betaFeedback

WorksPhysics and Astronomy

Shedding light on modified recombination scenarios with Rayleigh scattering of the CMB

Abstract

Rayleigh scattering (RS) of cosmic microwave background (CMB) photons by neutral hydrogen and helium, during and shortly after the last scattering surface, adds a frequency-dependent contribution to the CMB anisotropies that scales approximately as ν4ν^4 at leading order. The RS signal is most relevant at 200 GHz≲ν≲800 GHz200\,{\rm GHz}\lesssimν\lesssim800\,{\rm GHz} and could be detected using future CMB experiments. Because RS acts in a frequency-dependent manner which is sensitive to the ionisation history of the cosmological recombination process, it provides an additional probe of the recombination history. This makes it a natural test of modified recombination scenarios, such as variable electron mass mem_{\rm e} or fine-structure constant αα, which have been proposed to ease the Hubble tension. The correlation between the RS and primary CMB signal, both in temperature and polarisation, further enhances its usefulness for constraining cosmological parameters. Using a multi-frequency Fisher analysis with realistic noise and foregrounds, we forecast parameter constraints and biases for PICO and LiteBIRD + Planck. For PICO, adding RS improves the ΛΛCDM constraints on cosmological parameters by ≲7%\lesssim7\%, but improves the constraints on some parameters (H0H_0, mem_{\rm e}, αα, Ωbh2Ω_{\rm b}h^2, Ωch2Ω_{\rm c}h^2) by ≃31%\simeq31\% for some modified scenarios in the recombination history. However, the improvement for LiteBIRD + Planck is limited to less than 2%2\% for most of the parameters. Neglecting RS biases the cosmological parameters by up to ≃1σ\simeq1σ for PICO and below 1σ1σ for LiteBIRD + Planck. Therefore RS must be modelled in the analyses of future CMB data.

Prepared for submission to JCAP

Introduction

In modern cosmology, our understanding of the physics of the cosmic microwave background (CMB) and the precise measurement of its temperature and polarisation anisotropies with many ground- and space-based experiments [1, 2, 3, 4] has led to the demonstration that the standard ΛCDM model of the Universe provides an extremely successful description of the current data. However, several cosmological tensions still persist, e.g., (i) mismatch between the Hubble constant H0H_0 from CMB data [5] and the same obtained from Cepheid-calibrated Type Ia Supernovae [6]; (ii) discrepancy between the amplitude of the growth of structure (σ8\sigma_8) inferred from CMB data and the one from late-time large-scale structure observations [7, 8, 9, 10, 11]. Despite several efforts to attribute the Hubble tension to systematic uncertainties in the data [12, 13], it has persisted. Several new-physics models have attempted to alleviate the Hubble tension [14, 15, 16, 17, 18, 19, 20]. However, no conclusive resolution has yet been reached. Since the CMB precisely measures the angular size of the sound horizon, θ∗=rs/DM(z∗)\theta_* = r_s/D_M(z_*), the value of H0H_0 inferred from it depends both on the comoving sound horizon at recombination, rsr_s, which is set by pre-recombination physics, and on the comoving distance to the last-scattering surface, DM(z∗)D_M(z_*), which depends on the post-recombination expansion history [21]. Therefore, exploring modifications to the physics of the early or late Universe is particularly interesting to alleviate the Hubble tension.

Most attempts to alleviate the Hubble tension by modifying post-recombination physics rely on changes in the late-time expansion history, within the range allowed by BAO and Type Ia supernovae data [22, 23]. These typically invoke an evolving effective dark energy component while leaving the comoving sound horizon at recombination unchanged, and include dynamical dark energy (DDE) [24], interacting dark sector [25], and modified gravity models [26]. However, the freedom of these models to alleviate the H0H_0 tension is largely removed once BAO and supernovae data are included, since these models leave the sound horizon unchanged [21, 23]. Late-time modifications have nevertheless received renewed attention after the DESI BAO results [27] which prefer DDE at ∼2.8\sim2.8–4.2σ4.2\sigma significance to reduce the Ωm\Omega_m tension, although this has been challenged by [28], claiming such tension is due to unexplored systematics.

Another compelling possibility is to change the sound horizon in the pre-recombination era. A well-studied class of models that achieves this relies on introducing an early dark energy (EDE) component that decays before the recombination epoch (see [29] for a review). Another promising avenue is to invoke early recombination, for instance via time-varying fundamental constants; in particular, varying electron mass, mem_e or fine-structure constant α\alpha [15, 30, 31, 32, 33, 34]. A larger electron mass during the early Universe makes the recombination process faster, which leads to a shorter sound horizon and therefore a larger value of H0H_0. The varying electron mass model is well motivated in the literature and is consistent with other cosmological data sets that favour larger electron mass with greater significance than 2σ2\sigma [35, 36, 37, 38].

In the literature, the impact of varying fundamental constants is studied mainly through their influence on Thomson scattering of CMB photons off of free electrons during the recombination era, at redshift z∼1100z \sim1100. In this work, we discuss the impact of varying fundamental constants on the scattering of CMB photons with newly formed hydrogen and helium atoms during the recombination era, namely Rayleigh scattering (RS, [39, 40]). The effect of RS is often neglected in the literature, assuming that CMB photons can travel freely after recombination. RS dampens CMB anisotropies on sub-horizon scales because photons are further scattered out of the line of sight, which leads to mixing between photons at hot and cold spots. In addition, the large-scale EE-mode polarisation produced at recombination arises from the coherent scattering of the local temperature quadrupole into the line of sight. Since RS is increasingly important at higher frequencies and relevant only in the later stages of recombination, when the photons have free-streamed further, and the quadrupole has grown, the additional RS increases the polarisation amplitude on large scales at high frequencies. Thus RS introduces an asymptotic part in the CMB intensity and polarisation anisotropies which scales as ν4\nu^4 at leading order, making observations at frequencies 200 GHz≲ν≲800 GHz200\ \mathrm{GHz} \lesssim\nu\lesssim800\ \mathrm{GHz} particularly important for future CMB experiments. In this work, we model the effect of RS on CMB anisotropies and study its cosmological importance for future CMB experiments, e.g., LiteBIRD [41] and PICO [42].

The paper is organised as follows. In § 2, we describe the effect of RS on the CMB temperature and polarisation anisotropies, and show how a variation of mem_e or α\alpha modifies the RS opacity, the Thomson and Rayleigh visibility functions, and the resulting frequency-dependent power spectra. In § 3 we describe the multi-frequency Fisher formalism (§ 3.1) used to forecast parameter constraints and to estimate the biases arising from an incorrect model, and the foreground model (§ 3.2) included in the analysis. § 4 presents our results: the forecasts cover constraints with and without RS for Λ\LambdaCDM and the modified recombination scenarios (§ 4.1), and the parameter biases that arise when RS or the variation of the fundamental constants is not modelled consistently (§ 4.2). We summarise our conclusions in § 5. The derivation of the Fisher matrix and of the parameter bias is given in Appendix § 7, and the corresponding results for LiteBIRD + Planck are presented in Appendix § 8.

Rayleigh scattering in CDM and modified recombination scenarios

In this section, we illustrate the mechanism of RS and its impact on CMB power spectra at different frequencies, both in the standard Λ\LambdaCDM and modified recombination scenarios. The fiducial values of the six Λ\LambdaCDM cosmological parameters are set to the best-fit values of the Planck 2018 results [5] throughout this paper.

Rayleigh Scattering effects on CMB temperature and polarisation anisotropies

The elastic scattering of long-wavelength photons by neutral particles is known as Rayleigh scattering [43]. By the time recombination occurs, the vast majority of CMB photons have energies well below the hydrogen ionisation energy and therefore wavelengths well above the atomic sizes of the hydrogen and helium atoms that are produced during recombination.

The dipole induced in neutral hydrogen and helium scatters these long-wavelength photons with a cross-section that is proportional to the fourth power (∝ν4\propto\nu^4) of their frequency at leading order. The approximate Rayleigh scattering cross-section for photons at frequency ν\nu scattered by a hydrogen atom in the ground state is given by the Kramers–Heisenberg formula [44],

σR(ν)=σT(ννeff)4[1+2.626(ννeff)2+5.502(ννeff)4+⋯ ],(1)\sigma_{\mathrm{R}}(\nu) = \sigma_{\mathrm{T}} \left(\frac{\nu}{\nu_{\mathrm{eff}}}\right)^4 \left[1 + 2.626\left(\frac{\nu}{\nu_{\mathrm{eff}}}\right)^2 + 5.502\left(\frac{\nu}{\nu_{\mathrm{eff}}}\right)^4 + \cdots\right], \tag*{(1)}

where σT\sigma_{\mathrm{T}} is the Thomson scattering cross-section and νeff=89cR∞\nu_{\mathrm{eff}} = \sqrt{\frac{8}{9}cR_\infty} (where R∞R_\infty is the Rydberg constant). Therefore, the total comoving opacity of the photons at frequency ν\nu can be approximated as

τ˙(ν)=τ˙T+τ˙R(ν)=a neσT+a σR(ν)[nH+RHenHe],(2)\dot{\tau}(\nu) = \dot{\tau}_{\mathrm{T}} + \dot{\tau}_{\mathrm{R}}(\nu) = a\,n_e\sigma_{\mathrm{T}} + a\,\sigma_{\mathrm{R}}(\nu)\left[n_{\mathrm{H}} + R_{\mathrm{He}}n_{\mathrm{He}}\right], \tag*{(2)}

where τ˙T\dot{\tau}_{\mathrm{T}} and τ˙R(ν)\dot{\tau}_{\mathrm{R}}(\nu) are the comoving opacities for Thomson and Rayleigh scattering respectively, aa is the scale factor, nen_{\mathrm{e}}, nHn_{\mathrm{H}} and nHen_{\mathrm{He}} are the electron, neutral hydrogen and neutral helium number densities, respectively, and RHe≃0.1R_{\mathrm{He}}\simeq0.1 is the relative strength of Rayleigh scattering of helium with respect to hydrogen [45]. Rayleigh scattering becomes increasingly important relative to Thomson scattering as nen_{\mathrm{e}} drops and nHn_{\mathrm{H}} and nHen_{\mathrm{He}} increase during recombination. Furthermore, once recombination is complete, the number density of neutral hydrogen evolves with redshift as (1+z)3(1+z)^3 and the CMB photon frequency as (1+z)(1+z), so that the Rayleigh scattering rate changes as (1+z)7(1+z)^7 at leading order. This makes the process efficient near the last scattering surface only. This additional scattering reduces the photon mean free paths, thus increasing the diffusion damping acting on small scales [39].

The left panel of Figure 1 shows the comoving Thomson scattering opacity (solid line), together with the frequency-dependent RS contribution to τ˙\dot{\tau} in Eq. (2) at two representative frequencies: 321 GHz (dash-dotted line) and 799 GHz (dashed line), for the fiducial Λ\LambdaCDM cosmology. We have chosen these two frequencies since they correspond to PICO channels [42] and lie in the range where the RS signal is strongest. At low frequencies the RS comoving opacity is highly subdominant compared with Thomson scattering, but its steep ν4\nu^4 dependence makes it increasingly important at higher frequencies. The right panel shows the corresponding visibility functions, gT(η)=τ˙T(η)e−τT(η)g_{\mathrm{T}}(\eta)=\dot{\tau}_{\mathrm{T}}(\eta)e^{-\tau_{\mathrm{T}}(\eta)}, for Thomson scattering alone (solid line) and gR(ν,η)=τ˙R(ν,η)e−τR(ν,η)g_{\mathrm{R}}(\nu,\eta)=\dot{\tau}_{\mathrm{R}}(\nu,\eta)e^{-\tau_{\mathrm{R}}(\nu,\eta)} for RS at 321 GHz (dash-dotted line) and 799 GHz (dashed line). Since RS acts mainly in the later stages of recombination, it shifts the peak of the visibility function towards lower redshift, an effect present at all frequencies but most pronounced at high frequencies. Different observing frequencies therefore probe slightly different stages of the evolution of the anisotropies during recombination. At these later times, the local quadrupole is larger, which boosts the EE-mode polarisation on scales 10≲ℓ≲30010\lesssim\ell\lesssim300 [39, 46], while the sound horizon is slightly larger, resulting in a shift of the acoustic peaks to lower multipoles in the higher-frequency channels. This shift is potentially dependent on the ionisation history and the expansion rate of the Universe. By adding frequency-dependent information on the positions of the acoustic peaks, the RS signal can therefore help break parameter degeneracies, improving CMB constraints on the Hubble constant while also tightening constraints on the ionisation history. With the inclusion of the RS effect, CMB harmonic coefficients at frequency ν\nu can be perturbatively approximated as

Figure 1. Left panel: Comoving opacity of Thomson scattering, τ˙T\dot{\tau}_{\mathrm{T}} (solid line), and of Rayleigh scattering, τ˙R\dot{\tau}_{\mathrm{R}}, at 321 GHz (dash-dotted line) and 799 GHz (dashed line). At low frequencies, the RS contribution to the total opacity is negligible compared with Thomson scattering, whereas at high frequencies it becomes relevant towards the end of recombination, before decreasing rapidly as (1+z)7(1+z)^7 once recombination is complete. Right panel: The corresponding normalised visibility functions. For clarity, the visibility function at 799 GHz has been multiplied by a factor of 1.5.

αℓmX(ν)≃αℓmX+(νν0)4ΔαℓmX,4+(νν0)6ΔαℓmX,6+(νν0)8ΔαℓmX,8(3)\alpha_{\ell m}^{X}(\nu)\simeq\alpha_{\ell m}^{X}+\left(\frac{\nu}{\nu_{0}}\right)^{4}\Delta\alpha_{\ell m}^{X,4}+\left(\frac{\nu}{\nu_{0}}\right)^{6}\Delta\alpha_{\ell m}^{X,6}+\left(\frac{\nu}{\nu_{0}}\right)^{8}\Delta\alpha_{\ell m}^{X,8} \tag*{(3)}

where ν0\nu_0 is some reference frequency of RS and X∈{\in\{T, E}\}, aℓma_{\ell m} are the harmonic coefficients of the primary CMB due to Thomson scattering and ΔaℓmXn\Delta a_{\ell m}^{X_n} are the harmonic coefficients of the RS contribution at the reference frequency ν0\nu_0. Therefore, the cross-power spectra of the CMB computed between two channels can be written as

CℓXY(νi,νj)≃CℓXY+∑n=4,6,8(1ν0)n[νinCℓXΔYn+νjnCℓΔXnY]+(νiνjν02)4CℓΔX4ΔY4+…(4)C_\ell^{XY}(\nu_i,\nu_j) \simeq C_\ell^{XY}+\sum_{n=4,6,8}\left(\frac{1}{\nu_0}\right)^n\left[\nu_i^n C_\ell^{X\Delta Y_n}+\nu_j^n C_\ell^{\Delta X_nY}\right]+\left(\frac{\nu_i\nu_j}{\nu_0^2}\right)^4 C_\ell^{\Delta X_4\Delta Y_4}+\ldots \tag*{(4)}

The first, second, and third terms represent the primary CMB power spectrum, the cross-spectrum between the primary CMB1 and the RS signal, and the cross-spectrum of the RS signal between two observing frequencies, respectively.

Dependencies of Rayleigh scattering signal on varying fundamental constants

In this section, we illustrate the effect of varying fundamental constants focusing on the electron mass, mem_e, and the fine-structure constant, α\alpha, on the recombination history and CMB power spectra, with particular emphasis on the contribution from RS. We used the modified version of RECFAST++ [47, 48] to model the recombination history and propagate the resulting changes in the free-electron, nen_e, neutral hydrogen, nHn_\mathrm{H} and neutral helium nHen_\mathrm{He} number densities (cf. Eq. (2)) to a modified version of CAMB [49] to compute the CMB power spectra contributed by Thomson and Rayleigh scattering across frequencies.

