( k , η, η ) ≈ exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) (cid:20) g ad ( η, η ) + (cid:90) ηη d η (cid:48) g ab ( η, η (cid:48) )Ξ (is) bc ( k , η (cid:48) ) g cd ( η (cid:48) , η ) (cid:21) . (77)By plugging in this equation the expressions of the linear propagator g ab from eq. (63) and of Ξ (is) ab from eq. (65), andintegrating in time yields ξ ab ( k ; η, η ) ≈ (cid:104) g ab ( η, η ) + Ξ (is) ( k , e − η / C ab ( η, η ) (cid:105) exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) , (78)with C ab ( η, η ) ≡ f f e η − η − −
11 1 − −
11 1 − −
11 1 − − + e ( η − η )
12 8 12 f f f f f f f f − f f − f f − − − f f − f f − − + 2 f f − f f − − f f − f f f f − f f − − f f − f f + e − ( η − η ) − f f − f f f f − f f − − f f − − f f − f f − f f − f f f f − f f − f f − f f − − f f + 2 e − ( η − η ) f f − − f f − f f f f −
30 2 f f − − f f − f f f f − + e − ( η − η ) − − − − − − − − + 2 e − η − η ) − − f f f f − f f − f f f f − f f − − f f f f − . (79)Note that this result is written in terms of Ξ (is) takenat the initial time η in = 0, so that the time dependenceof Ξ (is) is included in eq. (78) and in the square bracketof eq. (79). As expected by eq. (65), the corrections tothe propagator due to the isodensity mode are invariantunder exchange of f ↔ − f and 1 , ↔ , a, b . Furthermore, C ab ( η, η ) = 0.The final expression of the nonlinear propa-gator is obtained after the ensemble average ofΞ (is) (0) exp (cid:16)(cid:82) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:17) has been taken. We recallhere that the different modes that enter in Ξ ab are notstatistically independent. The ensemble average can be written as (see appendix A), (cid:68) Ξ (is) (0) e (cid:82) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:69) = (cid:88) p x ,p − ( p − (cid:32) ∞ (cid:88) q =2 c q q ! (cid:33) , (80)where c q is the q -order cumulant of the adiabatic modesand x ,p − is a p -order cross-cumulant defined as x ,p − ≡ (cid:42) Ξ (is) (0) (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:19) p − (cid:43) c . (81)The explicit values of such coefficients depend on theprecise model. For Gaussian initial conditions only c x , are non-zero. Then, eq. (80) can be rewritten as (cid:28) Ξ (is) (0) exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:19)(cid:29) = x , exp (cid:16) c (cid:17) , (82)where c = (cid:90) ηη d η (cid:48) d η (cid:48)(cid:48) (cid:104) Ξ (ad) ( η (cid:48) )Ξ (ad) ( η (cid:48)(cid:48) ) (cid:105) , (83) x , = (cid:90) ηη d η (cid:48) (cid:104) Ξ (is) (0)Ξ (ad) ( η (cid:48) ) (cid:105) . (84)At late time Ξ (ad) is dominated by the growing mode.Thus, we can express it as on the right-hand side ofeq. (37) and we can use eq. (39) for c . For x , wefind x , = − k σ × ( e η − e η ) , (85) where σ × is the cross-correlation between the initial iso-density and the adiabatic modes, σ × ≡ (cid:90) d q C in ( q ) q , (86)with C in defined by (cid:68) Θ (is) ( q , (ad) ( q (cid:48) , (cid:69) = δ D ( q + q (cid:48) ) C in ( q ) . (87)Finally, the ensemble average in eq. (82) can be writtenas (cid:68) Ξ (is) e (cid:82) ηη d η (cid:48) Ξ (ad) ( η (cid:48) ) (cid:69) = − k σ × ( e η − e η ) e − k σ ( e η − e η ) / , (88)so that the nonlinear propagator reads, at first order, G ab ( k ; η, η ) ≈ (cid:104) g ab ( η, η ) − k σ × ( e η − e η ) e − η / C ab ( η, η ) (cid:105) e − k σ ( e η − e η ) / . (89)It is possible to compute the nonlinear propagator athigher orders in Ξ (is) . In particular, in appendix B wederive a recurrence formula for the most growing modeof the resummed propagator, to any order in Ξ (is) . Weare now in the position to illustrate the effect discussed inthis section in a practical case, i.e. the mixture of baryonsand cold dark matter after decoupling. IV. CDM AND BARYONS AFTERDECOUPLING
