SPT-3G D1: A Measurement of Secondary Cosmic Microwave Background Anisotropy Power

TL;DR

SPT-3G D1 delivers high-precision millimeter-wave power spectra over $1700 \le \ell \le 11{,}000$ across 95/150/220 GHz, enabling joint constraints on the thermal and kinematic SZ effects, the CIB, and radio galaxies. The work tests multiple emission models, from template SZ/CIB spectra to flexible, spline-based representations, to assess model-dependence in inferred SZ powers and CIB properties. The analyses yield $D_{3000}^{\rm tSZ}$ around 4.6–5.0 $\mu{\rm K}^2$ and $D_{3000}^{\rm kSZ}$ around 1.7–2.9 $\mu{\rm K}^2$ (at 143 GHz), with a detectable tSZ-CIB correlation and robust CIB power measurements; implications for reionization are drawn from the kSZ signal, though depend on assumptions about homogeneous kSZ power. These results advance foreground modeling for current and next-generation CMB experiments and set the stage for tighter constraints with the full SPT-3G survey and future observatories.

Abstract

We report new measurements of millimeter-wave temperature power spectra in the angular multipole range $1700 \le \ell \le 11,000$ (wavelengths $13^\prime \gtrsim λ\gtrsim 2^\prime$). We use two years of data in three observing bands centered near 95, 150, and 220 GHz from the SPT-3G receiver on the South Pole Telescope that cover a 1646 deg$^2$ region of the Southern sky. Using the measured power spectra, we present constraints on the thermal and kinematic Sunyaev-Zel'dovich (SZ) effects, radio galaxies, and cosmic infrared background (CIB). We find that inferred SZ powers are dependent on the detailed modeling of the thermal SZ-CIB correlation, and to a lesser extent on the assumed angular dependence of the SZ spectra. We report constraints for simulation-based model templates as well as fits where the angular dependencies of the SZ and CIB power spectra are allowed to vary. In the latter case at $\ell=3000$, we find thermal SZ power at 143 GHz of $D_{3000}^{\rm tSZ} = 4.91\pm0.37\, μ{\rm K}^2$ and kinematic SZ power of $D_{3000}^{\rm kSZ} =1.75\pm0.86\, μ{\rm K}^2$. We use the measured kinematic SZ power to estimate the duration of reionization, noting that the reionization inferences are sensitive to the model choices and assumed level of homogeneous kinematic SZ power from the late-time universe. We find a 95% limit on the duration from an ionization fraction of 25% to 75% of $Δ^{50} z_{\rm re} <\,3.8$ based on a semi-analytic model, or a limit on the duration from an ionization fraction of 5% to 95% of $Δ^{90} z_{\rm re} <\,6.1$ based on the AMBER simulations.

Figures (11)

• Figure 1: The six auto- and cross-spectra measured with the 95, 150, and 220 GHz SPT-3G data. To be more visible, the plotted uncertainties have been scaled up by a factor of ten. The plotted uncertainties are based on the diagonal elements of the covariance matrix, which includes noise and sample variance but not beam or calibration errors.
• Figure 2: The six auto- and cross-spectra measured with the 95, 150, and 220 GHz SPT-3G data (black circles) are plotted along with measurements by ACT DR6 (light green, upside-down triangles, louis25) and a combination of SPT-SZ and SPTpol (purple pentagons, reichardt21). Note that the plotted ACT bandpowers are array cross spectra and so have multiple points at the same $\ell$ for some frequency combinations. We stress that no correction has been made for the different observing bands or point source masking thresholds between the three measurements, which explains the observed differences in the power observed from AGN and dusty galaxies between the three experiments. ACT DR6 masked sources above 15 mJy at 150 GHz, while this work and reichardt21 masked sources above 6 mJy at 150 GHz. The plotted uncertainties are based on the diagonal elements of the covariance matrix, which include noise and sample variance, but not beam or calibration errors. Note that uncertainties for all three datasets have been increased by a factor of ten to be seen more easily.
• Figure 3: To illustrate where the data and best-fit models diverge, we show the bandpower differences between the model and data divided by the square root of the diagonal of the covariance matrix. The three cases shown are the best-fit points of the three models in Table \ref{['tab:chisq']}.
• Figure 4: Results for the median power (lines) and 68.3% confidence region (shaded area) for the tSZ power (left panel) and kSZ power (right panel). The orange dot-dashed lines and region mark the results when using the Agora templates for the tSZ, kSZ, and tSZ-CIB spectra, along with a CIB fit for the Poisson, 1-halo, and 2-halo terms. The other lines all use the free CIB model. The black solid line and grey region show the results for the $\ell^\alpha$ rescaling of the SZ templates used by G15. We show the unscaled shape of the G15 kSZ and tSZ templates with dotted lines; we stress that the amplitudes are set to match the black line at $\ell=3000$, rather than a fit. The $\ell^\alpha$ rescaling favors a tSZ power spectrum in between the Agora and G15 tSZ predictions, but a kSZ power spectrum that rises more at small angular scales than either template. The blue dashed line and region show the inferred SZ power in the free CIB+SZ case. We note that the inferred SZ power is fairly stable across model choices near $\ell=3000$, but we see variations at smaller angular scales, primarily in the better-constrained tSZ power spectrum. For reference, we show in red the constraints from Planck + ACT DR6 $\Lambda$CDM chains released with louis25. The DR6 results agree well with our results at $\ell=3000$. In yellow, we show the tSZ constraints from efstathiou25, which are within 2 $\sigma$ of all three fits of this work, except for the highest multipole bin.
• Figure 5: 1D and 2D posteriors for the kSZ power ($D_{3000}^{\rm kSZ}$), tSZ power at 143 GHz ($D_{3000}^{\rm tSZ}$), and tSZ-CIB correlation ($\xi_{3000}$), all evaluated at $\ell=3000$. The results when using a template fit based on the Agora tSZ, kSZ, and tSZ-CIB spectra are shown in orange, while the results for the more flexible free CIB model with an $\ell^\alpha$ rescaling of the SZ templates is shown in grey for the G15 kSZ and tSZ templates and in purple for the Agora kSZ and tSZ templates. The inferences are robust between these cases, with no significant shifts between the three model choices. There are clear degeneracies between all three parameters with higher levels of tSZ power correlating to lower levels of kSZ power and tSZ-CIB correlation. Equivalently, higher correlation between the tSZ and CIB emission reduces the tSZ power and increases the kSZ power.