Variations in mem_e or α\alpha modify the recombination process through several distinct physical mechanisms. These include changes in atomic binding energies, atomic transition rates, photoionisation/recombination rates, and the radiative transfer of photons through modified Thomson, Compton, and resonance scattering [50, 51, 52]. The resulting modifications in the recombination history in an effective three-level atom approach [53, 54] and their impact on the primary CMB power spectra are discussed in detail by [30]. We consider deviations of mem_e and α\alpha from their current values (me,0=511m_{e,0}=511 keV and α≃1/137\alpha\simeq1/137) during the recombination epoch, with both reaching their present values after recombination has completed. We parameterise these variations during the recombination era as, me=me,0(1+Δmeme,0)m_e=m_{e,0}(1+\frac{\Delta m_e}{m_{e,0}}) and α=α0(1+Δαα0)\alpha=\alpha_0(1+\frac{\Delta\alpha}{\alpha_0}). The most important effect of these variations is the increase in the atomic binding energies of hydrogen and helium, Ei∝meα2E_i\propto m_e\alpha^2, for Δmeme,0>0\frac{\Delta m_e}{m_{e,0}}>0 or Δαα0>0\frac{\Delta\alpha}{\alpha_0}>0, which makes photoionisation less efficient at a given temperature. Since CMB photon temperature drops with the expansion of the Universe as Tγ∝(1+z)T_\gamma\propto(1+z), CMB temperature falls below the ionisation threshold at a higher redshift for Δmeme,0>0\frac{\Delta m_e}{m_{e,0}} > 0 or Δαα0>0\frac{\Delta\alpha}{\alpha_0} > 0. Consequently, recombination occurs earlier, leading to a larger abundance of neutral hydrogen and helium at higher redshifts compared to the standard recombination history.

The earlier production of neutral atoms initially enhances the Rayleigh scattering opacity at high redshifts. In addition, variations in mem_e and α\alpha also modify the RS opacity through changes in the Rayleigh scattering cross-section, σR(ν)\sigma_{\mathrm{R}}(\nu). The Rayleigh scattering cross-section, to leading order in its frequency dependence, is given by

σR(ν)≃σT(ννeff)4∝ν4α−6me−6,(5)\sigma_{\mathrm{R}}(\nu) \simeq\sigma_{\mathrm{T}} \left(\frac{\nu}{\nu_{\mathrm{eff}}}\right)^4 \propto\nu^4\alpha^{-6}m_e^{-6}, \tag*{(5)}

where σT∝α2me−2\sigma_{\mathrm{T}} \propto\alpha^2m_e^{-2} is the Thomson scattering cross-section and the effective Rayleigh resonance frequency scales as νeff∝α2me\nu_{\mathrm{eff}} \propto\alpha^2m_e. Consequently, increasing mem_e or α\alpha decreases the Rayleigh scattering cross-section, which leads to a reduction in the Rayleigh scattering opacity with respect to the Λ\LambdaCDM case once recombination is essentially completed. Figure 2 shows the evolution of the Rayleigh comoving opacity, τ˙R(ν)\dot{\tau}_{\mathrm{R}}(\nu) (cf. Eq. (2)) for different values of Δme/me,0∈{−0.1,0.1}\Delta m_e/m_{e,0} \in\{-0.1, 0.1\} (left panel) and Δα/α0∈{−0.1,0.1}\Delta\alpha/\alpha_0 \in\{-0.1, 0.1\} (right panel) at two representative frequencies, 799 GHz (dashed curves) and 321 GHz (dashed-dotted curves). The corresponding curve for the Λ\LambdaCDM model is shown in grey. Increasing mem_e or α\alpha shifts the onset of Rayleigh scattering opacity to higher redshifts compared to the Λ\LambdaCDM scenario, due to earlier recombination. The effect is exactly the opposite for decreasing values of mem_e and α\alpha. The suppression of σR∝α−6me−6\sigma_{\mathrm{R}} \propto\alpha^{-6}m_e^{-6}, which becomes apparent only after recombination is complete, produces a steeper decline in comoving opacity than in Λ\LambdaCDM. As expected, the effect is slightly greater for changes in α\alpha than for mem_e, due to their different impact on the enhancement of energy transition levels and other intermediate processes of recombination as described in [30, 50]. At lower frequencies, τ˙R\dot{\tau}_{\mathrm{R}} responds in the same way to variations of mem_e and α\alpha, but its amplitude is substantially smaller since it decreases according to the scaling of the Rayleigh scattering cross-section σR(ν)∝ν4\sigma_{\mathrm{R}}(\nu) \propto\nu^4.

Figure 2. Effect of variations in mem_e and α\alpha on the Rayleigh comoving opacity, τ˙R(ν)\dot{\tau}_{\mathrm{R}}(\nu), at two representative observing frequencies, 321 GHz (dash-dotted lines) and 799 GHz (dashed lines). Left panel: For different values of Δme/me,0\Delta m_e/m_{e,0}. Right panel: For different values of Δα/α0\Delta\alpha/\alpha_0. The corresponding curves for the Λ\LambdaCDM scenario are highlighted in grey.

Effect of varying fundamental constants on visibility function and CMB power spectra

We investigate the changes in the RS visibility function, gR(η,ν)=τ˙R(η,ν)e−τR(η,ν)g_{\mathrm{R}}(\eta,\nu) = \dot{\tau}_{\mathrm{R}}(\eta,\nu)e^{-\tau_{\mathrm{R}}(\eta,\nu)}, and the Thomson visibility function, gT(η)=τ˙T(η)e−τT(η)g_{\mathrm{T}}(\eta) = \dot{\tau}_{\mathrm{T}}(\eta)e^{-\tau_{\mathrm{T}}(\eta)}, for different recombination histories, as these are governed by the dynamics of the recombination process and ultimately determine the Thomson and Rayleigh contributions to the CMB temperature and polarisation power spectra. We computed visibilities using a modified version of CAMB and in Figure 3 we present Thomson (solid lines) and Rayleigh (dashed lines) visibilities (normalised) at the observing frequency of 799 GHz for different values of Δme/me,0∈{−0.1,0.1}\Delta m_e/m_{e,0} \in\{-0.1, 0.1\} (left panel) and Δα/α0∈{−0.1,0.1}\Delta\alpha/\alpha_0 \in\{-0.1, 0.1\} (right panel). The visibilities corresponding to the Λ\LambdaCDM model are coloured grey. For Δme/me,0>0\Delta m_e/m_{e,0} > 0 and Δα/α0>0\Delta\alpha/\alpha_0 > 0, the visibility peak shifts towards higher redshifts compared to the Λ\LambdaCDM case, for Thomson and Rayleigh scattering. The peak shifts in the opposite direction for Δme/me,0<0\Delta m_e/m_{e,0} < 0 and Δα/α0<0\Delta\alpha/\alpha_0 < 0. Moreover, lowering mem_e or α\alpha broadens the Rayleigh visibility function more than the Thomson one, since the Rayleigh opacity is more sensitive to these variations.

Figure 3. Effect of variations in mem_e (left panel) and α\alpha (right panel), on the Thomson (solid lines) and Rayleigh (dashed lines) visibilities (normalised). For clarity, Rayleigh visibilities at 799 GHz are multiplied by a factor of 1.5. The corresponding curves for the Λ\LambdaCDM scenario are highlighted in grey.

We then propagate the effects of varying fundamental constants through a modified version of CAMB and calculate the auto- and cross-frequency CMB power spectra across the observing frequencies. We first investigate the power spectra for the Λ\LambdaCDM scenario. In Figure 4, we illustrate the total TTTT, EEEE, and TETE power spectra contributed by both Thomson and Rayleigh scattering at two representative frequencies, 321 GHz (dashed-dotted lines) and 799 GHz (dashed lines) and compare them with the primary CMB power spectra arising from Thomson scattering alone (solid grey lines). The lower panels show the relative difference between the total power spectra at each frequency and the primary CMB power spectra. At the lower frequency of 321 GHz, the total power spectra remain close to the primary CMB power spectra, since Rayleigh scattering is relatively inefficient at low frequencies. However, at higher frequencies, such as 799 GHz, the effects of Rayleigh scattering become more significant. In particular, the acoustic peaks of the total power spectra are shifted towards lower values of ℓ\ell (larger scales), reflecting the shift of the visibility peak towards lower redshifts, which causes an increase in the sound horizon and a decrease in the angular diameter distance to recombination at this frequency [55]. In addition, the amplitude of the anisotropies is suppressed since increased Rayleigh scattering leads to further mixing and redistribution of photons. We further investigate the frequency dependence of this effect by examining the angular power spectra at other observing frequencies and find that the most pronounced effect of Rayleigh scattering occurs in the EEEE power spectrum. Nevertheless, for observing frequencies below 500 GHz, the relative modification remains small, with ∣ΔDℓDℓ∣≲1%\left|\frac{\Delta D_\ell}{D_\ell}\right| \lesssim1\%.

Figure 4. Total power spectra including both Thomson and RS effects at two observing frequencies: 799 GHz (dashed line) and 321 GHz (dash-dotted line). The grey solid lines correspond to primary CMB power spectra due to Thomson scattering alone. The TTTT, EEEE and TETE spectra are shown in the upper left, upper right and lower left panels, respectively. All cases shown here are for the Λ\LambdaCDM model. The lower panel of each plot shows the relative difference between the total power spectrum at each frequency and the primary CMB-only power spectrum.

We then investigated the effects of varying fundamental constants on the Rayleigh-scattering contribution to the total CMB power spectra. In Figure 5, we illustrate the frequency-dependent contributions to the TTTT, EEEE, and TETE power spectra, as defined in Eq. (4), for Δme/me,0∈{−0.1,0.1}\Delta m_e/m_{e,0} \in\{-0.1, 0.1\} (left column) and Δα/α0∈{−0.05,0.05}\Delta\alpha/\alpha_0 \in\{-0.05, 0.05\} (right column) at a representative frequency channel of 799 GHz. These spectra are compared with the corresponding Λ\LambdaCDM case, shown by the solid grey curves. The lower panel of each subplot shows the relative Rayleigh-scattering contributions with respect to the primary CMB power spectra in the fiducial Λ\LambdaCDM scenario for the different cases. We find that increasing mem_e or α\alpha shifts the acoustic peaks towards higher ℓ\ell values (smaller scales). This behavior reflects the shift of the effective last-scattering surface towards higher redshifts [56, 57]. We also find that the shift of the acoustic peaks due to changes in mem_e or α\alpha remains largely independent of the observing frequency. The amplitudes of the acoustic peaks are also boosted when the fundamental constants are increased. For example, for Δα/α0=0.02\Delta\alpha/\alpha_0 = 0.02, we find that the relative change in the amplitude of acoustic peaks with respect to the Λ\LambdaCDM scenario is ≲12%\lesssim12\%.

Figure 5. Effect of varying the electron mass (left column) and α\alpha (right column) on the frequency-dependent part of the CMB power spectra (cf. Eq. (4)) i.e., the combined Rayleigh-primary cross-correlation and Rayleigh-Rayleigh auto-correlation contributions at observing frequency, 799 GHz. The upper, middle, and lower rows represent the TTTT, EEEE, and TETE power spectra, respectively. In each panel, the grey curves denote the power spectra for the Λ\LambdaCDM scenario. The lower panels in each plot present the Rayleigh scattering contribution with respect to the corresponding primary CMB power spectra for the Λ\LambdaCDM case as a reference, denoted by ΔDℓ/Dℓ\Delta D_{\ell}/D_{\ell} for different values of Δme/me,0∈{−0.1,0.1}\Delta m_{\mathrm{e}}/m_{\mathrm{e},0} \in\{-0.1, 0.1\} and Δα/α0∈{−0.05,0.05}\Delta\alpha/\alpha_{0} \in\{-0.05, 0.05\}. For the TETE power spectrum, the relative change is normalised as ΔDℓTE/DℓTTDℓEE\Delta D_{\ell}^{TE}/\sqrt{D_{\ell}^{TT}D_{\ell}^{EE}}.

Methodology

Fisher Formalism

We are interested in quantifying the improvement in the constraints on the parameters of the fiducial Λ\LambdaCDM cosmological model and several of its extensions due to the inclusion of the RS signal in addition to the primary CMB for future CMB experiments. To do so, we use a standard Fisher matrix approach [58, 59]. For a set of zero-mean Gaussian fields (in our case CMB spherical harmonic coefficients, aℓmXa_{\ell m}^{X} for X=T,EX = T,E) having parameter-dependent covariance matrix, Cℓ=diag⁡{CℓXY}\mathbf{C}_\ell= \operatorname{diag}\{C_\ell^{X Y}\}, the Fisher matrix is defined as the expectation value of the second derivative of the log-likelihood, L∝1det⁡C−1exp⁡(−12a†C−1a)\mathcal{L} \propto\frac{1}{\sqrt{\det\mathbf{C}^{-1}}}\exp\left(-\frac{1}{2}\mathbf{a}^{\dagger}\mathbf{C}^{-1}\mathbf{a}\right) with respect to the model parameters of interest (denoted here as θi\theta_i), evaluated at their fiducial values θi,fid\theta_{i,\mathrm{fid}},

Fij=−∂2ln⁡L∂θi∂θj∣fid=∑ℓ2ℓ+12fskyTr⁡[Cℓ−1∂Cℓ∂θiCℓ−1∂Cℓ∂θj].(6)F_{ij}=-\left.\frac{\partial^2\ln\mathcal{L}}{\partial\theta_i\partial\theta_j}\right|_{\mathrm{fid}}=\sum_{\ell}\frac{2\ell+1}{2}f_{\mathrm{sky}}\operatorname{Tr}\left[\mathbf{C}_{\ell}^{-1}\frac{\partial\mathbf{C}_{\ell}}{\partial\theta_i}\mathbf{C}_{\ell}^{-1}\frac{\partial\mathbf{C}_{\ell}}{\partial\theta_j}\right]. \tag*{(6)}

The CℓC_\ell matrix for CMB data can be written as,

Cℓ=(CℓTTCℓTECℓTECℓEE),(7)\mathbf{C}_{\ell}=\begin{pmatrix} C_{\ell}^{TT} & C_{\ell}^{TE} \\ C_{\ell}^{TE} & C_{\ell}^{EE} \end{pmatrix}, \tag*{(7)}

where CℓXYC_{\ell}^{XY} are Nν×NνN_{\nu} \times N_{\nu} matrices for NνN_{\nu} frequency channels and each element of CℓXYC_{\ell}^{XY} incorporates cross-frequency angular power spectra of CMB, foregrounds and noise. The noise power spectrum at frequency ν\nu is modelled as

NℓXY,νν′=ΔX2exp⁡[ℓ(ℓ+1)(θFWHM8ln⁡2)2]δν,ν′δX,Y,(8)N_{\ell}^{XY,\nu\nu'} = \Delta_{X}^{2} \exp\left[\ell(\ell+1)\left(\frac{\theta_{\mathrm{FWHM}}}{\sqrt{8\ln2}}\right)^{2}\right]\delta_{\nu,\nu'}\delta_{X,Y}, \tag*{(8)}

where ΔX\Delta_X (in μK⋅rad\mu\mathrm{K} \cdot\mathrm{rad}) is the noise sensitivity for X∈{T,E}X \in\{T,E\} and θFWHM\theta_{\mathrm{FWHM}} (in radians) is the beam FWHM at the respective frequency. In this analysis, we adopt the instrument configuration for PICO (see Table 3.2 of [42]), LiteBIRD and Planck (see Table 13 of [41] and Table 4 of [4]). We use only the frequency channels at ν≳90\nu\gtrsim90 GHz, since the RS signal is negligible at lower frequencies and since this choice avoids the need to model some of the low-frequency foregrounds, such as free-free and anomalous microwave emission, which dominate frequencies below 90 GHz. The modelling of foreground power spectra is discussed in § 3.2. Once the Fisher matrix is computed, we obtain the lower limit of the parameter covariance matrix,

