A Cosmic Symphony Without a Score: How Neural Networks Hear Galaxy Clusters ✛ ✦

Galaxy clusters are the greatest gravitational ensembles in the Universe. In X-ray glow from hot plasma, the story of cosmic expansion, hidden dark matter, and enigmatic dark energy is encoded. Traditionally, analyzing this data resembled reconstructing a symphony from notes snatched from different histogram bins — overtones and harmonies were inevitably lost. Yet each catalog contains thousands of objects, each with its own redshift, luminosity, and gravitational lensing signal.

Imagine a conductor who doesn't look at the score or count notes for each instrument. They simply listen to the entire orchestra, and at some point begin to understand what music is hidden in the chaos of sounds. That's exactly how the new approach works: a set-based neural network. Each cluster becomes an acoustic fingerprint, and an aggregator layer picks up collective harmonics — not just the sum and maximum, but also the correlation matrix between features, as if a conductor discerns the interplay of instruments. Then a Bayesian emulator, a masked autoregressive flow, learns to output the probability distribution for ΛCDM parameters, bypassing the heavy score of explicit likelihood.

Training on 200,000 simulated catalogs replicating eRASS1 selection (~3,300 clusters with 97% purity) yielded a stunning result. Numerical simulations showed: the mean uncertainty for Ωm was 11.5%, for σ8—4.4%. This is comparable to computationally expensive MCMC analysis but requires neither complex likelihood integrals nor rigid binning. SBC and TARP calibration tests confirmed no systematic shifts: the conductor doesn't miss a beat. And it seems to even capture correlations we haven't accounted for, as if discerning the whisper of instruments we haven't yet added to the orchestra.

This is more than accuracy. The method banishes the fear of high dimensionality: now complex systematics and scaling relations can be included in the model without simplifications. Petabyte streams from Euclid, Rubin, and Roman will become a new score, and SBI approaches may finally make out the faint voice of dark matter, resolving the long-standing tension between cosmic microwave background and large-scale structure. The cosmic symphony gains notes, and the neural network conductor performs it without a single sheet of music — listening to the very essence of matter and void.