Cosmic Microwave Background Anisotropy

Massachusetts Institute of Technology Department of Physics Physics 8.942 Cosmic Microwave Background Anisotropy 1 Introduction These notes present a simplified computation of cosmic microwave background anisotropy induced by primeval gravitational potential or entropy fluctuations. The basic treatment is equivalent to the calculation of anisotropy on large angular scales presented first by Sachs & Wolfe (1967). The discussion here goes beyond the Sachs-Wolfe treatment to include a discussion of the dominant contributions to anisotropy on small angular scales. The presentation given here is based in part on the PhD thesis of Sergei Bashinsky (Bashinsky 2001) and Bashinsky & Bertschinger (2001, 2002). More elementary treat­ ments of CMB anisotropy are given by Chapter 18 of Peacock and online by Wayne Hu at http://background.uchicago.edu/). We adopt several assumptions in order to simplify the algebra without losing the main physical effects: 1. Instantaneous recombination. We assume that, prior to hydrogen recombination at a = ar , photons scatter so frequently with electrons that the photon gas is a perfect gas with spatially-varying energy density ρr and fluid three-velocity v i (orthonormal components of peculiar velocity). This gas is tightly coupled to the “baryons” (electrons plus all ionization states of atomic matter). The gradual decoupling of photons and baryons is approximated by instantaneous decoupling at a = ar . We will indicate how to improve on this treatment by treating the radiation field prior to recombination as an imperfect gas coupled to baryons by Thomson scattering. An accurate treatment of hydrogen and helium recombination is needed in this case. With non-instantaneous recombination, photon polarization must also be considered. We ignore polarization here. The resulting errors are a few percent in rms temperature anisotropy at small angular scales. 2. No large-scale spatial curvature. We suppose that the universe is a perturbed Robertson-Walker spacetime. Although the curvature terms (proportional to 1−Ω) 1 will be included in the basic equations, our numerical calculations will not fully include them. This simplifies the harmonic decomposition and is consistent with inflation as the origin of fluctuations. The only important effect we miss is the anisotropy produced by the time-changing gravitational potential at redshift z < 10, the integrated Sachs-Wolfe effect. 3. Adopt a simplified description of matter. Aside from baryons prior to recombina­ tion, we will treat the matter as cold. This is appropriate for CDM and for other types of matter on scales larger than the Jeans length (an excellent approximation for CMB anisotropy). The complex gravitational interaction between photons, neutrinos, and dark matter prior to recombination will be treated in a simplified manner. However, we include correctly, without approximation, a possible nonzero cosmological constant (vacuum energy). 4. Neglect gravitational radiation. In the linearized treatment of small-amplitude fluc­ tuations, gravitational radiation is a distinct mode and so its effect on the CMB may be treated as a separate contribution computed independently of the density and entropy fluctuations studied here. 5. Assume that all perturbations have small (linear) amplitude. This is valid for “pri­ mary” anisotropies but not for “secondary” anisotropies caused by nonlinear struc­ tures forming at low redshift. Secondary anisotropies are negligible on angular scales larger than about 10 arcminutes. The justification for these simplifications is mainly pedagogical, although in general they do not introduce serious errors on angular scales larger than a few degrees of arc. (Gravitational radiation, if present, can contribute significantly to the CMB anisotropy on large angular scales. Curvature also has a significant effect on large scales.) Once the student understands CMB anisotropy in this simplified model, a more realistic treatment can be undertaken (Ma & Bertschinger 1995; Seljak & Zaldarriaga 1996; Zaldarriaga, Seljak, & Bertschinger 1998; Hu et al 1998, and references therein). Aside from the simplifications given above and some computational approximations stated later, the treatment given herein is rigorous and complete. We will discuss both the “adiabatic” (i.e. isentropic) and “isocurvature” (entropy) modes, the integrated Sachs-Wolfe effect, and both intrinsic and Doppler anisotropies. The computation of microwave background anisotropy has several ingredients. The rest of these notes give a systematic presentation of these ingredients. In Section 2 we present the Einstein equations for a perturbed Robertson-Walker spacetime. In Section 3 we derive and formally integrate the radiative transfer equation for CMB anisotropy. Section 4 gives a physical interpretation of the primary contributions to CMB anisotropy. These contributions cannot be calculated until the evolution of metric, matter, and radia­ tion perturbations is given, as they provide the source terms for CMB anisotropy. Section 2 5 presents a simplified set of evolution equations. Section 6 solves these equations on large scales to derive the famous Sachs-Wolfe formula. Section 7 presents a real-space Green’s function approach to solving and understanding the small-scale behavior, espe­ cially the acoustic peaks. Section 8 shows how to compute the angular power spectrum from the solution of the radiative transfer equation. Numerical results are presented in Section 9. 2 Perturbed Robertson-Walker Spacetimes This section summarizes the elementary treatment of a weakly perturbed RobertsonWalker spacetime. We write the spacetime line element as � ds2 = a2 (τ ) −(1 + 2φ)dτ 2 + (1 − 2ψ)dl2 � (1) where dl2 is the usual spatial line element in comoving coordinates (e.g. dl 2 = dχ2 + r2 dΩ2 , or dl2 = dx2 + dy 2 + dz 2 in Cartesian coordinates if the background space is flat). The metric perturbations are characterized by two functions φ(xi , τ ) and ψ(xi , τ ) which we assume are small (we neglect all terms quadratic in these fields). Equation (1) is a cosmological version of the standard weak-field metric used in linearized general relativity. It ignores gravitomagnetism (present in a g 0i metric term) and gravitational radiation (present in a transverse-traceless strain tensor h ij added to the spatial metric). Note that the Newtonian limit follows from equation (1) when ȧ = 0 and φ = ψ. Dots denote derivatives with respect to conformal time τ , while gradient symbols (e.g. �i ) represent gradients (covariant derivatives) with respect to the comoving spatial coordinates using the unperturbed 3-metric of dl 2 . In a flat background with Cartesian coordinates, �i = ∂i . As in the standard treatment of linearized general relativity, we describe spacetime by perturbations added to a background model. The coordinate-freedom allows us many ways to do this for a given set of physical perturbations. We will not discuss this gaugefixing problem in these notes. Bertschinger (1996) discusses this issue in detail. Equation (1) conveniently avoids gauge artifacts and enables us to focus on the physics. In the following, we often speak of “fundamental” (or “comoving”) observers by which we mean observers at fixed xi . The 4-velocity of a fundamental observer is V� = a−1 (1−φ)�eτ . In general, fundamental observers are not freely-falling (as one can see from the geodesic equation for dV i /dλ), nor do they correspond to the rest frame of a galaxy or to the frame in which the CMB dipole anisotropy vanishes. (See Peacock section 9.4 for a discussion of the dipole anisotropy caused by the Doppler effect.) However, they define a very convenient set of local rest frames in which to project tensor components. 3 As usual, we construct an orthonormal basis (a tetrad) at each spacetime point with �eˆ0 = V� and with the spatial axes being parallel to the coordinate axes assuming that we use orthogonal coordinates for the background Robertson-Walker spacetime. (The spatial curvature perturbation 1 − ψ does not affect the orthogonality of the comoving coordinate axes as it is a conformal factor for dl 2 .) Energy, momentum, three-velocity (i.e. proper peculiar velocity), energy density, entropy density, etc. all have the same meanings as in special relativity when obtained from tensor components with a funda­ mental observer’s orthonormal basis. Note that there is such an observer at every point in space; we don’t (yet) preferentially single out the observer at χ = 0. In the absence of anisotropic, relativistic stresses (produced mainly by neutrinos in the early universe), φ = ψ and the spacetime metric perturbations are described by a single Newtonian-like gravitational potential. In this case, equation (1) is identical to the usual weak-field metric in the Newtonian limit aside from the conformal factor a2 (τ ) arising from cosmic expansion. In the following, we will distinguish the two metric functions in order to keep track of which physical effects arise from the gravitational redshift part of the metric (φ) and which arise from the spatial curvature part (ψ). The unperturbed, or background, density ρ̄(t) obeys the Friedmann equation: � d ln a dτ �2 = 8π Gρ̄a2 − K . 3 (2) Note that because ρ¯ ∝ a−3 in the matter-dominated era and ρ¯ ∝ a−4 in the radiationdominated era, the curvature term is negligible at high redshift. However, spatial curva­ ture affects the geometry at low redshift and does have an effect on CMB anisotropy. For later use, we provide the solution to equation (2) for a model consisting of only matter (with present density parameter Ωm ) and radiation (whose energy density equals that of matter at a = aeq = 2.41 × 104 Ωm h2 ): a(τ ) τ 1 τ 2 aeq 1/2 = + , where H0 τe ≡ . (3) aeq Ωm τe 4 τe √ Note that a = aeq at τ = 2( 2 − 1)τe . For Ωm = 0.35 and h = 0.7, a−1 eq = 4100 and τe = 110 Mpc. This solution neglects curvature and a cosmological constant but should be accurate to better than 0.1% during recombination. � 3 � � � � � CMB Radiative Transfer We wish to determine the specific intensity Iν measured by the observer at χ = 0 and τ = τ0 as a function of photon direction and energy. Conceptually, the way we will do it is by tracing photon trajectories back from χ = 0 to recombination at radial coordinate 4 distance χe = τ0 − τe . The subscript “e” stands for emitter and reminds us that we are looking at radiation emitted from a photosphere, the edge of the plasma layer that exists for χ > χe , i.e. for times τ < τe . We see the CMB as though we live in a transparent spherical cavity of radius χe beyond which is a hot glowing plasma. Of course, χe really represents a boundary in time rather than space; every fundamental observer has a different photosphere. In the absence of emission, absorption, and scattering, conservation of photons im­ plies that the photon phase space density is conserved along null geodesics. In the language of basic astronomy, dIν /ds = 0 where Iν is the specific intensity and ds mea­ sures path length. However, unlike most other astronomical situations, the path length measures not just distance in space; d/ds measures the rate of change in phase space. The radiation field varies with time, energy (photon frequency) and photon direction as well as with position in space. We will work out these dependencies to calculate the CMB anisotropy. First, however, we must examine the cosmological radiative transfer equation more closely. 