A Planet Flashed Once — and Was Found ✦

Original: "One Transit Is All You Need: Detecting Exoplanets Through Learned Stellar Behaviour with EXOVEIL"

arXiv:2606.02778v3 · 2026-06-01 · CC BY 4.0 · ⏱ 1 min · Exoplanets Instrumentation Machine Learning

The ExoVeil system finds planets around other stars from a single eclipse.

Abstract

EXOVEIL is a system that teaches telescopes to spot planets, even if they pass in front of their star just once. Imagine a lantern whose light dims slightly when a moth flits by: EXOVEIL catches such moments without waiting for repeats. This is how we can find distant Earth-like worlds. What else is hiding in the flicker of stars?

Links in the knowledge graph 1

A street lamp shines every night, and you get used to its brightness. If something briefly blocks the bulb, the shadow is immediately noticeable. ExoVeil works on the same principle — a new method for finding planets around other stars. Before, astronomers waited for multiple transits, but the system catches even a single dip in light. It studies the behavior of thousands of stars in advance, including sun-like and tiny cool ones, and then predicts their normal shine. A brief drop in a measurement is a candidate for a planetary eclipse.

The first version was wrong more often than a coin toss: it confused eclipses in binary systems (for example, with a dead stellar core) with planets.

The method is especially valuable for searching for Earth-like worlds in the Galaxy. The future PLATO telescope will be able to find planets in distant orbits. And splitting light will show whether the atmosphere contains water, methane, or carbon dioxide. The discoveries are confirmed by the color shift according to Doppler and the laws of Rydberg — they reveal the chemical composition through dark lines.

🎯 The system trained on data from 16,499 stars — as many spectators as a good stadium can hold.

\delta = \left(\frac{R_p}{R_s}\right)^2

Transit depth—the fraction of light blocked by the planet—is set by the ratio of squared radii. Jupiter covers 1% of the Sun's disk area, Earth only 0.008%.

\text{SNR} = \frac{\sum_i (r_i \cdot m_i) / \sigma_i^2}{\sqrt{\sum_i m_i^2 / \sigma_i^2}}

The matched filter weights prediction residuals r_i with the transit template m_i and inverse noise variance. Quiet sections of the light curve get more weight, boosting sensitivity.

I am the author

A Planet Flashed Once — and Was Found ✦