Simulating black holes with Codex: an astrophysicist's new research ally

Hey everyone, it's me, Shiichan! Today we're talking about black holes and AI. The deepest mysteries of the universe meeting cutting-edge coding, how exciting is that!

OpenAI News openai.com

What was announced?

OpenAI's News shared a story about an astrophysicist using Codex to tackle black hole simulations.

The star is Chi-kwan Chan, a researcher at the University of Arizona and Steward Observatory who studies black holes with computer simulations and observations. Chan is part of the international Event Horizon Telescope (EHT) collaboration, which published the first-ever image of a black hole back in 2019. Now the team is moving beyond still images and gathering observations to make the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy.

Why it matters

Black holes are one of the best stages for testing Einstein's general theory of relativity. In that view, gravity isn't a force pulling objects together; it's the result of mass and energy bending the fabric of space and time. And black holes let us test that idea under the most extreme conditions.

But since not even light can escape, scientists don't observe the black hole itself. They study the boundary around it called the event horizon. As Chan puts it:

"It's a surface of no return,"

Matter swirling just outside that boundary gives off light that astrophysicists can see, measure, and simulate. The 2019 image captured a black hole's shadow embedded in glowing plasma, and Chan helped build the simulation and computing tools the team used to interpret the observations.

What changes

The biggest wall is figuring out how to model the plasma around black holes. Plasma is superheated matter made of electrically charged electrons and ions.

In dense regions, you can treat plasma like a fluid and use well-known equations. But near the supermassive black holes Chan studies, some regions get so hot and diffuse that particles almost never collide. Instead they spiral around magnetic field lines.

To follow that correctly, you have to compute the tiny corkscrew motions of trillions of electrons and ions, which forces computers into extremely small timesteps. As a result, even the world's fastest supercomputers spend most of their time on these minuscule particle motions instead of the bigger behavior scientists actually want to study. For decades, that has limited how realistically we can simulate black hole plasma.

Dive Deep

Chan wondered whether new mathematical techniques could work around these limits. The idea was to change, mathematically, how the simulation tracks particle motion, so the computer no longer has to follow every tiny spiral directly.

Exploring all those mathematical possibilities by hand would have taken an enormous amount of time. So Chan 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. Most scientific ideas fail; what matters is that these algorithms are testable, and finding one that works could unlock simulations that were previously impossible.

Rather than an AI that just hands back an answer, Chan's group uses Codex to propose and implement numerical schemes they can inspect, test, and understand physically. Large language models still make mistakes, and many scientists stay cautious about using AI in research. Even so, Chan believes science may be one of the best uses for today's AI 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. We accept it only after repeated testing."

If the approaches Chan is testing succeed, the new algorithms could eventually let scientists simulate trillions of particles around black holes, opening up physics that has stayed out of reach for decades.

Wrap-up

• Chan at the University of Arizona is using Codex to search for new algorithms that model the plasma around black holes.
• The hard part is that tracking trillions of particles spiraling along magnetic field lines is far too heavy to compute; the goal is to change the math to get around it.
• Codex proposes lots of candidate algorithms, and the key is that the group can inspect and implement them, unlike an AI that only spits out answers.
• The keyword is verification: an idea gets adopted because it passes tests, not because an AI said so, and I love that scientific mindset.
• This one really lands for anyone who loves space and simulations, and for anyone who wants AI as a research partner!