Astronomers build digital twin of Webb Camera to find hidden planets

Astronomers led by Northwestern University graduate student Rodrigo Ferrer-Chavez built software that creates a virtual model of the James Webb Space Telescope‘s Near Infrared Camera (NIRCam), so astronomers can subtract starlight from existing observations, revealing a hidden giant planet in archival data, according to a study published online Nov. 4, 2026, in Nature.

FLIP, or Physics-based Learning for Imaging Planets, simulates how the light of a target star travels through a virtual version of the telescope’s optics, then subtracts the modeled glare from the original image. Conventional processing instead borrows glare from a reference star observed at another time, but off-center stars, shifting temperatures, and flexing of Webb’s 6.5-meter mirror leave starlight smudges that hide faint planets.

Ferrer-Chavez and colleagues tested FLIP on Webb observations of three star systems and found it was two to 15 times more sensitive than standard processing techniques when checked against known planets. The researchers also used stars without known companions and inserted artificial planet signals to gauge improvement, though the gains varied by target and conditions, so the largest reported gain should not be assumed for every image. The gains are biggest close to stars, where planets like Earth tend to orbit.

Webb’s archive holds pictures of stars whose planets were too dim for earlier methods, and FLIP lets astronomers look back through those saved images. The method already revealed 51 Eridani b, a hidden giant planet, in Webb observations about two years old. FLIP detected it at about 4.3 micrometers, a wavelength where carbon dioxide absorbs light, making planets harder to spot there but potentially useful for studying atmospheric composition. Conventional processing had failed to find the planet at that wavelength.

Direct imaging is difficult because even the biggest planets can be more than 10,000 times fainter than the stars they circle, and an Earth-like world around a Sun-like star can be roughly 10 billion times fainter. Webb can detect Saturn-sized worlds beyond Pluto, but Ferrer-Chavez says Earth-sized planets would require a telescope 100 to 1,000 times more stable than Webb, a precision not yet achieved even in labs.

Webb is optically stable to the nanometer level—a nanometer is one billionth of a meter—but disturbances at that scale can still interfere with spotting planets beside bright stars. Wang said physical instruments alone cannot match the level of correction FLIP achieves when processing images, and Webb’s nanometer-level optical stability is still limiting.

The findings appeared in Nature in a paper titled “Imaging exoplanets below the telescope stability limit,” published online Nov. 4, 2026, as volume 659, issue 8139, page 58. Ferrer-Chavez is the lead author and a Northwestern graduate student in Jason Wang’s lab; Wang and Caltech professor Katie Bouman are co-authors. Authors include William T. Freeman and Brandon Y. Feng of MIT and Laurent Pueyo and Marshall D. Perrin of the Space Telescope Science Institute in Baltimore. Support came from the Space Telescope Science Institute, the Sloan Foundation, an Amazon AI4Science Discovery Award, and the National Science Foundation.

The work began when Ferrer-Chavez, then an undergraduate at Mexico’s Autonomous University of Yucatán, worked with Jason Wang in Dimitri Mawet’s laboratory. An initial thesis effort failed, and an early attempt to train machine-learning models to predict starlight patterns also proved unsuccessful before the team turned to physics-based optimization. Wang held a 51 Pegasi b Postdoctoral Fellowship in Mawet’s lab at the time; Mawet is the David Morrisroe Professor of Astronomy at Caltech and a senior research scientist at NASA’s Jet Propulsion Laboratory.

Building FLIP meant running computer searches across millions of possible flaws, small shifts, and other unknowns that might alter JWST’s observations. The software tunes a physical model of the observatory so differences between observations can be modeled better, yielding low-noise images of exoplanets.

Ars Technica describes the core challenge as the point spread function, the mathematical description of how a telescope distorts a point of light into a blur because of hardware imperfections and pointing limits. Imaging a known target can derive this function, but the telescope changes over time and the function may vary by wavelength. Modeling the hardware directly involves too many parameters to tune for computational tractability.

FLIP uses machine-learning optimization tools, but its model rests on known physics rather than on an artificial intelligence system that decides what a planet should look like. Researchers tweak millions of model parameters, so the predicted glare matches the real observation, and the software produces a highly precise model of the star’s appearance before subtracting that modeled starlight from the original data.

Dimitri Mawet, a senior research scientist at NASA’s Jet Propulsion Laboratory who was not involved in the study, called the result significant. He said researchers had discussed model-based glare removal for more than a decade without beating methods that physically measure reference stars, and that Ferrer-Chavez and colleagues show it can.

The researchers plan to refine FLIP and study its use with other instruments, including the Nancy Grace Roman Space Telescope and ground-based telescopes, because the method is not restricted to Webb and could work on any telescope. Roman carries a coronagraph instrument built by researchers at Caltech, JPL, and IPAC to suppress starlight, and a coronagraph mask inside Webb blocks most of the star’s light like an artificial eclipse, according to Ferrer-Chavez. FLIP could help shape NASA’s proposed Habitable Worlds Observatory, planned to launch in the 2040s as the first telescope designed to search for evidence of life on exoplanets.

The Habitable Worlds Observatory aims to find Earth-like planets and the habitable zones of Sun-like stars through direct imaging, with the habitable zone defined as the range of distances where temperatures could allow liquid water on a planet’s surface. Mawet said physics-based post-processing could plausibly ease part of HWO’s stability error budget, though HWO optics must still hold steady to within a trillionth of a meter to photograph Earth-like planets about 10 billion times fainter than their stars. The goal is to make it easier to build the observatory by letting hardware wobble more if software removes glare better.

Ferrer-Chavez put the challenge in perspective: “If our approach works, building this new observatory may be really hard but not really, really, really hard.” FLIP has not yet provided routine images of Earth-like worlds around other stars, but it could aid the search for fainter giant planets now and help inform the design of future observatories. Detecting a giant planet at 4.3 micrometers is not evidence of life, though it can improve measurements available for later analysis.

Bouman’s long-term aim is to capture direct images of Earth-like worlds and examine them for possible signs of life. Wang said FLIP has gone beyond the telescope stability limit for the first time and that researchers can apply the new paradigm to every instrument trying to image exoplanets. Ferrer-Chavez said this was the first demonstration that little imperfections or instabilities are not a limiting factor in the kinds of planets astronomers can image.