As an application, in this section we consider the caseof baryons and CDM particles just after decoupling. Thissituation is illustrative of the concepts that we introducedin this paper. Here we focus on the behavior of the propa-gators on scales which are interesting for PT calculations,i.e. k (cid:46) h Mpc − . We will see that for such statisticalobjects and such scales the impact of isodensity modesis very small. We leave the calculations of power spectrafor further studies.The first step of our analysis is to properly identify theisodensity modes after recombination. We will assumethat the primordial (i.e. before horizon crossing) large-scale perturbations are strictly adiabatic. In this caseeach fluid component is proportional to the same randomfield, for instance the primordial curvature perturbation ζ ( k ). We can assume that at the initial time η in = 0 thedifferent fluid variables are in the linear regime. Then,they can be written in terms of the initial linear transferfunctions T a ( k,
0) asΨ a ( k ,
0) = T a ( k, ζ ( k ) . (90) We will use CAMB [20] to generate the CDM and baryoninitial transfer functions, assuming the following cosmo-logical parameters: Ω c = 0 . b = 0 . Λ =0 . h = 0 . n s = 0 . A ζ = 2 . · − and mass-less neutrino species.A remark is in order here. Cosmological fluctuations,such as those described by the CAMB code, obey lin-ear general relativistic equations. On large scales, i.e. onscales comparable with the Hubble radius, these equa-tions may considerably deviate from the Newtonian equa-tions used in RPT, also at the linear level. Thus, one mayworry that the transfer functions generated by CAMBwill be affected by these deviations, which are gauge de-pendent. However, as shown in appendix C, for a set of pressureless fluids there exists a choice of variables forwhich at linear order the relativistic equations exactly reduce to the Newtonian equations. For the density con-trasts of cold dark matter and baryons, this choice cor-responds to take the energy density perturbations in agauge comoving to the total fluid. For the velocity diver-gences this corresponds to take them in the longitudinalgauge. In the limit where we can neglect radiation en-ergy and momentum, the dynamics of these variables iswell described by the Newtonian equations even on super-Hubble scales.Finally, note that even though Ω Λ (cid:54) = 0, we will use thelinear propagator derived in sec. III in a EdS universe.Indeed, as explained in [3] most of the cosmological de-pendence is encoded in the linear growth function D + and using the propagators derived for an EdS universe isa very good approximation.2 k (cid:64) h Mpc (cid:45) (cid:68) ∆ c (cid:45)Θ c ∆ b (cid:45)Θ b FIG. 4. Shape and amplitude of the transfer functions at z = 900. The transfer functions are plotted in units of the to-tal density transfer function. From top to bottom we have theCDM density transfer function (continuous line), the CDMvelocity transfer function (dotted line), the baryon densitytransfer function (dashed line) and the baryon velocity trans-fer function (dotted-dashed line). On super-horizon scalesthey are all approximately equal, denoting that f + (cid:39)
