Hyrax: An Extensible Framework for Rapid ML Experimentation and Unsupervised Discovery in the Era of Rubin, Roman, and Euclid

2026-05-18T18:00:06Z

The NSF-DOE Vera C. Rubin Observatory, Roman Space Telescope, Euclid, and other next-generation surveys will deliver imaging, spectroscopic, and time-domain data at scales that increasingly shift the bottleneck in astronomical machine learning (ML) projects from model design to infrastructure. We present Hyrax, an open-source, modular, GPU-enabled Python framework that supports the full ML lifecycle in astronomy: from data acquisition and training to inference and experiment comparison, with capabilities including multimodal dataset support, integrated vector databases for similarity search, and interactive two- and three-dimensional latent-space exploration for unsupervised discovery. We demonstrate Hyrax's versatility through five representative applications on real survey data: (i) unsupervised representation learning on $\sim 4\times10^5$ Rubin Legacy Survey of Space and Time (LSST) Data Preview 1 (DP1) galaxies, surfacing new merger and low-surface-brightness candidates missing from reference Euclid and Dark Energy Survey catalogs, while also isolating imaging artifacts -- all without labeled training data; (ii) hybrid density-based clustering for identifying cluster-scale gravitational lens candidates in DP1 data; (iii) multimodal early-time transient classification in the Zwicky Transient Facility leveraging light curves, spectra, images, and metadata; (iv) supervised false-positive filtering in shift-and-stack searches for distant solar system objects in the Dark Energy Camera Ecliptic Exploration Project survey; and (v) supervised detection of semi-resolved dwarf galaxies in Hyper Suprime-Cam and LSST-like imaging using synthetic source injection. Together, these results demonstrate that Hyrax provides astronomy-specific ML infrastructure that enables systematic discovery and rapid methodological iteration across next-generation astronomical surveys.

The pre-infall bias of subhalos

2026-05-18T18:00:00Z

Dark matter halos destined to fall into a more massive host differ from typical field halos of the same mass even before infall. In cosmological simulations, we find that the progenitor mass functions of these "future subhalos" are systematically shifted toward higher masses, with the shift growing as infall approaches. The bias takes a compact form within extended Press-Schechter theory: the collapse barrier is multiplied by a function $β(D/D_\mathrm{infall},a)$, where $D$ is the linear growth factor at scale factor $a$ and $D_\mathrm{infall}$ is the growth factor at infall. We find $β(x,a)=(1-x)^{1.20+0.14a}$ for the $M_{200\mathrm{c}}$ mass definition and $(1-x)^{1.20+0.05a}$ for $M_{200\mathrm{m}}$; the explicit scale-factor dependence captures the late-time influence of dark energy. One consequence is that halos shortly before infall are 10-15% more centrally concentrated than typical field halos of the same mass.

The Majoron Cosmological Window: Dark Matter and Thermal Leptogenesis

2026-05-18T18:00:00Z

The majoron is the Nambu-Goldstone boson associated with the spontaneous breaking of a global $B-L$ symmetry. Remarkably, the minimal majoron framework can simultaneously address three key empirical indications of physics beyond the Standard Model: neutrino masses, the matter-antimatter asymmetry, and dark matter. In this work, we identify the cosmologically viable region in which majoron dark matter and high-scale thermal leptogenesis can be realised simultaneously. We show that successful leptogenesis plays a central role in making this scenario predictive: by constraining the right-handed-neutrino mass scale, it determines the irreducible freeze-in contribution to the majoron abundance and fixes the size of the couplings relevant for visible dark matter decays. Combining the irreducible dark matter production mechanisms with warm dark matter limits and indirect searches for decaying dark matter, we map the resulting majoron cosmological window and show that future X- and gamma-ray telescopes can probe part of the surviving parameter space.

