US uses world’s first exascale supercomputer to model supernovae, fusion reactors

US scientists have turned to the world’s first exascale supercomputer, the Frontier, to unlock magnetic turbulence in plasma, widely regarded as one of physics’ most stubborn problems.

Researchers at the Department of Energy’s (DOE) Oak Ridge National Laboratory (ORNL) used AI models to train the world’s third fastest exascale supercomputer to precisely capture the chaotic behavior of plasma.

According to the results, the find could greatly reshape research and improve the way scientists model supernova explosions. It could also help build more efficient nuclear fusion reactors.

“This kind of capability has long been the dream of astrophysicists and many other scientists,” Eliu Huerta, PhD, a computational scientist at Argonne National Laboratory who oversaw the study, said.

“It’s the first time this level of insight via AI has been achieved for systems of this complexity.”

Modeling cosmic explosions

Magnetohydrodynamic (MHD) turbulence is a complex and chaotic process, that governs how electrically charged gases (plasmas), behave under the influence of magnetic fields. These turbulent flows are found throughout the universe.

They shape solar flares, supernova explosions, and even Earth’s own magnetic environment. However, accurately modeling MHD turbulence has remained a great challenge.

Traditional methods, like the Reynolds-Averaged Navier Stokes (RANS) approach, rely on simplified equations that smooth out fine details, and fail to account for all relevant physics.

“The more chaotic the system, the harder to simulate it,” Huerta noted. To tackle the challenge, the scientists came up with a two-stage approach. First, a physics-informed neural operator learns the behavior of plasma, while mapping how the system evolves over time.

Then, a diffusion-based model steps in to reconstruct the finer details, therefore regenerating the small eddies and rapid fluctuations that define turbulent flows.

But, training the models to generate thousands of detailed plasma simulations required massive computational power. For this purpose, the team received an allocation of computing time on the Frontier.

New simulations

Frontier, the third fastest exascale supercomputer in the world, after the Aurora and El Capitan, is capable of performing more than two quintillion calculations per second. This enabled simulations that were previously out of reach.

“Frontier was a lifesaver for us,” Semih Kacmaz, a graduate student who led the project, said.

“We used Frontier to generate high-fidelity datasets to train our diffusion model and to train our physics-informed neural operators.”

Its speed let the team train neural operators to capture overall MHD turbulence. Meanwhile, a diffusion model learned the finer details like small eddies and flows. Together, they worked as an AI tag team to simulate the turbulence.

This resulted a system that can produce highly detailed turbulence predictions in seconds, while reducing errors by more than half compared to previous methods.

“This is the first time AI has been able to faithfully model magnetized turbulence at such extreme conditions,” Huerta concluded in a press release. “By coupling the physics-informed neural operators with generative diffusion, we created a framework that respects the equations while recovering the full complexity of the plasma.”

The team aims to expand the model to handle more complex systems, including full 3D plasma simulations and astrophysical environments, as well as applications like modeling turbulence in nuclear fusion reactors.