1. Onecan observe that at this high redshift the baryon transfer func-tions are highly suppressed.
A. The linear modes after decoupling
In Fig. 4 we show the transfer functions for the dif-ferent fluid variables normalized to the transfer functionof the total matter perturbation δ m . We choose redshift z = 900 as initial time η in = 0. At this redshift the energydensity and momentum density of the radiation are stillimportant (of the order of 20% percents). However, wewill neglect their contributions in our treatment. More-over, since on super-horizon scales the transfer functionsare all approximately equal, then f + (cid:39) α = − θ α . Note that, contrary to what has been donein [7], one cannot consistently assume that the densityand velocity transfer functions are the same.The linear evolution of each mode can be constructedby applying the linear propagator g ab given in eq. (63).In particular, g (+) ab ( η, η ), g ( − ) ab ( η, η ), g (is) ab ( η, η ) and g (ci) ab ( η, η ) are the growing and decaying adiabatic, andthe decaying and constant isodensity time-dependentprojectors, given respectively by the first, second, thirdand fourth term on the right-hand side of eq. (63). Interms of these projectors one can define the transfer func-tion for each mode as T (+) ( k, η ) ≡ ( f c , , f b , g (+) ab ( η, T b ( k, ,T ( − ) ( k, η ) ≡ ( f c , , f b , g ( − ) ab ( η, T b ( k, ,T (is) ( k, η ) ≡ (0 , , , − g (is) ab ( η, T b ( k, ,T (ci) ( k, η ) ≡ (1 , , − , g (ci) ab ( η, T b ( k, . (91) k (cid:64) h Mpc (cid:45) (cid:68) T (cid:72) (cid:45) (cid:76) T (cid:72) is (cid:76) T (cid:72) ci (cid:76) FIG. 5. The transfer functions at z = 900, normalized to theadiabatic growing mode. From bottom to top, the adiabaticdecaying mode (dashed line), the isodensity decaying mode(continuous line) and the isodensity constant mode (dottedline). These definitions have been chosen in such a way that T a = T (+) u (+) a + T ( − ) u ( − ) a + T (is) u (is) a + T (ci) u (ci) a . (92)These quantities are shown on Fig. 5 where we plotthe amplitude of the transfer functions T ( − ) , T (is) and T (ci) at initial time, normalized to the amplitude of T (+) .Note that from these results one can compute the r.m.s.of Ξ (is) that appeared in the previous section. One findsthat (cid:68) Ξ (is)2 (cid:69) / = 8 . · − kh Mpc − , (93)at redshift z = 900, showing that for our scales of inter-est, k (cid:46) h Mpc − , the effects of the isocurvature modescan only be small. The explicit dependence of the propa-gators on the isodensity modes is shown in the following. B. The nonlinear propagators
In the presence of the decaying isodensity mode, theresummed propagator is no longer proportional to thefree field propagator. The effect of the isodensity modeon the resummed propagator is modulated by the matrix C ab ( η, η ) in eq. (79). To show this modulation, let usdefine the quantities R δ c ( k, η ) ≡ C a ( η, T a ( k, /T δ c ( k, η ) ,R δ b ( k, η ) ≡ C a ( η, T a ( k, /T δ b ( k, η ) , (94)where T δ c ( k, η ) ≡ g a ( η, T a ( k, ,T δ b ( k, η ) ≡ g a ( η, T a ( k, . (95)3 k (cid:64) h Mpc (cid:45) (cid:68) R ∆ c (cid:72) k , z (cid:76) z (cid:61) (cid:61) (cid:61) (cid:61) (cid:61) (cid:45) (cid:45) (cid:45) k (cid:64) h Mpc (cid:45) (cid:68) R ∆ b (cid:72) k , z (cid:76) z (cid:61) (cid:61) (cid:61) (cid:61) (cid:61) FIG. 6. The quantities R δ c ≡ C a ( z, z in = 900) T a ( k, z in =900) /T δ c ( k, z ) (upper pannel) and R δ b ≡ C a ( z, z in =900) T a ( k, z in = 900) /T δ b ( k, z ) (lower pannel) as a functionof scale at different redshifts. In Fig. 6 we have plotted these quantities as a functionof scale and for different redshifts z = 500 , , , , D + = 1 . , . , . , . , .