DataModelRSΩbh2\Omega_{\mathrm{b}}h^{2}Ωch2\Omega_{\mathrm{c}}h^{2}ln⁡(1010As)\ln(10^{10}A_{\mathrm{s}})H0H_{0}nsn_{\mathrm{s}}τ\taume/me,0m_{\mathrm{e}}/m_{\mathrm{e},0}
Primary CMBΛ\LambdaCDMno−0.95σ-0.95\sigma0.59σ0.59\sigma−0.16σ-0.16\sigma−0.57σ-0.57\sigma−0.49σ-0.49\sigma−0.12σ-0.12\sigma–
+ RS(−0.11%)(-0.11\%)(0.19%)(0.19\%)(−0.02%)(-0.02\%)(−0.12%)(-0.12\%)(−0.07%)(-0.07\%)(−0.39%)(-0.39\%)–
Primary CMBΛ\LambdaCDMno−11.96σ-11.96\sigma24.88σ24.88\sigma17.29σ17.29\sigma−54.46σ-54.46\sigma−5.49σ-5.49\sigma5.55σ5.55\sigma–
+ RS+xex_{\mathrm{e}}(−1.40%)(-1.40\%)(7.68%)(7.68\%)(1.80%)(1.80\%)(−11.45%)(-11.45\%)(−0.81%)(-0.81\%)(18.03%)(18.03\%)–
yes−8.52σ-8.52\sigma22.80σ22.80\sigma20.09σ20.09\sigma−54.34σ-54.34\sigma−4.74σ-4.74\sigma8.26σ8.26\sigma–
(−0.98%)(-0.98\%)(6.58%)(6.58\%)(1.98%)(1.98\%)(−10.66%)(-10.66\%)(−0.69%)(-0.69\%)(24.84%)(24.84\%)–
Λ\LambdaCDM + xex_{\mathrm{e}}no−0.81σ-0.81\sigma0.78σ0.78\sigma−0.30σ-0.30\sigma0.35σ0.35\sigma−0.89σ-0.89\sigma−0.22σ-0.22\sigma0.73σ0.73\sigma
(−0.10%)(-0.10\%)(0.24%)(0.24\%)(−0.03%)(-0.03\%)(0.16%)(0.16\%)(−0.18%)(-0.18\%)(−0.72%)(-0.72\%)(0.10%)(0.10\%)
Primary CMBΛ\LambdaCDMno−27.45σ-27.45\sigma29.17σ29.17\sigma20.71σ20.71\sigma−65.21σ-65.21\sigma−27.08σ-27.08\sigma5.64σ5.64\sigma–
+ RS+xex_{\mathrm{e}}(−3.20%)(-3.20\%)(9.01%)(9.01\%)(2.15%)(2.15\%)(−13.72%)(-13.72\%)(−4.00%)(-4.00\%)(18.30%)(18.30\%)–
+ σT+σR\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}}yes−26.27σ-26.27\sigma28.64σ28.64\sigma24.32σ24.32\sigma−67.38σ-67.38\sigma−26.35σ-26.35\sigma8.87σ8.87\sigma–
(−3.02%)(-3.02\%)(8.27%)(8.27\%)(2.40%)(2.40\%)(−13.22%)(-13.22\%)(−3.82%)(-3.82\%)(26.68%)(26.68\%)–
Λ\LambdaCDM + xex_{\mathrm{e}}no−17.08σ-17.08\sigma−5.17σ-5.17\sigma−0.06σ-0.06\sigma13.50σ13.50\sigma−1.26σ-1.26\sigma0.02σ0.02\sigma−13.25σ-13.25\sigma
(−2.09%)(-2.09\%)(−1.61%)(-1.61\%)(−0.01%)(-0.01\%)(−6.32%)(-6.32\%)(−0.26%)(-0.26\%)(0.05%)(0.05\%)(−1.91%)(-1.91\%)
yes−17.06σ-17.06\sigma−6.89σ-6.89\sigma1.24σ1.24\sigma−18.28σ-18.28\sigma−0.69σ-0.69\sigma1.42σ1.42\sigma−18.63σ-18.63\sigma
(−1.97%)(-1.97\%)(−2.02%)(-2.02\%)(0.12%)(0.12\%)(−6.08%)(-6.08\%)(−0.12%)(-0.12\%)(4.22%)(4.22\%)(−1.92%)(-1.92\%)
Λ\LambdaCDM + xex_{\mathrm{e}}no−0.02σ-0.02\sigma0.19σ0.19\sigma−0.15σ-0.15\sigma0.03σ0.03\sigma−0.45σ-0.45\sigma−0.13σ-0.13\sigma0.06σ0.06\sigma
+ σT+σR\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}}(−0.02%)(-0.02\%)(0.27%)(0.27\%)(−0.02%)(-0.02\%)(0.14%)(0.14\%)(−0.07%)(-0.07\%)(−0.41%)(-0.41\%)(0.08%)(0.08\%)

Table 4. Biases in units of σ\sigma, with the corresponding percentage shifts in parentheses, in the six Λ\LambdaCDM parameters and me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0} for PICO. The data (all including the RS signal) are generated under three scenarios: Λ\LambdaCDM, a modified recombination history with me/me,0=1.02m_{\mathrm{e}}/m_{\mathrm{e},0}=1.02 entering only through xex_{\mathrm{e}}, and the same variation also applied to σT\sigma_{\mathrm{T}} and σR\sigma_{\mathrm{R}}. Each data set is analysed with the models listed, without (“no”) or with (“yes”) modelling RS. When the model matches the data and RS is modelled, the biases vanish by construction and these rows are omitted.

Mij=(F−1)ij,(9)M_{ij} = \left(F^{-1}\right)_{ij}, \tag*{(9)}

and the marginalised 1σ1\sigma uncertainties of the parameters are σθi=Mii\sigma_{\theta_i} = \sqrt{M_{ii}}.

Moreover, we aim to quantify the bias in the parameter estimation if we incorrectly model the CMB power spectra, for instance ignore modelling the RS contribution in the covariance matrix CℓC_{\ell}. In that case, the spherical harmonic coefficients of the data are drawn from aℓm∼N(0,Cℓtrue)\boldsymbol{a}_{\ell m} \sim\mathcal{N}(0,C_{\ell}^{\mathrm{true}}) where CℓtrueC_{\ell}^{\mathrm{true}} is the true data covariance matrix. However, the log-likelihood, ln⁡L\ln\mathcal{L} assumes the model covariance CℓC_{\ell}, which differs from CℓtrueC_{\ell}^{\mathrm{true}}. Therefore, −ln⁡L-\ln\mathcal{L} is minimised with respect to the drifted parameter values, θ^i=θi,fid+δθi\hat{\theta}_{i}=\theta_{i,\mathrm{fid}}+\delta\theta_{i} with associated bias δθi\delta\theta_{i} for respective parameters. Minimising the first derivative of the log-likelihood, we can obtain the bias in each parameter as

δθi=∑j(F−1)ij∑ℓ(2ℓ+1)2fskyTr⁡[Cℓ−1∂Cℓ∂θjCℓ−1ΔCℓ],(10)\delta\theta_i = \sum_j (F^{-1})_{ij}\sum_\ell\frac{(2\ell+1)}{2}f_{\mathrm{sky}}\operatorname{Tr}\left[C_\ell^{-1}\frac{\partial C_\ell}{\partial\theta_j}C_\ell^{-1}\Delta C_\ell\right], \tag*{(10)}

where ΔCℓ=Cℓtrue−Cℓ(θfid)\Delta C_\ell=C_\ell^{\mathrm{true}}-C_\ell(\theta_{\mathrm{fid}}) is the difference in the covariance matrix due to the mismatch between the model and the data. A detailed derivation of Eq. (6) and Eq. (10) is shown in Appendix § 7.

Foreground modelling

We modelled the foreground contamination following the baseline model of the ACT DR6 analysis [60, 61]. In temperature, we include the kinematic and thermal Sunyaev–Zel’dovich (kSZ, tSZ) effects, the Poisson and clustered components of the cosmic infrared background (CIB), the tSZ×\timesCIB correlation, unresolved radio sources, Galactic dust and Galactic synchrotron emission. However, since we use only channels ≳90\gtrsim90 GHz in our analysis, we do not include a very detailed modelling of synchrotron (only for TETE), free-free and spinning dust emissions. The foreground power spectra are computed with the fgspectra{^2} code, assuming delta-function passbands at the channel frequencies. All components are normalised at ℓ0=3000\ell_0=3000 and ν0=150\nu_0=150 GHz except synchrotron. The synchrotron is normalised at ℓ0=80\ell_0=80 and ν0=30\nu_0=30 GHz. The functional forms and the values of the fiducial parameters are summarised in Table 1. The foreground spectra are added to the CMB and RS spectra in the covariance matrix of Eq. (7).

ComponentSpectraModel, DℓνiνjD_{\ell}^{\nu_i\nu_j} [μK2\mu\mathrm{K}^2]Fiducial parametersReferences
kSZTTTTakSZDℓ,ℓ0kSZa_{\mathrm{kSZ}}D_{\ell,\ell_0}^{\mathrm{kSZ}}akSZ=1.48a_{\mathrm{kSZ}}=1.48[60, 62]
tSZTTTTatSZDℓ,ℓ0tSZ(ℓℓ0)αtSZf(νi)f(νj)f2(ν0)a_{\mathrm{tSZ}}D_{\ell,\ell_0}^{\mathrm{tSZ}}\left(\frac{\ell}{\ell_0}\right)^{\alpha_{\mathrm{tSZ}}\frac{f(\nu_i)f(\nu_j)}{f^2(\nu_0)}}atSZ=3.3a_{\mathrm{tSZ}}=3.3, αtSZ=−0.53\alpha_{\mathrm{tSZ}}=-0.53[60, 63]
CIB, PoissonTTTTapℓ(ℓ+1)ℓ0(ℓ0+1)μ(νi;βp,Td)μ(νj;βp,Td)μ2(ν0;βp,Td)a_p\frac{\ell(\ell+1)}{\ell_0(\ell_0+1)}\frac{\mu(\nu_i;\beta_p,T_d)\mu(\nu_j;\beta_p,T_d)}{\mu^2(\nu_0;\beta_p,T_d)}ap=7.65a_p=7.65, βp=1.87\beta_p=1.87, Td=9.6T_d=9.6 K[60, 61]
CIB, clusteredTTTTacDℓ,ℓ0CIB−cμ(νi;βc,Td)μ(νj;βc,Td)μ2(ν0;βc,Td)a_cD_{\ell,\ell_0}^{\mathrm{CIB-c}}\frac{\mu(\nu_i;\beta_c,T_d)\mu(\nu_j;\beta_c,T_d)}{\mu^2(\nu_0;\beta_c,T_d)}ac=3.69a_c=3.69, βc=1.87\beta_c=1.87[64, 65]
tSZ×\timesCIBTTTT−ξatSZacDℓ,ℓ0tSZ×CIBf(νi)μ(νj;βc,Td)+f(νj)μ(νi;βc,Td)f(ν0)μ(ν0;βc,Td)-\xi\sqrt{a_{\mathrm{tSZ}}a_c}D_{\ell,\ell_0}^{\mathrm{tSZ}\times\mathrm{CIB}}\frac{f(\nu_i)\mu(\nu_j;\beta_c,T_d)+f(\nu_j)\mu(\nu_i;\beta_c,T_d)}{f(\nu_0)\mu(\nu_0;\beta_c,T_d)}ξ=0.09\xi=0.09[60, 66]
Radio sourcesTTTTasℓ(ℓ+1)ℓ0(ℓ0+1)(νiνjν02)βsg(νi)g(νj)g2(ν0)a_s\frac{\ell(\ell+1)}{\ell_0(\ell_0+1)}\left(\frac{\nu_i\nu_j}{\nu_0^2}\right)^{\beta_s}\frac{g(\nu_i)g(\nu_j)}{g^2(\nu_0)}as=2.86a_s=2.86, βs=−2.5\beta_s=-2.5[60, 61, 67]
Galactic dustTTTT, TETE, EEEEagXY(ℓ500)αdXYμ(νi;βd,Tdeff)μ(νj;βd,Tdeff)μ2(ν0;βd,Tdeff)a_g^{XY}\left(\frac{\ell}{500}\right)^{\alpha_d^{XY}}\frac{\mu(\nu_i;\beta_d,T_d^{\mathrm{eff}})\mu(\nu_j;\beta_d,T_d^{\mathrm{eff}})}{\mu^2(\nu_0;\beta_d,T_d^{\mathrm{eff}})}agTT=7.9a_g^{TT}=7.9, agTE=0.42a_g^{TE}=0.42, agEE=0.167a_g^{EE}=0.167, αdTT=−0.6\alpha_d^{TT}=-0.6, αdTE=−0.4\alpha_d^{TE}=-0.4, αdEE=−0.4\alpha_d^{EE}=-0.4, βd=1.5\beta_d=1.5, Tdeff=19.6T_d^{\mathrm{eff}}=19.6 K[60, 68]
SynchrotronTTTT, EEEEasyncXX(ℓ80)αsync(νiνjν02)βsyncg(νi)g(νj)g2(ν0)a_{\mathrm{sync}}^{XX}\left(\frac{\ell}{80}\right)^{\alpha_{\mathrm{sync}}}\left(\frac{\nu_i\nu_j}{\nu_0^2}\right)^{\beta_{\mathrm{sync}}}\frac{g(\nu_i)g(\nu_j)}{g^2(\nu_0)}, ν0=30\nu_0=30 GHzasyncTT=9.8a_{\mathrm{sync}}^{TT}=9.8, asyncEE=2.37a_{\mathrm{sync}}^{EE}=2.37, αsync=−0.81\alpha_{\mathrm{sync}}=-0.81, βsync=−3.1\beta_{\mathrm{sync}}=-3.1[69]

Table 1. Table 1: Foreground components included in the model, written as Dℓνiνj=ℓ(ℓ+1)Cℓνiνj/2πD_{\ell}^{\nu_i\nu_j}=\ell(\ell+1)C_{\ell}^{\nu_i\nu_j}/2\pi in CMB thermodynamic units for channels ii, jj. The pivot scale and pivot frequency are ℓ0=3000\ell_0=3000 and ν0=150\nu_0=150 GHz for all components except Synchrotron, which is normalised at ℓ0=80\ell_0=80 and ν0=30\nu_0=30 GHz. Where Dℓ,ℓ0XD_{\ell,\ell_0}^{X} denotes the template power spectrum of component X, normalised at the pivot multipole ℓ0\ell_0. Here g(ν)g(\nu) is the conversion factor from Rayleigh–Jeans units to CMB thermodynamic temperature unit, μ(ν;β,T)≡νβ+1g(ν)/(ehν/kBT−1)\mu(\nu;\beta,T)\equiv\nu^{\beta+1}g(\nu)/(e^{h\nu/k_{\mathrm{B}}T}-1) is a modified blackbody in CMB thermodynamic units, and f(ν)=xcoth⁡(x/2)−4f(\nu)=x\coth(x/2)-4 with x=hν/kBTCMBx=h\nu/k_{\mathrm{B}}T_{\mathrm{CMB}} is the tSZ spectral function. Fiducial values of the extragalactic and Galactic-dust parameters are taken from the given references.

