Lumina Simulation Recreates the Universe's Escape From Its Cosmic Dark Age

A team of researchers has built a virtual universe on one of the most powerful supercomputers in the world, and it starts in the dark. The simulation is called Lumina, after the Latin word for "lights." It begins about 45 million years after the Big Bang, when neutral hydrogen dominated all of space, and it runs forward for 2 billion years. The work, developed by researchers at MIT, Harvard University and other institutions, was published in the Monthly Notices of the Royal Astronomical Society.

The period it recreates has a name: the cosmic dark age. The story starts with the Big Bang, when the universe was an extremely hot, dense plasma filled with electrons and other subatomic particles, described as initially smaller than a single atom. As it expanded and cooled, particles slowed enough for electrons to combine with atomic nuclei, forming electrically neutral hydrogen. Around 380,000 years after the Big Bang, the dark age began. Space had become transparent to much of the leftover radiation from the Big Bang, but no stars had yet formed to light it. Neutral hydrogen filled the cosmos and could absorb certain kinds of radiation, including the ultraviolet light the first stars would later produce.

Lumina covers a region 500 megaparsecs across, roughly 1.6 billion light-years. Inside that volume it follows dark matter, gas, stars and radiation such as photons. It contains approximately half a trillion individual elements, which are particles of dark matter and cells of gas. That gives it about 30 times the resolution of IllustrisTNG, the earlier simulation it builds on, in a region 100 times larger in volume than previous leading simulations. The team upgraded the IllustrisTNG software to model galaxy formation in comparable detail over a far bigger space, and added a radiation transport model that calculates how light, including ultraviolet, moves through the universe and interacts with matter.

The main result concerns two transformations: hydrogen re-ionization and helium re-ionization. In the first, ultraviolet light from the earliest stars and galaxies stripped electrons from neutral hydrogen atoms. As more hydrogen was ionized, the universe grew more transparent to ultraviolet light, which could then travel much farther. Helium re-ionization needs substantially more energetic radiation. The simulation points to different drivers: early stars and galaxies appear primarily responsible for ionizing hydrogen, while actively feeding supermassive black holes supplied the intense radiation needed for helium. It is the first simulation to capture both processes together.

The computing demand was enormous. The team ran Lumina on Frontier, at Oak Ridge National Laboratory in Tennessee. MIT co-author Xuejian (Jacob) Shen said that at peak scale the calculation draws on the equivalent of more than 100,000 CPU cores and around 15,000 GPUs, and that getting all of those processors to communicate efficiently while evolving trillions of interacting pieces of information is itself a feat of computer engineering.

The model can be viewed at very different scales. The researchers have posted videos on the project website: one zooms from the whole simulated region down to a single disk-shaped galaxy, and another is a time-lapse of the full volume showing hydrogen re-ionization spreading over 2 billion years. First author Oliver Zier, a former postdoc in the Vogelsberger Lab, described the sequence: dark matter and gas clump into over-densities, gas falls into dark matter haloes, the first galaxies form there, and their stars emit radiation that the simulation tracks until it ionizes neutral hydrogen. Study author Mark Vogelsberger, a professor of physics at MIT, said other simulations cannot zoom in this far, and that the gap between the smallest and largest scales Lumina covers is larger than in any other.

The researchers are making Lumina available to other scientists so its predictions can be compared with observations from existing and future telescopes. They also plan to use it to study the large-scale arrangement of matter in the early universe and to help predict patterns of gravitational waves from ancient galaxies and black holes. Shen noted that its size lets it produce enormous populations of simulated galaxies, finally allowing comparison with the very large samples upcoming surveys will observe.

The team includes MIT authors Vogelsberger, Shen and Teodora-Elena Bluichi, plus Zier, Rongrong Liu, Sonja Koehler and Lars Hernquist of the Harvard and Smithsonian Center for Astrophysics, Aaron Smith of the University of Texas at Dallas, Rahul Kannan of York University, and Volker Springel and Rüdiger Pakmor of the Max Planck Institute for Astrophysics. NASA and the National Science Foundation provided partial support. The next test is whether Lumina's galaxy populations match what the coming telescope surveys actually find.