The temperature at the center of a galaxy cluster reaches a hundred million degrees. These largest gravitationally bound systems are natural laboratories where nature hides 85% of its mass in invisible dark matter. As early as the 1930s, Fritz Zwicky noticed that galaxies move too fast, and later Vera Rubin confirmed the mystery on the scale of individual galaxies. To read the true weight of such a galaxy cluster, one must understand the language of its hot gas, which shines in X-ray radiation.

Each cluster is a book written in plasma. Its pages are concentric layers from the core to the periphery, recording temperature, density, and pressure. The traditional hydrostatic method reads each page individually, assuming complete calm. But if a cluster undergoes a merger, equilibrium is disrupted and the text becomes riddled with errors.

The neural network with a GRU architecture (a gated recurrent unit network) reads the book differently: it remembers previous pages and picks up on the hidden plot. Trained on 16,495 virtual clusters from the TheThreeHundred project, the network processes radial profiles of entropy, temperature, and mass sequentially, step by step, and predicts the total enclosed mass without any assumptions about equilibrium. An ensemble of 50 models gives not only an accurate value but also an honest uncertainty.

The result speaks for itself: prediction error does not exceed 5% over most of the radius — twice as good as classical hydrostatic weighing. Even in the chaos of merging systems, there is no bias. When the network was tested on real data from the REXCESS and X-COP surveys (spectroscopy at redshifts up to z~0.1), it turned out that old methods systematically underestimate the most massive clusters by about 10%. Additional physics such as active galactic nucleus feedback does not throw the network off: it generalizes what it learned from simulations of varying complexity.

Now the neural network is ready to open the archives of eROSITA and CHEX-MATE — vast surveys that will capture hundreds of thousands of clusters. This will bring cosmology to a new level of precision: calibration of mass–observable relations becomes objective, and the parameters of dark matter and dark energy become less shaky. So the first book from the great library of the Universe has been read, and its plot is clear enough that it's time to turn the page.