Results

In this section we present the forecasted constraints on the cosmological parameters obtained with and without the inclusion of the RS signal, and quantify the parameter biases that arise when RS is neglected or when the recombination history is mis-modelled. We present the results for PICO [42] noise levels in the main text. The corresponding results for LiteBIRD + Planck [4, 41] are presented in Appendix § 8. The results for both LiteBIRD+Planck and PICO noise levels are obtained with fsky=0.7f_{\rm sky}=0.7, and with the foreground model of § 3.2, using the TTTT, TETE and EEEE spectra of all frequency channels and their cross-spectra. For PICO we set ℓmax⁡=3000\ell_{\max}=3000. We use ℓmax⁡=200\ell_{\max}=200 for LiteBIRD and for Planck we use ℓmax⁡=2500\ell_{\max}=2500 and ℓmax⁡=2000\ell_{\max}=2000 for temperature and polarisation, respectively.

We consider three cosmological models. Our baseline model is Λ\LambdaCDM with six parameters, {Ωbh2,Ωch2,ln⁡(1010As),H0,ns,τ}\{\Omega_bh^2,\Omega_ch^2,\ln(10^{10}A_s),H_0,n_s,\tau\} and fiducial values are set to the best-fit values of the Planck 2018 results [5]. Besides the Λ\LambdaCDM model, we study two extensions of the Λ\LambdaCDM model where we vary either me/me,0m_e/m_{e,0} or α/α0\alpha/\alpha_0 in two ways. In the model labelled Λ\LambdaCDM + xex_e, the variation of the fundamental constant enters only through the modified recombination history xe(z)x_e(z) computed with RECFAST++ [48, 70], while the Thomson and Rayleigh cross-sections are held at their standard values. In the model labelled Λ\LambdaCDM + xe+σT+σRx_e+\sigma_T+\sigma_R, the variation is propagated consistently to the scattering cross-sections as well, i.e. σT∝α2me−2\sigma_T\propto\alpha^2m_e^{-2} and σR∝ν4α−6me−6\sigma_R\propto\nu^4\alpha^{-6}m_e^{-6} (cf. Eq. (5)). The fiducial values are me/me,0=α/α0=1m_e/m_{e,0}=\alpha/\alpha_0=1 for the forecasts of parameter constraints in § 4.1, while the bias analysis of § 4.2 assumes me/me,0=1.02m_e/m_{e,0}=1.02 (or α/α0=1.01\alpha/\alpha_0=1.01) in the simulated data.

Forecast constraints with and without Rayleigh scattering

For PICO: Table 2 quotes the marginalised 68% uncertainties of the related parameters for all models, and Figure 6 shows the corresponding fractional improvement, 1−σθwith RS/σθwithout RS1-\sigma_{\theta}^{\rm with\ RS}/\sigma_{\theta}^{\rm without\ RS}, obtained by adding the RS signal to the primary CMB for PICO. The improvement factors are quoted in Table 3.

ModelΩbh2\Omega_b h^2Ωch2\Omega_c h^2ln⁡(1010As)\ln(10^{10}A_s)H0H_0nsn_sτ\taume/me,0m_e/m_{e,0}α/α0\alpha/\alpha_0
Λ\LambdaCDM1.46.54.96.71.87.3––
Λ\LambdaCDM + xex_e (for mem_e)6.55.56.531.616.49.031.6–
Λ\LambdaCDM + xex_e (for α\alpha)1.46.57.830.716.49.2–30.9
Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R (for mem_e)30.728.36.631.21.33.230.9–
Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R (for α\alpha)0.95.04.37.42.85.9–4.2

Table 3. Fractional improvement (in %), 1−σθwith RS/σθwithout RS1-\sigma_\theta^{\mathrm{with\ RS}}/\sigma_\theta^{\mathrm{without\ RS}}, in the marginalised 1σ1\sigma uncertainties quoted in Table 2 when the RS signal is included.

ModelΩbh2\Omega_bh^2Ωch2\Omega_ch^2ln⁡(1010As)\ln(10^{10}A_s)H0H_0nsn_sτ\taume/me,0m_e/m_{e,0}α/α0\alpha/\alpha_0
[×10−5][\times10^{-5}][×10−4][\times10^{-4}][×10−3][\times10^{-3}][×10−3][\times10^{-3}][×10−3][\times10^{-3}][×10−3][\times10^{-3}][×10−3][\times10^{-3}][×10−4][\times10^{-4}]
Λ\LambdaCDM
Without RS2.73.93.20.151.41.8––
With RS2.63.63.00.141.41.6––
Λ\LambdaCDM + xex_e
Without RS2.84.03.20.332.11.81.6–
With RS2.63.73.00.231.81.61.1–
Without RS2.73.83.20.312.11.7–5.6
With RS2.73.63.00.211.71.6–3.8
Λ\LambdaCDM + xe+σT+σRx_e+\sigma_T+\sigma_R
Without RS31183.32.81.41.714–
With RS21133.11.91.41.79.5–
Without RS2.73.93.30.242.11.8–4.1
With RS2.73.63.20.232.11.7–3.9

Table 2. 68% confidence intervals for the six Λ\LambdaCDM parameters and the additional parameter me/me,0m_e/m_{e,0} (and α/α0\alpha/\alpha_0) obtained for the Λ\LambdaCDM model and its possible extensions, comparing analyses with and without including the RS signal alongside the primary CMB for PICO.

Figure 6. Improvement in the error bars 1−σθwith RS/σθwithout RS1-\sigma_\theta^{\mathrm{with\ RS}}/\sigma_\theta^{\mathrm{without\ RS}} of the cosmological parameters and electron mass (left) or fine-structure constant (right) when we include the RS signal alongside the primary CMB signal for PICO. The blue, red and orange bars correspond to the Λ\LambdaCDM and two modified recombination cases, respectively.

We first investigate the improvement in the constraints on the six parameters for the standard Λ\LambdaCDM model. Including RS tightens the constraints on all parameters, but by different amounts (see Table 3). The largest gains are for τ\tau (7.3%), H0H_0 (6.7%) and Ωch2\Omega_ch^2 (6.5%), followed by ln⁡(1010As)\ln(10^{10}A_s) (4.9%). The gains for nsn_s (1.8%) and Ωbh2\Omega_bh^2 (1.4%) are marginal. The overall improvement is modest because in Λ\LambdaCDM, the recombination history is fixed by standard atomic physics, so RS does not open a new degree of freedom. It only provides a frequency-dependent view of the same acoustic structure. Moreover, the RS contribution becomes significant only at high frequencies, where instrumental noise and dust contamination are dominant. The improvement differs between parameters because RS adds information mainly through two effects. With RS, the peak of visibility shifts across frequency channels. Each channel therefore probes the sound horizon at a slightly different redshift around last scattering, producing a frequency-dependent shift of the acoustic peaks whose size depends on H(z)H(z), and hence on the matter density. This constrains Ωch2\Omega_c h^2 directly, and H0H_0 through their degeneracy, which is why the precision of both parameters improves by a similar amount (≃6.5%\simeq6.5\%). Second, RS boosts the EEEE power at 10≲ℓ≲30010 \lesssim\ell\lesssim300 (see Figure 4). This additional polarisation information tightens the constraints on τ\tau (7.3%) and, through its correlations with ln⁡(1010As)\ln(10^{10}A_s), the constraint on the primordial amplitude also improves by 4.9%. In contrast, RS produces no tilt-like change in the total power spectra in the Λ\LambdaCDM framework. Furthermore, although RS depends on the baryon density since nH∝Ωbh2n_{\mathrm{H}} \propto\Omega_b h^2, in the Λ\LambdaCDM framework this information is already encoded in the primary CMB acoustic peak structure. Since the primary CMB already measures nsn_s and Ωbh2\Omega_b h^2 precisely, from the overall tilt and the relative peak heights, their precision improves only by ≲2%\lesssim2\% with the addition of RS. Overall, RS shrinks the Λ\LambdaCDM posteriors without changing the degeneracy directions, as shown in the contour plots represented on the top panel of Figure 7.

Figure 7. 1σ1\sigma and 2σ2\sigma confidence contours obtained for the Λ\LambdaCDM model and its extensions when we analyse the data with primary CMB only (red) and primary CMB + RS signal (blue) for PICO.

Next we investigate the two cases of modified recombination history. First we discuss the results for the case in which only the ionisation history xe(z)x_e(z) is modified by varying mem_e or α\alpha, but scattering rates are unchanged. The impact of RS is considerably larger for some of the parameters once the recombination history is allowed to vary (see Table 3). For instance, for varying electron mass, the uncertainty on me/me,0m_e/m_{e,0} improves from 0.0016 to 0.0011 and that on H0H_0 from 0.33 to 0.23 km s−1^{-1}Mpc−1^{-1}, i.e. by ≃31%\simeq31\% in both parameters. Varying α\alpha leads to a very similar pattern: σ(α/α0)\sigma(\alpha/\alpha_0) decreases from 0.0006 to 0.0004 and σ(H0)\sigma(H_0) from 0.31 to 0.21 km s−1^{-1}Mpc−1^{-1} which is a ≃31%\simeq31\% improvement. These are the parameters RS probes most directly. In the primary CMB, a change in mem_e or α\alpha shows up only through the shift of the Thomson visibility. The changes in the Rayleigh visibilities are, however, more pronounced and frequency dependent. The multi-frequency RS signal provides additional, partly independent information on the timing and duration of recombination,

beyond what is available from the primary CMB alone. An earlier recombination caused by increased mem_e or α\alpha values reduces the sound horizon, and with the angular size of the sound horizon, θ∗\theta_* fixed by the data, this is compensated almost exactly by a larger H0H_0, giving the near-perfect mem_e–H0H_0 (and α\alpha–H0H_0) degeneracy as illustrated in the middle-left (and middle-right) panel of Figure 7. Since RS can sample the recombination timing, this gives improved constraints on mem_e and α\alpha and hence H0H_0 as well through mem_e–H0H_0 (or α\alpha–H0H_0) correlation. Therefore H0H_0 improves by almost the same amount as me/me,0m_e/m_{e,0} or α/α0\alpha/\alpha_0. The other parameters improve according to how strongly they are degenerate with the fundamental constants.

The constraint on the spectral index nsn_s improves by 16.4%, which is substantially more than in Λ\LambdaCDM (1.8%). This improvement is closely related to the changes in the damping tail of CMB power spectra in modified recombination scenarios. A change in the fundamental constant shifts the epoch of recombination and hence modifies the photon diffusion scale, thereby changing the suppression of power at high ℓ\ell. A change in nsn_s also alters the relative amount of power between large and small scales. From the primary CMB alone, these effects can therefore be partially degenerate, producing a correlation between nsn_s and the fundamental constants. RS helps to break this degeneracy because its contribution to the damping is frequency-dependent. In particular, RS adds extra diffusion damping that grows as ν4\nu^4, and therefore its damping effect on the anisotropies varies with frequency, while the primordial tilt is common to all frequency channels. Comparing the damping tail across frequencies therefore helps distinguish changes in the damping caused by the modified recombination history from those associated with nsn_s. This additional information reduces the nsn_s–mem_e (or nsn_s–α\alpha) degeneracy, leading to an improvement in the constraint on nsn_s. The slightly larger gain for α\alpha is consistent with its stronger impact on the recombination history and, consequently, on the damping and visibility structure. In contrast, Ωbh2\Omega_bh^2, ln⁡(1010As)\ln(10^{10}A_s) and τ\tau improve by 5–9%, similar to Λ\LambdaCDM. Because the scattering rates are unchanged in this model, the relative peak heights still determine the matter density and the amplitude independently of mem_e or α\alpha. These parameters are therefore only weakly correlated with the fundamental constants (quantitatively the correlation coefficients, ∣ρ∣≲0.3|\rho| \lesssim0.3), and their gain comes mainly from the same effects as in Λ\LambdaCDM. Ωbh2\Omega_bh^2 improves somewhat more than in Λ\LambdaCDM (6.5% for mem_e). This is because a shift of the recombination redshift changes the baryon loading, R(z∗)∝Ωbh2R(z_*) \propto\Omega_bh^2 at decoupling. Since the RS opacity is linked with nH∝Ωbh2n_H \propto\Omega_bh^2 this introduces a mild correlation between Ωbh2\Omega_bh^2 and mem_e, which gives RS more leverage on Ωbh2\Omega_bh^2 than in the standard Λ\LambdaCDM case. For α\alpha, Ωbh2\Omega_bh^2 is essentially uncorrelated with the fundamental constant already in the primary CMB, and its improvement is minimal.

When the variation of mem_e is consistently propagated to the scattering cross-sections, the constraints from the primary CMB degrade substantially. In particular, the marginalised uncertainties increase to σ(me/me,0)=0.0138\sigma(m_e/m_{e,0}) = 0.0138, σ(H0)=2.83 km s−1Mpc−1\sigma(H_0) = 2.83\ {\rm km\ s^{-1}Mpc^{-1}}, and σ(Ωbh2)=3.1×10−4\sigma(\Omega_bh^2) = 3.1 \times10^{-4}, σ(Ωch2)=1.81×10−3\sigma(\Omega_ch^2) = 1.81 \times10^{-3}, roughly an order of magnitude weaker than in the xex_e-only model. This happens because two effects of a larger mem_e partly cancel. First, recombination happens earlier, (1+z∗)∝me(1 + z_*) \propto m_e, when the Universe is denser, so there are more electrons available to scatter photons. Second, each electron scatters less, because σT∝me−2\sigma_T \propto m_e^{-2}. The net Thomson scattering rate at recombination therefore changes only mildly, roughly as ΓT(z∗)=neσT∝me\Gamma_T(z_*) = n_e\sigma_T \propto m_e. This remaining change can be absorbed by a faster expansion, H(z∗)∝Ωm(1+z∗)3H(z_*) \propto\sqrt{\Omega_m(1 + z_*)^3}, obtained by increasing Ωbh2\Omega_bh^2 and Ωch2\Omega_ch^2, with H0H_0 adjusted to keep the angular scale of the acoustic peaks fixed. These adjustments leave the ratio ΓT/H\Gamma_T/H unchanged, which decides the shape of the Thomson visibility function [32, 71]. As a result, the primary CMB can hardly distinguish a change in mem_e from a combined shift of H0H_0, Ωbh2\Omega_bh^2 and Ωch2\Omega_ch^2, and these four parameters become strongly degenerate, as demonstrated in the bottom-left panel of Figure 7.

Rayleigh scattering does not completely break this scaling. Although σR∝me−6\sigma_R \propto m_e^{-6} at fixed observed frequency ν\nu, the photon frequency at the last scattering epoch is νphys=ν(1+z∗)∝me\nu_{\rm phys} = \nu(1 + z_*) \propto m_e. Consequently, at a fixed observed frequency, σR(ν,z∗)∝ν4(1+z∗)4me−6∝ν4me−2\sigma_R(\nu,z_*) \propto\nu^4(1 + z_*)^4m_e^{-6} \propto\nu^4m_e^{-2}. Since nH∝(1+z∗)3∝me3n_H \propto(1 + z_*)^3 \propto m_e^3, the Rayleigh scattering rate scales approximately as ΓR(ν)∝nHσR∝ν4me\Gamma_R(\nu) \propto n_H\sigma_R \propto\nu^4m_e, which has the same leading dependence on mem_e as the Thomson scattering rate, ΓT∝me\Gamma_T \propto m_e. Thus, readjustment of other parameters to keep the primary CMB unchanged also preserves RS. However, the multi-frequency RS spectra provide additional, partly independent measurements of the same acoustic and damping structure, which are not completely redundant with the primary CMB. These modes provide additional information along the otherwise poorly constrained degeneracy direction, reducing its allowed extent without substantially changing its orientation, as shown in the bottom-left panel of Figure 7. As a result, the parameters that are most strongly aligned with this degeneracy improve by nearly the same percentage. In contrast, ln⁡(1010As)\ln(10^{10}A_s), τ\tau, and nsn_s are less strongly aligned with this degeneracy and improve by only 1–7%.

