ACT: microwave maps, calibration and lensing reconstruction

Microwave light above the Andes

The Atacama Cosmology Telescope observed the sky from the Andes in Chile. Its observations ended in 2022, but its measurements continued to yield new maps of the cosmic microwave background, including the images released in March 2025. This radiation carries information from approximately 380,000 years after the Big Bang, when cooling allowed neutral atoms to form and light could travel without continually scattering from free electrons.

ACT director Suzanne Staggs and analysis leader Jo Dunkley, both at Princeton University, connected the new intensity and polarization measurements with the early distribution and motion of gas. Deputy director Mark Devlin, at the University of Pennsylvania, explained how the apparent scale of patterns helps constrain cosmic expansion. The resulting early-universe measurements remained consistent with a lower expansion rate than nearby-galaxy methods, preserving the problem known as the Hubble tension.

Three frequency bands and a large footprint

The DR6 maps combine measurements collected with the AdvancedACT camera between 2017 and 2022. Three observing bands are centred on 98, 150 and 220 gigahertz. Together, the maps cover 19,000 square degrees, with a median combined depth of 10 microkelvin-arcminutes. They contain both temperature and polarization anisotropies at arcminute resolution. The observing period, frequency coverage and map footprint define this particular release; they are separate characteristics from the sky area selected for an individual statistical analysis.

Sigurd Naess and colleagues made the maps and derived products available through NASA's LAMBDA archive. An interactive web atlas and HiPS datasets provide additional ways to inspect the mapped sky. These products accompany the numerical maps rather than replacing them. The DR6 map publication, submitted on March 18, 2025, documents the instrument, mapmaking and map properties. Its three-band coverage supplies the starting point for later spectrum and component analyses.

Noise follows the atmosphere and scanning

ACT's map noise contains large-scale correlations arising from the atmosphere and the telescope's scanning strategy. Zachary Atkins and colleagues developed empirical Gaussian map-based models for these properties in March 2023. Their inputs are a small number of independent realizations covering the same sky region. A directional wavelet approach models correlated instrumental noise, alongside the other noise representations examined in the work. The models generate ensembles of noise realizations that can be compared with the actual DR6 data.

When these simulations entered the DR6 power-spectrum pipeline, the covariance-matrix diagonal was approximately 20 per cent higher than an analytic expression that treated the noise power spectrum as a sufficient description. The comparison uses covariance diagonals from simulated noise and the simplified analytic expression. Its denominator is the analytic covariance estimate, with the simulated realizations carrying the additional spatial correlations. The accompanying public code, mnms, supplies the map-based simulation machinery for these tests and for subsequent microwave-background analysis pipelines.

Estimating correlated statistical errors

A separate covariance calculation by Atkins and colleagues, submitted in December 2024, incorporated sharp features in signal and noise spectra, map-level anisotropies and uneven survey depth. Its analytic prescription accounts for these complicated map properties when estimating the covariance of the DR6 power spectra. Including the varying depth produces a structure similar to the approach used in Planck analysis. The team evaluated performance against Monte Carlo simulations rather than assuming that an analytic construction supplied sufficiently accurate errors.

The resulting prescription agreed with the simulations to better than 3 per cent, compared with a roughly 16 per cent difference for a simpler existing prescription. A further simulation-based correction brought both approaches into better than 1 per cent agreement. This correction also outperformed an alternative that assumed the analytic correlation matrix was already accurate. These percentages describe covariance-model comparisons under the stated tests; they are distinct from the noise-model comparison and do not share its numerical denominator.

Optics set an angle uncertainty

Detector polarization angles have their own instrumental uncertainty. Colin Murphy and colleagues used polarization-sensitive ray tracing in CODE V to estimate how perturbations in the optical system could affect the angles. Their framework included possible changes in lens positions and orientations, as well as the thickness and refractive index of anti-reflection coatings. The work first appeared in March 2024 and was revised in August that year.

For three ACT detector arrays operating between 90 and 220 gigahertz, millimetre-scale lens displacements and percentage-level coating perturbations produced a total systematic uncertainty on the scale of a tenth of a degree. This is an optical-model estimate under specified perturbations. More precise physical measurements of lens placement and orientation could reduce the estimated uncertainty. The angle calibration also matters for analyses sensitive to polarization leakage or a possible rotation of cosmic polarization.