43. InFig. 7 we have plotted R δ c and R δ b as a function of red-shift and for different scales k = 0 . , . , . , h Mpc − . At small redshift (large η ) R δ c and R δ b aredominated by the most growing mode of the matrix C ab ,i.e. the first term in eq. (79) which grows as e η , so thatthey are independent of redshift. At higher redshift thedecaying modes in the matrix C ab become important andfor z = 900, corresponding to the initial time η in = 0, R δ c and R δ b go to zero. Note that R δ b becomes infinite twicearound z ∼ T δ b ( k, η ) crosses zero twice.The entire effect of the isodensity mode on the prop-agator is represented by the second term in the squarebracket in eq. (89), which for η = η in = 0 is − k r × σ ( D + ( η ) − C ab ( η, . (96) R ∆ c (cid:72) k , z (cid:76) k (cid:61) (cid:61) (cid:61) (cid:61) (cid:45) (cid:45) R ∆ b (cid:72) k , z (cid:76) k (cid:61) (cid:61) (cid:61) (cid:61) FIG. 7. The quantities R δ c (upper pannel) and R δ b (lowerpannel) as a function of redshift at different scales. Here the parameter r × is the ratio of the isodensity-adiabatic displacement cross-correlation σ × to the vari-ance of the adiabatic displacement field σ , r × ≡ σ × σ = (cid:82) d ln q T (is) ( q, T (+) ( q, q n s − (cid:82) d ln q [ T (+) ( q, q n s − . (97)(In the second equality we have neglected the adiabaticdecaying mode.) At z = 900 this is r × (cid:39) .
85. For kσ d D + (cid:29) D + ( z ) −
1. For scales k < k d ≡ ( σ d D + ) − ,the corrective term (96) is always found to be extremelysmall.In order to be more quantitative, in Fig. 8 we show theeffect of the isodensity mode on the nonlinear propagatorby plotting G δ c /G (ad) δ c − G δ b /G (ad) δ b − k d , where G δ c ( k, η ) ≡ G a ( k ; η, T a ( k, ,G δ b ( k, η ) ≡ G a ( k ; η, T a ( k, , (98)and G (ad) δ c ( k, z ) and G (ad) δ b ( k, z ) are the same quantities4 − − z − δG δ c ( k d , z ) /G (ad) δ c ( k d , z ) δG δ b ( k d , z ) /G (ad) δ b ( k d , z ) FIG. 8. Effect of the isodensity mode on the nonlinear prop-agators normalized to the adiabatic nonlinear propagators, δG δ c /G (ad) δ c ≡ G δ c /G (ad) δ c − δG δ b /G (ad) δ b ≡ G δ b /G (ad) δ b − G (ad) δ c and G (ad) δ b are the CDM and baryon nonlinearpropagators in absence of isodensity mode, computed at fixedscale k d , as a function of redshifts. The oscillations appearingfor z (cid:46) k (cid:46) .
4. Note that the effect on the CDM prop-agator is plotted with the sign changed. in the adiabatic case – i.e. for Ξ (is) = 0. We have usedthat the value of the variance of the displacement fieldis σ d (cid:39) . × − h − Mpc at z = 900. For z ≤
50 wefind that the effect is less that ∼
1% and for z ≤ ∼ ‰ . Note that the effects are of different signsbetween CDM and baryons.Another example where these effects could be signifi-cant is when the isodensity modes are set at much lowerredshift. This is potentially the case for massive neutri-nos. However, massive neutrinos cannot be fully consid-ered as non-relativistic particles during their cosmologi-cal history as their behavior is determined by a wholeset of extra modes, such as pressure fluctuations andanisotropic stresses. We leave the study of this specialcase for the future. V. CONCLUSIONS
