We present a simple method for assessing the predictive performance of high-dimensional models directly in data space when only samples are available. Our approach is to compare the quantiles of observables predicted by a model to those of the observables themselves. In cases where the dimensionality of the observables is large (e.g., multiband galaxy photometry), we advocate that the comparison is made after projection onto a set of principal axes to reduce the dimensionality. We demonstrate our method on a series of two-dimensional examples. We then apply it to results from a state-of-the-art generative model for galaxy photometry () that generates predictions of colors and magnitudes by forward simulating from a 16-dimensional distribution of physical parameters represented by a score-based diffusion model. We validate the predictive performance of this model directly in a space of nine broadband colors. Although motivated by this specific example, we expect that the techniques we present will be broadly useful for evaluating the performance of flexible, nonparametric population models of this kind, and other settings where two sets of samples are to be compared.

We construct an antihalo void catalogue of 150 voids with radii 𝑅>10ℎ⁻¹Mpc in the Local Super-Volume (⁠<135ℎ⁻¹Mpc from the Milky Way), using posterior resimulation of initial conditions inferred by field-level inference with Bayesian Origin Reconstruction from Galaxies (BORG). We describe and make use of a new algorithm for creating a single, unified void catalogue by combining different samples from the posterior. The catalogue is complete out to 135ℎ⁻¹Mpc⁠, with void abundances matching theoretical predictions. Finally, we compute stacked density profiles of those voids which are reliably identified across posterior samples, and show that these are compatible with Λ cold dark matter expectations once environmental selection (e.g. the estimated ∼4 per cent underdensity of the Local Super-Volume) is accounted for.

Ultracold atomic gases can undergo phase transitions that mimic relativistic vacuum decay, allowing us to empirically test early Universe physics in tabletop experiments. We investigate the physics of these analog systems, going beyond previous analyses of the classical equations of motion to study quantum fluctuations in the cold-atom false vacuum. We show that the fluctuation spectrum of this vacuum state agrees with the usual relativistic result in the regime where the classical analogy holds, providing further evidence for the suitability of these systems for studying vacuum decay. Using a suite of semiclassical lattice simulations, we simulate bubble nucleation from this analog vacuum state in a 1D homonuclear potassium-41 mixture, finding qualitative agreement with instanton predictions. We identify realistic parameters for this system that will allow us to study vacuum decay with current experimental capabilities, including a prescription for efficiently scanning over decay rates, and show that this setup will probe the quantum (rather than thermal) decay regime at temperatures T≲10  nK. Our results help lay the groundwork for using upcoming cold-atom experiments as a new probe of nonperturbative early Universe physics.

The evolution of linear initial conditions present in the early Universe into extended halos of dark matter at late times can be computed using cosmological simulations. However, a theoretical understanding of this complex process remains elusive; in particular, the role of anisotropic information in the initial conditions in establishing the final mass of dark matter halos remains a long-standing puzzle. Here, we build a deep learning framework to investigate this question. We train a three-dimensional convolutional neural network to predict the mass of dark matter halos from the initial conditions, and quantify in full generality the amounts of information in the isotropic and anisotropic aspects of the initial density field about final halo masses. We find that anisotropies add a small, albeit statistically significant amount of information over that contained within spherical averages of the density field about final halo mass. However, the overall scatter in the final mass predictions does not change qualitatively with this additional information, only decreasing from 0.9 dex to 0.7 dex. Given such a small improvement, our results demonstrate that isotropic aspects of the initial density field essentially saturate the relevant information about final halo mass. Therefore, instead of searching for information directly encoded in initial conditions anisotropies, a more promising route to accurate, fast halo mass predictions is to add approximate dynamical information based e.g. on perturbation theory. More broadly, our results indicate that deep learning frameworks can provide a powerful tool for extracting physical insight into cosmological structure formation.

The abundance of dark matter haloes is a key cosmological probe in forthcoming galaxy surveys. The theoretical understanding of the halo mass function (HMF) is limited by our incomplete knowledge of the origin of non-universality and its cosmological parameter dependence. We present a deep-learning model which compresses the linear matter power spectrum into three independent factors which are necessary and sufficient to describe the z = 0 HMF from the state-of-the-art AEMULUS emulator to sub-per cent accuracy in a wCDM+Neff parameter space. Additional information about growth history does not improve the accuracy of HMF predictions if the matter power spectrum is already provided as input, because required aspects of the former can be inferred from the latter. The three factors carry information about the universal and non-universal aspects of the HMF, which we interrogate via the information-theoretic measure of mutual information. We find that non-universality is captured by recent growth history after matter-dark-energy equality and Neff for M ∼ 1013 M h−1 haloes, and by m for M ∼ 1015 M h−1. The compact representation learnt by our model can inform the design of emulator training sets to achieve high emulator accuracy with fewer simulations.

