Using the Hoffman2 Cluster to Model Dark Matter Subhalos

Standard cosmological models predict that thousands of low-mass, invisible dark matter clumps—or subhalos—cluster around massive galaxies. Because these minuscule structures are entirely devoid of stars, the Treu Research Group at UCLA bypasses traditional optical limitations by leveraging strong gravitational lensing, where a massive foreground galaxy acts as a cosmic magnifying glass that distorts light from a distant quasar into multiple images. To map these elusive signatures with minimal assumptions, the group pioneered macromodel-free frameworks using local lensing formalisms alongside an empirical subhalo tidal evolution pipeline. This workflow accelerates complex, semianalytic population predictions from hours to seconds, enabling the efficient simulation of complete dark matter distributions for high-resolution surveys utilizing the James Webb Space Telescope (JWST).

Extracting these subgalactic perturbations from lensed arcs requires an immense computational infrastructure capable of navigating high-dimensional parameter spaces. Conducting the group’s  Bayesian inferences demands millions of expensive likelihood evaluations, rendering the workflows computationally untenable on standard systems. UCLA’s Hoffman2 Cluster provides the critical high-throughput parallel processing and optimized job scheduling required to aggregate hundreds of thousands of CPU hours. By eliminating computational bottlenecks, the Hoffman2 Cluster empowered the Treu Research Group to constrain the minimum halo mass, directly test particle theories of dark matter, and accelerate discoveries in observational cosmology.