Researchers have developed a computational method that allows the James Webb Space Telescope (JWST) to peer through the glare of bright stars to reveal previously hidden exoplanets. The technique, known as Physics-based Learning for Imaging Planets (FLIP), improves the telescope’s sensitivity to faint celestial bodies by a factor of up to 15.
Building a Digital Twin of JWST
The FLIP method functions by creating a “digital twin” of the JWST’s Near Infrared Camera (NIRCam). By modeling the physical properties of the telescope—including its optics, mirrors, and detector responses—the software simulates how light travels through the hardware. The researchers then use gradient descent, a technique common in training large neural networks, to tune approximately one million parameters until the digital twin matches the actual observations recorded by the telescope. This approach allows the software to account for tiny mechanical, thermal, or optical fluctuations that define the telescope’s stability limit.
Unlike standard image processing, FLIP does not use artificial intelligence to identify planets. Instead, it utilizes an optimization strategy to solve for free parameters—such as star position, flux, and pixel responses—that are informed by wavefront sensing data. This process relies on a physical model of the telescope’s hardware, allowing for more precise starlight subtraction. The optimization runs on a GPU, such as the NVIDIA A30, and typically completes in roughly 35 minutes, though this is a benchmark rather than a fixed duration for every dataset.
His research trajectory included a Summer Undergraduate Research Fellowship project with Jason Wang, who was previously a 51 Pegasi b Postdoctoral Fellow in the lab of Dimitri Mawet, the director of instrumentation for the Caltech Optical Observatories. The study was supported by funding from the Space Telescope Science Institute, the Sloan Foundation, an Amazon AI4Science Discovery Award, and the National Science Foundation.
Rodrigo’s senior thesis was an exploration of possible approaches, none of which ended up being what we did for this paper, but it helped rule out what not to do and also helped us understand the problem better.
Jason Wang, assistant professor of physics and astronomy at Northwestern University
Recovering the Exoplanet 51 Eridani b
The team tested their method on archival JWST data of three star systems: HIP 65426, 51 Eridani, and TYC 5899. 51 Eridani b, a giant planet discovered by the ground-based Gemini Planet Imager, had remained undetected in previous analyses of JWST imagery due to the overwhelming glare of its host star. By applying FLIP, the researchers successfully recovered the signal of the planet at 4.3 microns, a wavelength where carbon dioxide absorbs strongly. This capability is significant, as carbon dioxide absorption reveals details about a planet’s atmospheric heavy-element composition.
For the star TYC 5899, which has no known exoplanets but exhibits significant variability between observations, the FLIP method boosted sensitivity by a factor of 15. The researchers observed that at smaller angular separations, where Solar-System-like planets reside, FLIP consistently outperforms the standard Karhunen-Loève Image Projection (KLIP) technique. The team noted that the gains FLIP provides are most impactful close to stars, where planets like Earth typically orbit.
Impact on the Habitable Worlds Observatory
The findings, published in the journal Nature (DOI: 10.1038/s41586-026-11213-z), suggest that software-based corrections could reduce the extreme engineering requirements for future missions. NASA’s proposed Habitable Worlds Observatory (HWO), which is planned for launch in the 2040s, aims to directly image Earth-like planets. Achieving this requires the telescope to maintain optical stability at the picometer scale—approximately 100 to 1,000 times more stable than the JWST. If image-processing software like FLIP can effectively clean up stellar glare, the hardware constraints for future observatories may be relaxed.
This is a significant result.
Dimitri Mawet, senior research scientist at NASA’s Jet Propulsion Laboratory
The team intends to continue refining FLIP to highlight where it makes the most dramatic changes to the model, which could potentially provide diagnostic information regarding the telescope’s hardware performance. By lowering the noise floor in existing observations, the researchers believe they will be able to spot dimmer, older, or lower-mass planets that have previously eluded detection. As Ferrer-Chavez noted, the goal remains to better understand our own place in the galaxy by finding more planets, as the current inability to see them does not mean they do not exist.
Earlier coverage: New software helps Webb Telescope image previously hidden exoplanets
Worth a look