3.1 Derivation of radiative transfer equation In place of the specific intensity, we use the radiation brightness temperature to character­ ize the phase space distribution. Let E be the photon energy measured by a fundamental observer and let ni = pi /E be a unit vector (with respect to the observer’s orthonormal basis) in the direction of photon travel. The brightness temperature Tbr (xi , τ, E, ni ) is defined implicitly through Iν = Bν (Tbr ) where Bν (T ) is the Planck function and Iν is the specific intensity measured by the fundamental observer at (xi , τ ). Because the radiation perturbations are small, we write Tbr = a−1 T0 (1 + Δ) where T0 is the present unperturbed blackbody temperature and Δ2 � 1. Our Δ coincides with Peacock’s δT /T ; most other workers write ΔT /T . Thus, the photon phase space distribution (summed over polarizations) is f (xi , τ, E, ni ) = aE c 2 Iν 4 3 = fP 1+Δ 2πh̄ ν � � where fP (�) ≡ 2h̄−3 [exp(�/T0 ) − 1]−1 . , (4) [The unperturbed temperature T0 can, in fact, never be measured because we see only the perturbed universe. Separating the radiation field into an unperturbed “back­ ground” plus perturbations is really a fiction chosen for the calculational convenience of linear perturbation theory. In practice, T0 is replaced by the average of the brightness temperature taken over the sky. Anisotropy measurements are made by subtracting pairs of brightness temperatures from different directions in the sky. Uncertainty in T 0 affects only the “monopole” anisotropy, which is unmeasurable and consequently ignored by observers of CMB anisotropy. The same is true of the “dipole” anisotropy, which re­ ceives contributions from both the observer’s motion and an intrinsic anisotropy due to 5 a possible large-scale gradient of brightness temperature across our Hubble volume. The choice of rest frame for fundamental observers is arbitrary; in practice it is chosen so that the dipole anisotropy vanishes at the solar system barycenter.] In general, the brightness temperature is a function of photon energy. However, we will see that for primary CMB anisotropies, the energy-dependence disappears: Δ = Δ(xi , τ, ni ). That is, in every direction, a fundamental observer finds that the specific intensity is a perfect black-body as a function of photon energy but with the tempera­ ture varying with photon direction. A simple example of this behavior is the Doppler anisotropy derived by Peacock in his Section 9.4. However, it holds also for all forms of CMB anisotropy that do not involve large-amplitude perturbations. Secondary anisotropies induced by nonlinear structures at low redshift often de­ pend on energy (i.e. the spectrum is distorted from blackbody). An example is the Sunyaev-Zel’dovich effect caused by scattering of CMB photons by hot ionized gas in clusters of galaxies. The spectral signature makes it easy to distinguish SunyaevZel’dovich anisotropy from primary CMB anisotropy. These notes consider only the primary anisotropies. Our goal is to determine Δ0 (ni ) = Δ(χ = 0, τ0 , ni ), the anisotropy measured by the fundamental observer at the origin. (The anisotropy measured by a moving observer at the origin, e.g. one on the Earth, follows by a simple Doppler shift following Peacock eq. 9.62.) We use the fact that the photon phase space density is conserved in the absence of emission, absorption, and scattering. From equation (4), the total change in the phase space density is (assuming Δ2 � 1) � where �� � aE aE d ln(aE) dΔ df − = fP� = dτ 1+Δ 1+Δ dτ dτ dΔ ∂Δ ∂Δ dxi ∂Δ dE ∂Δ dni = + i + + i . dτ ∂τ ∂x dτ ∂E dτ ∂n dτ � � � df dτ � , c (5) We may thus write the radiative transfer equation as d ln(aE) dΔ ∂Δ ∂Δ dxi ∂Δ dE ∂Δ dni + i + + i = + ∂τ ∂x dτ ∂E dτ ∂n dτ dτ dτ � � (6) c where the term with a subscript c accounts for collisions (emission, absorption, and scattering of photons). We can simplify equation (6) by linearizing it assuming small perturbations in both Δ and the phase space trajectories. From the metric, using the fact that we defined n i to be the photon direction measured by a fundamental observer in an orthonormal frame, we have (1 − ψ) dxi = ni . (7) (1 + φ) dτ 6 Thus, treating both the metric perturbations and Δ as being of the same order, we can replace dxi /dτ by ni in equation (6). A similar argument shows that we can drop the (∂Δ/∂ni )(dni /dτ ) term: both factors are first-order quantities. This is equivalent to neglecting gravitational lensing of anisotropies. Lensing by nonlinear (highly overdense) structures at low redshift has a small effect on the CMB on arcminute angular scales (Seljak 1996). We are left with calculating the rate of change of proper energy E, which is straight­ forward from the geodesic equation. The result is d ln(aE) = −ni ∂i φ + ∂τ ψ . dτ (8) The factor a accounts for the cosmological redshift while the first term on the right-hand side is the familiar gravitational redshift. (Both terms arise from g 00 .) The second term, due to spatial curvature fluctuations, is unfamiliar in the Newtonian limit but can be significant for photons and other relativistic particles. Because the right-hand side of equation (8) is independent of E, it follows that the d ln(aE)/dτ contribution to equation (6) cannot induce any energy-dependence of the brightness temperature perturbation. (As in gravitational lensing, gravity is “achro­ matic.”) The dominant collision process, Compton scattering, is also independent of energy (at temperatures much less than the electron mass). Thus, unless some other physical process causes a departure of the photon spectrum from blackbody, ∂Δ/∂E = 0 to an excellent approximation. None of the standard sources of primary anisotropy (primeval potential, density, or entropy perturbations) generates any significant black­ body distortion, so we assume ∂ΔE = 0 in the following. (Departures of order the baryon-to-photon ratio 10−10 are generated during recombination, but these are orders of magnitude smaller than the effects we retain.) Combining equations (6), (7), and (8), we obtain the fundamental equation of CMB anisotropy: � � dΔ dΔ = ∂τ Δ + ni ∂i Δ = −ni ∂i φ + ∂τ ψ + . (9) dτ dτ c Equation (9) has a very simple physical interpretation. Aside from changes due to gravitational redshift, time-varying spatial curvature fluctuations, or radiative processes (emission, absorption, and scattering), the CMB anisotropy (brightness temperature fluctuation) is constant along null geodesics. For completeness, we present without derivation the collision terms arising from non­ relativistic photon-electron scattering: � dΔ dτ � c 1 1 = ane σT −Δ + δγ + vei ni + Πij ni nj 3 2 � 7 � (10) where σT is the Thomson cross section and the factor outside the parentheses on the right-hand side is the Thomson scattering rate (up to a factor c = 1). The electrons have proper number density ne and peculiar velocity vei . (The unperturbed electron density may be used when considering primary anisotropies because the terms in parentheses are all first-order perturbations.) The photon number density fluctuation is δ γ ≡ δnγ /n̄γ = 3 δργ /ρ̄γ where nγ and ργ are the photon number density and energy density, respectively. 4 The perturbation can be obtained from an angular average of the anisotropy: δγ = � 3 Δ dΩ/(4π). (Note that nγ ∝ T 3 implies δγ = 3δ ln T .) Heuristically, equation (10) is easy to understand. Scattering corresponds to absorp­ tion (the −Δ term) and re-radiation in a different direction. To the extent that the scattering is isotropic in the electron frame, the emitted radiation is just the angular average of the absorbed radiation ( 13 δγ is the angular average of Δ). Scattering therefore isotropizes the radiation field, a phenomenon familiar to anyone driving a car through fog. The second term is due to the motion of the electron gas: scattering by moving targets introduces a dipole anisotropy. The third term, Πij ni nj , arises from the dependence of Thomson scattering on di­ rection and polarization. It is easy to understand qualitatively from Rayleigh’s Law of scattered light. When a photon scatters from an electron, the incoming and outgoing momenta lie in a plane called the scattering plane. The photon is polarized with its elec­ tric field orientation either in the plane or perpendicular to it. In the former case, the scattered flux is reduced by a factor cos2 θ compared with the perpendicular case where θ is the scattering angle. This result follows classically from the angular dependence of electric dipole radiation caused when the incoming light makes the electron oscillate. Another way to express the same result is to say that the scattering rate is proportional to (��1 ·��2 )2 where �1 and �2 are the polarization directions for the incoming and outgoing photons, respectively. These considerations imply that the differential cross section for unpolarized incident radiation is proportional to 1 + cos2 θ. Normalizing the total cross section, we find 3σT dσ = (1 + cos2 θ) . dΩ 16π (11) If the incident radiation is polarized, then the cross-section becomes a 2×2 matrix giving scattering rates for each orthogonal polarization. Thus, the dependence of the scattering rate on polarization and direction add corrections to the isotropic scattering implied by the −Δ + 31 δγ terms in equation (10). These corrections are only a few percent for the temperature anisotropy but they are crucial for polarization. Scattering induces a small polarization of the CMB. We will ignore the polarization and the quadrupolar scattering term Πij ni nj in the following. Note that the unperturbed (spatially homogeneous) electron density may be used in equation (10) because it multiplies a first-order quantity. Spatial variations in the 8 electron density only appear in second-order perturbation theory. In evaluating the primary CMB anisotropy, we need not take into account that recom­ bination takes place at different times in different places. The reason physically is that the radiation field is in nearly perfect thermal equilibrium with the electrons. For perfect (everywhere homogeneous and isotropic) blackbody radiation, nonrelativistic scattering by an electron gas at rest with respect to the radiation field has no effect whatsoever on the radiation. In the absence of pre-existing CMB fluctuations, scattering alone cannot produce any anisotropy no matter how inhomogeneous the electron distribution. The only possible effect occurs from the motion of the electron gas, and this is included already in equation (10). 3.2 Solution of the radiative transfer equation We can easily integrate equation (9) in the instantaneous recombination approximation, in which we assume that ne drops sharply to zero at τ = τe . Prior to this time the photon mean-free path is very short; we assume it is effectively zero. Under these conditions, integrating equation (9) along the backwards light cone to χe ≡ τ0 − τe gives the desired CMB anisotropy seen by the fundamental observer at χ = 0: Δ0 (ni ) = Δe + φe − φ0 + � χe 0 dχ ∂τ (φ + ψ)ret (12) where subscript “e” means to evaluate the quantity at τ = τe , χ = χe = τ0 − τe , in direction −ni (with a minus sign because the photon travels toward decreasing χ). Subscript “ret” means to evaluate the quantity in the integrand at retarded time τ = τ0 − χ. The first three terms on the right-hand side of equation (12) are all boundary terms which arise as constants of integration. The first two represent brightness temperature and gravitational potential fluctuations present at recombination and are the major contributors to CMB anisotropy. The third term, the gravitational potential at χ = 0, is an unobservable monopole contribution. The integral term is anisotropy produced after recombination due to time-changing gravitational potentials. Before examining the physical content of equation (12) in the next section, it is instructive to compare with the exact solution of the full radiative transfer equation. Including the Thomson scattering terms of equation (10), equation (9) has solution (Seljak & Zaldarriaga 1996) i Δ0 (n ) = � τ0 0 dχ e −τT (χ) � i −n ∂i φ + ∂τ ψ + ane σT where τT (χ) ≡ � χ 0 � 1 1 δγ + vei ni + Πij ni nj 3 2 dχ (ane σT )ret 9 �� (13) ret (14) is the Thomson optical depth (and not a conformal time). Equation (13) is the familiar solution of the radiative transfer equation. Emission (the terms in square brackets) at optical depth τT is reduced by an exponential absorption factor. In contrast with equation (12), there is no explicit indication of the photosphere. The radial integration is taken over the entire past lightcone to τ = 0 (χ = τ0 ). Contributions from χ > χe (which may be defined by τT (χe ) = 13 as in stellar astrophysics) are exponentially suppressed. Equation (13) may be clarified by defining the integral visibility function � ζ(τ ) ≡ exp − � τ0 τ � � � � dτ a(τ )ne (τ )σT = exp [−τT (τ0 − τ )] . (15) This function rises rapidly from zero to unity as τ increases through τe . The conformal time derivative dζ/dτ = ane σT exp(−τT ) is precisely the factor multiplying the scattering terms in equation (13). The spatial derivative term in equation (13) may be converted into a convective derivative through −ni ∂i = −(d/dτ ) + ∂τ . The convective derivative term (but not the partial derivative terms) may be integrated by parts. Using the function ζ(τ ) defined above, we obtain Δ0 (ni ) = 1 1 dχ ζ̇(τ0 − χ) φ + δγ + vei ni + Πij ni nj 3 2 0 � τ0 + dχ ζ(τ0 − χ)∂τ (φ + ψ)ret � τ0 � 0 � ret (16) where we have discarded the unobservable monopole −φ0 . Equation (16) simplifies in the instantaneous recombination approximation, for which ζ(τ ) is a unit step function at τ = τe and ζ̇ = δD (τ − τe ) is a Dirac delta function. In this limit, we recover equation (12) with 1 1 δγ + vei ni + Πij ni nj Δe = 3 2 � � . (17) e Equation (16) also shows us the effects of a finite width to the cosmic photosphere. The primary contributions multiplying ζ̇ are averaged over the finite timespan of recombi­ nation while the effects of the time derivatives of the potentials turn on gradually during recombination. (As a result, the time derivative terms make a significant contribution to anisotropy in models with low Ωh2 , because then recombination occurs only a little after the universe becomes matter-dominated, while the potentials are still changing.) 