The Lumina Project: CMB Optical Depth Fluctuations from Patchy Reionization

2026-05-18T18:00:00Z

Patchy reionization couples the ionized-bubble morphology to the underlying density field, making the CMB Thomson optical depth sensitive to both the global ionization history and anisotropic fluctuations on the sky. Using the large-volume radiation-hydrodynamical Lumina simulation, we compute $τ_{\rm CMB}$ in two ways: (i) from global volume- and mass-weighted ionization histories, and (ii) from explicit line-of-sight integrations through on-the-fly light cones. We find that the sightline-averaged optical depth in the light cone, $\langle τ_{\rm LOS} \rangle = 0.0550$, exceeds the value inferred from a global volume-weighted history, $τ_{{\rm CMB},V} = 0.0515$, by $\approx 7\%$. This enhancement is largely captured by the global mass-weighted prediction, $τ_{{\rm CMB},m} = 0.0544$, indicating that precision comparisons to CMB optical-depth constraints should use mass-weighted electron fractions or explicit light-cone integration rather than volume-weighted ionized fractions alone. The excess optical depth accumulates primarily near $z_{\rm LOS} = 8.0^{+1.9}_{-1.3}$, where the combination of high physical density and strong ionization-field patchiness is greatest. The resulting $τ_{\rm LOS}$ field is non-Gaussian and exhibits $\gtrsim 5\%$ sightline-to-sightline scatter, with fluctuations tracing rare early-ionized overdensities and large-scale structure. Coarse-graining experiments show that smoothing the ionization field on $\gtrsim 3 {\rm cMpc}$ scales suppresses the density-ionization correlation and biases $τ_{\rm CMB}$ low relative to the resolved calculation. Finally, angular power spectra and real-space correlation functions decomposed into HII, HeII, and HeIII auto- and cross-contributions reveal scale-dependent departures from simple hydrogen-helium co-tracing and evolving characteristic scales with redshift.

First detection of the moving lens effect with ACT and DESI LS

2026-05-18T17:59:32Z

The moving lens effect is a secondary CMB anisotropy induced by the transverse motion of gravitational potentials. We develop a Fourier-space cross-spectrum estimator that retains the scale dependence of the signal, and apply it to the Atacama Cosmology Telescope (ACT) DR6 CMB temperature maps and luminous red galaxies from the DESI Legacy Imaging Surveys. Using the foreground-reduced ACT NILC map, we find strong evidence for a non-zero amplitude of the cross-correlation $b_{\rm ML} = 1.24 \pm 0.26$ ($4.8σ$) for the extended sample and $0.93 \pm 0.25$ ($3.7σ$) for the main sample, both consistent with the halo-model prediction for the moving lens signal. Our Fourier-based pipeline enforces separation of scales between the reconstructed velocities and the cross-correlation, which we show is essential for foreground mitigation. The residual foreground contamination is expected to be significantly smaller than the signal from both simulations and the multi-frequency analysis presented in this paper. No curl-mode test exceeds $2σ$, and the results are robust across analysis variants. They constitute the first detection of the moving lens effect and unlock access to transverse velocities, a new cosmological probe. When combined with the kinematic Sunyaev-Zel'dovich effect, this provides a path toward mapping the three-dimensional velocity field of the Universe, opening a new avenue for probing the growth of structure and gravity on large scales.

Recoupled Dark Radiation reconciling CMB and DESI BAO measurements

2026-05-18T17:47:36Z

Recent DESI BAO measurements, combined with CMB data, reveal a tension within the $Λ$CDM model that leads to a discrepancy between cosmological and laboratory bounds on the summed neutrino mass. We show that a recoupled interacting radiation component can alleviate this cosmological tension, as well as the one with neutrino oscillation experiments. Sterile neutrinos interacting through a light pseudoscalar mediator provide a concrete realization of this scenario. The resulting interacting fluid modifies the CMB phenomenology, lowers the preferred matter density, and improves the consistency between CMB and DESI BAO measurements. Combining CMB with DESI DR2 BAO measurements, we find a $2.7 σ$ preference for a nonzero interacting sterile neutrino component, $\nsp=0.253 \pm 0.094$, corresponding to an improvement $Δχ^2=-8.98$ relative to $Λ$CDM. The model also reduces the tension with the SH0ES determination of the Hubble constant to the $2.4σ$ level.

The effects of dark energy on the matter-gravity coupling

2026-05-18T17:46:49Z

Dark energy perturbations are expected to modify the evolution of cosmological perturbations, producing different observational effects. We show that these effects can be encoded in a momentum, space and polarization dependent effective matter-gravity coupling. For scalar perturbations the effective gravitational coupling can be locally negative, in regions with local dark energy under-densities. For adiabatic perturbations the effective gravitational coupling for scalar perturbations can be negative for phantom dark energy, providing a possible explanation for the observed structure suppression at low redshift, consistent with other independent evidences of evolving dark energy. For gravitational waves the effective coupling can make the radiation emitted by compact binary convalescences polarization and frequency dependent.

Testing the consistency of gravitational waves and large scale structure constraints on dark energy

2026-05-18T17:40:49Z

Gravitational wave (GW) astronomy has opened a new window on the Universe, allowing to obtain constraints on dark energy and gravity independent from other electromagnetic waves observations, such as large scale structure (LSS). For the purpose of investigating the consistency between different observations the effective field theory (EFT) of dark energy is a useful tool, allowing to derive model and parametrization independent consistency relations (CR) between the effective gravitational constant, the slip parameter, the gravitational and electromagnetic luminosity (EM) distance, and the speed of GWs. We test the constant brading and no-slip CRs, by comparing for the first time the constraints on the effective gravitational coupling obtained from LSS observations with those from GW events with and without electromagnetic counterparts, confirming the validity of the CRs at the current level of experimental uncertainty. The event GW170817 and its electromagnetic counterpart provides a constraint of the effective gravitational constant with an accuracy comparable with LSS constraints, while the analysis of GW events without electromagnetic counterpart are consistent, but do not have a constraining power comparable to LSS observations. Beside allowing to test the consistency between independent observations, the CRs can be used to estimate the effective gravitational coupling with GWs at high redshift, where other observations are not available.

