A neural network tastes the atmospheres of distant planets express
When a planet passes in front of a star, the light pierces its atmosphere and absorbs the flavor of gases. Each gas leaves its color fingerprint in the light — a bitter or sweet stroke. Scientists try to read this recipe: what substances and in what proportions are mixed in the air of an alien world. Old computers sifted through millions of versions blindly, like a chef tasting a dish blindfolded.
The Exoformer neural network acts like an experienced taster. It quickly suggests the main ingredients and their proportions. Trained on thousands of simulated 'dishes'—model atmospheres of giant planets around other stars: it works 3–8 times faster than previous methods without losing accuracy. Remarkably, training on artificial recipes was enough for flawless work with real ones.
Tests on real data from the JWST telescope for worlds WASP-39b and WASP-17b confirmed: Exoformer yields the same compositions as lengthy calculations. The hybrid approach opens the way for rapid analysis of the avalanche of light fingerprints from new telescopes and, perhaps, brings us closer to the moment when we find an atmosphere suitable for breathing.
Hot Jupiters are not just scorching balls of gas. Due to the hellish heat, their clouds turn into vapors of molten metals, and there are rains of liquid glass.
In the movie 'Interstellar,' the heroes search for planets with breathable air: this is precisely why fast methods of atmosphere analysis are needed.