Separating signals across frequencies

Microwave maps contain more than the cosmic background. Galactic emission and the thermal Sunyaev–Zel'dovich signal also contribute. William Coulton and colleagues combined ACT DR4 and DR6 measurements at roughly 93, 148 and 225 gigahertz with Planck measurements spanning 30 to 545 gigahertz. Their July 2023 component-separated products cover approximately 13,000 square degrees, or about a third of the sky. Those frequencies and that footprint describe this earlier combined analysis, rather than the complete March 2025 map release.

The products include CMB intensity and E-mode polarization maps, together with an arcminute-resolution Compton-y map tracing electron pressure integrated along the line of sight. An internal linear-combination pipeline implemented in a needlet frame suppresses Galactic contamination and accounts for spatial variations in ACT noise. The different products retain different physical signals. The pressure map supports studies of cluster and galaxy astrophysics, while the separated microwave background can enter polarization and gravitational-lensing reconstruction analyses.

Testing the foreground model

Benjamin Beringue and colleagues examined the astrophysical-emission model used in DR6 multi-frequency power spectra in work submitted in June 2025 and revised in November. They tested different foreground templates, additional components and extra free parameters. They also compared the model with earlier ACT observations and measurements from Planck and the South Pole Telescope. These comparisons address the emission superimposed on the microwave background, with explicit checks of how modelling choices affect fitted quantities.

Across the tested scenarios, the baseline cosmological parameters and those in a model with a variable effective number of relativistic species shifted by less than half a statistical standard deviation. Some foreground parameters moved more, requiring their own modelling-related uncertainty estimates. Independent simulations included non-Gaussian, correlated Galactic and extragalactic fields. Processing those skies through the power-spectrum and likelihood pipeline recovered the input cosmological parameters. The simulation tests concern these constructed skies and model variations; their scope does not remove every possible instrumental uncertainty.

From maps to cosmological parameters

Thibaut Louis and colleagues estimated DR6 temperature and polarization power spectra over 10,000 square degrees. The TT, TE and EE spectra reach arcminute scales. A combination of microwave-background and foreground signals fits the measurements within the standard Lambda cold-dark-matter model. The joint P-ACT dataset combines ACT with larger-scale Planck measurements; ACT combined with WMAP provides another comparison. The spectrum-analysis area is smaller than the full 19,000-square-degree map footprint. The June 2025 revision distinguishes two external-data combinations:

• P-ACT, ACT and Planck lensing, and DESI DR1 baryon acoustic oscillations yield a Hubble constant of 68.22 ± 0.36 kilometres per second per megaparsec.
• Including DESI DR2 tightens the value to 68.43 ± 0.27 in the same units.

These are fitted results for specified combinations, rather than interchangeable ACT-only measurements. The analysis also found consistent constraints when comparing temperature and polarization within the joint dataset.

Cleaning a lensing reconstruction

ACT lensing analysis requires additional foreground controls. Niall MacCrann and colleagues tested mitigation strategies against two independent microwave-sky simulations in April 2023. Their combination of point-source subtraction, cluster-model subtraction and a profile bias-hardened lensing estimator reduced the inferred lensing-amplitude bias below 0.2 statistical standard deviations in those tests. A different approach, deprojecting a model of the cosmic infrared background while adding high-frequency Planck information, achieved similar performance.

Other frequency-cleaned options incurred larger noise costs or produced a biased recovered lensing spectrum. The team also tested ACT data directly, comparing foreground sensitivity between the two ACT frequencies while nulling the CMB lensing signal. Tests at both map and bandpower level passed. The infrared-background-deprojected measurement agreed with the baseline reconstruction. Together, these checks address specific routes by which extragalactic emission could contaminate lensing; simulation bias estimates retain their stated relationship to statistical uncertainty.

A distinct lensing spectrum

Frank Qu and colleagues measured CMB lensing over 9,400 square degrees using five observing seasons of DR6 temperature and polarization data. Their work first appeared in April 2023, with a further revision in December 2025. The lensing-spectrum amplitude reached 2.3 per cent precision, corresponding to a detection significance of 43 standard deviations. A blinded analysis framework accompanied null tests, consistency checks and systematic-error estimates, alongside pipeline choices intended to reduce sensitivity to foregrounds and noise properties.

