An international team, led by P. Ocvirk of the Strasbourg Astronomical Observatory, will use Summit, the world’s most powerful supercomputer, to perform a gigantic simulation of galaxy formation during the first billion years of the universe, a period that includes the birth of the very first stars, marking the end of the “dark ages” period. Thanks to the power of Summit, this simulation, called Cosmic Dawn III, will be the largest simulation of its kind ever produced. It will use more than a trillion (one thousand billion) calculation elements (particles and gas parcels), for a total size of 21 Petabytes. This is the storage capacity of more than 300,000 64GB smartphones. Cosmic Dawn III will generate a dataset that will be made public, and will be valuable for interpreting data from new observatories dedicated to the first billion years of the Universe, such as NenuFAR (Nancay Radio Astronomy Station), the James Webb Space Telescope, and the future giant SKA and ELT telescopes.
Shortly after the Big Bang, the matter that makes up the Universe is in the form of a very hot plasma. It is only 300,000 years later, when the Universe has cooled down enough, that the first hydrogen atoms can be formed. The original plasma then becomes a neutral and relatively cold gas. Gravity then allows this gas to assemble to form the very first stars, about 150 million years after the Big Bang. Their intense light breaks the hydrogen atoms into protons and electrons, bringing the intergalactic medium back to the plasma state that prevailed just after the Big Bang. It is this stuttering in the history of ionization of the Universe, which led to the term of “re-ionization”, since the Universe becomes ionized again with the appearance of the first stars. Reionization is accompanied by photo-heating with possibly serious consequences: the gas becomes hot enough to escape the low gravity of the less massive galaxies, depriving them of the material that allowed them to form stars. Thus, young stars produce ionizing radiation that tends, under certain conditions, to slow or prevent the formation of subsequent stellar generations. This type of feedback makes modelling the formation of galaxies particularly difficult, as soon as radiation is taken into account, which is essential to properly describe the epoch of reionization, because it results in scale coupling: the cores of small galaxies of the young Universe span a few hundred light years at most, and yet their ability to collect (by gravity) and keep gas capable of forming stars can be affected by the collective radiation of neighbouring galaxies and up to several tens of millions of light years away. It is this large range of spatial scales that the simulation must describe which is at the origin of its gigantism, it must be able to describe the small scales as well as the very large scales.
It is the sustained growth in computer power that makes it possible today to carry out calculations of such magnitude, unthinkable 20 years ago. On Titan, the predecessor to Summit, also the world’s most powerful supercomputer when it was built in 2012, Cosmic Dawn III would have required 540 million hours of computing time, more than 60,000 years on a single processor. Thanks to Summit’s almost 2.5 million cores, this time is reduced to one week. However, effectively orchestrating such a ballet of 0s and 1s is an enormous challenge, and has been an important part of the research effort of the Strasbourg team for more than 10 years. As a result, the team was awarded the ATOS – Joseph Fourier prize in 2019 for its pioneering and continuing work in the field, and in particular on hybrid architectures such as Summit. The exceptional access granted confirms the international recognition of the team’s know-how, acquired through regional computing resources, first at the centre de calcul de l’Universite de Strasbourg, then at the national level, thanks to GENCI (Grand Equipement National de Calcul Intensif) and IDRIS (Institut de Développement et des Ressources en Informatique Scientifique), and finally at the European level. Cosmic Dawn III will allow researchers to study the epoch of the reionization of the Universe as driven by the formation of the first galaxies, the appearance and the role of chemical elements and the formation of dust in the modulation of their ionizing emissivity and their interaction with the ultraviolet background, but also to better capture the properties and evolution of the population of absorbents of the intergalactic environment and the complex mixture of the different phases of hot and cold gas surrounding the galaxies.
Another particularity of Cosmic Dawn III is that it will use as initial conditions (i.e. as a “starting point”) a distribution of matter which, at the end of cosmic evolution under the effect of gravity, produces a virtual Local Universe, containing several of the large known structures of our cosmic environment, namely a Milky Way and its neighbour the Andromeda galaxy, the Virgo galaxy cluster, and several nearby galaxy groups. Despite these specificities, the large-scale distribution of matter has the same statistical properties as a random volume of universe. Thus, Cosmic Dawn III will not only describe the evolution of an “average” volume of Universe, but will also show how our own cosmic neighbourhood was formed and evolved during the first billion years of the
Although the $4 million cost of producing the simulation is entirely borne by the American computer centre (Oak Ridge Leadership Computing Facility), carrying out and analysing a simulation of this magnitude also requires significant human resources. In addition to P. Ocvirk, the project will involve D. Aubert, J. Chardin, J. Lewis, N. Gillet, also at Strasbourg Observatory, as well as a team of collaborators in France and all around the world, in particular, J. Sorce, member of the executive board of the CLUES project (Constrained Local UniversE Simulations), is responsible for the production of the initial conditions of Cosmic Dawn III.
References and links :
Oak Ridge Leadership Computing Facility: https://www.olcf.ornl.gov/
Pierre Ocvirk, firstname.lastname@example.org
A high resolution version of the figure is available :