Northwestern's FLIP Method Reveals Planet Hidden in Webb Data

A Northwestern University-led team of astrophysicists has developed a computational technique that cuts through a star's glare and reveals planets hiding inside it. The method, called Physics-based Learning for Imaging Planets, or FLIP, improved the James Webb Space Telescope's sensitivity by up to 15 times when the team applied it to existing Webb observations, and it pulled a giant exoplanet out of an image where conventional processing couldn't see one at all. The study was published in the journal Nature.

The problem is brightness. Planets are a million times fainter than their stars, and even relatively bright young giant planets can run 100,000 times fainter than the star they orbit. Webb's coronagraphs block the star itself, but some starlight still seeps through and leaves a haze. For the past 20 years, astronomers have handled that haze by photographing a different star and subtracting its image as a reference. The reference never matches perfectly, because no telescope holds perfectly still. Thermal changes can expand or contract a mirror by a picometer, and the cryocoolers that keep Webb's instruments cold carry small motors that vibrate the mirrors.

FLIP takes a different route. Instead of leaning on a reference photograph, it builds a digital twin of the telescope, a virtual replica that simulates how starlight travels through Webb's mirrors and reaches its detectors. The team fine-tuned millions of parameters in that twin, using optimization tools similar to the ones used to train AI models, until the simulation matched the real observation. Subtracting the simulated glare then left the planets behind. "We used the same computational tools that researchers use to train AI models, but we did not actually use AI," said Jason Wang, an assistant professor of physics and astronomy at Northwestern and a co-author of the study. "Because we already know the physics, we don't need AI to understand what's happening."

The test ran on past Webb observations of three stars: HIP 65426 and 51 Eridani, both with known planets, and TYC 5899, which has no known companions. Across those observations, FLIP beat conventional processing every time, improving sensitivity for faint planets by 2 to 15 times. With that gain, the team detected 51 Eridani b in two-year-old Webb data. Ground-based telescopes had discovered the planet earlier, but carbon dioxide in its atmosphere absorbs light at the infrared wavelength Webb observes, which made its signal even fainter and left it buried in residual glare until FLIP removed the glare.

"Planets are a million times fainter than their stars," said Rodrigo Ferrer-Chavez, the Northwestern doctoral student in astronomy who led the study. "So, to image exoplanets, we need to remove that glare. That is a very tricky problem because even the tiniest changes in the telescope can affect how that glare appears." Ferrer-Chavez and Wang are both members of Northwestern's Center for Interdisciplinary Exploration and Research in Astrophysics. The study was supported by the Space Telescope Science Institute, the Sloan Foundation and the National Science Foundation.

The wavelength matters beyond one planet. Webb can observe at 4.3 microns, where carbon dioxide strongly absorbs infrared light, and reading features at that wavelength can tell astronomers about a planet's atmosphere and composition. Looking ahead, NASA's proposed Habitable Worlds Observatory aims to photograph Earth-like planets directly, a job that could require optical stability 100 to 1,000 times tighter than Webb's. FLIP points at a different answer: build the correction into the analysis instead of demanding a perfectly still telescope. "We showed that we can actually recover a planet that people previously couldn't see," Wang said. "This is real data in a real scenario. It was science left on the table because we didn't have the right tools to analyze the data in the most optimal way."