4 Contributions to Primary Anisotropy In the instantaneous recombination approximation, there are a total of four contributions to primary CMB anisotropy measured by a fundamental observer: intrinsic, Doppler, 10 gravitational, and integrated Sachs-Wolfe. We will discuss them in turn so as to give a physical interpretation to the sources of CMB anisotropy. The intrinsic and Doppler anisotropies correspond to the first term in equation (12), i.e. the brightness temperature perturbation Δe present at the time of recombination. Equation (17) gives this term. However, further simplification results because of the tight coupling of photons and baryons (and electrons) prior to recombination. Under these conditions, vei = vγi . Also, the intrinsic quadrupole anisotropy source term due to polarization effects vanishes because multiple scattering prior to recombination damps the polarization caused by single scattering. This is the reason why CMB polarization is predicted to be so small (Zaldarriaga & Seljak 1997; Spergel & Zaldarriaga 1997). Thus, in the instantaneous recombination approximation, we may write Δe ≈ � 1 δγ + ni vγi 3 � . (18) e This gives the explicit form of the intrinsic ( 13 δγ ) and Doppler (ni vγi ) contributions. (Re­ call that δγ = 43 δργ /ρ̄γ .) It has a simple interpretation which does not require the full treatment of scattering used in deriving equation (16). In the instantaneous recombi­ nation approximation, the photon-baryon gas is a perfect fluid up to the moment of recombination. In the rest frame of this fluid, the radiation is isotropic blackbody radi­ ation at temperature T . Spatial variations of T translate into number density variations δγ = 3δ ln T , explaining the first term. But the photon-baryon fluid moves with a local three-velocity vγi . (Fluids flow!) For small velocities, the linear Doppler formula gives a brightness temperature shift ni vγi . In some directions, the photosphere is moving toward us (relative to unperturbed Hubble expansion) while in other places it recedes, producing anisotropy in the emitted radiation. The next contribution to anisotropy in equation (12) is the gravitational redshift φe − φ0 . Each photon travelling to us falls through a gravitational potential difference φe − φ0 and the energies are changed correspondingly. The literature contains some misstatements about the gravitational redshift effect. The primary anisotropy for isentropic fluctuations on large scales is Δ = 31 φe (Sachs & Wolfe 1967), a factor of 3 different from the gravitational redshift anisotropy to which it is sometimes erroneously ascribed. We will see below that the Sachs-Wolfe result comes from combining the intrinsic anisotropy ( 31 δγ ) and gravitational redshift effects. (Moreover, it is valid only for isentropic fluctuations but not for isocurvature ones. More on that below.) The final contribution to anisotropy comes from the integral term in equations (12) and (16), which is commonly called the “Integrated Sachs-Wolfe” term. If the gravi­ tational potential changes with time, photons suffer different amounts of gravitational redshift falling into and climbing out of potential wells between recombination and today. (This effect on the CMB was first pointed out by Rees & Sciama in 1968.) However, a 11 Newtonian-inspired calculation (Rees & Sciama 1968) would include only the ∂τ φ term because ∂τ ψ is due to space curvature and has no effect on nonrelativistic particles (nor does it have any Newtonian interpretation). But just as the deflection of light in a weak gravitational field is twice the Newtonian-inspired value, if ψ = φ the change in energy due to a time-varying potential is twice the result from ∂τ φ. Note that the integrated Sachs-Wolfe term is the only contribution to anisotropy in the linear regime (i.e. arising from linearized fluctuations) that is produced after recombination. In most models the contribution is small, although there is a measurable effect on large angular scale anisotropy in models with curvature and/or vacuum energy. 5 Evolution of Matter and Metric Perturbations To evaluate the solutions given above (e.g. eqs. 12 plus 18 in the instantaneous re­ combination approximation) we must determine the density, velocity, and gravitational potential fluctuations present at the time of recombination (and later, to get the inte­ grated Sachs-Wolfe contribution). This is non-trivial. Accurate results can be obtained only by integrating numerically two of equations (19)–(23) plus the coupled relativis­ tic perturbation equations for photons, baryons, neutrinos, dark matter, and any other components present in the universe (e.g. Ma & Bertschinger 1995; Seljak & Zaldarriaga 1996). This integration is performed numerically by the CMBFAST code of Seljak & Zaldarriaga as well as by its slower predecessors (e.g. Ma & Bertschinger 1995). Despite the complexity of this evolution, we can learn a great deal by examining the behavior on large scales and at early times in the universe. In this section we will derive the equations of motion for the metric and the coupled fluids responsible for metric perturbations. We will solve these analytically on large scales. This approach gives good physical insight to the various sources of CMB anisotropy. 5.1 Einstein equations The Einstein equations for metric (1) are given by Bertschinger (1996); beware that φ and ψ are reversed there. The 00 Einstein equation give � ȧ ȧ � + 3K ψ − 3 ∂τ ψ + φ = 4πGa2 δρ a a 2 � � � (19) where δρ = ρ−ρ̄(t) is the perturbation in the proper energy density at (xi , τ ) measured by a fundamental observer and K is the spatial curvature constant of a non-flat RobertsonWalker background spacetime. (We reserve the lower-case k for comoving wavenumber below.) Although we will later drop curvature, it is included here for completeness. 12 The 0i component of the Einstein equations helps us to understand the time-derivative terms in equation (19): � � ȧ �2 ∂τ ψ + φ = −4πGa2 �i (ρ + p)v i . (20) a The term in brackets is the momentum density measured by a fundamental observer, or T 0i in an orthonormal basis. In general relativity, momentum density (and momentum flux, i.e. stress) is a source of spacetime curvature. The momentum density, hence its effect on curvature, is typically smaller than δρ by a factor v/c. However, it is gravitationally significant on large scales (Hubble-length and beyond). Thus, the timederivative terms in equation (19) are unimportant on scales much less than the Hubble length but are important on larger scales. They are important for CMB anisotropy on scales larger than a degree or so. With the neglect of gravitomagnetism and gravitational radiation, the metric pertur­ bations follow from spatial scalars and not, for example, a vector potential (Bertschinger 1996). In the limit of small fluctuations, it follows that the velocity fields of the various matter and radiation components are the gradients of scalar fields. That is, the velocity fields have no vorticity (or are longitudinal, in the language of classical field theory). Using this fact, we may write vai = −γ ij �j ua (21) � � for the mean (fluid) velocity of any component a (e.g. photons, baryons, CDM, neutrinos) where u(xi , τ ) is the velocity potential and γ ij is the inverse of the unperturbed spatial 3­ metric γij defined so that the comoving spatial line element is dl 2 = γij dxi dxj in equation (1). This fact allows us to combine equations (19) and (20) into a cosmological Poisson equation, � � � �� ȧ (22) �2 + 3K φ = 4πGa2 δρa + 3 (ρa + pa )ua . a a Aside from the curvature and velocity potential terms, equation (22) is an obvious gen­ eralization of the Newtonian Poisson equation. The source for the metric perturbation must be δρ rather than ρ because φ = ψ = 0 for an unperturbed Robertson-Walker spacetime with ρ = ρ̄. The presence of the velocity potential shows that, in cosmology, momentum as well as mass-energy is a source for Newtonian gravity. This should not be surprising, given that momentum in one Lorentz frame transforms into energy in another frame. It can be shown that a small, spatially-varying transformation of the time coordinate can be performed which eliminates the momentum density (basically by transforming to the local fluid rest frame) and thereby converts the term in brackets to a pure density fluctuation (Bardeen 1980). However, it proves more convenient to use the coordinates implied by equation (1) and simply evaluate both the density and momen­ tum contributions to gravity in these coordinates. (For a discussion of other coordinate systems, see Bardeen 1980 and Bertschinger 1996.) 13 The summed ii components of the Einstein equations give a result that will be useful later: ∂τ2 ψ � ȧ d − Kψ + ∂τ (φ + 2ψ) + 2 a dτ � �2 � ȧ ȧ + a a � � 1 φ + �2 (φ − ψ) = 4πGa2 δp 3 (23) where δp = p − p̄ is the perturbation in the pressure (isotropic stress). Finally, the trace-free part of the ij components of the Einstein equations give a relation between the two gravitational potentials: � 1 �i �j − γij �2 (ψ − φ) = 8πGa2 Σij,⊥ 3 � (24) where Σij,⊥ is the longitudinal shear stress. Longitudinal shear stress arises from quadrupole anisotropy in the angular distribution of the particle momenta; for a weakly imperfect fluid it is the stress arising from shear viscosity. A perfect fluid has no shear stress. The major contribution to Σij,⊥ comes from massless neutrinos; photons make a small con­ tribution after recombination. When the universe becomes matter-dominated, Σ ij � δρ and therefore |ψ − φ| is small compared with φ. In the following we will neglect shear stress, with resulting errors of order one percent in the CMB anisotropy. For a full treatment including shear stress, see Ma & Bertschinger (1995). 