The eikonal approximation provides an efficient for-malism within which exact resummation in the high- k limit can be performed explicitly or numerically. We wereable to recover the standard results obtained in [2, 3, 5, 6]concerning the nonlinear 2-point and multi-point propa-gators describing the gravitational instabilities of a singlepressureless fluid. In particular, the propagators are cor-rected by an exponential cut-off whose scale is fixed bythe amplitude of the displacement field along the wave-mode k . We have shown this irrespective of the growthrate of the displacement field and of whether it followsthe linear regime, thus extending the standard resultspreviously quoted. q [ h Mpc − ] π q P i n ( q ) FIG. 9. Mode contribution per log interval to the variance ofthe displacement field from the adiabatic modes, eq. (40), at z = 0. Note that this formalism is based on a mode separa-tion between large-scales and small-scales. Indeed, wehave assumed that the large-scale modes with momen-tum q , collected in the random variable Ξ, are muchsmaller that the small-scale modes k . In Fig. 9 we showthe contribution from adiabatic modes to the variance ofthe displacement field, σ d , per logarithmic interval. Asone can see, most of the contribution comes from modeswith q (cid:46) . h Mpc − but that of smaller modes, with q ≈ . ∼ h Mpc − is not negligible. This suggeststhat, for k ≈ . ∼ . h Mpc − , a better description ofthe damping could be obtained by setting a UV cutofffor q in eq. (40).We have then extended the eikonal approximation tomultiple pressureless fluids. In this case one can identifytwo types of modes: Two adiabatic modes and two iso-density modes per added species. Isodensity modes areresponsible for new effects. Indeed, their large-scale flowchanges the phase but also (unlike the adiabatic modes)the amplitude of small scales. Thus, the growth of struc-ture and consequently the amplitude of propagators andspectra are affected in a more complex way than in thepurely adiabatic case. In this paper we focus our resultson the propagators, leaving the study of power spectrafor future work.In contrast to the single-fluid case, where the effect oflarge-scale adiabatic modes can be taken into account an-alytically, for the isodensity modes we have not been ableto find an analytic form for the resummed propagator. Inthis case, one should rely on a numerical or a perturbativeapproach. The latter is sufficient when one considers thecase of CDM-baryon mixing. For this example, we foundthat the impact of isodensity modes on the propagatorsis very small at low redshift and for scales of interest forstandard PT, i.e. for k (cid:46) h Mpc − . However, theremight be cases where the impact of large-scale modes ismore significant, for instance when the scale of interestare close to the non-linear regime at the time the isoden-sity modes are set in. This is expected to be the case5for massive neutrinos. Although we did not address thiscase explicitly, we stress that the eikonal method can beused irrespective of the field content of the system. Weleave the case of massive neutrinos for further studies. Acknowledgements:
We thank the participants ofthe PTchat workshop at IPhT in Saclay for interestingdiscussions. FV wishes to thank Antonio Riotto and RaviSheth for fruitful conversations.
Appendix A: Computation of moments
We want to compute the value of g n = (cid:104) Ξ n exp( D ) (cid:105) , (A1)where Ξ and D are two random variables whose statis-tical properties are entirely characterized by their jointcumulants, x p,q = (cid:104) Ξ p D q (cid:105) c . (A2)It is convenient to introduce the auxiliary functionexp( D + λ Ξ) and to notice that g n = d n d λ n (cid:104) exp( D + λ Ξ) (cid:105)| λ =0 . (A3)The ensemble average that appears in this expressioncan be written in terms of the cumulants generating func-tion of D + λ Ξ as g n = d n d λ n exp (cid:32) ∞ (cid:88) p =0 p (cid:88) q =0 q ! 