Speaker
Description
The growing volume of substellar spectra from JWST, including NIRSpec observations of brown dwarfs and directly imaged exoplanets, demands increasingly efficient tools for atmospheric characterization. Traditional spectral fitting approaches such as grid interpolation and Markov Chain Monte Carlo (MCMC) retrieval become significant computational bottlenecks when applied to large samples or dense model grids. For the Sonora Diamondback model (dback24; Morley et al. 2024), generating ~10,000 synthetic spectra via grid interpolation requires ~27 minutes, which can take weeks when fitting entire surveys with multiple model grids. Here we present a variational autoencoder (VAE) framework for generating synthetic atmospheric model spectra orders of magnitude faster than conventional methods. We trained a VAE on the Sonora Diamondback grid using the UCDMCMC fitting package (Burgasser et al. 2025), with training completing in approximately five minutes. The trained model generates ~10,000 synthetic spectra in under one second, over 1000x faster than traditional grid interpolation. VAE-generated models also enable denser parameter sampling than the original grid, improving fit accuracy. We validate the framework by fitting a NIRSpec source from the DAWN JWST Archive, recovering best-fit parameters (Teff = 1600 K, log g = 5.5, fsed = f8, [M/H] = 0.5) consistent with a parallel MCMC retrieval (Teff = 1577 K, log g = 5.38, fsed = f8, [M/H] = 0.49). This approach offers a practical path toward rapid atmospheric characterization across the large, heterogeneous samples that JWST is delivering. Future work will extend training to additional model grids including Sonora Elf Owl (Mukherjee et al. 2024), and determine if this framework can be used to accelerate the retrieval process.
| Do you plan to submit a 4-page extended abstract on OpenReview (only for Presentations/Posters)? | Maybe |
|---|