Cosmic Recipe: How a Neural Network Guesses What the Universe Is Made Of ✛ ✦
From galaxy clusters (giant clouds of thousands of star systems, superhot gas and invisible dark matter) we can calculate how much dark matter and dark energy drive the Universe's expansion. It's like trying to reconstruct a recipe from taste alone, without knowing the ingredients. Scientists used to build complex formulas by hand; now a neural network learns to guess the composition right away.
The method is based on training on thousands of simulated universes with different 'recipes.' The neural network doesn't memorize equations: it directly outputs likely parameters. Its main advantage is handling lists of clusters of any length: like a chef who can confidently assess a dish from twenty ingredients or a hundred. Accuracy already rivals traditional calculations.
The gas in them is heated to tens of millions of degrees — hotter than the Sun's core. The approach is universal: it can also be used to study gravitational lensing (where massive bodies bend light) or redshift (which measures distances to faraway galaxies).
X-ray emission from clusters is completely absorbed by Earth's atmosphere, so the eROSITA telescope operates in space, collecting signals invisible from the ground.