Astrophysicist Uses OpenAI's Codex to Model Plasma Around Black Holes

Black Hole Simulation Accelerated by AI-Assisted Code Generation

According to the OpenAI Blog, Chi-kwan Chan, an astrophysicist at the University of Arizona and Steward Observatory, is using Codex, OpenAI’s code-generation model, to overcome a critical computational bottleneck in black hole physics research. The challenge: simulating the behavior of plasma around supermassive black holes requires tracking the motion of trillions of individual charged particles—a calculation so demanding that current algorithms and hardware limit how realistic simulations can become. Codex is helping Chan and his collaborators write the specialized numerical code needed to model this extreme physics, accelerating a research program that feeds into real observational astronomy.

The Event Horizon Telescope’s Push Toward Black Hole Video

Chan is part of the Event Horizon Telescope (EHT) collaboration, which published humanity’s first image of a black hole in 2019. The team is now pursuing an ambitious next step: generating the first video of a supermassive black hole, focusing on the one at the center of the M87 galaxy. According to the OpenAI Blog, Chan helped develop the simulation and computing infrastructure that the EHT used to interpret that 2019 image, which showed a black hole’s shadow silhouetted against glowing plasma near the event horizon. Moving from static images to video requires orders of magnitude more data processing and more sophisticated modeling of how matter behaves under the universe’s most extreme conditions.

The Plasma Modeling Problem

The core obstruction is plasma—superheated matter composed of electrons and ions. In many regions of space, scientists can approximate plasma as a fluid and apply well-established equations to predict its motion. But near supermassive black holes, conditions diverge from this assumption. According to the OpenAI Blog, the plasma becomes so hot and diffuse that particles barely interact with each other; instead of colliding randomly as in denser plasmas, electrons and ions spiral around magnetic field lines. Accurately modeling this regime demands calculating every rapid corkscrew motion of trillions of particles in tandem—a computational task that standard simulation frameworks struggle to handle without either sacrificing realism or consuming prohibitive amounts of processing time.

Why This Matters

For observational astrophysics, the bottleneck is no longer telescopes or observation time but the ability to convert raw data into testable predictions. By reducing the engineering friction of writing complex particle-tracking code, Codex enables Chan’s team to iterate on simulation algorithms faster, refine comparisons between theory and the EHT’s observations, and potentially unlock new insights into Einstein’s general relativity in regimes inaccessible to Earth-bound experiments. If this pattern holds—where code generation models accelerate domain-specific simulation work—it suggests a broader frontier for AI-assisted scientific computing, particularly in high-energy physics, climate modeling, and molecular dynamics, where the computational bottleneck has been developer velocity rather than raw hardware throughput.