A Digital Twin of the Webb Telescope Pulls a Hidden Planet Out of Old Data and Makes JWST Up to 15 Times Sharper

FLIP, for Physics-based Learning for Imaging Planets, takes the opposite approach. Instead of borrowing someone else's starlight, it simulates the telescope's own: a model of how light travels through Webb's mirrors to its detectors, with millions of adjustable parameters covering tiny position shifts and optical distortions. The team tuned those parameters with the same optimisation tools used to train machine learning models until the simulated star matched the observed one, then subtracted the simulation. Wang, an assistant professor of physics and astronomy, stressed that no AI was used to find planets; the physics was already known, so the software only had to fit it.

What did FLIP find in the archive?

The team reprocessed Webb observations of three stars: HIP 65426 and 51 Eridani, each with a known planet, and TYC 5899, with none. The headline result is 51 Eridani b, a giant planet first imaged from the ground, recovered in the NIRCam F430M filter at above five sigma where the original 2025 analysis had seen nothing. That wavelength matters because carbon dioxide absorbs there, which is exactly why the planet is so faint and exactly why astronomers want to measure it. Universe Today's account quotes Ferrer-Chavez's summary of the problem: it is not that such planets do not exist, it is that we cannot see them. Wang's version is shorter: "We showed that we can actually recover a planet that people previously couldn't see."

What it means for finding another Earth

NASA's proposed Habitable Worlds Observatory, the next flagship after Webb, is meant to image Earth-like planets around Sun-like stars, a contrast far harsher than any young giant. The release puts the stability it would need at picometre scale, roughly 100 to 1,000 times steadier than Webb. Engineering that is enormously expensive. FLIP suggests a complementary route: accept some instability and correct for it in software after the fact.

The immediate payoff is cheaper. Webb's archive holds observations that were processed the old way, and some of them, like 51 Eridani, may be hiding detections. The obvious next step is to run the digital twin back through that archive and see what else was there all along.