OpenAI has published a case study on how Chi-kwan Chan, an astrophysicist at the University of Arizona, is using the AI coding tool Codex in black hole research[1]. Chan uses Codex to derive and test algorithms that simulate the movement of plasma around black holes, in an attempt to break through computational limits that have constrained researchers for decades.
A Researcher on the Team That Captured the First Black Hole Image
Chan is a researcher at the University of Arizona and Steward Observatory, and a member of the international Event Horizon Telescope (EHT) collaboration[1]. The EHT is known for publishing the first image of a black hole in 2019, and the team is currently gathering observations toward the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy[1].
The gravity around a black hole is so extreme that not even light can escape once it gets close enough. What researchers observe is therefore the region around a boundary called the event horizon. Chan describes it as "a surface of no return"[1]. The light emitted by matter swirling just outside this boundary is what astrophysicists can see, measure, and simulate.
The famous 2019 image also 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 those observations[1].
Spiraling Plasma: A Wall That Has Stood for Decades
The biggest roadblock for the team is modeling the plasma around black holes. Plasma is superheated matter made up of electrically charged electrons and ions. In many simulations, scientists approximate plasma as a fluid and model its movement with well-known equations. In denser plasma where particles constantly collide, this approximation works reasonably well[1].
However, around the supermassive black holes that Chan and his colleagues study, some regions become so hot and diffuse that particles rarely encounter each other. There, particles mostly spiral around magnetic field lines, and modeling that behavior correctly requires following trillions of electrons and ions through every rapid corkscrew turn[1].
Standard simulations must calculate each of these tiny turns, forcing computers to take extremely small timesteps. As a result, even the world's fastest supercomputers spend most of their time calculating minuscule particle motions rather than the larger phenomena scientists actually want to study. "For decades, this has limited how realistically we can simulate black hole plasma," Chan says[1].
Using Codex to Mass-Produce and Test Candidate Algorithms
Chan suspected that changing, mathematically, how the simulation tracks particle motion could free the computer from following every tiny spiral directly. But exploring all the mathematical possibilities by hand would have taken an enormous amount of time. This is where Codex comes in: he has it derive candidate algorithms and tests them against known solutions[1].
Not every approach Codex generates is correct. Even so, Chan takes a positive view: "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"[1].
Unlike AI systems that return only conclusions, Chan's group uses Codex to propose and implement numerical schemes that humans can inspect, test, and understand physically. "We don't accept an idea because it came from Einstein, from a bright student, or from an AI model. We accept it only after repeated testing," Chan says[1].
Summary
Chan, an EHT astrophysicist, is using Codex to search for new algorithms for simulating the plasma around black holes[1]. Rapidly producing testable ideas is a good example of how well AI fits scientific research, with its culture of rigorous verification. If this approach succeeds, simulations handling trillions of particles could become possible, bringing physics that has remained out of reach for decades within grasp.
Source: https://openai.com/index/using-codex-to-simulate-black-holes