5.2 Fluid equations The Einstein equations (19)–(23) require that we specify the density, velocity, and pres­ sure perturbations of matter and radiation. We will approximate the contents of the universe before recombination by two decoupled perfect fluids: cold dark matter and the photon-baryon plasma. Cold dark matter is a pressureless fluid that interacts only by gravity. Photons are tightly coupled to electrons by Thomson scattering, which are themselves tightly coupled to protons and helium atoms and ions by Coulomb scatter­ ing. We apply this description at temperatures below 1 MeV, after electron-positron annihilation and neutrino decoupling. Thus the neutrinos should be regarded as a third, decoupled gas. Their finite temperature and lack of collisions make the neutrinos an imperfect gas. We will discuss the neutrinos below. The fluid equations follow from �µ T µν = 0 applied to a perfect fluid with stressenergy tensor T µν = (ρ + p)V µ V ν + pg µν where V µ is the 4-velocity. We define the three-velocity v i = V i /V 0 . Working to first order in the metric perturbations and the fluid 3-velocity, we find � � ȧ − ∂τ ψ (ρ + p) + �i (ρ + p)v i = 0 , ∂τ ρ + 3 a � � ȧ ∂τ (ρ + p)v i + 4 (ρ + p)v i + �i p + (ρ + p)�i φ = 0 . a � � 14 (25) It is easy to interpret the various terms in equations (25). The terms proportional to ȧ/a are “Hubble damping” terms arising because we are using comoving coordinates. The ∂τ ψ term appears in the continuity equation because, from the metric (1), the homo­ geneous expansion factor a(τ ) is effectively modified by spatial curvature perturbations to become a(1 − ψ). The pressure p is present with ρ in the energy flux (momentum density) because we let ρ be the energy density (not the rest-mass density), which is affected by the work done by pressure forces in compressing the gas. For convenience in what follows, we define the perturbation in energy density relative to the enthalpy density ρ + p, δρ . (26) δ≡ ρ̄ + p¯ Beware, this is a non-standard notation: most authors (including the present one in the papers listed in the bibliography) define δ ≡ δρ/ρ̄. For cold dark matter, with p̄ = 0, there is no difference. However, for photons or other relativistic particles, p/ ¯ ρ¯ is nonzero. Our definition of δ agrees with Weinberg (2002). It has a simple physical interpretation: δ is the fractional perturbation in the number density or particles (rather than the energy density of particles). This choice simplifies the equations and calculations that follow. The unperturbed matter and radiation fluids are perfect fluids, i.e. they have zero shear stress. For these fluids, w = p/ρ depends only on ρ. The speed of sound for a fluid component is cs = (dp/dρ)1/2 . Linearizing equations (25) about the unperturbed solution and using equation (26), we now obtain (cf. Ma & Bertschinger 1995) ∂τ δ = (�2 u + 3∂τ ψ) , ȧ (27) ∂τ u + (1 − 3c2s )u = c2s δ + φ . a We have used the assumption of potential flow, equation (21). Our goal now is to obtain equations of motion for the density and velocity potentials of our two fluids, CDM and the photon-baryon plasma. The CDM case is simple with w = c2s = 0. The photon-baryon plasma is more complicated. First let us consider the perturbations of photons and baryons separately. Because of the large photon-tobaryon ratio (about 2 × 109 ), a negligible amount of heat is transferred to photons from baryons compared with the energy in the photons. Thus, photons with density perturbation δγ obey the first of equations (27) with w = 13 while baryons with density perturbation δb obey the same equation with w = 0 (since they are nonrelativistic). Now, the tight coupling due to Thomson scattering ensures that the photons and baryons have a common velocity potential, ub = uγ . Thus, if the photon-to-baryon ratio was unperturbed initially, it follows from the first of equations (27) that δ b = δγ . With equation of state w = p/ρ̄ ¯ = p̄γ /(ρ¯γ + ρ¯b ), we get c2s dp̄γ /dτ 1 3 = 1 + yb = d(ρ¯b + ρ¯γ )/dτ 3 4 � 15 �−1 , yb ≡ ρ̄b . ρ̄γ (28) Combining all our results, the perturbed fluid equations for CDM (subscript c) and the photon-baryon fluid (subscript γ) are ∂τ δc = �2 uc + 3∂τ ψ , ȧ ∂ τ uc = − uc + φ , a ∂τ δγ = �2 uγ + 3∂τ ψ , � � 9yb ȧ 2 ∂ τ uγ = c s δ γ − uγ + φ . 4 a (29) These agree with equations (3) of Seljak (1994) if neutrino shear stress is neglected so that ψ = φ. They are valid on all length scales provided that the photons are tightly coupled to the baryons. Equations (29) must be supplemented by an equation for the gravitational potential φ. We have a choice of equations (19)–(23). If we evolve the matter and radiation exactly, then all of these components of the Einstein equations are equivalent. We will choose a linear combination involving the effective sound speed squared c2w ≡ 1 3 d¯ p/dτ = 1+ y ρ/dτ d¯ 3 4 � �−1 , y≡ ρ̄b + ρ¯c . ρ̄γ + ρ̄ν (30) Note that cw would be the speed of sound if all matter (CDM and baryons) were tightly coupled to all radiation (photons and neutrinos). However, because this multi-component medium is not a single perfect fluid, cw is not a true sound speed. Nonetheless, it is a useful quantity for obtaining a physical evolution equation for the gravitational potential. We choose a linear combination of equations (19)–(23) that yields the simplest de­ scription of motion, by multiplying equation (19) by c2w and subtracting it from equation (23). Using equation (2), and setting ψ = φ, the resulting perturbed Einstein equation is (Bardeen 1980) ∂τ2 φ + 3(1 + ȧ c2w ) a ∂τ φ + � 3(c2w � �2 ȧ − w) a � 2 − (5 + 3w)K φ − cw �2 φ = 4πGa2 (δp − c2w δρ) ≡ c2w 2 �σ, y (31) where σ is a dimensionless entropy potential that will be discussed below, and w≡ p̄ 1 = . ρ̄ 3(1 + y) (32) In equation (31), δp and δρ are the total pressure and energy density perturbations summed over baryons, cold dark matter, photons, and neutrinos. Let us approximate 16 the behavior of the neutrinos by supposing that they trace photons, with δν = δγ and δpν /ρ̄ν = δpγ /ρ̄γ . This is not correct on scales smaller than the Hubble length, but it is a good approximation on larger scales because the neutrinos have then not yet had enough time to separate from the photons. Our assumption overestimates the neutrino perturbations on small scales because, in the absence of collisions, neutrinos easily dif­ fuse out of perturbations. However, the effects of this error are not large because the gravitational effect of neutrinos is important only during the radiation-dominated era, and during this time the photon perturbations rapidly oscillate on small scales so that their time average is small. Doing better would require that we solve the Boltzmann equation for neutrinos (Ma & Bertschinger 1995), which will not be discussed here. The source term in equation (31) is proportional to the entropy per unit mass of cold dark matter, � � Tγ3 δp − c2w δρ , (33) = δγ − δc = δ ln ρ̄c c2w ρc so that �2 σ = 4πGa2 ρ̄c y (δγ − δc ) . (34) Neutrino effects are included implicitly through the effective sound speed cw . Equation (31) is rather remarkable. It allows us to compute the gravitational po­ tential through a damped, driven wave equation whose source is the specific entropy perturbation. In Newtonian gravitation we think of the potential as being determined by action-at-a-distance (eq. 19 without the time-derivative terms). However, the Ein­ stein equations also enforce local energy-momentum conservation, providing us with alternatives for the computation of the metric perturbations. Heuristically, the potential φ is computed from the instantaneous distribution of energy density (and momentum density, as may be shown by combining eqs. 19 and 20). Because the source of the potential may have acoustic waves, then it, too, displays acoustic waves. Note that the wave speed is the effective sound speed cw . True gravitational waves (which propagate at c) are described by a transverse-traceless metric perturbation hij , which we are ignoring here. Thus, despite its appearance, equa­ tion (31) really is gravitational action-at-a-distance. Although gravitational waves are needed to restore causality, we are ignoring them here for the CMB anisotropy calcula­ tion. This is justified in linear perturbation theory by the fact that gravity waves have no coupling to the “scalar-mode” (density and entropy) perturbations we are consider­ ing here. In linear perturbation theory, the effects of density and entropy perturbations propagate at the speed of sound, not the speed of light. Our next step is to obtain an evolution equation for the entropy potential σ. Using the fact that a2 ρ̄c y is constant, from equations (34) together with equations (28)–(30) we obtain 17 � � ȧ ∂τ σ − (c2s − c2w )�2 σ a � � � � �� ȧ ȧ 3 c2w 2 2 2 ρ̄c δγ + 4 uγ . = 4πGa y(cs − cw ) ρ̄c δc + 3 uc + 2 a 4 cs − c2w a ∂τ2 σ + 1 + 3c2w − 3cs2 (35) Using equations (28) and (30), we obtain 3 c2w 3 ρ̄ν ρ̄c = ρ̄γ + ρ¯ν + ρ̄b 1 + 2 2 4 cs − c w 4 ρ̄γ � � �� � ρ̄b ρ̄ν 1− ρ̄c ρ̄γ �−1 . (36) Using this result, if we compare the right-hand sides of equations (22) and (35), we see that they would be identical for our multicomponent fluid if we neglect neutrinos. Including neutrinos requires approximation because we are not solving the Boltzmann equation for neutrinos. One approximation is to suppose that neutrinos have identical density and velocity perturbations as photons. Under this assumption, ργ would simply be replaced by ργ + ρν in equation (22) and similarly for pγ . This assumption was made in deriving equations (33) and (35). However, we know that this approximation is not exact, and so we are free to consider other approximations that make errors of the same order. That is, we can consider corrections to equation (36) that are of order ρ̄ ν . We choose to approximate the right-hand side of equation (36) by dropping the factors in parentheses (Bashinsky & Bertschinger 2002). This approximation is exact when neutrinos are neglected. The reason for making this choice is that now equations (35) becomes a simple wave equation that is very similar to equation (31), � ∂τ2 σ + 1 + 3c2w − 3cs2 � ȧ a � � 2 ∂τ σ − (c2s − cw )�2 σ = y(cs2 − c2w ) �2 + 3K φ . (37) This equation is exact only when neutrinos are neglected. However, as we will see in Section 7, when neutrinos are included it preserves the correct propagation of sound waves through the photon-baryon fluid at sound speed cs . The two-fluid approximation introduced here is equivalent to including neutrinos assuming that they have the same dynamics as photons. The baryon abundance is increased over the true abundance by a factor (1 + ρ̄ν /ρ̄γ ) so that the photon-baryon sound speed is unchanged. Finally, the CDM abundance is decreased so that the total matter density ρ̄c + ρ̄b is unchanged. This� model has the advantage of preserving the important time and length scales τeq and cs dτ while working with perfect fluids. For more discussion of the two-fluid approximations, see Bashinsky & Bertschinger (2002). Equations (31) and (37) completely characterize the evolution of matter, radiation, and gravity in our simplified two-fluid model (CDM and the photon-baryon fluid, with neutrinos added in a way that preserves the photon-baryon sound speed). Because 18 of the modification made in deriving equation (37), these equations are not exactly equivalent to the fluid equations (29) plus the Poisson equation (22) when neutrinos are included. They agree with the approximations made by Seljak (1994) and Bashinsky & Bertschinger (2001, 2002). We will find in Section 9 that the two-fluid approximation made here is remarkably accurate. 6 Large-scale anisotropy: the Sachs-Wolfe Effect Using the wave equations we have derived for the gravitational potential φ and entropy potential σ, it is possible analytically to solve for the perturbations on large scales. In this context, “large scale” means that we can neglect c2w �2 and c2s �2 relative to (ȧ/a)2 in equations (31) and (37). This is valid provided that we consider scales larger than the comoving acoustic length, defined by Lac ≡ cs . aH (38) The acoustic length subtends an angle of about one degree at recombination. We also will neglect the spatial curvature K, which makes a negligible contribution to equations (31) and (37) before and during recombination because |K | � (ȧ/a)2 . Under these conditions, equations (31) and (37) reduce to ordinary differential equa­ tions in time. We change variables from τ to y through equation (3), obtaining yφ�� + and � � 3c2w φ (cw τe )2 2 yφ� �σ + 3 1 + c2w φ� + = 2(1 + y) 4(1 + y) (1 + y) � � (c2s − c2y )τe2 2 2 yσ � 2 2 � yσ + + 1 + 3cw − 3cs σ = y �φ, 2(1 + y) 1+y �� (39) (40) where a prime denotes ∂y . We have retained the Laplacian terms that link the potential and density because we have said nothing yet about the relative sizes of φ and σ. As y → 0, the gravitational potential makes a negligible contribution to σ in equation (40), at least insofar as that contribution feeds back to equation (39). Thus, in solving for the gravitational potential it is safe to drop the right-hand side of equation (40), which then becomes an ordinary differential equation in y whose solutions as y → 0 are σ ∝ ln y and σ = σ(x) with ∂y σ = 0. The logarithmic solution is unphysical; it corresponds to a large separation between photons and CDM at the big bang, which does not occur in standard cosmology. Thus, the only possibility for the entropy is to have a constant or spatially-varying distribution of primeval entropy fluctuations that is constant in time early in the radiation era. 19 Given the time-independent solution