The behaviour for α\alpha is qualitatively different. Although α\alpha changes both the atomic binding energies and the scattering cross-sections, the different processes governing recombination do not scale with a common power of α\alpha. In particular, the rates associated with Lyman-α\alpha escape, two-photon decay, and photoionisation have different α\alpha dependences from σT∝α2\sigma_{\mathrm{T}} \propto\alpha^{2}. Therefore, no analogous rescaling of Ωbh2\Omega_{\mathrm{b}}h^{2}, Ωch2\Omega_{\mathrm{c}}h^{2}, and H0H_{0} can preserve the Thomson scattering history throughout recombination. The primary CMB consequently already constrains α/α0\alpha/\alpha_{0} tightly, with σ(α/α0)=0.0004\sigma(\alpha/\alpha_{0}) = 0.0004, leaving little degeneracy for RS to break. The inclusion of RS therefore produces only modest additional improvements, at the level of ≲8%\lesssim8\%.

For LiteBIRD + Planck: The corresponding results for LiteBIRD + Planck are given in Appendix § 8: Table 6 and Table 7 present the 68% confidence limits and relative improvement in the error bars of the parameters, respectively. Figure 10 displays the error-bar improvements as quoted in Table 7. The trends are the same as for PICO, but the gains are much smaller for most cases. In Λ\LambdaCDM no parameter improves by more than 0.7%, and in the Λ\LambdaCDM + xex_{\mathrm{e}} model the largest gains are 1.9% for H0H_{0} and 1.4% for me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0} (similarly 1.2% and 0.7% for H0H_{0} and α/α0\alpha/\alpha_{0}, respectively). This is expected as the RS signal is concentrated at high frequencies, where the noise of the Planck channels limits its detectability, and partially because we conservatively restrict the LiteBIRD multipole range to ℓmax⁡=200\ell_{\max}=200 in this analysis. The exception is the Λ\LambdaCDM + xe+σT+σRx_{\mathrm{e}} + \sigma_{\mathrm{T}} + \sigma_{\mathrm{R}} model for varying mem_{\mathrm{e}}, where Ωbh2\Omega_{\mathrm{b}}h^{2}, Ωch2\Omega_{\mathrm{c}}h^{2}, H0H_{0} and me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0} improve by 55–58%. With the higher noise of LiteBIRD + Planck as compared to PICO, the primary CMB alone constrains the degeneracies among these four parameters discussed above even more weakly, with σ(me/me,0)=0.046\sigma(m_{\mathrm{e}}/m_{\mathrm{e},0}) = 0.046 and σ(H0)=9.8 km s−1 Mpc−1\sigma(H_{0}) = 9.8\ \mathrm{km\ s^{-1}\ Mpc^{-1}}. Even the small amount of information that RS adds along this direction of me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0} therefore reduces these uncertainties substantially, to 0.019 and 4.1 km s−1 Mpc−14.1\ \mathrm{km\ s^{-1}\ Mpc^{-1}}. For α\alpha, where no such degeneracy exists as also discussed above, the improvement stays below 1%.

ModelΩbh2\Omega_b h^2Ωch2\Omega_c h^2ln⁡(1010As)\ln(10^{10}A_s)H0H_0nsn_sτ\taume/me,0m_e/m_{e,0}α/α0\alpha/\alpha_0
Λ\LambdaCDM
Without RS9.38.23.90.362.41.9––
With RS9.28.13.90.362.41.9––
Λ\LambdaCDM + xex_e
Without RS9.89.94.10.893.81.95.0–
With RS9.79.84.10.883.81.94.9–
Without RS9.39.44.40.794.41.9–17
With RS9.29.44.40.784.31.9–17
Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R
Without RS103574.39.82.61.946–
With RS44264.04.12.41.919–
Without RS9.58.94.40.594.21.9–13
With RS9.48.84.40.584.21.9–13

Table 6. 68% confidence intervals for the six Λ\LambdaCDM parameters and the additional parameter me/me,0m_e/m_{e,0} (and α/α0\alpha/\alpha_0) obtained for the Λ\LambdaCDM model and its possible extensions, comparing analyses with and without including the RS signal along with the Primary CMB for LiteBIRD + Planck.

ModelΩbh2\Omega_b h^2Ωch2\Omega_c h^2ln⁡(1010As)\ln(10^{10}A_s)H0H_0nsn_sτ\taume/me,0m_e/m_{e,0}α/α0\alpha/\alpha_0
Λ\LambdaCDM0.40.60.00.70.20.2––
Λ\LambdaCDM + xex_e (for mem_e)0.90.20.11.90.60.11.4–
Λ\LambdaCDM + xex_e (for α\alpha)0.50.20.21.20.40.2–0.7
Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R (for mem_e)57.454.56.757.86.61.557.6–
Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R (for α\alpha)0.30.40.00.70.10.2–0.2

Table 7. Fractional improvement (in %), 1−σwith RS/σwithout RS1-\sigma_{\mathrm{with\ RS}}/\sigma_{\mathrm{without\ RS}}, in the marginalised 1σ1\sigma uncertainties quoted in Table 6 when the RS signal is included for LiteBIRD + Planck.

Figure 10. Improvement in the error bars 1−σθwith RS/σθwithout RS1-\sigma_{\theta}^{\mathrm{with\ RS}}/\sigma_{\theta}^{\mathrm{without\ RS}} of the cosmological parameters and the electron mass (left) or the fine-structure constant (right) when we include the RS signal with the primary CMB signal for LiteBIRD + Planck. The blue, red and orange bars correspond to the Λ\LambdaCDM and the two modified recombination cases, respectively. For clarity, in the left-hand panel, the improvements for the Λ\LambdaCDM + xe+σT+σRx_{\mathrm{e}}+\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}} model are separately plotted since the values are much larger than in the other two cases.

Parameter biases

We now quantify the biases in the inferred parameters using Eq. (10). We generate three sets of mock data, all including the RS signal: (i) standard Λ\LambdaCDM; (ii) a modified recombination history in which the variation of the fundamental constant enters only through xe(z)x_e(z); and (iii) the same variation applied also to σT\sigma_T and σR\sigma_R. For the electron mass we adopt me/me,0=1.02m_e/m_{e,0} = 1.02 and for the fine-structure constant α/α0=1.01\alpha/\alpha_0 = 1.01, with all other parameters at their fiducial values. Each data set is analysed with Λ\LambdaCDM and with the extended models, with and without modelling the RS contribution. In each case, the bias is projected with the Fisher matrix of the model used in the analysis, and is quoted in units of that model’s marginalised 1σ1\sigma uncertainty (and in % in the parentheses). The results are shown in Table 4 (for mem_e) and 5 (for α\alpha) and illustrated in Figure 8 and Figure 9, where each point marks the recovered value θfid+δθ\theta_{\rm fid} + \delta\theta with the corresponding 1σ1\sigma error bar, and the vertical line marks the input value. When the model matches the data and RS is modelled, the bias vanishes by construction. Note that Eq. (10) is a linear approximation, valid for ∣δθ∣≲|\delta\theta| \lesssim a few σ\sigma; biases of tens of σ\sigma found below should therefore be read as indicating a severe mis-modelling rather than as precise values.

Figure 8. Data: Primary CMB + RS; Data: Primary CMB + RS + xex_e; Data: Primary CMB + RS + xe+σT+σRx_e + \sigma_T + \sigma_R; Model: Λ\LambdaCDM; Model: Λ\LambdaCDM + xex_e; Model: Λ\LambdaCDM + xe+σT+σRx_e + \sigma_T + \sigma_R; Without RS; With RS. Panel labels: Ωbh2\Omega_bh^2, Ωch2\Omega_ch^2, ln⁡(1010As)\ln(10^{10}A_s), H0H_0, nsn_s, τ\tau, and me/me,0m_e/m_{e,0}. Figure 8: Biased estimated values, θfid+δθ\theta_{\rm fid}+\delta\theta, with their marginalised 1σ1\sigma error bars, of the six Λ\LambdaCDM parameters and me/me,0m_e/m_{e,0} for PICO, when the simulated data are analysed with Λ\LambdaCDM and its extensions. The shaded horizontal bands correspond to the three simulated data sets, all including the RS signal: Λ\LambdaCDM (blue band), a modified recombination history with me/me,0=1.02m_e/m_{e,0}=1.02 entering only through xex_e (orange band), and the same variation also applied to σT\sigma_T and σR\sigma_R (cyan band). Within each band, the colour of the points indicates the model used in the analysis: Λ\LambdaCDM (blue), Λ\LambdaCDM + xex_e (red), and Λ\LambdaCDM + xe+σT+σRx_e+\sigma_T+\sigma_R (orange). Filled and open markers denote analyses without and with modelling RS, respectively. Each vertical line marks the input value used to generate the data.

Figure 9. Biased estimated values, θfid+δθ\theta_{\rm fid}+\delta\theta, with their marginalised 1σ1\sigma error bars, of the six Λ\LambdaCDM parameters and α/α0\alpha/\alpha_0 for PICO. The data, generated with α/α0=1.01\alpha/\alpha_0=1.01, include the RS signal, with the variation of α\alpha entering only through xex_e (orange band) or also through σT\sigma_T and σR\sigma_R (cyan band). Colours, markers and vertical lines are as in Figure 8. The Λ\LambdaCDM-data case, identical to Figure 8, is not repeated in this figure.

For PICO: We begin by discussing the parameter biases expected for the PICO instrument configuration.

Neglecting the RS in Λ\LambdaCDM model: If the data follow Λ\LambdaCDM but RS is omitted from the model, all biases remain below 1σ1\sigma. The largest shifts are in Ωbh2\Omega_b h^2 (−0.95σ-0.95\sigma), Ωch2\Omega_c h^2 (+0.59σ+0.59\sigma) and H0H_0 (−0.57σ-0.57\sigma), corresponding to fractional shifts of ≲0.2%\lesssim0.2\%. Although small, a bias of nearly 1σ1\sigma in Ωbh2\Omega_b h^2 would become relevant when combined with external data such as BBN, showing that RS must be modelled at the sensitivity of next-generation experiments.

Fitting a modified recombination history with Λ\LambdaCDM model: If the recombination history in the data is modified only through xex_e (data set ii) and the analysis assumes Λ\LambdaCDM, the inferred parameters are severely biased, whether or not RS is modelled. An increase in mem_e or α\alpha shifts recombination to earlier times and reduces the sound horizon; Λ\LambdaCDM can mimic this only by lowering H0H_0 to keep the angular acoustic scale fixed, together with a larger Ωch2\Omega_c h^2 and a correlated increase of ln⁡(1010As)\ln(10^{10}A_s) and τ\tau. For me/me,0=1.02m_e/m_{e,0} = 1.02, H0H_0 is shifted by ≃−54σ\simeq-54\sigma (≃−11%\simeq-11\%, i.e. H0≃60 km s−1 Mpc−1H_0 \simeq60\,\mathrm{km\,s^{-1}\,Mpc^{-1}}), Ωch2\Omega_c h^2 by ≃+23\simeq+23 to +25σ+25\sigma (≃7%\simeq7\%), ln⁡(1010As)\ln(10^{10}A_s) by ≃+17\simeq+17 to +20σ+20\sigma and Ωbh2\Omega_bh^2 by ≃−9\simeq-9 to −12σ-12\sigma, while τ\tau is shifted by 18–25%. For α/α0=1.01\alpha/\alpha_0=1.01, which shifts the atomic energy levels (∝α2me\propto\alpha^2m_e) by the same 2%, the shifts are even larger, reflecting the additional, non-uniform α\alpha-dependence of the recombination rates: H0H_0 by ≃−78σ\simeq-78\sigma (H0≃56H_0\simeq56–57 km s−1^{-1} Mpc−1^{-1}), Ωch2\Omega_ch^2 by ≃+40σ\simeq+40\sigma (12–14%) and τ\tau by 29–42%. In this case the model mis-specification dominates, and modelling RS within a Λ\LambdaCDM fit does not reduce the bias; it only increases its significance, since RS tightens the error bars. Once the fundamental constant is included in the model (Λ\LambdaCDM +xe+x_e), the biases from neglecting RS fall to ≲1σ\lesssim1\sigma for all parameters (at most −0.89σ-0.89\sigma in nsn_s and 0.73σ0.73\sigma in me/me,0m_e/m_{e,0}, and −1.05σ-1.05\sigma in Ωbh2\Omega_bh^2 and 0.88σ0.88\sigma in α/α0\alpha/\alpha_0), and they vanish when RS is also modelled.

Mis-modelling the scattering cross-section dependence: For data set (iii), in which the variation also rescales σT\sigma_T and σR\sigma_R, a Λ\LambdaCDM fit again gives very large biases. For mem_e these reach ≃−65σ\simeq-65\sigma in H0H_0 (≃−13.5%\simeq-13.5\%), ≃+29σ\simeq+29\sigma in Ωch2\Omega_ch^2, ≃−27σ\simeq-27\sigma in both Ωbh2\Omega_bh^2 and nsn_s, and 18–27% in τ\tau; for α\alpha, H0H_0 is shifted by ≃−70σ\simeq-70\sigma (≃−15%\simeq-15\%) and Ωch2\Omega_ch^2 by ≃+37σ\simeq+37\sigma, whereas Ωbh2\Omega_bh^2 and nsn_s are much less affected (≲3σ\lesssim3\sigma).

More importantly, analysing these data with the Λ\LambdaCDM +xe+x_e model, i.e. allowing for a varying constant but neglecting its effect on the cross-sections, leads to qualitatively different outcomes for mem_e and α\alpha. For mem_e, large biases remain: ≃−17σ\simeq-17\sigma in Ωbh2\Omega_bh^2, −13.5σ-13.5\sigma (−18.3σ-18.3\sigma) in H0H_0 and −13.3σ-13.3\sigma (−18.6σ-18.6\sigma) in me/me,0m_e/m_{e,0} without (with) RS. The recovered electron mass is shifted by ≃−1.9%\simeq-1.9\%, to me/me,0≃1.000m_e/m_{e,0}\simeq1.000, so that a true 2% variation would be inferred as consistent with no variation at all. This is a direct consequence of the scaling degeneracy discussed in § 4.1: the reduction of σT∝me−2\sigma_{\mathrm{T}} \propto m_{\mathrm{e}}^{-2} largely compensates the earlier recombination, so that the data resemble a standard recombination history with shifted Ωbh2\Omega_{\mathrm{b}}h^2 and H0H_0. The fractional shifts are almost identical with and without RS, but their significance increases with RS because the error bars are tighter. For α\alpha, where no such compensation exists, the recovered value remains close to the input: α/α0≃1.0135\alpha/\alpha_0 \simeq1.0135 (+6.5σ+6.5\sigma) without RS and ≃1.0092\simeq1.0092 (−2.1σ-2.1\sigma) with RS, accompanied by biases of ≃5\simeq5–6σ6\sigma in Ωbh2\Omega_{\mathrm{b}}h^2 and in H0H_0 (without RS) or nsn_{\mathrm{s}} (with RS). In both cases, only the fully consistent Λ\LambdaCDM +xe+σT+σR+x_{\mathrm{e}}+\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}} model recovers the input parameters; neglecting RS in this model produces biases of ≲0.5σ\lesssim0.5\sigma for mem_{\mathrm{e}} and ≲1.2σ\lesssim1.2\sigma for α\alpha.

