John F. Wu

The environments around star clusters evolve as stellar feedback reshapes the interstellar medium and dynamical processes reorganize the structure of the surrounding stellar field. As approximately single-age populations, star clusters can serve as clocks to trace these environmental changes. In this exploratory study, we test whether convolutional neural networks (CNNs) can identify age-dependent changes in cluster environments. We take cluster ages as given from basic spectral energy distribution (SED) fitting of five-band UV-optical aperture photometry from the Physics at High Angular resolution in Nearby Galaxies Hubble Space Telescope survey. We first show that CNNs can be trained on image cutouts centered on clusters to recover ages directly from imaging. This demonstration provides the foundation for this study, which examines whether the information used by CNNs to predict age is coherent and physically meaningful. We perform controlled image occlusion experiments as an explainable AI method. These show that the CNNs extract age-predictive environmental cues in the absence of cluster light and when information on SED shape is removed by combining the five filters into one image. We find that reliance on environmental information increases at the youngest (1 Gyr) ages, where clusters can exhibit similarly red colors. Our results are consistent with the long-recognized picture that cluster environments evolve systematically with age. We demonstrate that this information is encoded at a level detectable by machine learning and recoverable from broadband imaging. This establishes a path for using new techniques to connect image-based age inference to the physical evolution of cluster environments.

Properties of massive galaxy clusters, such as mass abundance and concentration, are sensitive to cosmology, making cluster statistics a powerful tool for cosmological studies. However, favoring a more simplified, spherically symmetric model for galaxy clusters can lead to biases in the estimates of cluster properties. In this work, we present a deep learning approach for estimating the triaxiality and orientations of massive galaxy clusters (those with masses ≳10 ¹⁴ M _⊙ h ⁻¹ ) from 2D observables. We utilize the flagship hydrodynamical volume of the suite of cosmological-hydrodynamical MillenniumTNG (MTNG) simulations as our ground truth. Our model combines the feature extracting power of a convolutional neural network and the message passing power of a graph neural network in a multimodal, fusion network. Our model is able to extract 3D geometry information from 2D idealized cluster multiwavelength images (soft X-ray, medium X-ray, hard X-ray, and tSZ effect) and mathematical graph representations of 2D cluster member observables (line-of-sight radial velocities, 2D projected positions and V -band luminosities). Our network improves cluster geometry estimation in MTNG by 30% compared to assuming spherical symmetry. We report an R ² = 0.85 regression score for estimating the major axis length of triaxial clusters and correctly classifying 71% of prolate clusters with elongated orientations along our line of sight.

We investigate how Milky Way (MW)–like environments influence the sizes and structural properties of low-mass galaxies by comparing satellites of MW analogs from the Satellites Around Galactic Analogs (SAGA) Survey with two control samples: an environmentally agnostic population from the SAGA background sample and isolated galaxies from the Sloan Digital Sky Survey NASA-Sloan Atlas. All sizes and structural parameters are measured uniformly using pysersic to ensure consistency across samples. We find the half-light sizes of SAGA satellites are systematically larger than those of isolated galaxies, with the magnitude of the offset ranging from 0.05 to 0.12 dex (10%–24%) depending on the comparison sample and completeness cuts. This corresponds to physical size differences between 85 and 200 pc at log 10 ( M ⋆ / M ⊙ ) = 7.5 and 220–960 pc at log 10 ( M ⋆ / M ⊙ ) = 10 . This offset persists among star-forming galaxies, suggesting that environment can influence the structure of low-mass galaxies even before it impacts quenching. The intrinsic scatter in the size–mass relation is lower for SAGA satellites than isolated galaxies, and the Sérsic index distributions of satellites and isolated galaxies are similar. In comparison to star-forming satellites, quenched SAGA satellites have a slightly shallower size–mass relation and rounder morphologies at low mass, suggesting that quenching is accompanied by structural transformation and that the processes responsible differ between low- and high-mass satellites. Our results show that environmental processes can imprint measurable structural differences on satellites in MW-mass halos.

Star formation in galaxies is regulated by the interplay of a range of processes that shape the multiphase gas in the interstellar and circumgalactic media. Using the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) suite of cosmological simulations, we study the effects of varying feedback and cosmology on the average star formation histories (SFHs) of galaxies at z ∼ 0 across the IllustrisTNG, SIMBA, and ASTRID galaxy formation models. We find that galaxy SFHs in all three models are sensitive to changes in stellar feedback, which affect the efficiency of baryon cycling and the rates at which central black holes grow, whereas the effects of varying active galactic nucleus (AGN) feedback depend on model-specific implementations of black hole seeding, accretion, and feedback. We also find strong interaction terms that couple stellar and AGN feedback, usually by regulating the amount of gas available for the central black hole to accrete. Using a double power law to describe the average SFHs, we derive a general set of equations relating the shape of the SFHs to physical quantities like baryon fraction and black hole mass across all three models. We find that a single set of equations (albeit with different coefficients) can describe the SFHs across all three CAMELS models, with cosmology dominating the SFH at early times, followed by halo accretion, and feedback and baryon cycling at late times. Galaxy SFHs provide a novel, complementary probe to constrain cosmology and feedback, and can connect the observational constraints from current and upcoming galaxy surveys with the physical mechanisms responsible for regulating galaxy growth and quenching.