Relative to the best-fitting Planck 2018 microwave-background model, the baseline lensing amplitude was 1.013 ± 0.023. Relative to the ACT DR4 plus WMAP best fit, it was 1.005 ± 0.023. Both amplitudes are normalized to their respective best-fitting reference models. The measurements probe lensing across redshifts of approximately 0.5 to 5, and agreed with standard-model growth predictions extrapolated from much earlier microwave-background anisotropies. The tested spectrum did not provide evidence of suppressed structure amplitude at lower redshifts.

Combining lensing with distance information

Mathew Madhavacheril and colleagues used a reconstructed lensing mass map from ACT's 2017–2021 observations to constrain cosmological parameters. The analysis, submitted in April 2023 and revised in August 2024, combined ACT lensing with baryon acoustic oscillation measurements from SDSS and 6dF. That combination gave a Hubble constant of 68.3 ± 1.1 kilometres per second per megaparsec. Adding Planck lensing produced 68.1 ± 1.0, with the same units and a different combined dataset. The team also revisited KiDS, DES and HSC galaxy-survey results under a uniform set of assumptions. Their structure-amplitude estimates were lower than the ACT-plus-Planck lensing value by between 1.7 and 2.1 statistical standard deviations. The comparison involved differing scales and redshift sensitivity: CMB lensing mainly probes approximately redshifts 0.5–5 on mostly linear scales, whereas galaxy lensing samples smaller scales around redshift 0.5. The comparison therefore retains the different scale and redshift ranges of the CMB and galaxy measurements.

Galaxies provide another cross-check

Gerrit Farren and colleagues cross-correlated ACT and Planck lensing maps with the unWISE Blue and Green galaxy samples. The samples span approximate redshift ranges of 0.2–1.1 and 0.3–1.8 respectively. Their September 2023 analysis, revised in May 2024, incorporated additional spectroscopic information for redshift calibration and a more flexible theoretical model. Systematic and null tests formed part of a blind analysis, while the two galaxy samples supplied comparisons at different cosmic epochs.

ACT cross-correlations alone yielded a conventional structure-amplitude parameter S8 of 0.813 ± 0.021. Combining Planck and ACT cross-correlations gave 0.810 ± 0.015. Adding baryon acoustic oscillation information broke the degeneracy between matter density and fluctuation amplitude, allowing separate constraints on the latter. Measurements from the two samples, centred around redshifts 0.6 and 1.1, were consistent with standard growth expectations. These are cross-correlation constraints with particular galaxy selections, separate from the lensing power-spectrum amplitude and its reference-model normalization.

Testing alternatives with specified data

Erminia Calabrese, J. Colin Hill and colleagues used DR6 to test extensions of the standard cosmological model. Their March 2025 work, revised in June, used ACT power spectra alone and in combinations with Planck. Lensing and DESI Year-1 distance information helped break geometric degeneracies. For models affecting late-time expansion, they also included Pantheon+ supernova measurements. The choice of external information therefore depends on the model being tested.

The analyses found no statistically significant preference for departure from the baseline model. Tested possibilities included additional relativistic species, neutrino self-interactions, early dark energy, primordial magnetic fields, modified recombination and changes in fundamental constants. Neutrino properties remained consistent with standard expectations within the tested combinations. The extended-model analysis did not favour the examined modifications intended to increase the inferred Hubble constant or lower the inferred density-fluctuation amplitude. These constraints apply to the tested models and data combinations, rather than every conceivable explanation of the tensions.

Polarization rotation retains a systematic caveat

P. Diego-Palazuelos and E. Komatsu separately examined cosmic birefringence using DR6 polarization. Their work first appeared in September 2025 and was revised in April 2026. A Bayesian analysis used EB and TB correlations, with instrumental angle priors derived from ACT optical modelling. It also marginalized over residual leakage from intensity into polarization. Cosmic birefringence would rotate the plane of linear polarization, making angle calibration directly relevant to this particular measurement.

The fitted rotation was 0.215 ± 0.074 degrees at 68 per cent confidence, a 2.9-standard-deviation departure from zero. However, the authors explicitly identified remaining, insufficiently understood ACT systematics that prevented strong cosmological conclusions. Their polarization-rotation analysis found a sign and magnitude consistent with earlier WMAP and Planck results. The statistical preference and unresolved instrumental issues belong together: the result is a separate, qualified polarization measurement rather than an established detection of new physics.