SEDONA-GesaRaT: an AI-Accelerated Radiative Transfer Program for 3-D Supernova Simulations

When a massive star dies in a violent explosion, it becomes a supernova, a cosmic event that outshines entire galaxies for a brief moment. To understand what happens inside these explosions, astronomers rely on computer simulations that track how light travels through the expanding cloud of debris. This process, known as radiative transfer, is like following a beam of light as it bounces off, gets absorbed by, or passes through billions of atoms. The challenge is that the atoms inside a supernova are not in a calm, stable state; they are constantly being excited and ionized by intense radiation, a condition scientists call non-local thermodynamic equilibrium. Calculating the exact state of these atoms for every point in a three-dimensional explosion is so computationally heavy that, until now, researchers have been forced to simplify their models, often ignoring the complex 3D structure or the detailed atomic physics to get a result in a reasonable amount of time.

A team of researchers has now broken through this bottleneck with a new computer program called SEDONA-GesaRaT. This software combines a traditional method for simulating light with a powerful artificial intelligence tool designed to solve the difficult atomic physics problems instantly. By training this AI on 119 previous simulations of Type Ia supernovae—a specific kind of stellar explosion that serves as a standard candle for measuring cosmic distances—the team created a system that can predict the behavior of atoms without performing the slow, step-by-step calculations required by older methods. The result is a program that can simulate the full, complex three-dimensional structure of a supernova explosion, including the detailed physics of how light interacts with matter, in a fraction of the time it used to take.

The researchers tested this new tool using a detailed model of a Type Ia supernova known as N100. They ran simulations to see how light would travel through the explosion and how it would appear to an observer from different angles. The program successfully generated a massive dataset showing not just the brightness of the light, but also its polarization, which describes the orientation of the light waves. This polarization data acts like a fingerprint of the explosion's internal shape, revealing whether the debris is smooth or clumpy. The team found that their AI-accelerated program produced results that matched the older, slower methods very closely for the overall brightness and color of the light. However, they did notice specific discrepancies where the program overestimated the opacity of certain chemical elements, specifically the silicon and calcium spectral lines. These minor errors suggest that while the AI is incredibly fast and generally accurate, it still needs refinement to perfectly capture every nuance of the atomic interactions.

What makes this achievement particularly significant is the sheer scale of the simulation. A single run of this three-dimensional, high-precision model took about 3,000 hours of computing time on a standard processor. While this sounds like a lot, previous methods would have required resources far beyond what is currently available to achieve the same level of detail and accuracy. In fact, older codes could only manage to simulate a simplified, one-dimensional version of the explosion with this same level of physical detail, or a full three-dimensional version with much simpler physics. The new program allows scientists to see the explosion as a complex, lumpy, three-dimensional object rather than a smooth sphere. For instance, the simulations revealed how the density of calcium in the explosion correlates with the polarization of light, showing that the light's orientation changes depending on where the dense gas is located within the expanding shell.

The ability to run these detailed simulations opens the door to a new era of understanding supernovae. Because the program is so efficient, researchers can now systematically study the internal structures of these explosions, comparing different models to see which ones best match the light we actually observe from Earth. This is crucial for interpreting data from telescopes that capture the polarization of supernova light, which holds clues about the explosion's geometry and the distribution of elements. The researchers noted that while their current tool is trained specifically on Type Ia supernovae, the same approach could eventually be applied to other types of stellar explosions, provided the AI is trained on the specific chemical ingredients found in those events. By making the impossible computationally feasible, SEDONA-GesaRaT transforms how astronomers can probe the violent, intricate hearts of dying stars.