The application of deep machine learning methods in astronomy has exploded in the last decade, with new models showing remarkably improved performance on benchmark tasks. Not nearly enough attention is given to understanding the models' robustness, especially when the test data are systematically different from the training data, or "out of domain." Domain shift poses a significant challenge for simulation-based inference, where models are trained on simulated data but applied to real observational data. In this paper, we explore domain shift and test domain adaptation methods for a specific scientific case: simulation-based inference for estimating galaxy cluster masses from X-ray profiles. We build datasets to mimic simulation-based inference: a training set from the Magneticum simulation, a scatter-augmented training set to capture uncertainties in scaling relations, and a test set derived from the IllustrisTNG simulation. We demonstrate that the Test Set is out of domain in subtle ways that would be difficult to detect without careful analysis. We apply three deep learning methods: a standard neural network (NN), a neural network trained on the scatter-augmented input catalogs, and a Deep Reconstruction-Regression Network (DRRN), a semi-supervised deep model engineered to address domain shift. Although the NN improves results by 17% in the Training Data, it performs 40% worse on the out-of-domain Test Set. Surprisingly, the Scatter-Augmented Neural Network (SANN) performs similarly. While the DRRN is successful in mapping the training and Test Data onto the same latent space, it consistently underperforms compared to a straightforward Yx scaling relation. These results serve as a warning that simulation-based inference must be handled with extreme care, as subtle differences between training simulations and observational data can lead to unforeseen biases creeping into the results.

We test the Platonic Representation Hypothesis (PRH) in astronomy by measuring representational convergence across a range of foundation models trained on different data types. Using spectroscopic and imaging observations from JWST, HSC, Legacy Survey, and DESI, we compare representations from vision transformers, self-supervised models, and astronomy-specific architectures via mutual $k$-nearest neighbour analysis. We observe consistent scaling: representational alignment generally increases with model capacity across our tested architectures, supporting convergence toward a shared representation of galaxy astrophysics. Our results suggest that astronomical foundation models can use pre-trained general-purpose architectures, allowing us to capitalise on the broader machine learning community's already-spent computational investment.

We present results from Atacama Large Millimeter/submillimeter Array (ALMA) spectral line-scan observations at 3 mm and 2 mm bands of three near-infrared-dark (NIR-dark) galaxies behind two massive lensing clusters MACS J0417.5-1154 and RXC J0032.1+1808. Each of these three sources is a (sub)millimeter faint (delensed S _{1.2 mm} < 1 mJy) triply lensed system originally discovered in the ALMA Lensing Cluster Survey. We have successfully detected CO and [C i] emission lines and confirmed that their spectroscopic redshifts are z = 3.652, 2.391, and 2.985. By utilizing a rich multiwavelength data set, we find that the NIR-dark galaxies are located on the star formation main sequence in the intrinsic stellar mass range of log (M _*/M _⊙) = 9.8–10.4, which is about 1 order of magnitude lower than that of typical submillimeter galaxies (SMGs). These NIR-dark galaxies show a variety in gas depletion times and spatial extent of dust emission. One of the three is a normal star-forming galaxy with gas depletion time consistent with a scaling relation, and its infrared surface brightness is an order of magnitude smaller than that of typical SMGs. Since this galaxy has an elongated axis ratio of ∼0.17, we argue that normal star-forming galaxies in an edge-on configuration can be heavily dust-obscured. This implies that existing deep WFC3/F160W surveys may miss a fraction of typical star-forming main-sequence galaxies due to their edge-on orientation.

We present measurements of the neutral atomic hydrogen (H i) mass function (HiMF) and cosmic H i density (Ω_{H I}) at 0 ≤ z ≤ 0.088 from the Looking at the Distant Universe with MeerKAT Array (LADUMA) survey. Using LADUMA Data Release 1 (DR1), we analyze the HiMF via a new “recovery matrix” method that we benchmark against a more traditional modified maximum likelihood (MML) method. Our analysis, which implements a forward modeling approach, corrects for survey incompleteness and uses extensive synthetic source injections to ensure robust estimates of the HiMF parameters and their associated uncertainties. This new method tracks the recovery of sources in mass bins different from those in which they were injected and incorporates a Poisson likelihood in the forward modeling process, allowing it to correctly handle uncertainties in bins with few or no detections. The application of our analysis to a high-purity subsample of the LADUMA DR1 spectral line catalog in turn mitigates any possible biases that could result from the inconsistent treatment of synthetic and real sources. For the surveyed redshift range, the recovered Schechter function normalization, low-mass slope, and “knee” mass are ϕ * = 3.5 6 − 1.92 + 0.97 × 1 0 − 3 Mpc⁻³ dex⁻¹, α = − 1.1 8 − 0.19 + 0.08 , and log ( M * / M ⊙ ) = 10.0 1 − 0.12 + 0.31 , respectively, which together imply a comoving cosmic H i density of Ω H I = 3.0 9 − 0.47 + 0.65 × 1 0 − 4 . Our results show consistency between recovery matrix and MML methods and with previous low-redshift studies, giving confidence that the cosmic volume probed by LADUMA, even at low redshifts, is not an outlier in terms of its H i content.

Galaxy appearances reveal the physics of how they formed and evolved. Machine learning (ML) models can now exploit galaxies’ information-rich morphologies to predict physical properties directly from image cutouts. Learning the relationship between pixel-level features and galaxy properties is essential for building a physical understanding of galaxy evolution, but we are still unable to explicate the details of how deep neural networks represent image features. To address this lack of interpretability, we present a novel neural network architecture called a Sparse Feature Network (SFNet). SFNets produce interpretable features that can be linearly combined in order to estimate galaxy properties like optical emission line ratios or gas-phase metallicity. We find that SFNets do not sacrifice accuracy in order to gain interpretability, and that they perform comparably well to cutting-edge models on astronomical ML tasks. Our novel approach is valuable for finding physical patterns in large data sets and helping astronomers interpret ML results.