With a huge and still-growing catalog of exoplanets, researchers are anxious to take the next steps: trying to infer what conditions are like on them, either by imaging them directly or by capturing the light that has passed through their atmosphere. Both are substantial challenges, given the presence of stars that are far, far brighter than light that has come to us via the planet.

While telescopes can be fitted with devices called coronagraphs that block out much of a star’s light, there are limits to their effectiveness, caused by noise from subtle variations in the telescope’s configuration and pointing between observations. Now, researchers have developed a new software package that optimizes a physical model of the telescope in a way that allows us to better model these differences, producing low-noise imaging of exoplanets.

Point spreads

Central to this work is the concept of a point spread function. If you use a telescope to image a single point of light, it will not appear that way in the actual image. Instead, it will be slightly blurry and spread out, due to imperfections in the telescope hardware and the limits of our ability to keep it pointed accurately through the time it takes to image something. The mathematical description of how a telescope distorts a point of light is called its point spread function.

There are two ways to figure out what the point spread function is. One is to use the telescope to image a known target and use those observations to derive a mathematical description of the point spread function. This is effective at reducing noise but has its limits, as the behavior of the telescope hardware can change over time, and the point spread function may vary with the wavelengths being imaged.

The alternative is to try to model the hardware’s behavior and use that model to derive a point spread function. But modern telescopes are incredibly complicated instruments, with many, many factors feeding into their performance in ways that can be difficult to ascertain, even with the full engineering schematics. As a result, these models simply contain too many parameters that need to be tuned to make this route computationally tractable.

But anyone paying attention to the world of AI developments will recognize that we’ve become quite successful at tuning billions of parameters in neural networks, in part through a mathematical technique called gradient descent. And, because this is central to training large neural networks, people have put a lot of work into optimizing the performance of gradient descent on GPUs.

So, a team of researchers has developed a process they’re calling FLIP (a non-acronym for Physics-based Learning for Imaging Planets) that uses a physical model of the hardware (in this case, the NIRCam on the James Webb Space Telescope) with 10⁶ possible parameters, and then uses software to optimize all its parameters. The optimization starts with a simulated point spread function, then uses gradient descent to optimize its performance on data from an actual star. Run on a current GPU, this process only takes a bit more than half an hour.

The results of that are then fine-tuned on the actual scientific image they are analyzing. This last part of the process is critical, since some of the variables will differ between observations.

Good results

To test how this works, the researchers used observations of two stars with known exoplanets (HIP 65426 and 51 Eridani), along with a third (TYC 5899) that doesn’t have any known exoplanets but is known to vary a lot between observations. These are young stars, and two of them have massive planets that are still warm from their gravitational collapse, allowing them to have been imaged previously.

Using Webb data, the team used FLIP to boost the signal-to-noise ratio by a factor of 1.5 to 2, depending on the observations. At a wavelength that can be used to detect carbon dioxide, they were able to spot one of the planets in the Webb data for the first time. For TYC 5899, the approach boosted the sensitivity of the observations by a factor of 15.

The researchers point out that the applications go beyond getting a better view of known planets. By lowering the noise, it should also be possible to spot dimmer (and hence older or lower-mass) ones that had previously eluded detection. They also think they can modify FLIP to highlight where it’s making the most dramatic changes to the model, which can potentially tell us useful things about what’s going on with the telescope’s hardware.

Nature, 2026. DOI: 10.1038/s41586-026-11213-z  (About DOIs).