for the entropy σ, it is easy to solve equation (39) for φ(y). The homogeneous (σ = 0) solutions for φ(y) are (Kodama & Sasaki 1984, 1986) √ √ 1+y 2 8 16 16 1 + y − 2− 3+ , φ2 (y) = . (41) φ1 (y) = 1 + 3 y y3 9y 9y 9y 9 1 For y � 1, φ1 = 10 (1 − 16 y + · · ·). The density perturbations associated with φ1 grow in 9 the matter-dominated era y � 1 (hence φ1 is often called the growing mode) although φ1 itself does not. The other mode, φ2 , is called the decaying mode. From the these homogeneous solutions, we can construct particular solutions subject to any desired initial conditions. Using the Green’s function method we can also construct solutions with nonzero �2 σ. Before presenting the solutions, we recall from equation (13) that the CMB anisotropy computation requires δγ and vγi in addition to φ. On large scales, where pressure-gradient forces are unimportant so that dark matter and photons move together, we can deter­ mine these quantities from the Einstein equations without having to solve the coupled equations of motion for all the matter and radiation species. This is because, on scales larger than the acoustic length, the matter plus radiation behaves as a single fluctu­ ating perfect fluid. The Einstein equations have built-in redundancy — they enforce local energy-momentum conservation — so we can use them to get the net energy den­ sity and velocity perturbations without having to solve the fluid equations. (That is a radical concept to a Newtonian physicist, but it is a direct consequence of the Einstein equations!) From equation (19), dropping the curvature and Laplacian terms (they are both small on scales larger than the Hubble length) and neglecting anisotropic stress (i.e. setting φ = ψ), we obtain δρ = −2 (a∂a φ + φ) = −2∂y (yφ) . (42) ρ̄ Note that we are using a gravitational field equation to solve for the mass density pertur­ bation, having used time-evolution equations to solve for the potential. By contrast, in Newtonian physics we are used to solving time evolution equations for the fluid variables and then solving a gravitational field equation for the potential. In general relativity, we have more flexibility in how to compute gravity. Equation (42) has a simple physical interpretation. By assumption, the density is smoothly varying on scales of the Hubble length. Thus, each Hubble volume behaves like a homogeneous universe (slightly closed or slightly open depending on the sign of δρ). In a homogeneous model, the density and cosmic proper time t measured by a fundamental observer are related by ρ(x, τ ) ∝ (Gt2 )−1 . Because of the long-wavelength perturbations in g00 , however, the relative clock rate between proper time t(x, τ ) and conformal time depends on position: dt/dτ = a(τ )(1 + φ). Thus, we may write t = t0 (τ ) + Δt(x, τ ) 20 � � where t0 (τ ) = a dτ and Δt = φa dτ . The density perturbation is computed at fixed τ using δρ/ρ̄ = (d ln ρ/dt)Δt = −2(Δt)/t0 . If ∂τ φ = 0, then Δt = φt0 giving δρ/ρ̄ = −2φ. This is simply the gravitational redshift: clocks run slower in a gravitational potential well. If ∂τ φ �= 0, the gravitational redshift is time-dependent and a more careful calculation is needed. Matters are complicated because of the particular, somewhat arbitrary choice of constant-τ hypersurfaces implied by our metric equation (1). Our intuition is not always a good guide for physics on scales larger than the Hubble length. Fortunately, the Einstein equations will guide us to the correct solution even if we cannot guess it from Newtonian considerations. The photon density perturbation on large scales follows simply from combining equa­ tion (42) with equation (34). (One must combine the photon and matter perturbations with appropriate weights to obtain the net density perturbation.) The result is � 9 � δγ = c2w −2(1 + y)∂y (yφ) + (2y/3)τe2 �2 σ . 4 (43) Similarly, the large-scale peculiar velocity follows from equation (20). For a matter plus radiation universe on scales much larger than the curvature and acoustic scales, uc = uγ . The velocity potential is then √ 1 y 1+y uγ = τe ∂y (yφ) . (44) 2 1 + (3/4)y 6.1 Isentropic and Isocurvature modes We now have all the ingredients needed to compute the CMB anisotropy on large angular scales. The treatment here is equivalent to that of Sachs & Wolfe (1967) for isentropic initial fluctuations and extends their treatment to more general initial conditions. Equation (39) has two physically relevant quantities: the gravitational potential φ and specific entropy perturbation σ. Because the equation is second-order, one might think there are two physical solutions for any σ, and therefore a vast family of solutions. However, we have seen that σ cannot change with time over distance scales larger than the acoustic length. Moreover, there is only one combination of the homogeneous solutions of equations (41) that remains finite as y → 0. Consequently, physical initial conditions set in the very early universe must be a linear combination of the finite solution with �2 σ = 0 but φ = � 0 and of the finite solution with �2 σ = � 0 but φ = 0. Because the evolution equations are all linear in the perturbations, then general solution is a linear combination of the solutions starting from these two types of initial conditions. The two sets of solutions we have identified are called isentropic and isocurvature. In both cases we assume that the fluctuations are produced in the very early universe. 21 This assumption is consistent with inflation but not with some other models of structure formation where the density fluctuations are generated much later. Isentropic fluctuations are defined by the condition σ = 0 but φ �= 0. They cor­ respond to the energy density of all components (photons, baryons, dark matter, etc) varying together in such a way that the entropy perturbation vanishes. In particular, the photon-to-baryon number density ratio is a universal constant in this model (until after recombination, when photons and baryons decouple), as is the initial photon-todark matter particle number density ratio. This it exactly the condition resulting from standard inflation plus reheating. Inflationary fluctuations are often called “adiabatic” although isentropic (i.e. σ = 0 initially) is more accurate and descriptive. Isocurvature fluctuations have φ = 0 initially (i.e. no curvature perturbation) but σ �= 0. Physically, they correspond to an equation of state p(ρ, S) with spatially-varying entropy S but initially constant ρ. Isocurvature fluctuations can be created by a firstorder phase-transition or any other process that produces spatial variations in the photon/baryon or photon-to-dark matter ratio while leaving the net energy density unper­ turbed. As the universe expands, as long as particle collisions are rapid, the specific entropy of each fluid element is conserved. That is, prior to decoupling for any particle species, the ratio of number densities of that particle to photons is conserved. If the ratio varies spatially, then pressure gradients develop which move matter and cause a gravitational potential perturbation to develop. Nonstandard models of inflation can be constructed in which entropy perturbations are generated during or after reheating. In our particular case we are considering “isocurvature CDM” perturbations, since σ �= 0 corresponds to a spatially varying photon-to-CDM ratio in equation (33). It is also possible to consider models with a fixed Tγ3 /ρc but with spatially varying Tγ3 /ρb , or “isocurvature baryon” models. Indeed, isocurvature-type perturbations can be produced that have the ratio of any two or more species varying in space initially, as long as the total energy density is constant so that the metric is unperturbed (φ = 0). It is easy to recognize the isentropic and isocurvature “modes” (as they are often called in the literature) as homogeneous and particular solutions of equation (39). We impose initial conditions at τ = τi such that yi � 1. Isentropic initial conditions have φ(x, τi ) = φi (x) and σ(x, τi ) = ∂τ φ(x, τi ) = 0. (For example, the fluctuations produced by inflation are frozen in and unchanging during the radiation-dominated era on scales larger than the acoustic length.) Isocurvature initial conditions have φ(x, τi ) = ∂τ φ(x, τi ) = 0 and σ(x, τi ) = σi (x), with ∂τ σ = 0 on scales larger than the acoustic length. We leave it as an exercise for the student to show that the desired solutions of equation (39) are � 9 φ1 (y)φi (x) , isentropic ; (45) φ(x, τ ) = 10 2y 2 2 φ (y)τ � σ (x) , isocurvature . i e 5(4+3y) 1 (The φ2 solution is not included here because it is the decaying mode. Its amplitude 22 must be negligible at late times because φ2 diverges as y → 0.) In the matter-dominated era, y � 1, the general solution including both isentropic and isocurvature modes is a constant potential, 9 1 2 2 2 φ(x, τ ) ≈ φi (x) + τ � σi 10 5 3 e � � for y � 1 . (46) The entropy term in parentheses is simply the relative variation of the photon-to-CDM ratio from equation (34). The relative sizes of the coefficients in equation (46) (0.9 vs 0.2) shows how inefficient entropy is in producing a gravitational potential fluctuation. This will cause the isocurvature mode to have a much larger CMB anisotropy than the isentropic mode, for a given post-recombination gravitational potential. The Sachs-Wolfe anisotropy follows from substituting into equations (12) and (18) the potential, density and velocity perturbations computed on large scales assuming that recombination occurs at y � 1. First, from equations (18), (43), (44), and (46), we get Δe = � −(2/3)φe − (2/3ηe )ni �i φe , isentropic ; φe − (2/3ηe )ni �i φe , isocurvature , (47) where ηe ≡ (ȧ/a)e and subscript e refers to recombination. Note that the velocity term becomes small for comoving wavelengths much larger than the Hubble distance ηe−1 at recombination. Note also the critical difference between isentropic and isocurvature initial conditions as far as the post-recombination potential. Combining equation (12 with (47), we get the final Sachs-Wolfe effect for the anisotropy in direction (θ, ϕ): � � � χe 2 ∂ dχ ∂τ φ(χ, θ, ϕ, τ0 − χ) , Δ(θ, ϕ) = C + φ(χe , θ, ϕ, τe ) + 2 3ηe ∂χ 0 (48) where C = 1/3 for isentropic fluctuations and C = 2 for isentropic fluctuations. Note that equation (48) includes only the gravitational potential at and after recombination — by ignoring all acoustic effects (and neutrinos) we have simplified the evolution sufficiently to express the primary anisotropy entirely in terms of the potential and its derivatives for τ ≥ τe . To the extent that the first two terms (intrinsic and Doppler anisotropy) dominate the anisotropy, maps of the CMB anisotropy reveal (on scales larger than the acoustic horizon) the gravitational potential at recombination. The coefficient C is six times larger for isocurvature fluctuations than for isentropic ones. (This is, of course, an approximation — the solution assumed zrec � zeq which is not exact.) For a given amplitude of gravitational potential fluctuations, hence dark mat­ ter fluctuations (dark matter being the dominant component at recombination), isocur­ vature fluctuations produce a much larger CMB anisotropy. This is because the initial entropy fluctuations directly perturb the radiation with no compensating time dilation 23 effect. For isentropic fluctuations, we saw that the large-scale photon density perturba­ tion has opposite sign to the gravitational potential (because of the time dilation effect discussed following eq. 42), which results in a cancellation of the CMB anisotropy. Hence the famous Sachs-Wolfe formula Δ = 31 φ. Although Sachs & Wolfe (1967) did not con­ sider isocurvature initial conditions, we have seen that it is just as easy to estimate their contribution to large-scale anisotropy. The much larger ratio of CMB fluctuations to dark matter perturbations has led to a rejection of most isocurvature models. 