Glare-Removal Method Reveals Giant Planet in Two-Year-Old JWST Data
The research is led by Northwestern University astrophysicists: doctoral student Rodrigo Ferrer-Chavez and assistant professor Jason Wang, both members of the Center for Interdisciplinary Exploration and Research in Astrophysics. Their work develops FLIP, a physics-based learning method for imaging planets, and is set to appear in Nature on Nov. 4. The team created a computational “digital twin” of NASA’s James Webb Space Telescope to simulate how starlight passes through the observatory’s optics and reaches its detectors. Unlike earlier methods that remove residual glare by subtracting reference images of other stars, FLIP uses machine-learning-style optimization to fine-tune millions of physical parameters, accounting for tiny telescope imperfections caused by thermal shifts, cryocooler vibrations and other minute mechanical effects. By matching the simulated starlight to actual observations, the researchers subtract the modeled glare with greater precision, revealing faint planets that conventional processing can miss. When applied to existing JWST observations of stars with and without known planets, FLIP improved detection sensitivity by 2 to 15 times. In a two-year-old image of the star 51 Eridani A, the method recovered the giant exoplanet 51 Eridani b, which standard glare-removal techniques had failed to identify in infrared observations where carbon dioxide strongly absorbs light. The detection shows that valuable planet science may already be hiding in previously collected data. The advance could help astronomers find smaller, colder and fainter exoplanets without changing telescope hardware. It may also improve atmospheric studies at key infrared wavelengths and ease future engineering demands for proposed missions such as the Habitable Worlds Observatory, which would require extreme optical stability to directly image Earth-like worlds. More broadly, the team argues that many planetary systems remain unseen not because planets are absent, but because current analysis tools have not fully extracted their signals.
