An astrophysicist at the University of Arizona is using Codex to help refine the algorithms behind black hole simulations, aiming to model one of the universe’s most difficult environments with greater accuracy.
Chi-kwan Chan, a researcher at Steward Observatory, studies the hot plasma that swirls around black holes. In work tied to the OpenAI tool, he is testing mathematical approaches that could make it possible to simulate the motion of electrons and ions without requiring computers to track every tiny particle spiral in real time.
Black holes remain central to efforts to test Einstein’s general theory of relativity, which describes gravity as the bending of space and time by mass and energy. But the physics around them is so extreme, and the computing demands so high, that existing models still leave important gaps. Chan said current methods and available computing power limit how realistic these simulations can be.
The challenge is the plasma that surrounds many black holes. In denser regions, scientists often model plasma as a fluid, using standard equations that approximate how matter moves. That approach works when particles collide frequently.
Near supermassive black holes, however, the plasma can become so hot and diffuse that electrons and ions barely collide at all. Instead, they move in tight loops around magnetic field lines. Capturing that behavior requires following huge numbers of particles as they rapidly spiral, which forces simulations to use very small time steps.
That level of detail can consume enormous computing resources. Chan said the result is that even powerful supercomputers may spend most of their time calculating the smallest particle motions rather than the large-scale behavior researchers actually want to study.
Chan said he turned to Codex because exploring new mathematical methods by hand would have taken too long. The goal was to generate candidate algorithms and then test them against known solutions.
Not every suggestion produced by the system was correct, according to Chan, but that was part of the process. He said scientific work involves many failed ideas, and the value of the tool lies in helping researchers search through more possibilities more quickly.
The approach differs from AI systems that produce answers without revealing their reasoning. Chan said his team uses Codex to propose and implement numerical schemes that can be inspected, checked, and understood physically. That transparency matters in a field where results must be reproducible and grounded in known physics.
Chan is also part of the international Event Horizon Telescope collaboration, which released the first image of a black hole in 2019. The group is now collecting data aimed at producing the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy.
He helped develop some of the simulation and computing tools used to interpret the EHT observations, and he said the effort is continuing as the collaboration improves its instruments and moves from still images toward motion studies.
If the new algorithms Chan is testing prove successful, they could eventually let scientists simulate trillions of particles around black holes. That would open the door to studying plasma behavior and other extreme phenomena that have been out of reach for decades.
Chan said he sees AI as a way to speed up scientific discovery, but only when its output is tested carefully. In his view, the standard for acceptance remains unchanged. Ideas are only useful if repeated testing shows they work.