Directly imaging worlds beyond our solar system presents a fundamental challenge in modern astronomy. Even though advanced observatories can peer back at galaxies from billions of years ago, the blinding glare of a host star routinely overpowers the faint light reflected or emitted by orbiting planets. Because exoplanets can be a million times dimmer than their parent stars, separating the two signals requires extraordinary precision.

To address this obstacle, researchers at Northwestern University developed a new computational method that successfully recovered a giant planet hidden within two-year-old data from the James Webb Space Telescope (JWST). Published in the journal Nature, the technique improves the telescope's effective sensitivity to faint exoplanets by two to fifteen times without requiring hardware upgrades.

Overcoming the Challenge of Starlight Glare

Astronomers routinely use specialized hardware called coronagraphs to physically block starlight when searching for companions. However, residual light still leaks around these instruments, creating intricate patterns of glare that can obscure nearby planets. For the past two decades, researchers have attempted to clear this interference by using reference photographs of different stars, subtracting them from target observations to eliminate the unwanted glow.

The core difficulty lies in minute hardware variations. Telescopes never remain entirely still; thermal shifts, mechanical vibrations from cryogenic cooling motors, and tiny optical adjustments can alter how starlight spreads across a detector. A microscopic shift of just a few nanometers in a mirror changes the light pattern, causing standard reference subtractions to leave behind residual noise capable of hiding a planet.

How Physics-Based Learning Unlocks Hidden Data

To bypass these physical limitations, the Northwestern team created a software solution called Physics-based Learning for Imaging Planets (FLIP). Instead of depending entirely on external reference stars, the digital approach builds a computational twin of the telescope:

  • Digital Modeling: The computer model accurately simulates how starlight travels through the telescope's mirrors and hits its detectors.
  • Parameter Adjustment: Millions of parameters within the simulation are iteratively tuned to match optical distortions, vibrations, and minor movements.
  • Precise Subtraction: By carefully predicting the exact starlight pattern for a given observation, the system can remove the glare with high accuracy.

Although the technique leverages optimization tools typically associated with machine learning, it relies strictly on established physical principles rather than artificial intelligence to recognize celestial bodies.

Recovering 51 Eridani b in JWST Data

The research team tested FLIP on existing JWST observations of three stars, including HIP 65426 and 51 Eridani, which harbor known planets, alongside TYC 5899, which has no known companions. The method excelled particularly when processing light from exceptionally faint targets.

A prime example involved the giant planet 51 Eridani b. Although originally found via ground-based telescopes, the world remained invisible in JWST observations captured at a 4.3-micron infrared wavelength. This specific wavelength is scientifically vital because carbon dioxide heavily absorbs light there, offering clues about atmospheric compositions. Unfortunately, that same absorption made the planet's already faint profile even more difficult to distinguish.

When FLIP was applied to those archival observations, the target emerged clearly from the stellar glare. The success demonstrates that valuable scientific data has been left sitting in existing archives, waiting for more optimal analysis tools.

Broader Implications for Future Space Missions

Refined starlight removal techniques could significantly impact future astronomy efforts, including NASA's proposed Habitable Worlds Observatory. That future mission aims to photograph Earth-like planets directly, a goal that would otherwise demand hardware stability at the picometer scale—making instruments roughly 100 to 1,000 times more stable than JWST.

By using computational models like FLIP to compensate for unavoidable optical imperfections, researchers may expand the variety of worlds they can study.