7 Green’s function solution for small-scale anisotropy As we have seen, the coupled Einstein and fluid equations can be solved analytically on large scales, with the result that the temperature anisotropy basically traces the gravitational potential fluctuations on the CMB photosphere. However, this is not the full story for CMB anisotropy. On smaller scales, the potential fluctuations set the photonbaryon fluid into motion, sending sound waves propagating through the universe. These sound waves leave a characteristic signature in the correlations of CMB anisotropies on angular scales of about one degree. This signature is reflected in the acoustic peaks of the angular power spectrum. On scales smaller than the acoustic length, it is hard to solve the coupled Einstein and fluid equations for a multicomponent medium. The usual procedure is to work in Fourier space so that the partial differential equations become ordinary differential equations in time. Seljak (1994) solved equations (29) and (20) numerically in Fourier space. Ma & Bertschinger (1995) solved the complete coupled system including the Boltzmann equa­ tion for photons and neutrinos, which are necessary for imperfect fluids or collisionless plasmas. Seljak & Zaldarriaga (1996) introduced CMBFAST, a fast computer code for solving the complicated system of equations. These developments are valuable, especially CMBFAST because it has revolutionized CMB data analysis and model testing. Here, however, we seek physical insight more than numerical results. The approach is based on work by Sergei Bashinsky (Bashinsky 2001, Bashinsky & Bertschinger 2001, 2002). Instead of solving the fluid and Einstein equations in Fourier space, we solve the coupled wave equations (31) and (37) in configuration space (x, τ ). Our model for CMB anisotropy is based on two fluids coupled by gravity: the photonbaryon fluid (with sound speed cs ) and CDM (with vanishing sound speed). On physical grounds, we would expect waves to propagate through the photon-baryon fluid with speed cs but no waves to propagate through the CDM. It is intriguing that equations (31) and (37) have almost the expected form, namely a coupled system of damped wave equations with the appropriate sound speeds. However, the fundamental variables appearing there are the gravitational and entropy potentials, which do not cleanly separate the two fluids. If we ignore curvature by setting K = 0 (as is justified at high redshift), then the 24 Laplacian operators in equations (31) and (37) both operate on the same quantity, namely φ + σ/y. This suggests that we consider linear combinations of φ and σ to separate the wave components. Amazingly, there is a simple set of variables that does the trick: cw φr = cs � �2 � σ φ+ y � , φc = φ − φ r . (49) We use the subscripts r for radiation (the photon-baryon fluid, with an approximate inclusion of neutrinos) and c for CDM. The reason for doing so is that the Poisson equation now holds separately for each potential with a corresponding source (for K = 0): ȧ �2 φr = 4πGa2 ρ̄r δγ + 3 u¯ , a � � ȧ �2 φc = 4πGa2 ρ̄c δc + 3 u¯ , a � where u¯ ≡ ρ̄r uγ + ρ̄c uc , ρ̄r + ρ̄c � 4 ρ̄r ≡ (ρ¯γ + ρ¯ν ) + ρ¯b . 3 (50) (51) Thus, we may interpret φr as the gravitational potential produced by “radiation” (pho­ tons, neutrinos, and baryons) while φc is the gravitational potential produced by CDM. (One must be careful not to push the Newtonian analogy too far, since here we have contributions from the peculiar velocity potentials.) Equations (50) are equivalent to equation (22) applied separately to each of our two fluids. Equations (31) and (37) now separate into a pair of coupled wave equations with the expected behavior: 2 ȧ 3(c2w + ϕ) 2c−2 ȧ s − 4 + 3cw φ − φc = c2s �2 φr , ∂τ2 φr + 3(1 + c2s ) ∂τ φr + 3c2w ∂τ φc + r a a 4τe2 y 3τe2 y 2 2c−2 − 3 ȧ ȧ 3ϕ (52) ∂τ2 φc + 3(c2w − cs2 ) ∂τ φr + 3 ∂τ φc − 2 φr + s 2 2 φc = 0 , a a 4τe y 3τe y where ϕ(y) is a ratio of two quartic polynomials given by Bashinsky & Bertschinger (2002). The first of equations (52) shows that perturbations in the photon-baryon fluid indeed propagate as sound waves with speed cs . As the sound wave passes by a CDM fluid element, its gravity acts on the CDM, causing the CDM to evolve as given by the second equation. 25 7.1 Radiation era Equations (52) are still too complicated to solve analytically. To gain insight, let us examine the solutions in the radiation era, y � 1. In this limit, equations (52) become ∂τ2 φr + 1 ∂τ (4φr + φc ) − τ 3 ∂τ2 φc + ∂τ φc + τ 1 φc = c2s �2 φr , τ2 � � 1 3 1 4 (y − yb ) φc = ∂τ φ r + 2 φ r . τ2 4 τ τ (53) We will solve these for isentropic initial conditions, σ = 0 as y → 0. From equations (49), φc /φr ∝ y. Thus, CDM has negligible effect on the dominant radiation component in the radiation era. On the other hand, the radiation perturbations provide the source for φc . Thus, we must solve the simple damped wave equation ∂τ2 φr + 4 ∂τ φr = c2s �2 φr τ (54) with constant sound speed cs = 3−1/2 . Since this is a partial differential equation, we must specify the initial conditions for all space. However, the problem simplifies because it is a linear partial differential equation. The linear superposition principle means that we can expand the solution in any convenient set of spatial basis functions. The common choice in cosmology for flat (Euclidean) space is the plane wave basis exp(i�k · �x ), so that �2 = −k 2 for any given plane wave component. We use the usual notation of flat space in which �k and �x are 3-vectors. The Fourier space solution to equation (54) is φr (k, τ ) = 3 j1 (kτ ) , kτ (55) where j1 (x) = (sin x − x cos x)/x2 is a spherical Bessel function. Note that the normal­ ization chosen so that φr = 1 for τ = 0. Suppose that the initial potential fluctuations at the end of the inflationary era are φi (�x ). The Fourier transform is φi (�k ). (Note the common practice of using the same function name to denote the real-space function and its Fourier transform. The meaning becomes clear from the argument: �x is always a spatial position and �k is always a wavevector.) Linear superposition means that the Fourier transform at any later time is φr (�k, τ ) = φr (k, τ )φi (�k ) (56) By analogy with linear filters in electrical systems, φr (k, τ ) is called the transfer function for the radiation potential. It is the factor by which every Fourier component of φ r (�k, τ ) 26 changes through linear evolution. Every linear field has a transfer function (e.g. φ c , δγ , etc.). The transfer function is spherically symmetric in �k-space because of the rotational invariance of the equations of motion. Note carefully the notation: φr (k, τ ) is the transfer function, i.e. the solution subject to initial condition φr = 1, while φr (�k, τ ) is the solution subject to the initial condition φr = φi (�k ). The point is that once the transfer function is known, then the solution for arbitrary initial condition follows by simple multiplication. The transfer function is the solution to the evolution equation with initial condition exp(i�k · �x ). Plane wave initial conditions are but one possible choice of initial condition. Another useful choice is a point-like perturbation at the origin, (3) φ(3) x, τ ) → δD (�x ) as τ → 0 r (� (57) (3) where δD (�x ) is the three-dimensional Dirac delta function. The superscript (3) is placed on φr to remind us that the initial condition is point-like in three dimensions. The solution subject to a point-like initial condition is called a Green’s function. The solution at any later time is then simply a convolution of the initial field by the Green’s function: � φr (�x, τ ) = d3 x� φr (�x � , 0)φ(3) x − �x � , τ ) . (58) r (� Heuristically, we can imagine decomposing the initial field into a sum of delta functions. Each delta function evolves into the Green’s function. The evolved field is therefore a superposition of Green’s functions. The Green’s function description is mathematically equivalent to the Fourier space description with a transfer function. In fact, the Green’s function is simply the inverse Fourier transform of the transfer function: φ(k, τ ) = � � φ(3) (r, τ ) e−ik·�x d3 x . (59) (We remove the subscript r since this relation is not restricted to the radiation potential but is true for any transfer function and Green’s function pair.) Note that since the transfer function is spherically symmetric in k-space, the Green’s function is also spher­ ically symmetric in position space. That is, φ(3) depends only on the length r = |�x |. (Assuming a flat cosmology, we use r interchangeably with the radial comoving coor­ dinate χ.) The spherical Green’s function and transfer function are Fourier transform pairs. In the Fourier domain, the evolved field follows by multiplication with the transfer function. In the position domain, the evolved field follows by convolution with the Green’s function. The Fourier transform of a convolution is the product of Fourier transforms, and the Fourier transform of a product becomes a convolution. So what is the Green’s function corresponding to the radiation transfer function φr (k, τ )? It is easily found either by Fourier transformation of equation (55) or by 27 solution of equation (54) in spherical coordinates with delta function initial condition. The result is 3 (cs τ )−3 θ(cs τ − r) φ(3) (60) r (r, τ ) = 4π where θ(x) is the Heaviside step function, θ(x) = 0 for x < 1 and θ(x) = 1 for x > 1. The Green’s function for the radiation potential is remarkably simple: it is a spatially uniform wave expanding at the speed of sound, with unit volume integral (from eq. 59 for k = 0). In short, it is a spherical sound wave! The oscillations of the transfer function 3j1 (kτ )/(kτ ) arise because of the finite extent of the Green’s function and they are characteristic of causal behavior. This phenomenon is essentially the origin of the acoustic peaks of the CMB power spectrum, which we will discuss below. This beautifully simple result suggests that we apply the Green’s function method to the full problem of equations (52) for all times. However, there is a practical problem with the three-dimensional Green’s functions. The Poisson equation tells us that the photon density fluctuation δγ is given, in part, by �2 φ(3) r . The derivative of the Heaviside function is a Dirac delta function. The Laplacian takes one more derivative, so that δγ(3) (r, τ ) has a term proportional to dδD (r − cs τ )/dr. Great care must be exercised when dealing with derivatives of delta functions, especially when φ(3) r is determined numerically. The numerical challenges are serious enough to dissuade us from using three-dimensional Green’s functions. These delta function singularities are a consequence of the perfect fluid assumption. If photon diffusion is included, then the acoustic wavefront will spread slightly. Even with diffusion, however, the spherical transfer function for δγ is sharply peaked at the radius of the acoustic sphere. This means that an initial point-like perturbation in the potential, located on or near the CMB photosphere, will produce a ring of temperature anisotropy on the sky. The observed pattern is a superposition of such rings like the ripples from stones dropped in a pond. The maximum size of the rings — twice the acoustic radius at recombination — imprints a characteristic scale in the CMB anisotropy. 