Faster CMB lensing with control variates

2026-05-18T17:05:30Z

We present a new method for fast computation of the realization-dependent bias, a major computational bottleneck in measurements of the cosmic microwave background (CMB) lensing power spectrum. The method accelerates the bias calculation by differencing two correlated estimates: one based on fully realistic masked simulations and the other on isotropic simulations, for which the bias is analytically tractable. We show that our algorithm reduces the total computational cost of a lensing power spectrum measurement by approximately a factor of five for Atacama Cosmology Telescope- or Simons Observatory-like noise levels, or by a factor of three if current anisotropic filtering methods are left unchanged. Owing to its simplicity, the method can be readily implemented in existing CMB lensing analysis pipelines.

Non-Gaussianity in SMICA

2026-05-18T16:28:39Z

We develop a new formalism for the component separation method Spectral Matching Independent Component Analysis (SMICA) in order to include the information contained in the foregrounds beyond second-order statistics. We also develop a binned bispectrum estimator that works directly using maps of different frequency channels, capable of determining the bispectrum of multiple components at the same time, shifting the traditional approach to non-Gaussianity estimation from a cleaned map to the component separation step, for a better handling of foreground uncertainty. We test our method on 400 E and B polarization simulations based on the LiteBIRD experiment, containing the two main sources of contamination for CMB polarization experiments: polarized dust and synchrotron emission. We show that the bispectrum does not improve the precision of the power spectrum estimation or of the spectral parameters. However, we are capable of recovering the correct 3-point correlator of the foregrounds and standard constraints on primordial non-Gaussianity in a coherent multi-frequency and multi-component framework. The advantage of our approach is that it combines data in an optimal way accounting for the power spectrum and the bispectrum of the various components, which is not true for the standard approach.

Primordial power spectrum reconstructions from BOSS + eBOSS

2026-05-18T16:25:02Z

We reconstruct the primordial power spectrum $P_{\mathcal{R}}(k)$ from the BOSS DR 12 LRG and eBOSS DR 16 QSO catalogs with a non-parametric Bayesian method. The $P_{\mathcal{R}}(k)$ is reconstructed by linearly interpolating $N$ knots in the $\{ \log k, \log P_{\mathcal{R}}(k) \}$ plane. We use a parametric model to describe the galaxy power spectra of the BOSS+eBOSS catalogs, assuming any power-law deviations and BAO contributions separately from the matter power spectrum template, composed of seven parameters $Θ_{\text{model}}$. This parametric model enables us to reconstruct $P_{\mathcal{R}}(k)$ at non-linear scales, reaching $k = 0.3 \text{ h} \text{ Mpc}^{-1}$. The method is validated by applying it to different Primordial Features (PF) templates and by recovering the input power law of \textsc{MD-Patchy} and \textsc{EZmock} mock catalogs, representative of the BOSS and eBOSS data. These mocks provide additional information on $Θ_{\text{model}}$, enabling us to impose Gaussian correlated priors on $Θ_{\text{model}}$. This prior set allows us to reconstruct $P_{\mathcal{R}}(k)$ more precisely and to alleviate the degeneracies between the model and knot parameters. The results for both individual and combined $z$-bins and galactic caps of the BOSS and eBOSS catalogs are consistent, showing no evidence of the presence of PF in $P_{\mathcal{R}}(k)$ and pointing to a quasi-scale-invariant power law as the preferred model for $P_{\mathcal{R}}(k)$, as predicted by most slow-roll inflationary models. With a different prior set that relaxes the Gaussian constraints on $Θ_{\text{model}}$ and imposes Planck-based priors on the extreme knots, the results also favor the power law. From the knot reconstructions, we robustly constrain the spectral index $n_s = 0.976 \pm 0.021$, compatible with the Planck value.