For LiteBIRD + Planck: The corresponding biases for LiteBIRD + Planck are listed in Table 8 (for mem_{\mathrm{e}}) and 9 (for α\alpha) in Appendix § 8. The corresponding biases are also shown in Figure 11 and 12 respectively. The pattern is the same as for PICO, but the significances are smaller because the uncertainties are larger. Neglecting RS in a Λ\LambdaCDM analysis of Λ\LambdaCDM data gives biases of at most 0.13σ0.13\sigma, so at this sensitivity omitting RS is harmless for the standard model. In contrast, fitting Λ\LambdaCDM to data with a modified recombination history still gives strong biases as expected. When the cross-section scaling is neglected (data set for Λ\LambdaCDM +xe+σT+σR+x_{\mathrm{e}}+\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}} analysed with Λ\LambdaCDM +xe+x_{\mathrm{e}}), the scaling degeneracy again biases the electron mass: me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0} is shifted by ≃−4σ\simeq-4\sigma (≃−2.1%\simeq-2.1\%, i.e. me/me,0≃0.999m_{\mathrm{e}}/m_{\mathrm{e},0} \simeq0.999), together with ≃−4.5σ\simeq-4.5\sigma in Ωbh2\Omega_{\mathrm{b}}h^2 and ≃−4.9σ\simeq-4.9\sigma in H0H_0. For α\alpha, the corresponding biases remain ≲2σ\lesssim2\sigma. Once the correct model is used, neglecting RS biases all parameters by less than 0.15σ0.15\sigma.

DataModelRSΩbh2\Omega_{\mathrm{b}}h^{2}Ωch2\Omega_{\mathrm{c}}h^{2}ln⁡(1010As)\ln(10^{10}A_{\mathrm{s}})H0H_{0}nsn_{\mathrm{s}}τ\taume/me,0m_{\mathrm{e}}/m_{\mathrm{e},0}
Primary CMB + RSΛ\LambdaCDMno−0.05σ-0.05\sigma<br>(−0.02%)(-0.02\%)0.00σ0.00\sigma<br>(0.00%)(0.00\%)−0.07σ-0.07\sigma<br>(−0.01%)(-0.01\%)0.00σ0.00\sigma<br>(0.00%)(0.00\%)−0.13σ-0.13\sigma<br>(−0.03%)(-0.03\%)−0.01σ-0.01\sigma<br>(−0.04%)(-0.04\%)–
Primary CMB + RS+xex_{\mathrm{e}}Λ\LambdaCDMno−3.88σ-3.88\sigma<br>(−1.61%)(-1.61\%)0.40σ0.40\sigma<br>(0.27%)(0.27\%)2.72σ2.72\sigma<br>(0.35%)(0.35\%)−12.23σ-12.23\sigma<br>(−6.60%)(-6.60\%)2.83σ2.83\sigma<br>(0.71%)(0.71\%)0.40σ0.40\sigma<br>(1.38%)(1.38\%)–
yes−3.76σ-3.76\sigma<br>(−1.55%)(-1.55\%)0.31σ0.31\sigma<br>(0.21%)(0.21\%)2.80σ2.80\sigma<br>(0.36%)(0.36\%)−12.23σ-12.23\sigma<br>(−6.55%)(-6.55\%)3.02σ3.02\sigma<br>(0.76%)(0.76\%)0.46σ0.46\sigma<br>(1.59%)(1.59\%)–
Λ\LambdaCDM + xex_{\mathrm{e}}no−0.08σ-0.08\sigma<br>(−0.04%)(-0.04\%)−0.05σ-0.05\sigma<br>(−0.05%)(-0.05\%)−0.05σ-0.05\sigma<br>(−0.01%)(-0.01\%)−0.08σ-0.08\sigma<br>(−0.11%)(-0.11\%)−0.02σ-0.02\sigma<br>(−0.01%)(-0.01\%)0.00σ0.00\sigma<br>(−0.02%)(-0.02\%)−0.09σ-0.09\sigma<br>(−0.05%)(-0.05\%)
Primary CMB + RS+xex_{\mathrm{e}} + σT+σR\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}}Λ\LambdaCDMno−8.27σ-8.27\sigma<br>(−3.42%)(-3.42\%)1.37σ1.37\sigma<br>(0.93%)(0.93\%)2.09σ2.09\sigma<br>(0.27%)(0.27\%)−16.16σ-16.16\sigma<br>(−8.71%)(-8.71\%)−2.94σ-2.94\sigma<br>(−0.74%)(-0.74\%)0.15σ0.15\sigma<br>(0.53%)(0.53\%)–
yes−8.15σ-8.15\sigma<br>(−3.36%)(-3.36\%)1.26σ1.26\sigma<br>(0.86%)(0.86\%)2.18σ2.18\sigma<br>(0.28%)(0.28\%)−16.16σ-16.16\sigma<br>(−8.65%)(-8.65\%)−2.74σ-2.74\sigma<br>(−0.69%)(-0.69\%)0.24σ0.24\sigma<br>(0.83%)(0.83\%)–
Λ\LambdaCDM + xex_{\mathrm{e}}no−4.57σ-4.57\sigma<br>(−2.02%)(-2.02\%)−2.58σ-2.58\sigma<br>(−2.19%)(-2.19\%)−0.18σ-0.18\sigma<br>(−0.02%)(-0.02\%)−4.91σ-4.91\sigma<br>(−6.66%)(-6.66\%)−0.03σ-0.03\sigma<br>(−0.01%)(-0.01\%)0.09σ0.09\sigma<br>(0.32%)(0.32\%)−4.17σ-4.17\sigma<br>(−2.13%)(-2.13\%)
yes−4.50σ-4.50\sigma<br>(−1.98%)(-1.98\%)−2.52σ-2.52\sigma<br>(−2.13%)(-2.13\%)−0.15σ-0.15\sigma<br>(−0.02%)(-0.02\%)−4.84σ-4.84\sigma<br>(−6.52%)(-6.52\%)−0.05σ-0.05\sigma<br>(−0.02%)(-0.02\%)0.08σ0.08\sigma<br>(0.28%)(0.28\%)−4.07σ-4.07\sigma<br>(−2.07%)(-2.07\%)
Λ\LambdaCDM + xex_{\mathrm{e}} + σT+σR\sigma_{\mathrm{T}}+\sigma_{\mathrm{R}}no0.01σ0.01\sigma<br>(0.02%)(0.02\%)0.01σ0.01\sigma<br>(0.05%)(0.05\%)−0.06σ-0.06\sigma<br>(−0.01%)(-0.01\%)0.01σ0.01\sigma<br>(0.14%)(0.14\%)−0.12σ-0.12\sigma<br>(−0.03%)(-0.03\%)−0.01σ-0.01\sigma<br>(−0.05%)(-0.05\%)0.01σ0.01\sigma<br>(0.04%)(0.04\%)

Table 8. Same as Table 4, but for LiteBIRD + Planck: biases in units of σ\sigma, with the corresponding percentage shifts in parentheses, in the six Λ\LambdaCDM parameters and me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0}, for data generated with me/me,0=1.02m_{\mathrm{e}}/m_{\mathrm{e},0}=1.02 and analysed with different models.

Figure 11. Recovered biased values, θfid+δθ\theta_{\mathrm{fid}}+\delta\theta of the six Λ\LambdaCDM parameters and me/me,0m_e/m_{e,0} with their marginalised 1σ1\sigma error bars for LiteBIRD + Planck. The data sets, models, colours and markers are as in Figure 8.

Conclusions

In this work, we have studied Rayleigh scattering of CMB photons as a probe of the cosmological recombination process, with particular focus on recombination scenarios modified by a variation of the electron mass, mem_{\mathrm{e}}, or the fine-structure constant, α\alpha. We modelled the RS signal with modified versions of Recfast++ and CAMB, and used a multi-frequency Fisher analysis including realistic noise and foregrounds to forecast parameter constraints and biases for PICO and LiteBIRD + Planck.

Increasing mem_{\mathrm{e}} or α\alpha increases atomic binding energies, so recombination occurs earlier. Neutral hydrogen and helium therefore appear at higher redshift, which shifts the onset of RS opacity and both the Thomson and Rayleigh visibility functions to earlier times. Because the RS contribution scales as ν4\nu^4, these changes imprint a frequency-dependent signature on the TTTT, TETE and especially the EEEE spectra, which is most pronounced in the highest-frequency channels.

For PICO, adding RS to the primary CMB improves the Λ\LambdaCDM constraints by ≲7%\lesssim7\%, since RS does not open any new degree of freedom when the recombination history is fixed. This finding is in good agreement with [55]. The gain is substantially greater once the recombination history is allowed to vary. In the Λ\LambdaCDM +xe+x_{\mathrm{e}} model, the uncertainties in me/me,0m_{\mathrm{e}}/m_{\mathrm{e},0}, α/α0\alpha/\alpha_0 and H0H_0 shrink by ≃31%\simeq31\%, and the error bars of other parameters also improve, because RS samples the timing of recombination independently at each frequency. When the variation of mem_{\mathrm{e}} is propagated consistently to the Thomson and Rayleigh cross-sections, the primary CMB alone suffers from an almost exact scaling degeneracy between mem_{\mathrm{e}}, Ωbh2\Omega_{\mathrm{b}}h^2, Ωch2\Omega_{\mathrm{c}}h^2 and H0H_0. RS reduces the extent of this degeneracy and improves all four parameters by ≃30%\simeq30\%. For α\alpha, there is no such degeneracy and the gain is ≲8%\lesssim8\%. For LiteBIRD + Planck the improvements are below 2%, except along the same mem_{\mathrm{e}} scaling degeneracy, where they reach 55–58% because the primary-CMB constraints are much weaker.

The bias analysis shows that RS must be modelled in future analyses. For PICO, neglecting RS biases the Λ\LambdaCDM parameters by up to ≃1σ\simeq1\sigma, while for LiteBIRD + Planck the biases remain below 0.15σ0.15\sigma. A modified recombination history analysed with Λ\LambdaCDM produces very large biases for both experiments. Most importantly, if the effect of a varying mem_{\mathrm{e}} on the scattering cross-sections is neglected, the scaling degeneracy drives the inferred electron mass back to me/me,0≃1.0m_{\mathrm{e}}/m_{\mathrm{e},0 \simeq1.0}. A genuine 2% variation of mem_{\mathrm{e}} would then be completely missed because the data is fitted by readjusting the other parameters. A variation of α\alpha, in contrast, is still recovered close to its input value. Only when the variation is modelled consistently, including σT\sigma_{\mathrm{T}} and σR\sigma_{\mathrm{R}}, are the input parameters recovered, with residual biases from neglecting RS of ≲1σ\lesssim1\sigma.

Our forecasts rely on the Fisher formalism, and the bias estimates use a linear expansion. Biases of tens of σ\sigma should therefore be read as signs of severe mis-modelling rather than as precise values, and the constraints along the near-degenerate mem_{\mathrm{e}} direction would benefit from a full likelihood analysis. More accurate foreground modelling and calibration parameters, and exploiting the high-frequency channels of future experiments, will be essential to fully realise the potential of RS as an independent probe of recombination physics and of proposed modified recombination-based solutions to the Hubble tension.

Acknowledgments

DA, RTGS, CHM and JARM acknowledge financial support from the projects PID2023-150398NB-I00 and PID2023-151567NB-I00, funded by the Ministry of Science, Innovation and Universities of Spain, AEI/10.13039/501100011033 and by FEDER, EU. This work is part of grant CEX2025-001609-S, awarded to the Instituto de Astrofísica de Canarias under the Severo Ochoa Centre of Excellence program and funded by MI-CIU/AEI/10.13039/501100011033. JC acknowledges the support of the Fundación Occident and the Instituto de Astrofísica de Canarias under the Visiting Researcher Programme 2025-2028 agreed between both institutions. EV acknowledges that the project that gave rise to these results received the support of a fellowship from the a “Caixa” Foundation (ID B006373). The fellowship code is LCF/BQ/DI24/12070001. JC was also supported by the UKSA grant: LiteBIRD UK ST/Y005945/1. We thank the Simons Observatory collaboration for making the fgspectra code publicly available, and Antony Lewis for providing the modified version of CAMB to include the Rayleigh scattering. The paper uses the python packages, numpy [72], matplotlib [73], scipy [74].

Derivation of Fisher matrix and parameter bias due to incorrect fiducial model

The log-likelihood of a Gaussian random vector, a\mathbf{a}, having a covariance matrix, C\mathbf{C} is given by

ln⁡L=−12[ln⁡det⁡C+a†C−1a]+constant,(11)\ln\mathcal{L} = -\frac{1}{2}\left[\ln\det\mathbf{C}+\mathbf{a}^{\dagger}\mathbf{C}^{-1}\mathbf{a}\right]+\text{constant}, \tag*{(11)}

where C\mathbf{C} is a block-diagonal matrix, with the matrix Cℓ\mathbf{C}_{\ell} of size 2ℓ+12\ell+1 as its elements for each multipole ℓ\ell. Taking the derivative with respect to parameter θi\theta_i and using the log-determinant matrix identities ∂iln⁡det⁡C=Tr⁡(C−1∂iC)\partial_i\ln\det\mathbf{C}=\operatorname{Tr}(\mathbf{C}^{-1}\partial_i\mathbf{C}) and ∂iC−1=−C−1∂iCC−1\partial_i\mathbf{C}^{-1}=-\mathbf{C}^{-1}\partial_i\mathbf{C}\mathbf{C}^{-1}, we obtain

∂ln⁡L∂θi=−12[Tr⁡(C−1∂iC)−a†C−1∂iCC−1a].(12)\frac{\partial\ln\mathcal{L}}{\partial\theta_i}=-\frac{1}{2}\left[\operatorname{Tr}(\mathbf{C}^{-1}\partial_i\mathbf{C})-\mathbf{a}^{\dagger}\mathbf{C}^{-1}\partial_i\mathbf{C}\mathbf{C}^{-1}\mathbf{a}\right]. \tag*{(12)}

Taking the second derivative with respect to the parameter θj\theta_j and using the matrix identity, ⟨a†Aa⟩=Tr⁡(AC)\langle\mathbf{a}^{\dagger}\mathbf{A}\mathbf{a}\rangle=\operatorname{Tr}(\mathbf{A}\mathbf{C}) where ⟨a†a⟩=C\langle\mathbf{a}^{\dagger}\mathbf{a}\rangle=\mathbf{C}, we derive the Fisher matrix as

Fij=−⟨∂2ln⁡L∂θi∂θj⟩=12Tr⁡(C−1∂iCC−1∂jC).(13)F_{ij}=-\left\langle\frac{\partial^2\ln\mathcal{L}}{\partial\theta_i\partial\theta_j}\right\rangle=\frac{1}{2}\operatorname{Tr}\left(\mathbf{C}^{-1}\partial_i\mathbf{C}\mathbf{C}^{-1}\partial_j\mathbf{C}\right). \tag*{(13)}

For CMB data, since the covariance matrix C\mathbf{C} is block-diagonal for each ℓ\ell of size (2ℓ+1)(2\ell+1), the Fisher matrix for a sky-fraction of fskyf_{\mathrm{sky}} is given by

Fij=∑ℓ(2ℓ+1)fsky2Tr⁡(Cℓ−1∂Cℓ∂θiCℓ−1∂Cℓ∂θj).(14)F_{ij}=\sum_{\ell}\frac{(2\ell+1)f_{\mathrm{sky}}}{2}\operatorname{Tr}\left(\mathbf{C}_{\ell}^{-1}\frac{\partial\mathbf{C}_{\ell}}{\partial\theta_i}\mathbf{C}_{\ell}^{-1}\frac{\partial\mathbf{C}_{\ell}}{\partial\theta_j}\right). \tag*{(14)}

Now we show the derivation of the bias in the parameters due to the mismatch between model Cℓ\mathbf{C}_{\ell} and true data Cℓtrue\mathbf{C}_{\ell}^{\mathrm{true}}. Due to the systematic difference ΔCℓ=Cℓtrue−Cℓ\Delta\mathbf{C}_{\ell}=\mathbf{C}_{\ell}^{\mathrm{true}}-\mathbf{C}_{\ell} between the true power spectrum and our model’s expectation, the maximum of the likelihood will be shifted to θ^\hat{\theta} from the true parameter values θfid\theta_{\mathrm{fid}} by a bias δθ\delta\theta. Therefore the estimator θ^\hat{\theta} satisfies the condition,

