pop-cosmos: Insights from generative modeling of a deep, infrared-selected galaxy population

TL;DR

The paper extends the forward-modeling framework pop-cosmos to infrared-selected galaxies up to $z\sim6$ by training a diffusion-based population prior over 16 SPS parameters on 26-band COSMOS2020 data. It couples a fast SPS emulator with a diffusion-based uncertainty model and a robust inference pipeline to produce realistic mock catalogs and precise object-level posteriors, including redshifts and stellar masses, for hundreds of thousands of galaxies. The authors validate the model against COSMOS photometry and spectroscopy, demonstrating accurate redshift recovery ($\text{median}\Delta z\approx-8\times10^{-4}$, $\sigma_{\text{MAD}}=0.0132$, outlier $\sim6.19\%$) and sensible scaling relations (mass function, SFS, mass–metallicity, dust, and AGN trends). They release software, two million-galaxy mock catalogs, and COSMOS-like posteriors to support forward-modeling for Stage IV surveys, enabling realistic, large-scale cosmological analyses with IR-selected populations.

Abstract

We present an extension of the pop-cosmos model for the evolving galaxy population up to redshift $z\sim6$. The model is trained on distributions of observed colors and magnitudes, from 26-band photometry of $\sim420,000$ galaxies in the COSMOS2020 catalog with Spitzer IRAC $\textit{Ch. 1}<26$. The generative model includes a flexible distribution over 16 stellar population synthesis (SPS) parameters, and a depth-dependent photometric uncertainty model, both represented using score-based diffusion models. We use the trained model to predict scaling relationships for the galaxy population, such as the stellar mass function, star-forming main sequence, and gas-phase and stellar metallicity vs. mass relations, demonstrating reasonable-to-excellent agreement with previously published results. We explore the connection between mid-infrared emission from active galactic nuclei (AGN) and star-formation rate, finding high AGN activity for galaxies above the star-forming main sequence at $1\lesssim z\lesssim 2$. Using the trained population model as a prior distribution, we perform inference of the redshifts and SPS parameters for 429,669 COSMOS2020 galaxies, including 39,588 with publicly available spectroscopic redshifts. The resulting redshift estimates exhibit minimal bias ($\text{median}[Δ_z]=-8\times10^{-4}$), scatter ($σ_\text{MAD}=0.0132$), and outlier fraction ($6.19\%$) for the full $0<z<6$ spectroscopic compilation. These results establish that pop-cosmos can achieve the accuracy and realism needed to forward-model modern wide--deep surveys for Stage IV cosmology. We publicly release pop-cosmos software, mock galaxy catalogs, and COSMOS2020 redshift and SPS parameter posteriors.

Figures (31)

• Figure 1: Broadband magnitudes (logarithmic) predicted by the new pop-cosmos generative model, compared to COSMOS2020 weaver22 for $\textit{Ch.\,1}<26$.
• Figure 3: Broadband colors (logarithmic, alsing24 band combinations) predicted by the new pop-cosmos generative model, compared to COSMOS2020 weaver22 for $\textit{Ch.\,1}<26$. Contours enclose the 68 and 95% highest density regions.
• Figure 4: Flux uncertainty vs. (logarithmic) magnitude for the COSMOS broad bands, compared to the flux errors reported in the COSMOS2020 catalog weaver22 for $\textit{Ch.\,1}<26$. Note that the pop-cosmos uncertainty predictions are made conditional on the pop-cosmos model magnitudes. Contours enclose the 68 and 95% highest probability density regions.
• Figure 5: Comparison of $z^\text{phot}$ vs. $z^\text{spec}$ based on $z^\text{phot}$ (posterior median $z$) inferred under the new pop-cosmos prior. Each galaxy is weighted by its spectroscopic confidence level from khostovan25. Solid and dashed lines show $z^\text{phot}=z^\text{spec}$ and $|\Delta_z|=0.15$, respectively.
• Figure 6: Comparison of stellar mass estimates for the 423,262 COSMOS2020 galaxies with $\textit{Ch.\,1}<26$. The vertical axis shows the posterior median $\log_{10}(M/M_\odot)$ under the new pop-cosmos prior. The horizontal axis shows (left) the LePhare stellar mass estimates from weaver22, and (right) the posterior median under the Prospector prior.