Two Record-Breaking Universe Simulations Track 13.4 Trillion Dark Matter Particles

Cosmologists have a problem that sounds almost like a paradox: the better their telescopes get, the harder it becomes to interpret what they see. Dark matter, the invisible substance thought to make up most of the mass in the universe, leaves no light of its own. Its structure only becomes visible indirectly, through how it bends light and shapes where galaxies form. To connect those observations to theory, researchers build vast virtual universes and let them evolve.

That is what a team at the U.S. Department of Energy’s Argonne National Laboratory has just done at unprecedented scale. Using Aurora, one of the world’s first exascale supercomputers, the scientists completed the two largest cosmological simulations ever run, according to an announcement reported by Chinese outlet Science and Technology Daily. Each simulation followed the evolution of 13.4 trillion dark matter particles across a region 28.6 billion light-years wide.

The number is hard to picture. For comparison, earlier flagship simulations tracked particles numbering in the billions or low trillions. Jumping to more than 13 trillion particles in a single run means the virtual cosmos can resolve much finer structures, small clumps of dark matter that might otherwise blur into a statistical haze. Those small structures matter because they leave fingerprints on how galaxies cluster and how dark matter distorts the light of distant objects.

The simulations are aimed at a key target: cosmic inflation, the theory that the very early universe went through a brief, violent burst of expansion. That process would have stretched tiny quantum fluctuations into the seeds of today’s large-scale structure. By comparing simulated universes with different inflation scenarios against real observations, astronomers can narrow down which versions of the theory survive.

There is a practical payoff too. A new generation of sky surveys is mapping billions of galaxies and measuring subtle distortions in their shapes. These surveys produce enormous datasets, and squeezing cosmological information out of them requires equally enormous theoretical predictions. The Aurora runs are intended to serve as a reference library for that analysis, helping researchers separate genuine signals from noise and modeling artifacts.

Exascale computing, meaning machines capable of more than a quintillion calculations per second, has been a long-sought milestone. Aurora, based at Argonne near Chicago, is among the first systems to cross that threshold. Cosmology has become one of its showcase workloads, since gravity-only simulations of dark matter are computationally demanding but relatively clean to model, making them an ideal stress test for new hardware.

For global readers, the story is a reminder that modern astronomy is increasingly a computational enterprise. Telescopes gather the photons, but supercomputers turn them into understanding. As observatories in Chile, Hawaii and eventually space generate ever-larger catalogues of galaxies, the bottleneck is shifting from collecting data to interpreting it.

The Argonne team’s results are still simulations, not discoveries, and their scientific value will depend on how well they match reality. But each doubling of particle count brings the virtual universe a little closer to the real one, and the largest structures in the cosmos are now being recreated on a machine in Illinois, one particle at a time.