Software squeezes more data out of Webb Telescope for imaging exoplanets

Directly imaging exoplanets is difficult because their host stars overwhelm the faint light from the planets. Coronagraphs block much of the starlight, but subtle changes in a telescope’s shape and pointing between observations leave residual noise. A key way to remove that noise is to model the telescope’s point spread function: how its optics turn a point of light into a blurred image. Deriving this function from observations can fail when the instrument’s behavior changes or when observations use different wavelengths. Modeling the hardware instead has been computationally unwieldy because so many interacting parameters affect its performance.

Researchers developed FLIP, a software method that uses a physical model of Webb’s NIRCam instrument and gradient descent to tune 106 parameters. The process begins with a simulated point spread function, optimizes it against observations of a star, then fine-tunes the result for the specific science image. Running on a modern GPU takes a little over 30 minutes. Tests on Webb observations of stars with known planets improved the signal-to-noise ratio by 1.5 to 2 times, and revealed one planet at a wavelength useful for detecting carbon dioxide. For a variable star without known planets, sensitivity improved fifteenfold. The method could help find fainter, older, or lower-mass planets, and analysis of its adjustments may also reveal changes in the telescope’s hardware.