The Primordial power spectrum from the largest to smallest CMB scales

Technical Summary: The Primordial Power Spectrum from the Largest to Smallest CMB Scales

Problem Statement
The primordial power spectrum (PPS) encodes the initial conditions of the Universe generated during inflation. While the standard $\Lambda$CDM model assumes a nearly scale-invariant power-law PPS, deviations from this form (such as oscillatory features, bumps, or changes in spectral tilt) could reveal specific physics of the inflationary paradigm, such as sharp features in the potential or variations in the sound speed. Previous studies have identified potential anomalies in Cosmic Microwave Background (CMB) data, including hints of a "blueward tilt" at small scales in Atacama Cosmology Telescope (ACT) data and localized features. However, the consistency of these features across different CMB missions (Planck, ACT, and South Pole Telescope SPT) and their statistical significance remain subjects of investigation. This paper aims to reconstruct the PPS across the widest possible range of cosmological scales using model-independent methods and to rigorously test the consistency between these datasets.

Methodology
The authors employ a "top-down" approach to reconstruct the non-parametric PPS directly from observed CMB angular power spectra ($C_\ell$). The core methodology involves:

1. Modified Richardson-Lucy (MRL) Deconvolution: The authors utilize the Modified Richardson-Lucy algorithm to invert the integral relation connecting the primordial power spectrum $P_R(k)$ to the observed $C_\ell$. This iterative algorithm updates the power spectrum estimate to minimize the residual between theoretical and observed spectra.
   • Uncertainty Handling: Unlike the standard RL algorithm, the MRL version incorporates a $\tanh^2$ weighting term to account for data uncertainties ($C_\ell$ errors), preventing the algorithm from over-fitting noise.
   • Regularization: To mitigate unphysical sharp oscillations arising from noise over-fitting, three regularization schemes are applied:
     • Gaussian Regularization (GR): Convolution with a Gaussian kernel in log-space.
     • Diffusive Regularization (DR): Adds a second-derivative term inspired by the heat diffusion equation to smooth sharp features while preserving major structures.
     • Total Variation Regularization (TVR): Uses an $L_1$-norm penalty on the derivative to preserve edges while smoothing.
2. Datasets: The analysis utilizes the latest data releases:
   • Planck PR4 (CamSpec-NPIPE): Covering $\ell \approx 2-2500$.
   • ACT-DR6: Covering $\ell \approx 593-4810$.
   • SPT-3G D1: Covering $\ell \approx 400-3000$ (truncated for consistency with TT data).
   • Combination: The authors concatenate TT, TE, and EE spectra, selecting the dataset with lower uncertainties in overlapping multipole ranges to create combined reconstructions (e.g., Planck+ACT covering $\ell=2-4810$).
3. Parametric Validation: To quantify specific deviations found in the non-parametric reconstructions, the authors perform a Bayesian analysis using a "double-tilt" power-law model. This model allows the spectral index to change from $n_{s1}$ to $n_{s2}$ at a break scale $k_{break}$, testing the significance of the parameter $\Delta n = n_{s2} - n_{s1}$.

Key Contributions

• Comprehensive Reconstruction: This work presents the first model-independent reconstruction of the PPS combining Planck, ACT, and SPT data, covering the full accessible CMB scale range from large angular scales ($\ell \sim 2$) to small scales ($\ell \sim 4810$).
• Cross-Dataset Consistency Check: The study explicitly tests the consistency of reconstructed features across different missions and background cosmologies (using ACT, Planck PR3, and Planck PR4 best-fit cosmologies as backgrounds).
• Statistical Rigor: The authors employ simulation-based error analysis (generating 1000 power-law realizations) to establish confidence intervals and p-values for any observed features, distinguishing between physical signals and reconstruction artifacts.

Results

1. Non-Parametric Reconstructions:

   • Planck (CamSpec): The reconstruction shows no statistically significant deviations from a power-law spectrum.
   • SPT: Consistent with a power-law spectrum when using its own best-fit cosmology.
   • ACT: When reconstructed using ACT's own best-fit cosmology, the PPS is consistent with a power law. However, when reconstructed using Planck PR3 or PR4 background cosmologies, the ACT data shows deviations, particularly a localized bump around $k \approx 0.33-0.34$ Mpc$^{-1}$ and a deviation in the range $k \approx 0.15-0.24$ Mpc$^{-1}$.
   • Correlation: A strong correlation is found between Planck and ACT reconstructions in the overlapping range ($0.07 < k < 0.14$ Mpc$^{-1}$), even at the level of localized features. In contrast, SPT reconstructions often exhibit out-of-phase behavior relative to Planck and ACT.
   • Significance: None of the features observed in individual datasets survive the simulation-based error analysis as statistically significant (i.e., they fall within the 99% confidence intervals of power-law simulations).

2. Parametric Analysis (Double-Tilt Model):

   • The authors investigate the "blueward tilt" preference in ACT data at small scales using a double-tilt model.
   • Planck PR3 + ACT: The preference for a double tilt (where $n_{s2} > n_{s1}$) is marginal, reaching significance levels of roughly $1\sigma$ to $2\sigma$ depending on the likelihood combination and whether nuisance parameters are sampled or fixed.
   • Planck PR4 + ACT: The discrepancy is slightly more pronounced. When nuisance parameters are fixed to their best-fit values, the deviation from a single-tilt spectrum reaches up to $\sim 5\sigma$ for the CamSpec+ACT combination. However, when nuisance parameters are marginalized (sampled), the significance drops to the $1\sigma$ level.
   • ACT Only: Analysis of ACT data alone (with a fixed break scale at $k \approx 0.30$ Mpc$^{-1}$) shows no significant evidence for a double tilt; the data remains consistent with a single-tilt power law.

Significance and Claims
The paper concludes that there is no decisive evidence for significant deviations from a power-law primordial spectrum across the full range of CMB scales.

• Consistency: The Planck, ACT, and SPT datasets are generally consistent with a featureless, nearly scale-invariant power-law PPS.
• Tensions: While a marginal disagreement exists between Planck and ACT data (particularly when using Planck's background cosmology for ACT reconstruction), this tension is not statistically robust enough to rule out the standard power-law model. The observed "blueward tilt" in ACT data is preferred only at the $1\sigma$ level in a rigorous Bayesian analysis with marginalized nuisance parameters.
• Conclusion: The study reinforces the standard inflationary paradigm's prediction of a nearly scale-invariant spectrum, suggesting that previously reported anomalies are likely statistical fluctuations or artifacts of specific background assumptions rather than definitive signatures of new inflationary physics.