1( p − q )! x q,p − q λ q (cid:33)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) λ =0 , (A4)which can be re-written as g n = exp (cid:32) ∞ (cid:88) p =0 x ,p p ! (cid:33) d n d λ n exp (cid:32) n (cid:88) q =1 X q q ! λ q (cid:33)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) λ =0 , (A5)where X q = ∞ (cid:88) p = q p − q )! x q,p − q . (A6) g n can then be formally expressed in terms of X q as g = exp (cid:32) ∞ (cid:88) p =0 x ,p p ! (cid:33) , (A7) g = X g , (A8) g = (cid:0) X + X (cid:1) g , (A9) g = (cid:0) X + 3 X X + X (cid:1) g , (A10) . . . . Note that for a unit g the relation between X q and g n is exactly the one relating cumulants of order q with mo-ments of order n . Thus, in general this relation is ob-tained by the Arbogast-Fa`a di Bruno formulae.These relations greatly simplify in the case of Gaussianinitial conditions, assuming – without loss of generality– that x , = x , = x , = 0. Indeed, in this case only X and X are non-zero, X = x , , X = x , , (A11)so that g = exp (cid:16) x , (cid:17) , (A12) g = x , g , (A13) g = (cid:0) x , + x , (cid:1) g , (A14) g = (cid:0) x , + 3 x , x , (cid:1) g , (A15) . . . . Appendix B: Higher-order time dependence
It is possible to compute the nonlinear propagator athigher orders in Ξ (is) using the expansion ξ af ( k , η, η ) = g af ( η, η ) exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) + ξ (1) af ( k , η, η ) + ξ (2) af ( k , η, η ) + . . . , (B1)where ξ (1) af ≡ exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) (cid:90) ηη d η (cid:48) g ab ( η, η (cid:48) )Ξ (is) bc ( k , η (cid:48) ) g cf ( η (cid:48) , η ) ,ξ (2) af ≡ exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) (cid:90) ηη d η (cid:48) (cid:90) η (cid:48) η d η (cid:48)(cid:48) g ab ( η, η (cid:48) )Ξ (is) bc ( k , η (cid:48) ) g cd ( η (cid:48) , η (cid:48)(cid:48) )Ξ (is) de ( k , η (cid:48)(cid:48) ) g ef ( η (cid:48)(cid:48) , η ) ,. . . . (B2)6In particular, here we derive an explicit expression forthe most growing solution to the, in terms of a recurrenceformula.The key point is to show that the fastest growing modeof the amplitude of the corrections to the resummed prop- agator goes as the adiabatic growing mode, i.e. ∝ e η − η ,to all orders in Ξ (is) . Let us consider the amplitude ofthe n th -order correction. From eq. (B2) and using that g ( λ ) ab ( η, η ) ∝ e λ ( η − η ) and Ξ (is) ab ( η ) ∝ e − η ( η in = 0), itstime dependence is proportional to n nested integrals,given by I ( n ) ≡ (cid:90) ηη d η e λ ( η − η ) e − η (cid:90) η η d η e λ ( η − η ) e − η · · · (cid:90) η n − η d η n e λ n − ( η n − − η n ) e − η n e λ n ( η n − η ) , (B3)where each λ i can take the values { , , − / , − / } . Aswe are only interested in the fastest growing mode, wetake λ = 1.We define α ij ≡ λ j − λ i − j − i , (B4)satisfying α ij + α jk = α ik . It is important to note that α i < i (cid:54) = 0. Then, we can rewrite the integralsabove as C (cid:90) ηη d η e α η (cid:90) η η d η e α η (cid:90) η n − η d η n e α n − ,n η n (B5)with C ≡ e λ η e − λ n η . By performing an integration byparts on η , we find the boundary term C (cid:20) e α η α (cid:90) η η d η e α η · · · (cid:90) η n − η d η n e α n − ,n η n (cid:21) η = ηη = η , (B6)and the remaining integrals − C (cid:90) ηη d η e α η α e α η · · · (cid:90) η n − η d η n e α n − ,n η n . (B7)The lower bound for the boundary term obviously van-ishes. Therefore, the boundary term contains an e α η factor making it subleading with respect to the remain-ing integrals, which for large η become constant. To seethis, we perform ( n −