Constraints on primordial non-Gaussianity from Quaia

2026-05-18T16:14:42Z

We analyse the large-scale angular clustering of quasars in the \gaia-\unwise quasar catalog, \quaia, and their cross-correlation with maps of the lensing convergence of the Cosmic Microwave Background (CMB), to constrain the level of primordial non-Gaussianity (PNG). Specifically, we target the scale-dependent bias that would be induced by PNG on biased tracers of the matter inhomogeneities on large scales. The \quaia sample is particularly well suited for this analysis, given the large effective volume covered, and our ability to map out the main potential sources of systematic contamination and mitigate their impact. Using the universality relation to characterise the response of the quasar overdensity to PNG ($p_φ=1$), we report constraints on the local-type PNG parameter $f_{\rm NL}$ of $f_{\rm NL} =-20.5^{+19.0}_{-18.1}$ (68\% C.L.) by combining the quasar auto-correlation and its cross-correlation with CMB lensing in two tomographic redshift bins (or $f_{\rm NL} =-28.7^{+26.1}_{-24.6}$ if assuming a lower response for quasars, $p_φ=1.6$). The error on $f_{\rm NL}$ can be further improved if the cross-correlation between the tomographic redshift bins is included. Using the CMB lensing cross-correlations alone, we find $f_{\rm NL} =-13.8^{+26.7}_{-25.0}$ and $f_{\rm NL} = -15.6^{+42.3}_{-34.8}$ for $p_φ=1$ and $p_φ=1.6$ respectively. These are the tightest constraints on $f_{\rm NL}$ to date from angular clustering statistics and cross-correlations with CMB lensing.

FLAMINGO: Tracing the co-evolution of hot gas and black holes in galaxy groups and clusters

2026-05-18T15:35:11Z

The gas mass fraction of galaxy groups and clusters is a key physical quantity for constraining the impact of feedback processes on large-scale structure. While several modern cosmological simulations use the gas fraction-halo mass relation to calibrate their feedback implementations, we note that this relation exhibits substantial intrinsic scatter whose origin has not been fully elucidated. Using the large-volume FLAMINGO hydrodynamical simulations, we examine the role of both central and satellite supermassive black holes (BHs) in shaping this scatter, probing higher halo masses than previously possible. For haloes with M500 < 10^13 Msun, we find that central BH mass correlates strongly and negatively with gas fraction, such that higher BH masses give rise to lower gas fractions at fixed halo mass, consistent with previous studies. Interestingly, however, for 10^13 Msun < M500 < 10^14.5 Msun the correlation reverses and becomes positive, with overmassive BHs residing in haloes with above-average gas fractions. By tracing progenitor BHs and haloes through cosmic time, we show that this behaviour is driven by the expulsion and subsequent re-accretion of halo gas, regulated by the timing of BH growth and feedback. Specifically, haloes that collapse earlier form BHs earlier, leading to earlier gas expulsion and re-accretion and a high gas fraction compared to haloes of the same present-day mass that formed later. Our results demonstrate that present-day scatter in the gas fraction-halo mass relation is strongly shaped by the early growth history of BHs and their haloes, a prediction that can be tested with future observational measurements.

(An)Isotropy in Pantheon+ and Type Ia supernova samples: intrinsic limits of directional tests

2026-05-18T14:28:04Z

The use of methods that investigate the value of the Hubble constant H$_0$ in different patches (60 $ or 90$^ size) across the sky to probe the statistical isotropy of the Universe using large SNe Ia databases has led to contradictory claims of either anisotropy or isotropy. The anisotropy directions vary amongst research works. The objective of this paper is to clarify the abovementioned claims and study the lack of basis for depicting directions of anisotropy with the present SNe Ia samples. We explain the type of limitation embedded in the SN Ia lightcurve method to determine the isotropy of H_0 and the corresponding consequences. The widely used analysis through the Region Fitting and the Hemisphere Comparison methods is done here using the Pantheon+ database, simulating 2000 distinct directions in the sky within a Bayesian Markov Chain Monte Carlo approach. We also study a smaller SNe Ia database, the Carnegie Supernova Project sample, leading to a similar kind of result as that from the Pantheon+ sample. We investigate the validity of the directions found for anisotropy within these analyses. We have found that within the tests used here, the Region Fitting method and the Hemisphere Comparison method, one can not determine with robustness the direction of an anisotropy of H$_0$ using the present SNe Ia large data samples. This is intrinsic to the way H$_0$ is obtained with the SN Ia lightcurve method. Achieving robust constraints will require a quite uniform sky coverage from larger SNe Ia samples with improved systematics.

Cosmological perturbations of TDiff fields

2026-05-18T13:56:45Z

We study scalar field theories that break diffeomorphism invariance down to the subgroup of transverse diffeomorphisms through the matter sector in cosmological backgrounds. We focus on single- and multi-field models and develop the corresponding cosmological perturbation theory. We analyze the different contributions to the pressure perturbation, discussing the adiabaticity and the effects in the perturbation coefficients of the interactions that arise in the multi-field case as a consequence of the symmetry breaking. We also consider the stability of the perturbations in terms of the effective speed of sound and present particular models that could be of phenomenological interest.