7.2 Plane-parallel Green’s functions For numerical integration purposes it is preferable to modify the Green’s function method to avoid derivatives of delta functions. The singular behavior is reduced if we superpose many spherical waves to create a plane wavefront using the Huyghens’ construction. This works for any Green’s function as follows: φ(1) (x, τ ) = � ∞ |x| φ(3) (r, τ ) 2πrdr . (61) The integral ensures that φ(1) is less singular than φ(3) at the wavefronts. From equation (59) it follows that φ(1) is the one-dimensional inverse Fourier transform of the transfer 28 function, φ(k, τ ) = � ∞ −∞ φ(1) (x, τ ) e−ikx d3 x (62) where r = |�x |. The transfer function contains the same information as either its one­ dimensional or three-dimensional Fourier transform. Thus, the choice of approach is purely a matter of convenience, numerical ease or accuracy, and/or ease of interpretation. Recalling the initial condition φ(k, 0) = 1 for the transfer function, from equation (62) it follows that φ(1) (x, τ ) is the Green’s function for initial condition φ(�x, 0) = δD (x). The initial perturbation is sheet-like (i.e., constant on planes of constant Cartesian coordinate x) rather than point-like. We use the superscript (1) to distinguish such plane-parallel Green’s functions from the spherical variety. Analytic solutions for the plane-parallel Green’s functions in the radiation era are given by Bashinsky & Bertschinger (2002). Because the dynamics involve two potentials (for gravity and entropy), initial conditions must be set for both. The calculations have all assumed isentropic initial conditions with σ(x, τ ) = 0 at τ = 0. It would be interesting to work out Green’s functions for isocuravture initial conditions with φ(x, 0) = 0 and σ(x, 0) ∝ 1/r (so that the entropy perturbation �2 σ is a delta function) but this has not been done. Bashinsky (2001) constructed numerical solutions for the isentropic case by evolv­ ing equations (52) in one space dimension and time, with �2 = ∂ 2 /∂x2 . Usually it is much harder to accurately solve partial differential equations than ordinary differential equations, and this is an argument for the Fourier space approach. However, the equa­ tions in question are linear wave equations and they may be solved by the method of characteristics. This method is fast and accurate. Figure 1 shows the results of this numerical integration for the radiation and CDM gravitational potentials at recombination. The initial conditions were φ (1) r (x, 0) = δD (x) (1) and φc (x, 0) = 0. The delta function separated into left-going and right-going waves, whose evolution spread φ(1) over space and diminished the central peak. The gravita­ r tional potential hill of the radiation caused outward-directed gravitational forces which expelled the CDM away from x = 0. The CDM has a central cusp reflecting the initial repulsive singularity in the gravitational potential φ(1) r ; this cusp is preserved because the CDM particles have no thermal motion. In fact, once the universe becomes matterdominated φ(1) c (x, τ ) stops evolving, as may be seen from the second of equations (52) in the limit y � 1. Figure 2 shows three snapshots of the time evolution of the Green’s functions for the radiation and CDM potentials. Only the range x > 0 needs to be plotted since the Green’s functions depend only on |x|. Note that while the amplitude of φr decreases rapidly with time, φc changes much more slowly and becomes constant in the matter era as expected. We also see the effect of changing the baryon content. This changes the photon-baryon sound speed in equation (28), thereby changing the distance to the 29 (1) (1) φr (x) and φc (x) 2 1.5 1 0.5 0 −0.6 −0.4 −0.2 0 x / cτ 0.2 0.4 0.6 rec Figure 1: Plane-parallel Green’s functions for the potentials φr (solid) and φc (dashed) at recombination, in the fluid approximation. The cosmological parameter values are Ωm = 0.35, ΩΛ = 0.65, Ωb h2 = 0.02, and h = 0.65. By recombination the universe has become matter-dominated. From Bashinsky (2001). wavefront. Adding baryons decreases the sound speed. The density and velocity perturbations of photons, baryons, and the CDM fluid follow from φr , φc and their derivatives. The photon perturbations contribute directly to CMB anisotropy while the CDM perturbations eventually seed the formation of galaxies. Figure 3 shows the plane-parallel Green’s functions at recombination for the photon and CDM density fluctuations. (Recall that δb = δγ in the fluid approximation.) The singularities of δγ and δc arise from the d2 φ/dx2 term of the Poisson equation. Comparing with Figure 1, we see why these singularities occur at the wavefronts for δ γ but at x = 0 for CDM. The central spike for the CDM is negative because of the repulsive sign of the initial gravitational potential peak (top row of Fig. 2). It is surrounded by positive tails because of mass conservation: the CDM pushed out from x = 0 piles up into the region between x = 0 and the acoustic wavefront. Note from Figures 1–3 that our spacetime is unperturbed outside of the acoustic wavefront. This is a consequence of our initial condition that left the metric unperturbed everywhere except at x = 0. Had we instead made δγ a Dirac delta function with zero velocity perturbation, then the gravitational potential φ would have been nonzero for all x. However, our treatment in the earlier sections shows that the gravitational potential, not the density, is the fundamental physical quantity for CMB anisotropy. That is why our Green’s functions are chosen so that φ(x, 0) = δD (x). The Dirac delta function contributions of δγ (x, τ ) at the acoustic wavefronts make 30 φ(1) r φ(1) c 0.1 τe 15 3 10 2 5 1 0 0 0.2 0.4 0 0.6 3 1 2 0.5 1 τe 1.5 0 0 0.2 0.4 0 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0 0.2 0.4 0.6 3 0.4 τrec 2 0.2 0 1 0 0.2 x / cτ 0.4 0 0.6 rec x / cτ rec Figure 2: Time evolution of the plane-parallel Green’s functions for the potentials φ r (left) and φc (right) in the fluid approximation. The solid (dashed) lines show the results for Ωb h2 = 0.03 (Ωb h2 = 0). The three rows give snapshots √ at the times (0.1τe , τe , and τrec = 2.5τe ). The speed of sound is decreased below c/ 3 by the inclusion of baryons. From Bashinsky (2001). a significant contribution to the total CMB anisotropy. So does the dip at x = 0. To understand the physics of this dip, let us examine the combination of photon density and gravitational potential appearing in the Sachs-Wolfe effect on large scales, which is given by equation (48) without the Doppler and ∂τ φ (integrated Sachs-Wolfe) contributions: Δeff ≡ 1 δγ + φ . 3 (63) (Recall that δγ is defined here as the fractional perturbation in photon number density.) The plane-parallel Green’s function for Δeff at recombination is shown in Figure 4. Note that the relation of Δeff to potential on small scales in position space is quite different from the Δeff ≈ 31 φ of the large-scale Sachs-Wolfe effect. However, this relation still holds if one integrates the Green’s functions over x to get the long wavelength contributions. The positive Dirac delta functions more than compensate for the negative δγ (x) behind � � the acoustic wavefronts so that Δeff (x) dx ≈ 13 (φr + φc ) dx. 31 20 5 10 δ(1)(x) 0 c r δ(1)(x) 10 0 −5 −10 −10 −20 −0.5 0 x / cτ 0.5 −0.5 0 x / cτ rec 0.5 rec Figure 3: Plane-parallel Green’s functions for the density fluctuations of photons (δ r = δγ , solid) and CDM (dashed) at recombination, in the fluid approximation, for the same parameters shown in Fig. 1. Thick vertical lines represent Dirac delta functions of δ γ (x) at its acoustic wavefronts and of δc (x) at the origin. From Bashinsky (2001). Figure 4 suggests that baryons are responsible for the central dip in Δeff . This dip is easy to understand. From Figure 1, the total gravitational potential has a sharp peak at x = 0. The photon-baryon fluid has a high sound speed and therefore quickly adjusts to this potential by settling into hydrostatic equilibrium near x = 0. (Obviously hydrostatic equilibrium does not apply at the acoustic wavefronts, but it does hold as x → 0 where the elapsed time is many sound-crossing times.) From the second of equations (25) we obtain the equation of hydrostatic equilibrium, �i ρ + (ρ + p)�i φ = 0 . (64) Linearizing this for the photon-baryon fluid gives �i � � 1 3 ρ̄b δγ + 1 + 4 4 ρ̄γ � � � φ = �i Δeff � 3 ρ̄b + φ =0. 4 ρ̄γ (65) Thus, if ρb = 0, Δeff has zero gradient in hydrostatic equilibrium. If Ωb �= 0, on the other hand, Δeff (0, τ ) = − 34 (ρ̄b /ρ̄γ )φ + constant as x → 0. The positive cusp of the CDM potential becomes a negative cusp of CMB anisotropy whose amplitude is proportional to Ωb . Thus, baryons have two main effects on CMB anisotropy: they slow the sound speed and thereby decrease the acoustic distance compared with a pure photon gas, and they create a central dip in the Green’s function. 8 Angular Power Spectrum The goal of CMB integrations is to predict the anisotropy in a given cosmological model. We must ask carefully what this means. Does it mean to predict the actual pattern of 32 (1) (1) (1) Δeff = 1/3 δγ + φ −2 2 Ωb= 0.03 h 1 Ω =0 b 0 −1 −2 | | − cτ rec | − 0.5 cτ rec | 0 x | 0.5 cτ rec cτ rec Figure 4: Plane-parallel Green’s function for the combined intrinsic ( 13 δγ ) plus gravita­ tional redshift contributions to CMB anisotropy at recombination in the fluid approxi­ mation, for the same parameters as shown in Fig. 2. The solid (dashed) lines show the results for Ωb h2 = 0.03 (Ωb h2 = 0). The quantity Δeff not only gives the Sachs-Wolfe contribution to CMB anisotropy on large scales, it also measures the departure of the photon-baryon fluid from hydrostatic equilibrium. Based on Bashinsky & Bertschinger (2001). fluctuations that will be measured in the sky? No. That is impossible because the fluc­ tuations originate from random fluctuations in the early universe (quantum fluctuations during inflation or perhaps thermal fluctuations in some first-order phase transition). At best, we can predict only the statistical properties of the CMB anisotropy. But then what statistical properties should we calculate? Inflation predicts a gaussian random field of primeval potential fluctuations. That is, in a Fourier decomposition (here assuming a K = 0 background) φ(�x, τ ) = � � d3 k eik·�x φ(�k, τ ) , (66) each mode φ(k) is a zero-mean, normally distributed random variable. Actually, the Fourier transform of a real field is a complex field, so each mode is a pair of normally distributed variables with the same variance. Also note that our definition of the Fourier transform here differs from that in equation (59). For a Gaussian random field, the two-point correlation, i.e. the covariance of field values at different points, completely characterizes the statistical properties of the field. For φ(k), its covariance defines the power spectrum (or spectral density) P φ (k): � φ(�k)φ∗ (�k � ) ≡ Pφ (k) δD (�k − �k � ) . � (67) Had we used the alternative Fourier transformation definition implied by equation (59), the right-hand side would be larger by a factor (2π)3 . Note the three-dimensional Dirac 33 delta function δD . The complex conjugation may be removed provided that the argument of the delta function is changed to �k + �k � . It is easy to check that the spatial covariance, i.e. the two-point correlation function, is the Fourier transform of the power spectrum: �φ(�x1 )φ(�x2 )� = � � d3 k eik·(�x1 −�x2 ) Pφ (k) . (68) Note that the correlations of different points in space are generally nonzero — in par­ ticular, scale-invariant fluctuations are a statistical fractal, with correlations extending over all scales — but that the correlation at different points vanishes in k-space because of the Dirac delta function in equation (67). This is a general feature of “stationary” random fields, i.e. random fields whose statistical properties are invariant under spatial translation. It is a natural outcome of inflationary models, where the fluctuations are the zero-point fluctuations of a nearly massless free field; each mode is independent. Our calculations have shown that the primary CMB anisotropy Δ is linear in the gravitational potential (and its derivatives and integrals). Thus, the anisotropy at any point in the sky is a linear combination of gaussian-distributed random numbers. A linear combination of normal (gaussian) random numbers is itself normal. Thus, the anisotropy Δ0 (�n) should be a gaussian random field on the sphere. Thus, to fully characterize its statistical properties we need only to calculate the analogue of the power spectrum on a sphere. The angular power spectrum is straightforward to define by analogy with equations (66) and (67). We first expand the temperature anisotropy in the orthonormal basis functions appropriate for a sphere, spherical harmonics: Δ0 (�n) = ∞ � l � alm Ylm (�n) (69) l=0 m=−l where Ylm (�n) is shorthand for Ylm (θ, ϕ) for spherical angles (θ, ϕ) giving the observational direction �n. (Note that here we are defining �n to have opposite sign to the rest of these notes, for consistency with usage by experimentalists: −�n is the direction the photon travels.) The angular power spectrum is now defined by the covariance of the expansion coefficients: �alm a∗l� m� � ≡ Cl δll� δmm� (70) where δll� is now the Kronecker delta. Each angular coefficient is independent for a random field with rotational invariance on a sphere. This is the angular analogue to translational invariance, which led to a Dirac delta function in equation (67). The compactness of the sphere makes the sum over harmonics discrete rather than continuous, leading to a Kronecker delta rather than a Dirac delta function. 