2) more integration by parts, ev-ery time dropping the boundary term for the very samereason, until we are left with( − n − Cα α · · · α ,n − (cid:90) ηη d η n e α n η n . (B8)The leading term of this integral is independent of η ,( − n n (cid:89) i =1 Cα i e α n η , (B9)and by replacing C by its definition, we have I ( n ) = n (cid:89) i =1 α i e − n η e η − η . (B10) Now that we have studied the time dependence, wecan reintroduce the time independent matrices h ab g ( λ i ) bc in(B2) and sum over λ i in each of the integrals of eq. (B3).Let us define the time independent matrices A (0) ab ≡ g (1) ab ,A ( i ) ab ≡ (cid:88) λ =1 , , − / , − / h ac g ( λ ) cb i/ − λ , i ≥ , (B11)where h ab is defined in eq. (65), g ( λ ) ab ≡ g ( λ ) ab ( η , η ) arethe time independent projectors on the right-hand sideof eq. (63), and the sum runs over λ = 1 , , − / , − / − . Thedenominator (1 + i/ − λ ) is exactly the α i of eq. (B10).It is easy to verify that the most growing solution for ξ ( n ) ab in eq. (B1) is then given by ξ ( n ) ab ( k , η, η ) = (cid:104) Ξ (is) ( k , e − η / (cid:105) n (cid:104) A (0) n (cid:89) i =1 A ( i ) (cid:105) ab × e η − η exp (cid:18)(cid:90) ηη d η (cid:48) Ξ (ad) ( k , η (cid:48) ) (cid:19) . (B12)This equation formally generalizes the expression for themost growing mode contained in eqs. (78) and (79) toany order. Note that the time dependence is the same asthe linear growing mode, i.e. ∝ e η . Ensemble averages ofthese quantities can be taken using the equations givenin appendix A. Appendix C: Treating super-horizon scales
Let us consider a linearly perturbed FLRW metric inlongitudinal gauge with only scalar perturbations. In theabsence of anisotropic stress the two metric potentials areidentical and the metric simply readsd s = a ( τ ) (cid:2) − (1 + 2 φ lon )d τ + (1 − φ lon )d x (cid:3) , (C1)where τ is the conformal time defined by d τ = d t/a ( t ).7Combining the 00 and the 0 i components of Einstein’sequations gives, in Fourier space, [21] k φ lon = − H (cid:0) δ lon + 3 H θ lon /k (cid:1) , (C2)where δ lon and θ lon are the total density contrast and di-mensionless velocity divergence [22], respectively, in lon-gitudinal gauge and H ≡ d ln a/ d τ is the conformal Hub-ble rate. Moreover, in this gauge, the continuity andEuler equations for a single-fluid read, at linear order,[21] δ (cid:48) lon = −H θ lon + 3 φ (cid:48) lon , (C3) θ (cid:48) lon = − ( H (cid:48) / H + H ) θ lon + k φ lon / H , (C4)where a prime denotes the derivative with respect to con-formal time.One can check that, at the linear level, the NewtonianEuler equation (2) is the same as its relativistic version,eq. (C4), while the continuity equations (1) and (C3)differ by the term 3 φ (cid:48) lon . Indeed, for instance, using that a ∝ τ in matter dominance, one can check that thesolutions to the above equations are φ lon ( k, τ ) = φ + ( k ) + ( kτ ) − φ − ( k ) , (C5)and δ lon = − (cid:18) kτ ) (cid:19) φ + − (cid:18)
16 ( kτ ) − − kτ ) − (cid:19) φ − , (C6) θ lon = 16 ( kτ ) φ + −
14 ( kτ ) − φ − . (C7)Thus, eq. (C7) correctly describes the growing and decay-ing solutions of θ in the Newtonian limit, eq. (11), with θ + ∝ a and θ − ∝ a − / , even on super-Hubble scales,while for δ lon we recover the Newtonian case, eq. (10),only in the limit kτ (cid:29) δ com ≡ δ lon + 3 H θ lon /k = − (cid:0) ( kτ ) φ + + ( kτ ) − φ − (cid:1) , (C8)does this job. Indeed, replacing φ (cid:48) lon using the 0 i com-ponents of Einstein’s equation [21], and using eq. (C4),eq. (C3) reads δ (cid:48) com = −H θ lon , (C9) thus reproducing the linear part of the continuity equa-tions in the Newtonian limit, eq. (1). Moreover, in termsof this variable eq. (C2) becomes a Poisson-like equation, k φ lon = − H δ com , (C10)reproducing eq. (3).We conclude that at linear level the Newtonian equa-tions (1), (2) and (3) describe the relativistic dynamicsonce we interpret the Newtonian potential φ as the met-ric potentials in longitudinal gauge φ lon , the Newtoniandensity contrast δ as the comoving energy density per-turbation δ com and θ as the dimensionless velocity diver-gence in longitudinal gauge, θ lon .The case of many fluids is not very different. In