∂ln⁡L∂θi∣θ^=0.(15)\left.\frac{\partial\ln\mathcal{L}}{\partial\theta_i}\right|_{\hat{\theta}}=0. \tag*{(15)}

Expanding Eq. A.5 around the true parameter values θfid\theta_{\mathrm{fid}} we get,

∂ln⁡L∂θi∣θ^≃∂ln⁡L∂θi∣θfid+δθj∂2ln⁡L∂θi∂θj∣θfid=0(16)\left.\frac{\partial\ln\mathcal{L}}{\partial\theta_i}\right|_{\hat{\theta}}\simeq\left.\frac{\partial\ln\mathcal{L}}{\partial\theta_i}\right|_{\theta_{\mathrm{fid}}}+\delta\theta_j\left.\frac{\partial^2\ln\mathcal{L}}{\partial\theta_i\partial\theta_j}\right|_{\theta_{\mathrm{fid}}}=0 \tag*{(16)}

Taking the expectation value of Eq. A.6 and using the definition of the Fisher matrix, Fij=−⟨∂2ln⁡L∂θi∂θj∣θfid⟩F_{ij}=-\left\langle\left.\frac{\partial^2\ln\mathcal{L}}{\partial\theta_i\partial\theta_j}\right|_{\theta_{\mathrm{fid}}}\right\rangle, we obtain

⟨∂ln⁡L∂θi∣θfid⟩−∑jFijδθj=0.(17)\left\langle\left.\frac{\partial\ln\mathcal{L}}{\partial\theta_i}\right|_{\theta_{\mathrm{fid}}}\right\rangle-\sum_jF_{ij}\delta\theta_j=0. \tag*{(17)}

Therefore, the bias vector is

δθi=∑jFij−1⟨∂ln⁡L∂θj∣θfid⟩.(18)\delta\theta_i=\sum_jF^{-1}_{ij}\left\langle\left.\frac{\partial\ln\mathcal{L}}{\partial\theta_j}\right|_{\theta_{\mathrm{fid}}}\right\rangle. \tag*{(18)}

Using the expression in Eq. A.2 and considering ⟨a†a⟩=Ctrue\langle\mathbf{a}^{\dagger}\mathbf{a}\rangle=\mathbf{C}^{\mathrm{true}} and the matrix identity, ⟨a†Aa⟩=Tr⁡(ACtrue)\langle\mathbf{a}^{\dagger}\mathbf{A}\mathbf{a}\rangle=\operatorname{Tr}(\mathbf{A}\mathbf{C}^{\mathrm{true}}), Eq. A.8 can be expressed as,

⟨∂ln⁡L∂θi⟩=−12[Tr⁡(C−1∂iC)−Tr⁡(C−1∂iCC−1Ctrue)].(19)\left\langle\frac{\partial\ln\mathcal{L}}{\partial\theta_i}\right\rangle=-\frac{1}{2}\left[\operatorname{Tr}(\mathbf{C}^{-1}\partial_i\mathbf{C})-\operatorname{Tr}(\mathbf{C}^{-1}\partial_i\mathbf{C}\mathbf{C}^{-1}\mathbf{C}^{\mathrm{true}})\right]. \tag*{(19)}

Using Ctrue=C+ΔC\mathbf{C}^{\mathrm{true}}=\mathbf{C}+\Delta\mathbf{C} and considering that CMB power spectra are diagonal in ℓ\ell and mm space, from Eq. A.8, we get the final expression for the parameter bias vector,

δθi=∑j(F−1)ij∑ℓ(2ℓ+1)2fskyTr⁡[Cℓ−1∂Cℓ∂θjCℓ−1ΔCℓ],(20)\delta\theta_i=\sum_j(\mathbf{F}^{-1})_{ij}\sum_{\ell}\frac{(2\ell+1)}{2}f_{\mathrm{sky}}\operatorname{Tr}\left[\mathbf{C}_{\ell}^{-1}\frac{\partial\mathbf{C}_{\ell}}{\partial\theta_j}\mathbf{C}_{\ell}^{-1}\Delta\mathbf{C}_{\ell}\right], \tag*{(20)}

where ΔCℓ=Cℓtrue−Cℓ(θfid)\Delta\mathbf{C}_{\ell}=\mathbf{C}_{\ell}^{\mathrm{true}}-\mathbf{C}_{\ell}(\theta_{\mathrm{fid}}).

Results with LiteBIRD + Planck

Figure 12. Recovered biased values, θfid+δθ\theta_{\mathrm{fid}} + \delta\theta of the six Λ\LambdaCDM parameters and α/α0\alpha/\alpha_0 with their marginalised 1σ1\sigma error bars, for LiteBIRD + Planck. The data sets, models, colours and markers are as in Figure 9.

DataModelRSΩbh2\Omega_b h^2Ωch2\Omega_c h^2ln⁡(1010As)\ln(10^{10} A_s)H0H_0nsn_sτ\tauα/α0\alpha/\alpha_0
Primary CMB + RS + xex_eΛ\LambdaCDMno−4.30σ-4.30\sigma1.63σ1.63\sigma5.51σ5.51\sigma−16.49σ-16.49\sigma5.40σ5.40\sigma0.66σ0.66\sigma–
(−1.78%)(-1.78\%)(1.11%)(1.11\%)(0.71%)(0.71\%)(−8.89%)(-8.89\%)(1.36%)(1.36\%)(2.26%)(2.26\%)–
yes−4.16σ-4.16\sigma1.52σ1.52\sigma5.60σ5.60\sigma−16.48σ-16.48\sigma5.62σ5.62\sigma0.74σ0.74\sigma–
(−1.71%)(-1.71\%)(1.03%)(1.03\%)(0.72%)(0.72\%)(−8.82%)(-8.82\%)(1.41%)(1.41\%)(2.53%)(2.53\%)–
Λ\LambdaCDM + xex_eno−0.06σ-0.06\sigma−0.04σ-0.04\sigma−0.04σ-0.04\sigma−0.07σ-0.07\sigma−0.02σ-0.02\sigma−0.01σ-0.01\sigma−0.08σ-0.08\sigma
(−0.03%)(-0.03\%)(−0.03%)(-0.03\%)(−0.01%)(-0.01\%)(−0.09%)(-0.09\%)(−0.01%)(-0.01\%)(−0.02%)(-0.02\%)(−0.01%)(-0.01\%)
Primary CMB + RS + xex_e + σT+σR\sigma_T + \sigma_RΛ\LambdaCDMno−1.89σ-1.89\sigma0.99σ0.99\sigma5.77σ5.77\sigma−14.36σ-14.36\sigma8.27σ8.27\sigma0.78σ0.78\sigma–
(−0.78%)(-0.78\%)(0.67%)(0.67\%)(0.75%)(0.75\%)(−7.75%)(-7.75\%)(2.09%)(2.09\%)(2.69%)(2.69\%)–
yes−1.76σ-1.76\sigma0.90σ0.90\sigma5.85σ5.85\sigma−14.38σ-14.38\sigma8.49σ8.49\sigma0.85σ0.85\sigma–
(−0.73%)(-0.73\%)(0.61%)(0.61\%)(0.76%)(0.76\%)(−7.70%)(-7.70\%)(2.13%)(2.13\%)(2.92%)(2.92\%)–
Λ\LambdaCDM + xex_eno1.73σ1.73\sigma0.96σ0.96\sigma−0.29σ-0.29\sigma2.14σ2.14\sigma−0.20σ-0.20\sigma−0.08σ-0.08\sigma1.70σ1.70\sigma
(0.76%)(0.76\%)(0.78%)(0.78\%)(−0.04%)(-0.04\%)(2.57%)(2.57\%)(−0.09%)(-0.09\%)(−0.28%)(-0.28\%)(0.30%)(0.30\%)
yes1.78σ1.78\sigma0.97σ0.97\sigma−0.22σ-0.22\sigma2.17σ2.17\sigma−0.13σ-0.13\sigma−0.06σ-0.06\sigma1.74σ1.74\sigma
(0.78%)(0.78\%)(0.79%)(0.79\%)(−0.03%)(-0.03\%)(2.58%)(2.58\%)(−0.06%)(-0.06\%)(−0.21%)(-0.21\%)(0.31%)(0.31\%)
Λ\LambdaCDM + xex_eno−0.04σ-0.04\sigma−0.03σ-0.03\sigma−0.04σ-0.04\sigma−0.06σ-0.06\sigma−0.01σ-0.01\sigma−0.01σ-0.01\sigma−0.08σ-0.08\sigma
+ σT+σR\sigma_T + \sigma_R(−0.02%)(-0.02\%)(−0.02%)(-0.02\%)(−0.01%)(-0.01\%)(−0.06%)(-0.06\%)(−0.01%)(-0.01\%)(−0.02%)(-0.02\%)(−0.01%)(-0.01\%)

Table 9. Same as Table 5, but for LiteBIRD + Planck: biases in units of σ\sigma, with the corresponding percentage shifts in parentheses, in the six Λ\LambdaCDM parameters and α/α0\alpha/\alpha_0, for data generated with α/α0=1.01\alpha/\alpha_0 = 1.01 and analysed with different models.

DataModelRSΩbh2\Omega_b h^2Ωch2\Omega_c h^2ln⁡(1010As)\ln(10^{10}A_s)H0H_0nsn_sτ\tauα/α0\alpha/\alpha_0
Primary CMB + RS + xex_eΛ\LambdaCDMno−10.74σ-10.74\sigma42.10σ42.10\sigma29.76σ29.76\sigma−77.66σ-77.66\sigma−9.90σ-9.90\sigma9.01σ9.01\sigma–
(−1.28%)(-1.28\%)(13.55%)(13.55\%)(3.11%)(3.11\%)(−16.96%)(-16.96\%)(−1.47%)(-1.47\%)(29.30%)(29.30\%)–
yes−6.47σ-6.47\sigma39.79σ39.79\sigma34.80σ34.80\sigma−77.81σ-77.81\sigma−8.73σ-8.73\sigma14.01σ14.01\sigma–
(−0.76%)(-0.76\%)(11.94%)(11.94\%)(3.47%)(3.47\%)(−15.80%)(-15.80\%)(−1.27%)(-1.27\%)(42.26%)(42.26\%)–
Λ\LambdaCDM + xex_eno−1.05σ-1.05\sigma0.66σ0.66\sigma−0.44σ-0.44\sigma0.47σ0.47\sigma−1.03σ-1.03\sigma−0.25σ-0.25\sigma0.88σ0.88\sigma
(−0.13%)(-0.13\%)(0.21%)(0.21\%)(−0.05%)(-0.05\%)(0.21%)(0.21\%)(−0.22%)(-0.22\%)(−0.80%)(-0.80\%)(0.05%)(0.05\%)
Primary CMB + RS + xex_e + σT+σR\sigma_T + \sigma_RΛ\LambdaCDMno−2.53σ-2.53\sigma38.02σ38.02\sigma27.43σ27.43\sigma−70.48σ-70.48\sigma1.27σ1.27\sigma8.76σ8.76\sigma–
(−0.30%)(-0.30\%)(12.24%)(12.24\%)(2.87%)(2.87\%)(−15.39%)(-15.39\%)(0.19%)(0.19\%)(28.50%)(28.50\%)–
yes0.08σ0.08\sigma37.27σ37.27\sigma31.76σ31.76\sigma−72.14σ-72.14\sigma2.61σ2.61\sigma12.89σ12.89\sigma–
(0.01%)(0.01\%)(11.18%)(11.18\%)(3.17%)(3.17\%)(−14.65%)(-14.65\%)(0.38%)(0.38\%)(38.90%)(38.90\%)–
Λ\LambdaCDM + xex_eno5.45σ5.45\sigma2.83σ2.83\sigma−0.83σ-0.83\sigma6.13σ6.13\sigma−0.77σ-0.77\sigma−0.29σ-0.29\sigma6.53σ6.53\sigma
(0.66%)(0.66\%)(0.90%)(0.90\%)(−0.09%)(-0.09\%)(2.74%)(2.74\%)(−0.17%)(-0.17\%)(−0.93%)(-0.93\%)(0.35%)(0.35\%)
yes4.95σ4.95\sigma3.55σ3.55\sigma1.62σ1.62\sigma−1.89σ-1.89\sigma6.49σ6.49\sigma0.55σ0.55\sigma−2.12σ-2.12\sigma
(0.59%)(0.59\%)(1.08%)(1.08\%)(0.16%)(0.16\%)(−0.62%)(-0.62\%)(1.20%)(1.20\%)(1.63%)(1.63\%)(−0.08%)(-0.08\%)
Λ\LambdaCDM + xex_e + σT+σR\sigma_T + \sigma_Rno−1.19σ-1.19\sigma0.61σ0.61\sigma−0.40σ-0.40\sigma0.27σ0.27\sigma−1.00σ-1.00\sigma−0.25σ-0.25\sigma0.83σ0.83\sigma
(−0.15%)(-0.15\%)(0.20%)(0.20\%)(−0.04%)(-0.04\%)(0.10%)(0.10\%)(−0.22%)(-0.22\%)(−0.82%)(-0.82\%)(0.03%)(0.03\%)

Table 5. Biases in units of σ\sigma, with the corresponding percentage shifts in parentheses, for data generated with α/α0=1.01\alpha/\alpha_0 = 1.01, entering either only through xex_e or also through σT\sigma_T and σR\sigma_R for PICO. Each data set is analysed with the models listed, without (“no”) or with (“yes”) modelling RS. When the model matches the data and RS is modelled, the biases vanish by construction and these rows are omitted. The case of Λ\LambdaCDM data is identical to Table 4.

