Neural Network Cleans Stellar Noise from the Search for Distant Worlds express
For six straight years, the HARPS-N telescope on La Palma island stared at the Sun as if it were a distant star — taking spectra, tracking brightness, and recording every quiver of spectral lines. All this was to understand how to distinguish a star's subtle sway from a genuine planetary signal.
A star doesn't sit still: its surface boils like a pot of soup. Hot bubbles rise, spots appear, and magnetic fields shift. All of this slightly displaces spectral lines, mimicking the gentle wobble caused by a planet. Previously, astronomers tried to subtract this 'stellar noise' manually, but it's too complex and irregular.
The neural network took over this task. It was trained on six years of solar data, with hints about which changes in brightness, magnetic field, and spectral line shapes are linked to surface boiling. It turned out that not all hints are equally useful. For instance, line tilt or the width of the sodium band barely help. But magnetic flux, overall brightness and its rate of change, and the shape of the hydrogen line are highly informative.
What remains is just under a meter per second, and this appears to be the fundamental limit set by giant convective cells on the Sun, each the size of entire continents. To catch an Earth-like planet, even this noise must be removed. So next time, the neural network will learn from even subtler traces of boiling.
Original abstract
as written by the authors arXiv:2602.17760
The measurement of exoplanet masses using the radial velocity (RV) technique is currently limited by stellar activity, which introduces quasiperiodic variability signals that must be modeled and removed to enhance the sensitivity of the RV measurements to exoplanet signals. Neural networks have previously been demonstrated effective in modeling stellar activity signals in HARPS-N solar data using white light cross correlation functions (CCFs). Building on this work, we train a neural network on six years of HARPS-N solar data with additional parameters commonly associated to stellar activity, including chromatic CCFs, line shape metrics, spectral activity indicators, total solar irradiance (TSI) light curves from SORCE and TSIS-1, and TSI time derivatives. Our results show that parameters such as the bisector inverse slope and Na D equivalent widths do not significantly improve the neural network's ability to predict activity-induced RV variations compared to using the white light CCFs alone. However, parameters such as unsigned magnetic flux, the TSI and its time derivative, S-index, H-alpha equivalent width, chromatic CCFs, contrast, and full width at half maximum do improve the neural network's ability to predict RV scatter. Our new model reduces the RV scatter in a held-out test set from 147.1 cm/s to 93.3 cm/s, consistent with supergranulation noise levels reported in previous studies. These results suggest that finding effective tracers for (super)granulation will be critical to train models capable of further mitigating RV jitter, and necessary for characterizing Earth analogues.
A speed of 93 centimeters per second is slower than a garden snail's crawl (about 1.5 cm/s, but a snail can't sustain that pace for long, while a star 'trembles' continuously).
In the movie 'Interstellar', the heroes search for planets around a black hole, but even around a calm star like the Sun, its own noise hinders spotting Earth — now a neural network helps 'hear' the quiet signal.