Galaxies are considered to be the largest building blocks in the universe. But how were they born? How did they evolve? In order to get closer to the answers to these questions, an international team of scientists - including researchers from the Max Planck Institutes for Astronomy and Astrophysics - packed space into the computer. The result is Illustris TNG50, the most detailed large-scale cosmological simulation to date. For the first time, it reveals that the geometry of the cosmic gas flows around galaxies determines galaxy structure, and vice versa. Conversely, the properties of the gas flows result from the evolution of the galaxies.
Astronomers running cosmological simulations face a fundamental trade-off: with finite computing power, typical simulations so far have been either very detailed or have spanned a large volume of virtual space, but not both. Detailed simulations with limited volumes can model no more than a few galaxies, making statistical deductions difficult. Large-volume simulations, in turn, typically lack fine details on smaller scales, which are important for describing individual galaxies. The TNG50 simulation, which has just been published, manages to avoid this trade-off. For the first time, it combines the idea of a large-scale cosmological simulation – a Universe in a box – with the computational resolution of “zoom” simulations, at a level of detail that had previously only been possible for studies of individual galaxies.
In a simulated cube of space that is more than 230 million light-years across, TNG50 can discern physical phenomena that occur on scales one million times smaller, tracing the simultaneous evolution of thousands of galaxies over 13.8 billion years of cosmic history. It does so with more than 20 billion particles representing dark matter, stars, cosmic gas, magnetic fields, and supermassive black holes. The calculation itself required 16,000 cores on the Hazel Hen supercomputer in Stuttgart working together, 24/7, for more than a year – the equivalent of fifteen thousand years on a single processor, making it one of the most demanding astrophysical computations to date.