How Webb recovered a planet that starlight had hidden

A signal recovered from an existing observation

The James Webb Space Telescope looked at 51 Eridani on October 18, 2023. Its camera collected the infrared light, and astronomers knew where to look for the young gas giant orbiting the star. Yet in one filter, centered at 4.3 microns, the planet disappeared into the residual light from its host. Researchers have now pulled it out of the same data by modeling how the telescope formed the image.

The result, published online in Nature on November 4, is a new detection of a known planet in an earlier observation at a particular wavelength. Astronomers first reported 51 Eridani b in 2015 using the Gemini Planet Imager in Chile. Webb had already detected it around 4.1 microns. The longer-wavelength recovery matters because it adds a measurement in a band that can help researchers study the planet's atmosphere. It also gives them a way to ask what other signals might survive in archives of observations that once looked unpromising.

Source notes: 1, 4, 5, 6.

The instrument leaves a pattern behind

Webb's Near-Infrared Camera, or NIRCam, has a coronagraph that suppresses light from a star so its much fainter surroundings can be measured. The mask removes much of the brightest light, but diffraction and imperfections in the optics spread some of it across the detector. Near the star, a fragment of this pattern can look like the spot astronomers hope to find.

One established way to remove that pattern uses images of reference stars. A computer combines reference patterns to estimate the glare in the science image and subtracts it. The method works best when the references describe what the telescope actually did during the observation. Even a slight change in how precisely the star sits behind the mask, or in how light travels through the optics, can leave a different pattern. Subtraction then leaves residuals whose shapes and brightness compete with the planet's signal.

The new method, called FLIP, starts with a physical model of the telescope and uses measurements of its optics to simulate how starlight spreads across the detector. It first calibrates that model on a reference observation. Then it adjusts a smaller set of parameters against the science image itself. Automatic differentiation helps the software calculate how changes to model parameters affect the mismatch between the prediction and the observed image. The fit can account for some changes that a fixed library of reference images misses.

There is a danger in giving any model that much flexibility. A fit that keeps improving its match to the whole image can also start absorbing the faint planetary signal into the estimate of the glare. The researchers checked for this by adding simulated planets of known brightness and position to the data and tracking how much of each signal survived. Those tests helped them decide when to stop fitting and estimate what kinds of signals their processing could recover.

Source notes: 1, 2.

What the sensitivity gain measures

The researchers compared FLIP with the established subtraction method KLIP on archival observations of three stars. Two systems, 51 Eridani and HIP 65426, have known planets. The third, TYC 5899, has no known planetary companion. It provided a test of how well the method could recover injected signals when the reference and science observations were taken months apart.

The reported gains varied with the data and with the distance from the star in the image. The roughly fifteenfold improvement was a contrast result at half an arcsecond in the TYC 5899 test. In the 4.3-micron observation of 51 Eridani, the paper reports about a fourfold improvement in sensitivity at the planet's location, where earlier processing did not recover it. The telescope did not collect more photons. The analysis lowered the brightness a source needed to have to stand out against what remained after subtraction.

Recovering a planet at a known position gives the team a useful check against measurements from other filters and instruments. Searching for an unknown planet is a harder test. A newly visible spot could still be a background object or a remnant of the instrument's response. The authors themselves flag a fainter feature in the 4.3-micron image whose nature remains uncertain. A convincing new candidate would need corroboration, such as observations over time that show it moving with the star.

Source notes: 1, 3, 5.

What astronomers can do with the extra wavelength

The new detection lies in a region of the infrared spectrum where carbon dioxide absorbs light. Astronomers can combine the brightness in this band with measurements at other wavelengths and compare them against atmospheric models. That can help constrain which models fit the data and, in turn, how the gas giant may have formed. This study establishes the recovery of the signal. The authors describe a fuller atmospheric analysis as future work.

The immediate opportunity is to rerun the method on other archived coronagraph observations and see where the result holds up. Different targets, filters and reference data will expose different failure modes, and the paper reports that the method still falls short of the sensitivity limit imposed by photon noise. Future observatories might eventually make use of the same approach, but their instruments will have to be tested under their own observing conditions. For Webb, the practical result is already concrete: the instrument had recorded usable light at 4.3 microns, and a more faithful account of the star's glare made it possible to detect the planet in that band.

Source notes: 1, 2, 6.