arXiv subjects

Azton I. Wells

Publications and source records attributed to Azton I. Wells.

5 recordsLinked to original sources

arXiv2025

Multi-modal Foundation Model for Cosmological Simulation Data

We present a multi-modal foundation model for astrophysical galaxy data, designed to map between simulation- and observation-based galactic features. Our encoder-only transformer flexibly ingests scalar quantities (e.g., redshifts, galaxy masses) and vectors (e.g., star formation histories, spectra), supporting multi-task training that includes within-modality reconstruction and cross-modality prediction. With a dynamic masking strategy, the model can query arbitrary galaxy properties from partial inputs -- including predicting spectra from redshift and mass, or estimating photometric redshifts from broadband magnitudes -- while also recovering missing segments within a modality. Trained on 185,000 simulated galaxies from a gigaparsec-scale Cosmology simulation, the model yields a 50% improvement in redshift estimation when combining LSST and SPHEREx photometry over LSST photometry alone, and a 63% improvement in stellar mass inference when combining late-time SFH with LSST photometry over early-time SFH with LSST photometry. The model demonstrates strong generalization across multi-modal tasks and lays the groundwork for future integration of higher-dimensional and structured data such as images, merger trees, and 3D fields. This approach provides a unified framework for connecting simulations and observations, advancing the development of generalizable astrophysical foundation models.

astro-ph.GA↗

arXiv2024

Galaxies and Their Environment at $z \gtrsim 10$ -- I: Primordial Chemical Enrichment, Accretion, Cooling, and Virialization of Gas in Dark Matter Halos

Recent observations made using the James Webb Space Telescope have identified a number of high-redshift galaxies that are unexpectedly luminous. In light of this, it is clear that a more detailed understanding of the high redshift, pre-reionization universe is required for us to obtain the complete story of galaxy formation. This study is the first in a series that seeks to tell the story of galaxy formation at $z \gtrsim 10$ using a suite of large-scale adaptive mesh refinement cosmological simulations. Our machine-learning-accelerated surrogate model for Population III star formation and feedback, StarNet, gives us an unprecedented ability to obtain physically accurate, inhomogeneous chemical initial conditions for a statistically significant number of galaxies. We find that of the 12,423 halos in the mass range of $10^6\,\,M_\odot < M_\mathrm{vir} < 10^9\,\, M_\odot$ that form in our fiducial simulation, $16\%$ are chemically enriched by Population III supernovae by $z\sim12$. We then profile and compare various cooling processes at the centers of halos, and find a complete absence of atomic cooling halos. All of our halos with central cooling gas are dominated by H$_2$ cooling, metal cooling, or a mixture of the two, even in the presence of a strong H$_2$-photodissociating Lyman-Werner background. We also find that gas accretion through the virial radius is not driven by cooling. We find that gas virialization in halos with $M_\mathrm{vir}\gtrsim10^7\,\,M_\odot$ is supported by bulk turbulent flows, and that thermal energy accounts for only a small fraction of the total kinetic energy. Because of this, the mean gas temperature is well below the virial temperature for these halos. We then compute the mass of gas that is available for Population II star formation, and infer star formation rates for each potential star-forming halo.

astro-ph.GA↗

arXiv2022

The First Galaxies and the Effect of Heterogeneous Enrichment from Primordial Stars