References

  1. [1]J. J. McMahon, K. A. Aird, B. A. Benson, L. E. Bleem, J. Britton, J. E. Carlstrom et al., SPTpol: an instrument for CMB polarization, in American Institute of Physics Conference Series, B. Young, B. Cabrera, & A. Miller, ed., vol. 1185 of American Institute of Physics Conference Series, pp. 511–514, Dec., 2009, DOI.DOI
  2. [2]M. D. Niemack, P. A. R. Ade, J. Aguirre, F. Barrientos, J. A. Beall, J. R. Bond et al., ACTPol: a polarization-sensitive receiver for the Atacama Cosmology Telescope, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, vol. 7741 of Presented at the Society of Photo-Optical Instrumentation Engineers (SPIE) Conference, July, 2010, 1006.5049, DOI.DOI
  3. [3]C. L. Bennett, D. Larson, J. L. Weiland, N. Jarosik, G. Hinshaw, N. Odegard et al., Nine-year wilkinson microwave anisotropy probe (wmap) observations: Final maps and results, The Astrophysical Journal Supplement Series 208 (2013) 20.
  4. [4]Planck Collaboration, N. Aghanim, Y. Akrami, F. Arroja, M. Ashdown, J. Aumont et al., Planck 2018 results. I. Overview and the cosmological legacy of Planck, A&A 641 (2020) A1 [1807.06205].
  5. [5]Planck Collaboration, N. Aghanim, Y. Akrami, M. Ashdown, J. Aumont, C. Baccigalupi et al., Planck 2018 results. VI. Cosmological parameters, A&A 641 (2020) A6 [1807.06209].
  6. [6]A. G. Riess, W. Yuan, L. M. Macri, D. Scolnic, D. Brout, S. Casertano et al., A comprehensive measurement of the local value of the hubble constant with 1 km s−1 Mpc−1 uncertainty from the hubble space telescope and the shoes team, The Astrophysical Journal Letters 934 (2022) L7.
  7. [7]E. Di Valentino, L. A. Anchordoqui, Ö. Akarsu, Y. Ali-Haimoud, L. Amendola, N. Arendse et al., Cosmology Intertwined III: fσ8 and S8, Astroparticle Physics 131 (2021) 102604 [2008.11285].
  8. [8]I. d. O. C. Pedreira, M. Benetti, E. G. M. Ferreira, L. L. Graef and L. Herold, Visual tool for assessing tension-resolving models in the H0–σ8 plane, Phys.Rev.D 109 (2024) 103525 [2311.04977].
  9. [9]ACT, DES collaboration, Cosmology from cross-correlation of ACT-DR4 CMB lensing and DES-Y3 cosmic shear, Mon. Not. Roy. Astron. Soc. 528 (2024) 2112 [2309.04412].
  10. [10]F. J. Qu, Q. Hang, G. Farren, B. Bolliet, J. N. Aguilar, S. Ahlen et al., Atacama cosmology telescope dr6 and desi: Structure growth measurements from the cross-correlation of desi legacy imaging galaxies and cmb lensing from act dr6 and planck pr4, Phys. Rev. D 111 (2025) 103503.
  11. [11]T. Karim, S. Singh, M. Rezaie, D. Eisenstein, B. Hadzhiyska, J. S. Speagle et al., Measuring σ8 using desi legacy imaging surveys emission-line galaxies and planck cmb lensing, and the impact of dust on parameter inference, Journal of Cosmology and Astroparticle Physics 2025 (2025) 045.DOI
  12. [12]E. Mörtsell, A. Goobar, J. Johansson and S. Dhawan, Sensitivity of the Hubble Constant Determination to Cepheid Calibration, ApJ 933 (2022) 212 [2105.11461].DOI
  13. [13]E. Mörtsell, A. Goobar, J. Johansson and S. Dhawan, The Hubble Tension Revisited: Additional Local Distance Ladder Uncertainties, ApJ 935 (2022) 58 [2106.09400].DOI
  14. [14]C.-T. Chiang and A. Slosar, Inferences of H₀ in presence of a non-standard recombination, arXiv e-prints (2018) arXiv:1811.03624 [1811.03624].arxiv.org/abs/1811.03624
  15. [15]L. Hart and J. Chluba, Updated fundamental constant constraints from Planck 2018 data and possible relations to the Hubble tension, MNRAS 493 (2020) 3255 [1912.03986].DOI
  16. [16]S. Roy Choudhury, S. Hannestad and T. Tram, Updated constraints on massive neutrino self-interactions from cosmology in light of the H₀ tension, JCAP 03 (2021) 084 [2012.07519].DOI
  17. [17]K. Jedamzik and L. Pogosian, Relieving the Hubble Tension with Primordial Magnetic Fields, Phys.Rev.Lett 125 (2020) 181302 [2004.09487].DOI
  18. [18]Ruchika, H. Rathore, S. Roy Choudhury and V. Rentala, A gravitational constant transition within cepheids as supernovae calibrators can solve the Hubble tension, JCAP 06 (2024) 056 [2306.05450].DOI
  19. [19]E. Calabrese, J. C. Hill, H. T. Jense, A. La Posta, I. Abril-Cabezas, G. E. Addison et al., The Atacama Cosmology Telescope: DR6 constraints on extended cosmological models, JCAP 2025 (2025) 063 [2503.14454].arxiv.org/abs/2503.14454
  20. [20]D. Wang and D. Mota, Did DESI DR2 truly reveal dynamical dark energy?, European Physical Journal C 85 (2025) 1356 [2504.15222].arxiv.org/abs/2504.15222
  21. [21]L. Knox and M. Millea, Hubble constant hunter’s guide, Phys.Rev.D 101 (2020) 043533 [1908.03663].DOI
  22. [22]E. Di Valentino, O. Mena, S. Pan, L. Visinelli, W. Yang, A. Melchiorri et al., In the realm of the Hubble tension-a review of solutions, Classical and Quantum Gravity 38 (2021) 153001 [2103.01183].DOI
  23. [23]N. Schöneberg, G. Franco Abellán, A. Pérez Sánchez, S. J. Witte, V. Poulin and J. Lesgourgues, The H0 Olympics: A fair ranking of proposed models, Phys. Rept. 984 (2022) 1 [2107.10291].
  24. [24]G.-B. Zhao, M. Raveri, L. Pogosian, Y. Wang, R. G. Crittenden, W. J. Handley et al., Dynamical dark energy in light of the latest observations, Nature Astronomy 1 (2017) 627 [1701.08165].DOI
  25. [25]E. Di Valentino, A. Melchiorri and O. Mena, Can interacting dark energy solve the H₀ tension?, Phys. Rev. D 96 (2017) 043503 [1704.08342].DOI
  26. [26]L. Pogosian, M. Raveri, K. Koyama, M. Martinelli, A. Silvestri, G.-B. Zhao et al., Imprints of cosmological tensions in reconstructed gravity, Nature Astron. 6 (2022) 1484 [2107.12992].arxiv.org/abs/2107.12992
  27. [27]A. G. Adame, J. Aguilar, S. Ahlen, S. Alam, D. M. Alexander, M. Alvarez et al., DESI 2024 VI: cosmological constraints from the measurements of baryon acoustic oscillations, JCAP 2025 (2025) 021 [2404.03002].DOI
  28. [28]E. Ó Colgáin and M. M. Sheikh-Jabbari, DESI and SNe: dynamical dark energy, Ωₘ tension or systematics?, MNRAS 542 (2025) L24 [2412.12905].DOI
  29. [29]M. Kamionkowski and A. G. Riess, The Hubble Tension and Early Dark Energy, Annual Review of Nuclear and Particle Science 73 (2023) 153 [2211.04492].arxiv.org/abs/2211.04492
  30. [30]L. Hart and J. Chluba, New constraints on time-dependent variations of fundamental constants using planck data, Monthly Notices of the Royal Astronomical Society 474 (2017) 1850 [https://academic.oup.com/mnras/article-pdf/474/2/1850/22636771/stx2783.pdf].DOI
  31. [31]L. Hart and J. Chluba, Improved model-independent constraints on the recombination era and development of a direct projection method, MNRAS 495 (2020) 4210 [1912.04682].DOI
  32. [32]T. Sekiguchi and T. Takahashi, Early recombination as a solution to the H₀ tension, Phys.Rev.D 103 (2021) 083507 [2007.03831].DOI
  33. [33]Y. Toda, W. Giarè, E. Özülker, E. Di Valentino and S. Vagnozzi, Combining pre- and post-recombination new physics to address cosmological tensions: Case study with varying electron mass and sign-switching cosmological constant, Physics of the Dark Universe 46 (2024) 101676.DOI
  34. [34]N. Schöneberg and L. Vacher, The mass effect — variations of the electron mass and their impact on cosmology, JCAP 2025 (2025) 004 [2407.16845].
  35. [35]K. Hoshiya and Y. Toda, Electron mass variation from dark sector interactions and compatibility with cosmological observations, Phys.Rev.D 107 (2023) 043505 [2202.07714].DOI
  36. [36]O. Seto and Y. Toda, Big bang nucleosynthesis constraints on varying electron mass solution to the Hubble tension, Phys.Rev.D 107 (2023) 083512 [2206.13209].DOI
  37. [37]Y. Toda and O. Seto, Constraints on the simultaneous variation of the fine structure constant and the electron mass in light of desi bao data, Physics of the Dark Universe 49 (2025) 102035.
  38. [38]S. Ghosh, K. K. Boddy and T.-T. Yu, Early Universe Constraints on Variations in Fundamental Constants Induced by Ultralight Scalar Dark Matter, arXiv e-prints (2025) arXiv:2511.14532 [2511.14532].arxiv.org/abs/2511.14532
  39. [39]Q. Yu, D. N. Spergel and J. P. Ostriker, Rayleigh scattering and microwave background fluctuations, The Astrophysical Journal 558 (2001) 23.DOI
  40. [40]A. Lewis, Rayleigh scattering: blue sky thinking for future CMB observations, JCAP 2013 (2013) 053 [1307.8148].
  41. [41]LiteBIRD collaboration, Probing Cosmic Inflation with the LiteBIRD Cosmic Microwave Background Polarization Survey, PTEP 2023 (2023) 042F01 [2202.02773].DOI
  42. [42]S. Hanany, M. Alvarez, E. Artis, P. Ashton, J. Aumont, R. Aurlien et al., PICO: Probe of Inflation and Cosmic Origins, arXiv e-prints (2019) arXiv:1902.10541 [1902.10541].arxiv.org/abs/1902.10541
  43. [43]L. Rayleigh, X. On the electromagnetic theory of light, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 12 (1881) 81 [https://doi.org/10.1080/14786448108627074].
  44. [44]H.-W. Lee, Exact low-energy expansion of the rayleigh scattering cross-section by atomic hydrogen, Monthly Notices of the Royal Astronomical Society 358 (2005) 1472 [https://academic.oup.com/mnras/article-pdf/358/4/1472/18660986/358-4-1472.pdf].DOI
  45. [45]S. P. Tarafdar and M. S. Vardya, The rayleigh scattering cross-sections of he, c, n and o, Monthly Notices of the Royal Astronomical Society 145 (1969) 171 [https://academic.oup.com/mnras/article-pdf/145/2/171/9360740/mnras145-0171.pdf].DOI
  46. [46]F. Takahara and S. Sasaki, Effect of rayleigh scattering on cosmic microwave background anisotropies, Progress of Theoretical Physics 86 (1991) 1021 [https://academic.oup.com/ptp/article-pdf/86/5/1021/5430029/86-5-1021.pdf].DOI
  47. [47]J. Chluba and R. M. Thomas, Towards a complete treatment of the cosmological recombination problem, MNRAS 412 (2011) 748 [1010.3631].
  48. [48]J. Chluba, G. M. Vasil and L. J. Dursi, Recombinations to the Rydberg states of hydrogen and their effect during the cosmological recombination epoch, MNRAS 407 (2010) 599 [1003.4928].
  49. [49]A. Lewis, A. Challinor and A. Lasenby, Efficient Computation of Cosmic Microwave Background Anisotropies in Closed Friedmann-Robertson-Walker Models, ApJ 538 (2000) 473 [astro-ph/9911177].DOI
  50. [50]C. G. Scóccola, S. J. Landau and H. Vucetich, WMAP 5-year constraints on time variation of alpha and m_e, Memorie della Societ Astronomica Italiana 80 (2009) 814 [0910.1083].arxiv.org/abs/0910.1083
  51. [51]S. Seager, D. D. Sasselov and D. Scott, How Exactly Did the Universe Become Neutral?, ApJS 128 (2000) 407 [astro-ph/9912182].
  52. [52]J. Chluba and Y. Ali-Haïmoud, COSMOSPEC: fast and detailed computation of the cosmological recombination radiation from hydrogen and helium, MNRAS 456 (2016) 3494 [1510.03877].DOI
  53. [53]Y. B. Zeldovich, V. G. Kurt and R. A. Syunyaev, Recombination of Hydrogen in the Hot Model of the Universe, Zhurnal Eksperimental noi i Teoreticheskoi Fiziki 55 (1968) 278.
  54. [54]S. Seager, D. D. Sasselov and D. Scott, A New Calculation of the Recombination Epoch, ApJ 523 (1999) L1.DOI
  55. [55]B. Beringue, P. D. Meerburg, J. Meyers and N. Battaglia, Cosmology with Rayleigh scattering of the cosmic microwave background, JCAP 2021 (2021) 060 [2008.11688].DOI
  56. [56]M. Kaplinghat, R. J. Scherrer and M. S. Turner, Constraining variations in the fine-structure constant with the cosmic microwave background, Phys.Rev.D 60 (1999) 023516 [arXiv:astro-ph/9810133].DOI
  57. [57]R. A. Battye, R. Crittenden and J. Weller, Cosmic concordance and the fine structure constant, Phys.Rev.D 63 (2001) 043505 [astro-ph/0008265].DOI
  58. [58]M. Tegmark, A. N. Taylor and A. F. Heavens, Kahrünen-Löeve Eigenvalue Problems in Cosmology: How Should We Tackle Large Data Sets?, ApJ 480 (1997) 22 [astro-ph/9603021].
  59. [59]J. Carron, On the assumption of Gaussianity for cosmological two-point statistics and parameter dependent covariance matrices, A&A 551 (2013) A88 [1204.4724].DOI
  60. [60]T. Louis, A. La Posta, Z. Atkins, H. T. Jense, I. Abril-Cabezas, G. E. Addison et al., The Atacama Cosmology Telescope: DR6 power spectra, likelihoods and ΛCDM parameters, JCAP 2025 (2025) 062 [2503.14452].
  61. [61]B. Beringue, K. M. Surrao, J. C. Hill, Z. Atkins, N. Battaglia, B. Bolliet et al., The Atacama Cosmology Telescope: DR6 power spectrum foreground model and validation, JCAP 2025 (2025) 082 [2506.06274].DOI
  62. [62]N. Battaglia, A. Natarajan, H. Trac, R. Cen and A. Loeb, Reionization on large scales. iii. predictions for low-ℓ cosmic microwave background polarization and high-ℓ kinetic sunyaev–zel’dovich observables, The Astrophysical Journal 776 (2013) 83.
  63. [63]N. Battaglia, J. R. Bond, C. Pfrommer and J. L. Sievers, On the Cluster Physics of Sunyaev-Zel’dovich and X-Ray Surveys. II. Deconstructing the Thermal SZ Power Spectrum, ApJ 758 (2012) 75 [1109.3711].DOI
  64. [64]Planck Collaboration, P. A. R. Ade, N. Aghanim, C. Armitage-Caplan, M. Arnaud, M. Ashdown et al., Planck 2013 results. XXX. Cosmic infrared background measurements and implications for star formation, A&A 571 (2014) A30 [1309.0382].
  65. [65]S. K. Choi, M. Hasselfield, S.-P. Ho, B. Koopman, M. Lungu, M. H. Abitbol et al., The Atacama Cosmology Telescope: a measurement of the Cosmic Microwave Background power spectra at 98 and 150 GHz, JCAP 2020 (2020) 045 [2007.07289].
  66. [66]G. E. Addison, J. Dunkley and D. N. Spergel, Modelling the correlation between the thermal Sunyaev Zel’dovich effect and the cosmic infrared background, MNRAS 427 (2012) 1741 [1204.5927].DOI
  67. [67]J. Dunkley, R. Hlozek, J. Sievers, V. Acquaviva, P. A. R. Ade, P. Aguirre et al., The Atacama Cosmology Telescope: Cosmological Parameters from the 2008 Power Spectrum, ApJ 739 (2011) 52 [1009.0866].DOI
  68. [68]Planck Collaboration, Y. Akrami, M. Ashdown, J. Aumont, C. Baccigalupi, M. Ballardini et al., Planck 2018 results. XI. Polarized dust foregrounds, A&A 641 (2020) A11 [1801.04945].
  69. [69]Planck Collaboration, Y. Akrami, M. Ashdown, J. Aumont, C. Baccigalupi, M. Ballardini et al., Planck 2018 results. IV. Diffuse component separation, A&A 641 (2020) A4 [1807.06208].
  70. [70]J. Chluba, D. Paoletti, F. Finelli and J.-A. Rubino-Martin, Effect of primordial magnetic fields on the ionization history, ArXiv:1503.04827 (2015) [1503.04827].arxiv.org/abs/1503.04827
  71. [71]K. L. Greene and F.-Y. Cyr-Racine, Thomson scattering: one rate to rule them all, JCAP 2023 (2023) 065 [2306.06165].DOI
  72. [72]C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau et al., Array programming with NumPy, Nature 585 (2020) 357 [2006.10256].

Paper details

vlnc:W_kh9h75hs
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents