DESI Strong Lens Foundry V: A Sample of HST-Observed Strong Lenses Modeled with GIGA-Lens

TL;DR

This paper introduces a novel full forward modeling approach with GIGA-Lens for analyzing HST strong lens images, demonstrating robust parameter inference on complex galaxy-scale lenses, and paving the way for larger future surveys.

Contribution

It presents the first HST strong lens sample modeled with simultaneous inference and explicit convergence validation, ensuring statistically robust results for complex lenses.

Findings

Achieved convergence-validated modeling with high ESS and low $ ext{R}$ for all systems.

Demonstrated GIGA-Lens's capability to handle complex galaxy-scale lenses.

Set the stage for scaling to larger high-resolution strong lens samples.

Abstract

We present six galaxy-scale strong lenses with HST imaging modeled using GIGA-Lens. This is Paper V of the DESI Strong Lens Foundry series. These systems were discovered in the DESI Legacy Imaging Surveys using ML/AI methods and confirmed with DESI, Keck/NIRES, and VLT/MUSE spectroscopy. They span $z_d = 0.39 - 1.1$ and $z_s = 1.4 - 3.3$. This is the first HST strong lens sample modeled with full forward modeling -- all lens and source parameters sampled simultaneously in a single inference -- with explicit convergence validation using both $\widehat{R}$ and effective sample size (ESS) for each system. All inferred parameters satisfy $\widehat{R} < 1.1$ and ${\rm ESS} \gtrsim 10,000$, demonstrating that GIGA-Lens achieves statistically robust inference even for some of the most complex galaxy-scale lenses known. These results pave the way for scaling to much larger, high-resolution strong lens samples from HST, Euclid, JWST, and Roman. Convergence-validated modeling will be critical for key science goals, including constraining the mass-density profile of galaxies, detecting low-mass dark matter (sub)halos, and delivering precise and accurate cosmological constraints.