Cornell Researchers Use Exascale Computing to Rethink Cosmic Ray Transport

Feb. 14, 2025 — More than a century ago, scientists pondered how evolution might be driven by mutations caused in part by cosmic rays that bombard Earth after the explosive demise of faraway stars. Yet how cosmic rays move through space has been an open question in astrophysics for decades.

Because it exerts such powerful pressure in galaxies, cosmic ray scattering is vital to understanding the universe, says Fielding, an assistant professor of astronomy. “There’s as much energy in cosmic rays as there is in thermal pressure, magnetic pressure or turbulent pressure, and yet we have no idea how the cosmic rays actually work. By understanding cosmic ray pressure, we hope to understand how galaxies are able — or are not able — to turn gas into stars, which then allows us to understand the overall structure of the universe.”

The new results come from the research group’s early simulations on Oak Ridge Leadership Computing Facility’s Frontier supercomputer. Fielding and his colleagues received a Department of Energy (DOE) Innovative and Novel Computational Impact on Theory and Experiment (INCITE) award of 900,000 node-hours on Frontier and its counterpart at Argonne Leadership Computing Facility, Aurora. These DOE machines are the first and second in the world to achieve exascale, more than a million trillion operations per second. Philipp Kempski, Lyman Spitzer Jr. Postdoctoral Fellow in astrophysical sciences at Princeton, is co-principal investigator.

The longstanding model of cosmic ray transport has held sway since the early 1970s. Based on earthbound cosmic ray detectors that capture data on hydrogen nuclei, or protons, the predominant cosmic ray type, the model centers on an interstellar medium — the matter and energy between stars and galaxies — striated with wavy magnetic field lines. As protons race along their tracks, they induce waves in the magnetic field that lead to scattering.

Their new model calls for mostly smooth, not wavy, magnetic field lines that have infrequent but dramatic bends and kinks arising from large-scale interstellar gas turbulence that pulls the field lines with it, creating clumps or tangles. In this picture, cosmic rays take long, uneventful journeys along field lines, then plunge into the tangles, which scatter them. Higher-energy cosmic rays would need larger tangled regions to be scattered, and such enormous tangles are expected to be rare. “And so our model naturally has this energy-dependent scattering, which is empirically constrained by the boron-to-carbon ratio,” Fielding says.

Smaller simulations on Rusty, a computing cluster at the Simons Foundation’s Flatiron Institute in New York City, gave an inkling of the new concept, but the overall picture didn’t come into focus until the team got access to Frontier. Only an exascale machine could include the relevant large-scale turbulent motion, small-scale magnetic tangling and the interactions between the two. At the large scale is the galactic dynamo, astronomically long magnetic field lines that roughly follow the galaxy’s spiral as the rotating galactic mass drags them along. At the small scale, tangled field lines reverse polarity as they bend, break and reconnect, scattering the rays.

“Our simulations are pointing to a paradigm shift,” Fielding says. “What we already see is that the small scale in our simulations, which include the large scale simultaneously, looks really different than the small scale everybody thought was always going to be the case. And that’s because there is an interplay between the big and the small that’s only been possible to capture on Frontier.”

The team is pushing the limit of Frontier’s compute power. “I am using as much as they’ll let me use, 95% of the machine,” Fielding says. “It’s the biggest computer in the world, I’m using almost the whole thing, and I’m using it for a week. It’s really the absolute edge of what’s never been possible before.”

Fielding looks forward to working on Aurora, which will surpass Frontier as the most powerful machine. “What we’re looking to do with these simulations is to not just validate that this idea is plausible, but to really try to push it into a regime that’s large enough that we can start to see if it’s really the best explanation for what we see in the real universe.”