You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Independent Reanalysis of Atacama Cosmology Telescope Data Release 6: Daytime and Nighttime Power Spectra, Map-Level Null Tests, Beam Systematics, and Differentiable Cosmological Inference
We present an independent end-to-end reanalysis of the Atacama Cosmology Telescope Data Release 6 (ACT DR6) daytime and nighttime sky maps across the PA4 (150, 220 GHz), PA5 (90, 150 GHz), and PA6 (90, 150 GHz) detector arrays, spanning angular scales $600 \le \ell \le 2500$ and combining cross-split power-spectrum reconstruction, map-level time-split null tests, Spin-2 polarization diagnostics, beam-systematic modeling, and GPU-accelerated Bayesian cosmological inference.
Evaluating mode-decoupled pseudo-$C_\ell$ estimators on the Needlet Internal Linear Combination (NILC) analysis footprint ($f_{\mathrm{sky}} = 0.312$) with point-source-subtracted maps, nominal instrument beams, and exact SACC bandpower window functions, we reproduce the official ACT DR6 nighttime auto-frequency bandpowers to within $0.1%\text{--}1.4%$ ($|\Delta D_\ell / D_\ell^{\mathrm{off}}| \le 0.006$ across $800 \le \ell \le 2000$ for PA5 $90,\mathrm{GHz}$, PA5 $150,\mathrm{GHz}$, and PA6 $150,\mathrm{GHz}$) and all six cross-frequency channels to within $0.2%\text{--}2.0%$ once co-temporal split pairs ($i = j$) are excluded to eliminate correlated atmospheric noise.
Map-level four-way time-split null tests $\langle (s_0 - s_1) \times (s_2 - s_3) \rangle$ and $\langle (s_0 - s_2) \times (s_1 - s_3) \rangle$ pass across the full multipole range with probabilities to exceed $\mathrm{PTE} = 0.45$ and $0.30$, respectively.
For the daytime dataset, we derive unbiased relative gain calibration factors $c_{\mathrm{day}} = \langle D_\ell^{n_0 \times n_1} / D_\ell^{d \times n} \rangle$ from day$\times$night cross-spectra across all six array-frequency bands, obtaining $c_{\mathrm{day}} = 1.030 \pm 0.003$ (PA5 $90,\mathrm{GHz}$) and $c_{\mathrm{day}} = 0.992 \pm 0.003$ (PA5 $150,\mathrm{GHz}$) on the NILC footprint, along with unit split-gain invariants $r_{\mathrm{split}}$ confirming phase coherence between the daytime and nighttime sky maps. Direct Spin-2 spherical harmonic analysis of the daytime Stokes $(Q, U)$ splits reveals parity-conserving and parity-violating residuals ($|D_\ell^{BB}| > |D_\ell^{EE}| \sim 10^2\text{--}10^3,\mu\mathrm{K}^2$, $|D_\ell^{EB}| \neq 0$) two to three orders of magnitude above the cosmological $E$-mode signal and uncorrelated with nighttime polarization, whereas nighttime temperature-to-polarization leakage remains below $0.036%$, establishing from first principles why daytime cosmological analyses must be restricted to temperature.
We further investigate a $2\sigma$ high-$\ell$ deficit in PA6 $90,\mathrm{GHz}$ nighttime auto-spectra, showing that a $1%$ linear beam-transfer correction above $\ell = 1000$ improves the full-covariance goodness of fit from $\chi^2/\mathrm{ndf} = 2.066$ ($\mathrm{PTE} = 5.3 \times 10^{-4}$) to $\chi^2/\mathrm{ndf} = 0.675$ ($\mathrm{PTE} = 0.910$), and demonstrate via a six-pair split covariance analysis and 14-mode beam principal-component projection of PA5 $90,\mathrm{GHz}$ that a localized $-6.7%$ residual at $\ell = 2375.5$ originates from map-level compact-source subtraction rather than smooth beam modes.
Incorporating our calibrated daytime and nighttime bandpowers into a differentiable neural-emulated $\Lambda\mathrm{CDM}$ likelihood, we find that daytime temperature data tighten $TT$-only parameter posteriors by $11.3%$ in $H_0$ and $12.1%$ in $n_s$, while contributing a $\simeq 1.0%$ tightening in the full $TT+TE+EE$ likelihood where nighttime polarization dominates the Fisher weight. Marginalizing over the 14 PA5 $90,\mathrm{GHz}$ beam uncertainty modes shifts $n_s$ by $+0.83\sigma$ ($0.9689 \pm 0.0070 \to 0.9765 \pm 0.0092$), highlighting the coupling between high-$\ell$ $90,\mathrm{GHz}$ beam calibration and the scalar spectral index.