🤖 AI Summary
This study addresses the failure of simulation-based inference when applied to real James Webb Space Telescope (JWST) spectra, identifying that this breakdown stems from omitting planetary radius as a latent variable rather than from simulator inaccuracies. To resolve this, we propose MIRAGE, a framework that incorporates radius as a critical latent variable and integrates flow matching, optimal transport mappings, and importance sampling to rectify the inference pipeline, achieving physically consistent retrievals through nested sampling calibration. By reducing the goodness-of-fit metric from 301 to 0.06, our approach successfully demonstrates generalization across instruments and multiple targets, enabling end-to-end retrieval of real exoplanetary atmospheric data. This work reveals that latent variable misspecification, not simulator error, underlies prior inference failures, offering a robust methodology for future spectroscopic analyses.
📝 Abstract
Amortized simulation-based inference (SBI), which is trained on radiative-transfer simulators, recovers exoplanet atmospheres accurately on synthetic James Webb Space Telescope (JWST) spectra but collapses when it comes to real reduced spectra. The flow posterior collapsed on real WASP-39b (importance-sampling effective sample size (ESS) = 1, best-fit \c{hi}2/N = 301), and such a failure is usually blamed on missing the forward-model physics, but ruling these levers out with nested sampling first (a temperature gradient, SO2 opacity, and a high-fidelity opacity set) leaves the fit unchanged, meaning the collapse is not from the simulator misspecification but instead from a missing latent, the planet radius. To fix this, we build MIRAGE, a radius-augmented flow-matching posterior calibrated against an independent nested-sampling reference with importance sampling and an optimal-transport map, which yields a physical and literature-consistent retrieval of real WASP-39b (with \c{hi}2/N from 301 to 0.06). This same method transfers unchanged across two instruments and three real JWST targets, including one spectrum self-reduced end-to-end from raw Mikulski Archive for Space Telescopes(MAST) data. The lesson is cross-domain, as a latent the simulator encodes but the inference omits can masquerade as misspecification.