34 The angular analogue of equation (68) is the angular correlation function ∞ 1 � C(θ) ≡ �Δ(�n1 )Δ(�n2 )� = (2l + 1)Cl Pl (�n1 · �n2 ) 4π l=0 (71) where �n1 · �n2 = cos θ and Pl (x) is the Legendre polynomial of degree l. Notice that the angular correlation function C(θ) and power spectrum Cl are Fourier-Legendre transfor­ mations of each other and therefore contain the same information. To calculate the angular power spectrum we must relate the anisotropy to the po­ tential in k-space. We do this by expanding the anisotropy in plane waves. In the Sachs-Wolfe approximation, from equation (48) we get Δ0 (�n) = � ⎡ ⎛ ⎞ ⎤ � χe 2i�k · �n ⎠ � � n⎝ 3 ⎣ −iχe�k·� φ(k, τe ) + 2 dχ e−iχk·�n ∂τ φ(�k, τ0 − χ)⎦ (72) dk e C− 3ηe 0 with C ≈ 13 for the isentropic mode and C ≈ 2 for isocurvature modes. The minus sign in the plane waves occurs because −�n is the radial unit vector. Now we use the spherical wave expansion of a plane wave, eixµ = ∞ � il (2l + 1)jl (x)Pl (µ) (73) l=0 where jl (x) is the spherical Bessel function (e.g. Jackson Classical Electromagnetism). Substituting into equation (72) yields Δ0 (�n) = � 3 dk ∞ � l=0 (−i)l (2l + 1)Δl (�k, τ0 )Pl (kˆ · �n) (74) where k̂ = �k/k and, in the Sachs-Wolfe approximation, � � � χe 2 � jl (kχe ) φ(�k, τe ) + 2 dχ jl (kχ) ∂τ φ(�k, τ0 − χ) . Δl (�k, τ0 ) = Cjl (kχe ) + 3ηe 0 (75) (The derivative term jl� came from differentiating eq. 73 with respect to x, which brings down the factor iµ = −ikˆ · �n needed for the Doppler term in eq. 72.) We need one more mathematical result, the addition theorem for spherical harmonics, l 4π � ∗ Pl (�n1 · �n2 ) = Ylm (�n1 )Ylm (�n2 ) . 2l + 1 m=−l (76) Using equations (69), (74), and (76), we obtain l alm = (−i) 4π � ∗ d3 k Ylm (k̂) Δl (�k, τ0 ) . 35 (77) To get the angular power spectrum we must now relate the potentials in equation (75) to the initial random field of potential or entropy fluctuations that induced the CMB anisotropy. We define the CMB transfer function Dl (k) ≡ Δl (�k, τ0 ) φi (�k) (78) where φi (�k) is the Fourier transform of the primeval potential field of equation (46). (This assumes the initial fluctuations were isentropic; if they were isocurvature, then φ i should be replaced by the initial entropy perturbation field.) The universe is like a linear amplifier in electronics: each harmonic component is modulated by an analog filter. Note that the transfer function depends on spatial frequency but not direction, because the equations of motion are rotationally invariant. Using equations (70), (76), (77), and (78), plus the orthonormality of the spherical � ˆ = δll� δmm� ), we get the key result for the angular power ˆ ∗� � (k) harmonics ( dΩ Ylm (k)Y lm spectrum of the CMB: � Cl = 4π d3 k Pφ (k)Dl2 (k) . (79) Here, Pφ (k) is the primeval power spectrum. For scale-invariant initial fluctuations, Pφ ∝ k −3 . The dominant contribution to large-scale anisotropy is the Sachs-Wolfe effect, for which Dl = C(φe /φi )jl (kχe ). For a scale-invariant spectrum, equation (79) can be integrated exactly giving Cl ∝ 1/[l(l + 1)], corresponding to equal power per logarithmic interval of angular degree l, i.e. a flat spectrum on the sphere. (In d dimensions, a scale-invariant spectrum is k −d . For large l, the effective wavenumber is k ∝ l.) This is why plots of the CMB power spectrum always show l(l + 1)Cl . We can go beyond the Sachs-Wolfe approximation by replacing equation (72) with the exact equation (16). Each field such as δγ (�x, τ ) is written as a Fourier integral with the Fourier components proportional to the initial gravitational potential through a transfer function, e.g. δγ (�k, τ ) = δγ (k, τ )φi (�k ), a straightforward generalization of equation (56). Thus, we may generalize equation (72) to (Seljak & Zaldarriaga 1996; note the modified coefficient of δγ due to our different definition here) � ∂ 1 1 + Πij ni nj Dl (k) = dχ ζ̇(τ0 − χ) δγ + φ + uγ 3 ∂χ 2 0 � τ0 ∂ + dχ ζ(τ0 − χ) (φ + ψ)(k, τ0 − χ) , ∂τ 0 � τ0 � jl (kχ) ret (80) where ζ(τ ) = exp[−τT (τ0 − τ )] is the “visibility function” of equation (15) and subscript ret means to evaluate the quantity in brackets at (k, τ = τ0 − χ). The functions δγ , φ, 36 uγ , etc., are all transfer functions, i.e. they are solutions to the equations of motion in Fourier space subject to the initial conditions φ(k) = 1, σ(k) = 0. The transfer functions may be computed directly in �k-space or by Fourier transfor­ mation of the plane-parallel Green’s functions by equation (59). The publicly available CMBFAST code (Seljak & Zaldarriaga 1996) computes them directly in �k-space, while Bashinsky (2001) and Bashinsky & Bertschinger (2002) have computed them (in the fluid approximation) by transforming the Green’s functions. 9 Numerical results In Section 6 we derived approximate results for the CMB anisotropy on large scales in the Sachs-Wolfe approximation, equation (48). Later, in Section 7, we worked out the plane-parallel Green’s function solutions in the two-fluid approximation. Section 8 showed how to compute the angular correlation function and angular power spectrum from these ingredients. Here we present results for the two-fluid approximation and compare with the exact computation by CMBFAST. The CMBFAST code includes three physical processes that are excluded by the two-fluid approximation: a full treatment of neutrinos; photon diffusion; and the polarization-dependence and anisotropic scattering contributions of Thomson scattering. Figure 5 shows the angular correlation function on small scales (θ < 2◦ ). There are four notable phenomena. First, the curves all show a prominent dip. This feature is associated with the overlap of acoustic rings in the CMB photosphere. Recall that the spherical Green’s functions correspond to acoustic waves expanding at the speed of sound. By the time of recombination, these waves reach a comoving distance of about 0.5τrec (Fig. 4). At the distance of recombination, τrec (the diameter of the acoustic sphere) subtends an angle of about 1.2◦ . Bashinsky & Bertschinger (2002) show how the sharp feature occurring at this angle (thin curve in Fig. 5) arises from the singularities of δγ at the acoustic wavefronts. For angular separations less than this, sound waves are able to establish acoustic contact. For larger angular separations, the correlations reflect the correlations present in the unmodified inflationary spectrum. (Causal effects extend to about twice the acoustic length because the speed of sound is about half the speed of light.) The second notable result from Figure 5 is that the two-fluid approximation with instantaneous recombination (thin curve) recovers much of the qualitative behavior of the CMBFAST result (thick curve) but underestimates the anisotropy on small scales. The main reason for this is that the Doppler contribution has been neglected here. Thus, the largest source of CMB anisotropy is accounted for by the intrinsic ( 13 δγ ) and gravitational redshift contributions, but the Doppler contribution is appreciable on scales less than the acoustic length. 37 2 ΛCDM , Ωbh = 0.02 C (θ) , 10−9 K 2 8 1/3 δ + φ , inst. rec. γ ΛCDM , Ω h2= 0.03 b 6 Ω = Ω = 0.35 m 4 2 0.4 0.8 1.2 θ, degrees 1.6 2 Figure 5: Angular CMB temperature correlation function. The parameters of the ΛCDM model are Ωm = 0.35, ΩΛ = 0.65, h = 0.65 with a scale-invariant spectrum Pφ ∝ k −3 . For this set of parameters and the displayed Ωb h2 values, the acoustic sphere subtends an angle θs � 1.2◦ . The thin solid curve is calculated using the two-fluid approximation with instantaneous recombination. All the other curves are calculated using CMBFAST. The dip in C(θ) is shifted to smaller angles for the open model (dashed-dotted line). The slope for θ < θs /2 is dependent on the baryon density (dashed vs. solid line). Based on Bashinsky & Bertschinger (2001). The third notable result is that the acoustic length in the open model (dashed-dotted curve in Fig. 5) subtends a smaller angle. This is simply because the angular distance r(χ) to the CMB photosphere is larger for an open model (Weinberg 2000). Finally, Figure 5 shows that the baryon contribution has the most significant effect for angles less than half the diameter of the acoustic sphere. The baryon effect arises from the interplay between the wavefront singularity and the central valley in Figure 4. As we showed, the depth of that valley is proportional to Ωb . This translates into a roughly linear dependence of the inner slope of C(θ) on Ωb . The angular correlation function has the attractive property of associating acoustic phenomena with localized features. The localization in angle means that a FourierLegendre representation will produce “ringing” (oscillations) in the angular frequency. Because the sphere is a compact manifold, the angular frequency l is discrete. Figure 6 plots the angular power spectrum Cl to show us that this ringing corresponds to the famous “acoustic peaks.” The fluid approximation yields surprisingly accurate prediction of the CMB anisotropy 38 5 4 l (l+1) Cl 3 2 1 0 0 500 1000 l 1500 2000 Figure 6: The angular power spectrum of CMB temperature anisotropy calculated using the two-fluid approximation (solid line, including the Doppler contribution) and using CMBFAST (dashed line). The ΛCDM model parameters are the same as for the thick solid curve in Fig. 5. From Bashinsky (2001). spectrum. With a phenomenological correction for photon diffusion, the error of our nu­ merical calculations in fluid approximation is within 5% on angular scales from 2π to a few arcminutes, see Fig. 9. It can also compute the gravitational radiation (tensor mode) and gravitational lensing (secondary) contributions to anisotropy. The agreement between the two-fluid approximation and CMBFAST is much better in Figure 6 than in Figure 5. There are several reasons for this improvement. First, all the anisotropy contributions in equation (80) have now been included except for Π ij ni nj (polarization and anisotropic scattering terms). Second, photon diffusion was included approximately in a phenomenological manner. Third, recombination was gradual rather than instantaneous. Seljak (1994) included these effects in his earlier two-fluid treat­ ment, however he did not include the integrated Sachs-Wolfe effect and he used the instantaneous recombination approximation, with some analytic corrections that are not as accurate as our numerical integrations. Our two-fluid model achieves accuracy better than 5% in Cl for l ≤ 1500. It appears that the approximate description given by equations (52) captures well the essential physics of the cosmological dynamics of the CMB photons and the other components with which they interact. Although CMBFAST remains the method of choice for accurate calculations and data analysis, the simplified model presented in 39 these notes provides a useful way to comprehend the dynamics of CMB anisotropy. I would like to thank Sergei Bashinsky for his collaboration and for many helpful discussions, without which these notes would be impossible. References [1] Bardeen, J. M. 1980, PRD, 22, 1882. [2] Bashinsky, S. 2001, MIT PhD thesis. [3] Bashinsky, S., & Bertschinger, E. 2001, PRL, 87, 081301. [4] Bashinsky, S., & Bertschinger, E. 2002, PRD, 65, 123008. [5] Bertschinger, E. 1996, in Cosmology and Large Scale Structure, proc. Les Houches Summer School, Session LX, ed. R. Schaeffer, J. Silk, M. Spiro, and J. Zinn-Justin (Amsterdam: Elsevier Science), 273-347. [6] Hu, W., Seljak, U., White, M., & Zaldarriaga, M. 1998, PRD, 57, 3290. [7] Kodama, H., & Sasaki, M. 1984, Prog. Theor. Phys. Suppl., 78, 1. [8] Kodama, H., & Sasaki, M. 1986, Int. J. Mod. Phys., A1, 265. [9] Ma, C.-P., & Bertschinger, E. 1995, ApJ, 455, 7. [10] Rees, M. J., & Sciama, D. W. 1968, Nature, 517, 611. [11] Sachs, R. K., & Wolfe, A. M. 1967, ApJ, 147, 73. [12] Seljak, U. 1994, ApJ, 435, L87. [13] Seljak, U. 1996, ApJ, 463, 1. [14] Seljak, U., & Zaldarriaga, M. 1996, ApJ, 469, 437. [15] Spergel, D. N., & Zaldarriaga, M. 1997, PRL, 79, 2180. [16] Weinberg, S. 2000, PRD, 62, 127302. [17] Weinberg, S. 2002, astro-ph/0207375. [18] Zaldarriaga, M., & Seljak, U. 1997, PRD, 55, 1830. [19] Zaldarriaga, M., Seljak, U., & Bertschinger, E. 1998, ApJ, 494, 491. 40