Taking a picture of a planet around another star is mostly a problem of getting rid of light. Young giant planets, the easiest kind to image directly, can be about 100,000 times fainter than the stars they orbit, according to Northwestern University. Even a coronagraph that blocks the star's disk leaves behind a field of scattered glare, and that glare can hide a planet. For about two decades, astronomers have handled this in much the same way: they observe a reference star, treat its glare as a template and subtract it.
A paper published Nov. 4 in Nature argues that this approach has quietly been costing the James Webb Space Telescope part of its reach. The Northwestern-led team describes a method called FLIP. Instead of copying the glare pattern from another star, it models the telescope that produces the glare. When the researchers applied it to existing JWST data, they recovered a known giant planet at a wavelength where conventional processing could not distinguish it from residual glare.
The problem with borrowed glare
Reference-star subtraction only works if the telescope behaves the same way during both observations. Webb doesn't quite manage that, and very little would. As the Northwestern announcement explains, shifts of just a few nanometers in the optical path change the pattern of stray starlight. A template taken on one star at one moment won't match the glare around another star at a different moment, and even a slight mismatch can leave behind enough glare to hide a faint planet.
The paper's title, "Imaging exoplanets below the telescope stability limit," names the issue. The stability limit is the point beyond which mechanical, thermal or optical fluctuations make clean images impossible. Below that floor, a conventional subtraction struggles to tell a planet apart from the telescope's own drift. Universe Today frames this limit as the barrier that FLIP is designed to get past.
A digital twin instead of a template
FLIP replaces the template with physics. According to the Nature paper, the technique pairs a flexible physical model of the instrument with a multistep optimization. Automatic differentiation, the same kind of computational tool used to train machine-learning models, lets the team optimize millions of model parameters together so that the model reproduces the glare actually recorded in each observation. Space.com describes the result as a digital twin of the telescope: a simulated Webb whose small imperfections are tuned to match the real instrument. Wang is explicit that this is not AI finding planets; the team used the same optimization tools but relied on known physics.
The modeled glare comes from the same physical system that produced the image, not from a different star observed at another time. That means the subtraction can follow the nanometer-scale changes that defeat a borrowed template. Whatever is left after the model is removed is more likely to be real signal, such as a planet.
The team tested the method on Webb's NIRCam coronagraph, reanalyzing past observations of three stars: HIP 65426 and 51 Eridani, both with known planets, and TYC 5899, with none. The paper reports sensitivity 2 to 15 times better than standard techniques. Northwestern summarizes this as an improvement of up to 15 times in JWST's sensitivity.
51 Eridani b, recovered
The test case was 51 Eridani b, a gas giant discovered in 2015 by ground-based telescopes. Space.com puts it at roughly 11 Jupiter masses and about 97 light-years from Earth. Conventional processing of the JWST observations could not pick the planet out of the glare at the wavelength Webb uses to study carbon dioxide in planetary atmospheres. With FLIP, the team reports the first detection of 51 Eridani b at 4.3 microns.
The key point is that this did not require a new observing program. The photons were already in the archive. Better modeling of the telescope separated them from the glare. Jason Wang, a Northwestern assistant professor and co-author, called the result "science left on the table because we didn't have the right tools to analyze the data in the most optimal way."
The study was led by Rodrigo Ferrer-Chavez, a Northwestern PhD student, with Wang as co-author.
Why It Matters
There are three reasons to pay attention, and they get bigger in scope.
First, Webb's existing data. If one well-known planet was missing from earlier processing at 4.3 microns, other faint signals may be sitting in the archive too. Northwestern says the method could help astronomers discover smaller, colder and fainter planets without changing the telescopes themselves. Software can be applied to old data, which makes this an unusually cheap upgrade for a flagship observatory.
Second, the next telescopes. The Nature paper says the approach could be extended to other current and future observatories, including the Nancy Grace Roman Space Telescope and the proposed Habitable Worlds Observatory. The authors call testing on Roman a critical next step.
Third, the engineering requirements for finding Earth-like worlds. The Habitable Worlds Observatory is meant to directly image Earth-like planets around sun-like stars. Northwestern says that may require optical stability at the picometer scale, roughly 100 to 1,000 times more stable than JWST. The authors write that if the method can be extended to HWO contrast levels, the ultrastability requirement may be partially lifted. If software can model and remove some instability that would otherwise have to be engineered out, part of the difficulty moves from hardware to computation. That would be a meaningful change for a mission whose stability goals are among its hardest problems.
The caveats
The demonstration so far is one instrument and three target stars, with 51 Eridani b as the showcase planet. The 2-to-15 range is wide, and the gain will depend on the observation. The claims about Roman and the Habitable Worlds Observatory are conditional extensions the authors flag as needing further testing, not results anyone has shown. Any software substitute for ultrastability will have to hold up against the much tighter contrast that those missions require. Still, the main result is concrete: a planet that conventional processing of Webb's data could not detect at 4.3 microns is now detected in it, using the same data.
