Astrophysicist Uses Codex to Break Supercomputing Barriers in Black Hole Simulation

Chi-kwan Chan, an astrophysicist at the University of Arizona and Steward Observatory, is using Codex to attack a computational problem that has limited black hole simulations for decades: how to track the countless charged particles spiraling around supermassive black holes without burning the world's fastest supercomputers on microscopic details.

Chan is part of the international Event Horizon Telescope (EHT) collaboration, which published the first image of a black hole in 2019. The team is now gathering observations to produce the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy. Turning those observations into scientific understanding requires enormous amounts of data processing, large-scale computing workflows, and simulations capable of modeling some of the most extreme physics in the universe.

The stakes are fundamental. "Black holes are among the best places to test Einstein's general theory of relativity," Chan said. The theory remains the best available explanation of gravity: rather than a force pulling objects together, gravity results from mass and energy bending the fabric of space and time.

A spiraling problem

Because light cannot escape a black hole, scientists study the region around it called the event horizon, a boundary beyond which nothing can escape. "It's a surface of no return," said Chan. Matter swirling just outside this boundary emits light that astrophysicists can see, measure, and simulate. The 2019 EHT image showed a black hole's shadow embedded in glowing plasma near the event horizon. Chan helped develop the simulation and computing tools the team used to interpret the observations, and his group has kept improving those instruments as the collaboration moves from still images toward videos.

The biggest roadblock is modeling the plasma itself. Plasma is superheated matter made up of electrically charged electrons and ions. Many simulations simplify plasma by treating it as a fluid, using well-known equations to model its movement around a black hole. That approximation works reasonably well in dense plasma, where electrons and ions constantly collide.

But near the supermassive black holes Chan studies, some regions become so hot and diffuse that particles rarely meet. "They don't really collide with each other," he said. Instead, the particles mostly spiral around magnetic field lines.

Modeling that behavior correctly means following trillions of electrons and ions as they corkscrew rapidly around a black hole. Standard simulations must calculate every tiny turn, forcing computers to take extremely small timesteps. The result: even the world's fastest supercomputers spend most of their time computing minuscule particle motions instead of the larger behavior scientists actually want to study.

"For decades, this has limited how realistically we can simulate black hole plasma," Chan said.

A testable pipeline, not a black box

Chan suspected new mathematical techniques could work around those limitations. The core idea was to change, mathematically, how the simulation tracked particle motion, so the computer no longer had to follow every tiny spiral directly.

"But exploring all the mathematical possibilities by hand would have taken an enormous amount of time," Chan said. So he turned to Codex to help derive candidate algorithms and test them against known solutions.

Codex generated many potential approaches, and not all of them were correct. "But that's okay," Chan said. "Most scientific ideas fail. What matters is that these algorithms are testable. Once you find one that works, it can potentially unlock simulations that were previously impossible."

That emphasis on testability distinguishes Chan's use of AI from typical generative workflows. Some AI systems return results without showing the steps behind their conclusions. Chan's group instead uses Codex to propose and implement numerical schemes that researchers can inspect, test, and understand physically.

Large language models still make mistakes, and many scientists remain cautious about using AI in research. Chan argues that science may be one of the best uses for today's AI systems precisely because scientific ideas can be tested rigorously. "We don't accept an idea because it came from Einstein, from a bright student, or from an AI model," he said. "We accept it only after repeated testing."

He sees AI as a tool that helps researchers explore more ideas, test them faster, and accelerate discovery while remaining grounded in verification and reproducibility. That framing matters as research institutions debate how to integrate AI into scientific practice without letting unverified outputs contaminate the literature.

If the approaches Chan is testing with Codex succeed, the new algorithms could eventually allow scientists to simulate trillions of particles around black holes. That would open physics that has remained out of reach for decades just as the EHT prepares to release the first video of a supermassive black hole.