Planck Data Reconsidered

TL;DR

This paper challenges Planck-derived cosmological parameters by reanalyzing Planck data using map-based foreground cleaning with high-frequency channels and survey cross-spectra. It finds that the 217×217 spectrum likely carries detector-related systematics, and that, once cleaning and cross-spectra are employed, the inferred ΛCDM parameters shift toward pre-Planck values, reducing tensions with external measurements. The authors show that the main drivers of discrepancy are data-handling choices—particularly detector-set spectra and single-survey regions—rather than foreground modeling or sky coverage per se. Their approach underscores the importance of independent analyses and cross-spectra in robust cosmological inference and highlights remaining questions about Planck systematics once full-five-season data are released.

Abstract

The tension between the best fit parameters derived by the Planck team and a number of other astronomical measurements suggests either systematics in the astronomical measurements, systematics in the Planck data, the need for new physics, or a combination thereof. We re-analyze the Planck data and find that the $217\,\text{GHz}\times 217\,\text{GHz}$ detector set spectrum used in the Planck analysis is responsible for some of this tension. We use a map-based foreground cleaning procedure, relying on a combination of 353 GHz and 545 GHz maps to reduce residual foregrounds in the intermediate frequency maps used for cosmological inference. For our baseline data analysis, which uses 47% of the sky and makes use of both 353 and 545 GHz data for foreground cleaning, we find the $Λ$CDM cosmological parameters $Ω_c h^2 = 0.1170 \pm 0.0025$, $n_s = 0.9686 \pm 0.0069$, $H_0 = 68.0 \pm 1.1\,\mathrm{km} \mathrm{s}^{-1}\mathrm{Mpc}^{-1}$, $Ω_b h^2 = 0.02197 \pm 0.00026$, $\ln 10^{10}A_s = 3.082 \pm 0.025$, and $τ= 0.090 \pm 0.013 $. While in broad agreement with the results reported by the Planck team, these revised parameters imply a universe with a lower matter density of $Ω_m=0.302\pm0.015$, and parameter values generally more consistent with pre-Planck CMB analyses and astronomical observations. We compare our cleaning procedure with the foreground modeling used by the Planck team and find good agreement. The difference in parameters between our analysis and that of the Planck team is mostly due to our use of cross-spectra from the publicly available survey maps instead of their use of the detector set cross-spectra which include pixels only observed in one of the surveys. We show evidence suggesting residual systematics in the detector set spectra used in the Planck likelihood code, which is substantially reduced for our spectra.

Figures (12)

• Figure 1: Means for $n_s$, $\Omega_M$ and $H_0$ derived from the publicly available Planck likelihood code (black) and without the $217\times217$ data (gray), both for the Planck+WP data set. The contours indicate the regions that contain $68$ and $95\%$ of our $500$ simulations with $217\times217$ spectra drawn from the conditional probability \ref{['eq:predict']}.
• Figure 2: Predicted versus observed power spectra: This figure shows the difference between the published $217\times217$ spectrum binned with $\Delta \ell = 25$ (points) and the spectrum predicted from the published Planck $100\times 100$, $143\times 143$, and $143\times217$ spectra (filled band indicates 68% CL). Note that the points are systematically below the model for $1700<\ell <1900$ and above the model for $2100<\ell< 2400.$ In the boxed insert, we show the predicted spectrum and observed spectrum for each multipole between $500 < \ell< 600$ where predicted and measured spectra agree remarkably well.
• Figure 3: This figure shows the more conservative mask used in the Planck analysis in which foregrounds are modeled, SA24 (left) and the less conservative mask used in our main analysis, SA47 (right). These masks are a product of a point source mask, a mask that excludes pixels observed only in a single survey, and galactic masks with $f_\text{sky}=35\%$ for SA24 and $70\%$ for SA47. With the SA24 mask, there is $24.3\%$ of the sky available for analysis, with the SA47 mask there is $46.8\%$ of the sky available for analysis.
• Figure 4: This figure shows the amplitude of the optimal coefficient to minimize foregrounds based on minimizing $((1+\alpha)T_{217} - \alpha T_{353})^2$ as a function of multipole number $\ell$ and sky cut. The blue dots are for mask SA24 and the orange squares for SA47. The dashed line shows the value chosen for map cleaning which was fixed to minimize the extragalactic foregrounds. There is no significant variation in the cleaning coefficient with multipole or sky cut.
• Figure 5: Planck spectra before and after foreground cleaning for mask SA47. Left panel: Planck spectra from the uncleaned maps, masked to remove point sources and galactic emission. Right panel: Foreground-cleaned Planck spectra using the 545 GHz data to clean the maps used in the cosmological analysis. The solid line shows the best-fit CMB spectrum while the dashed line indicates the residual Poisson foreground emission after cleaning with an amplitude of $458~(\mu$K$)^2$ at $\ell=2000$, as shown in Figure \ref{['fig:resid']}.