We incorporate new scale-intelligent models of metal-enriched star formation (\starss) with surrogate models of primordial stellar feedback (\starnet) into the astrophysics simulation code \enzo to analyze the impact of heterogeneous metal enrichment on the first galaxies. Our study includes the earliest generations of stars and the protogalaxies ($10^6 \lesssim M_v/M_\odot \lesssim 10^8$) containing them. We compare results obtained with the new methods to two common paradigms of metallicity initial conditions in simulations: ignoring the metallicity initial condition and assuming a uniform metallicity floor. We find that ignoring metallicity requirements for enriched star formation results in a redshift-dependent excess in stellar mass created and compounding errors consisting of stars forming in pristine gas. We find that using a metallicity floor causes an early underproduction of stars before $z=21$ that reverses to overproduction by $z=18$. At the final redshift, $z=14.95$, there is $\sim 20\%$ excess stellar mass with 8.6\% increased protogalaxy count. Heterogeneous metallicity initial conditions greatly increase the range of halo observables, e.g., stellar metallicity, stellar mass, and luminosity. The increased range leads to better agreement with observations of ultra-faint dwarf galaxies when compared to metallicity-floor simulations. \starnet generates protogalaxies with low stellar mass, $M_* \lesssim 10^3 M_\odot$, so may also serve to model low-luminosity protogalaxies more effectively than a metallicity floor criterion at similar spatial and mass resolution.

astro-ph.GA↗

arXiv2021

Connecting Primordial Star Forming Regions and Second Generation Star Formation in the Phoenix Simulations

We introduce the {\em Phoenix Simulations}, a suite of highly resolved cosmological simulations featuring hydrodynamics, primordial gas chemistry, Population III and II star formation and feedback, UV radiative transfer, and saved outputs with $Δt$=200 kyr. The suite samples 73,523 distinct primordial star formation events within \npiii distinct regions, forming \ngii second-generation enriched star clusters by $z \geq 12$ within a cumulative 156.25 Mpc$^3$ volume. The regions that lead to enriched star formation contain up to $167$ primordial stars, with 78.7 \% of regions having experienced multiple types of primordial supernovae. The extent of a primordial region, measured by its metal-rich surrounding cloud, is highly variable: the average region has radius $\sim 3$ kpc, with 95 \% confidence limit on the distribution of measured radii is $\sim 5-7$ kpc. For continuing star formation, we find that the metallicity distribution of second generation stars is similar to that of subsequent Population II star formation, with both distributions spanning hyper metal-deficient ([Z/H]$\sim-7$) to super-solar ([Z/H]$\sim0.8$). We find that the metallicity of second generation stars has no strong dependence on the configuration of progenitor supernovae, with the mean metallicity of second-generation stars having $-1.73 < $[Z/H]$<-2.15$. Finally, we create an interpretable regression model to predict the radius of metal-rich influence of \piii star systems within the first 7-18 Myr after the first light. The model predicts the radius with $R_2 \gtrsim 0.4$ and mean squared error $\leq 0.06$. The probability distribution function of predicted radii compares well to that of observed radii with Jensen-Shannon distance $\lesssim 0.2$ for all modelled times.

astro-ph.GA↗

arXiv2021

Predicting Localized Primordial Star Formation with Deep Convolutional Neural Networks

We investigate applying 3D deep convolutional neural networks as fast surrogate models of the formation and feedback effects of primordial stars in hydrodynamic cosmological simulations of the first galaxies. Here, we present the surrogate model to predict localized primordial star formation; the feedback model will be presented in a subsequent paper. The star formation prediction model consists of two sub-models: the first is a 3D volume classifier that predicts which (10 comoving kpc)$^3$ volumes will host star formation, followed by a 3D Inception-based U-net voxel segmentation model that predicts which voxels will form primordial stars. We find that the combined model predicts primordial star forming volumes with high skill, with $F_1 >0.995$ and true skill score $>0.994$. The star formation is localized within the volume to $\lesssim5^3$~voxels ($\sim1.6$~comoving kpc$^3$) with $F_1>0.399$ and true skill score $>0.857$. Applied to simulations with low spatial resolution, the model predicts star forming regions in the same locations and at similar redshifts as sites in resolved full-physics simulations that explicitly model primordial star formation and feedback. When applied to simulations with lower mass resolution, we find that the model predicts star forming regions at later redshift due to delayed structure formation resulting from lower mass resolution. Our model predicts primordial star formation without halo finding, so will be useful in spatially under-resolved simulations that cannot resolve primordial star forming halos. To our knowledge, this is the first model that can predict primordial star forming regions that match highly-resolved cosmological simulations.

astro-ph.GA↗