thiscase one can show that for each species α the relativis-tic perturbation variable which satisfies the Newtoniancontinuity equation is the energy density perturbationcomoving to the total fluid, δ α, com ≡ δ α, lon + 3 H θ lon /k , (C11)where θ lon is the total dimensionless velocity divergencein longitudinal gauge, θ lon ≡ (cid:88) α f α θ α, lon . (C12)Using this variable, eq. (C2) becomes a Poisson equationwhile the velocity divergence in longitudinal gauge foreach species α , θ α, lon , naturally satisfies the NewtonianEuler equation.Let us connect these variables with those of CAMB inthe case of a mixture of cold dark matter and baryons.CAMB uses synchronous gauge comoving with cold darkmatter [20]. Using the gauge transformation betweensynchronous and longitudinal gauge [21] we have that δ α, CAMB = δ α, lon + 3 H σ ,θ α, CAMB = θ α, lon − σk / H , (C13)where σ is the shear. This is related to the usual syn-chronous metric perturbation variables h syn and η syn ,respectively the trace and traceless part of the spatialmetric, by σ = ( h (cid:48) syn + 6 (cid:48) syn ) / k . Using these equa-tions, since in CAMB the dark matter velocity vanishes, θ c , CAMB = 0, we have θ c , lon = σk / H ,θ b , lon = θ b , CAMB + σk / H ,δ α, com = δ α, CAMB + 3 H f b θ b , CAMB /k . (C14) [1] F. Bernardeau, S. Colombi, E. Gazta˜naga, and R. Scoc-cimarro, Phys. Rep. , 1 (Sep. 2002) [2] M. Crocce and R. Scoccimarro, Phys. Rev. D , 063519 (Mar. 2006), astro-ph/0509418[3] M. Crocce and R. Scoccimarro, Phys. Rev. D , 063520(Mar. 2006), astro-ph/0509419[4] M. Crocce and R. Scoccimarro, ArXiv e-prints (Apr.2007), 0704.2783[5] F. Bernardeau, M. Crocce, and R. Scoccimarro, Phys.Rev. D , 103521 (Nov. 2008), arXiv:0806.2334[6] F. Bernardeau, M. Crocce, and E. Sefusatti, Phys. Rev.D , 083507 (Oct. 2010), arXiv:1006.4656 [astro-ph.CO][7] G. Somogyi and R. E. Smith, Phys. Rev. D , 023524(Jan. 2010), arXiv:0910.5220 [astro-ph.CO][8] M. Pietroni, Journal of Cosmology and Astro-ParticlePhysics , 36 (Oct. 2008), arXiv:0806.0971[9] P. Valageas, Astr. & Astrophys. , 79 (Jun. 2008),arXiv:0711.3407[10] F. Bernardeau and P. Valageas, Phys. Rev. D , 083503(Oct. 2008), arXiv:0805.0805[11] H. D. I. Abarbanel and C. Itzykson, Phys. Rev. Lett. ,53 (1969)[12] M. Levy and J. Sucher, Phys. Rev. , 1656 (1969)[13] Another case studied in PT with two components follow-ing geodesic motion is a mixture of CDM and clusteringquintessence [23]. However, in this case the initial condi-tions are such that the two fluids remain comoving duringtheir evolution and isodensity modes do not develop.[14] D. Tseliakhovich and C. Hirata, Phys. Rev. D , 083520 (Oct. 2010), arXiv:1005.2416 [astro-ph.CO][15] R. Scoccimarro, in The Onset of Nonlinearity in Cosmol-ogy , New York Academy Sciences Annals, Vol. 927, editedby J. N. Fry, J. R. Buchler, and H. Kandrup (2001) pp.13–+[16] It is straightforward to show that for ΛCDM f − = − Ω m and f + /f − = 1 − aD + .[17] In this paper we will indistinguishably use the term ofpropagator for both ξ ab and G ab although the latter isthe ensemble average of the former.[18] We are here in a situation comparable to that encoun-tered in Lagrangian space where the displacement isfound to be non-potential at order three and beyond inPT.[19] J. R. Bond, G. Efstathiou, and J. Silk, Physical ReviewLetters , 1980 (Dec. 1980)[20] A. Lewis, A. Challinor, and A. Lasenby, Astrophys. J. , 473 (2000), astro-ph/9911177[21] C.-P. Ma and E. Bertschinger, Astrophys. J. , 7 (Dec.1995), arXiv:astro-ph/9506072[22] Note that contrary to the notation of [21], here θ de-notes the dimensionless velocity divergence so that θ our ≡ θ MB / H .[23] E. Sefusatti and F. Vernizzi, Journal of Cosmol-ogy and Astro-Particle Physics3