Strongly lensed supernovae are rare and valuable probes of cosmology and astrophysics. Upcoming wide-field time-domain surveys, such as the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST), are expected to discover an order-of-magnitude more lensed supernovae than have previously been observed. In this work, we investigate the cosmological prospects of lensed type Ia supernovae (SNIa) in LSST by quantifying the expected annual number of detections, the impact of stellar microlensing, follow-up feasibility, and how to best separate lensed and unlensed SNIa. We simulate SNIa lensed by galaxies, using the current LSST baseline v3.0 cadence, and find an expected number of 44 lensed SNIa detections per year. Microlensing effects by stars in the lensing galaxy are predicted to lower the lensed SNIa detections by ∼8 per cent. The lensed events can be separated from the unlensed ones by jointly considering their colours and peak magnitudes. We define a 'gold sample' of ∼10 lensed SNIa per year with time delay >10 d, >5 detections before light curve peak, and sufficiently bright (mi < 22.5 mag) for follow-up observations. In 3 yr of LSST operations, such a sample is expected to yield a 1.5 per cent measurement of the Hubble constant.

We use explainable neural networks to connect the evolutionary history of dark matter halos with their density profiles. The network captures independent factors of variation in the density profiles within a low-dimensional representation, which we physically interpret using mutual information. Without any prior knowledge of the halos’ evolution, the network recovers the known relation between the early time assembly and the inner profile and discovers that the profile beyond the virial radius is described by a single parameter capturing the most recent mass accretion rate. The results illustrate the potential for machine-assisted scientific discovery in complicated astrophysical datasets.

Cold-atom analog experiments are a promising new tool for studying relativistic vacuum decay, enabling one to empirically probe early-Universe theories in the laboratory. However, existing proposals place stringent requirements on the atomic scattering lengths that are challenging to realize experimentally. Here we eliminate these restrictions and show that any stable mixture between two states of a bosonic isotope can be used as a faithful relativistic analog. This greatly expands the landscape of suitable experiments, and will expedite efforts to study vacuum decay with cold atoms.

Gravitational-wave (GW) observations of neutron star-black hole (NSBH) mergers are sensitive to the nuclear equation of state (EOS). We present a new methodology for EOS inference with nonparametric Gaussian process priors, enabling direct constraints on the pressure at specific densities and the length-scale of correlations on the EOS. Using realistic simulations of NSBH mergers, incorporating both GW and electromagnetic selection to ensure sample purity, we find that a GW detector network operating at O5 sensitivities will constrain the radius of a 1.4M⊙ NS and the maximum NS mass with 1.6% and 13% precision, respectively. With the same sample, the projected constraint on the length-scale of correlations in the EOS is ≥3.2 MeV fm-3. These results demonstrate strong potential for insights into the nuclear EOS from NSBH systems, provided they are robustly identified.

We present pop-cosmos: a comprehensive model characterizing the galaxy population, calibrated to 140,938 (r < 25 selected) galaxies from the Cosmic Evolution Survey (COSMOS) with photometry in 26 bands from the ultraviolet to the infrared. We construct a detailed forward model for the COSMOS data, comprising: a population model describing the joint distribution of galaxy characteristics and its evolution (parameterized by a flexible score-based diffusion model); a state-of-the-art stellar population synthesis model connecting galaxies’ intrinsic properties to their photometry; and a data model for the observation, calibration, and selection processes. By minimizing the optimal transport distance between synthetic and real data, we are able to jointly fit the population and data models, leading to robustly calibrated population-level inferences that account for parameter degeneracies, photometric noise and calibration, and selection. We present a number of key predictions from our model of interest for cosmology and galaxy evolution, including the mass function and redshift distribution; the mass-metallicity-redshift and fundamental metallicity relations; the star-forming sequence; the relation between dust attenuation and stellar mass, star formation rate, and attenuation-law index; and the relation between gas-ionization and star formation. Our model encodes a comprehensive picture of galaxy evolution that faithfully predicts galaxy colors across a broad redshift (z < 4) and wavelength range.