Astronomers are harnessing the power of deep learning to simulate cosmic explosions that occur during the death of massive stars. By improving their understanding of galactic formation and evolution, these researchers believe that more accurate modeling of challenging-to-simulate aspects of supernovas could provide insight into how the necessary chemical elements for life are dispersed throughout the cosmos.
This breakthrough comes from a team led by Keiya Hirashima, an astronomer from the University of Tokyo, who found inspiration in applying deep learning, a technology that enables computers to recognize patterns across various datasets, to weather forecasting during a Hackathon event.
“Weather may be a complex phenomenon, but ultimately, it boils down to fluid dynamics calculations,” said Hirashima in a statement. “So, I wondered if we could modify deep learning models used for weather forecasting and apply them to another fluid system, one that exists on a vastly larger scale and that we lack direct access to – my field of research, supernova explosions.”
Related: AI just spotted its 1st supernova. Could it replace human explosion hunters?
How supernovas connect generations of stars
Throughout their lives, stars produce chemical elements within their cores through the process of nuclear fusion. This fusion involves the collision of atoms, creating heavier elements while releasing energy that allows the star to shine. The ability to forge heavier elements depends on the mass of the star, but all stellar bodies have their limitations.
When a star can no longer continue fusing heavier elements, nuclear fusion stops, and so does the outward pressure that has protected the star from its own gravitational force for millions or billions of years. The core of the star collapses, causing the outer layers to be ejected in a powerful explosion known as a supernova. This explosion disperses the elements created by the star throughout space.
Over time, this material becomes incorporated into the next generation of stars and the planets that orbit them. Eventually, these “star stuff” elements are part of the building blocks of any life forms that evolve on those worlds, including humans. Therefore, supernovas play a crucial role in unraveling the origin story of humanity.
Supernovas also have a significant influence on the galaxies surrounding them, affecting various aspects of galactic evolution beyond chemical enrichment. Understanding the behavior of supernovas is essential to comprehending how galaxies change over time.
RELATED STORIES:
– Hundreds of supernova remnants remain hidden in our galaxy. These astronomers want to find them
– Supernova photos: Great images of star explosions
– How artificial intelligence is helping us explore the solar system
“The challenge lies in the time it takes to calculate the explosion of supernovas. Many models of galaxies simplify things by assuming that supernovas explode spherically, as this is relatively easy to calculate,” explained Hirashima. “However, in reality, they are highly asymmetric, with certain regions of the explosion’s shell being more complex than others.”
Deep learning enabled the scientists to identify which parts of the explosion required more attention during a simulation and which parts required less. This approach ensured the highest level of accuracy while reducing the overall computational time.
“This technique of dividing the problem is known as Hamiltonian splitting,” added Hirashima. “Our new model, 3D-MIM, can decrease the number of computational steps for calculating 100,000 years of supernova evolution by 99%. Therefore, it can significantly help to overcome computational bottlenecks.”
Although this may seem straightforward, deep learning demands extensive training. To train the system, the team had to run hundreds of simulations, which consumed millions of hours of computer time. Despite the monumental effort, the team hopes that the methodology behind 3D-MIM can be applied to other astrophysical phenomena influencing galactic evolution, including the birth of expansive star-forming regions known as starburst regions.
The team has already applied 3D-MIM to the end stages of stars’ lives and is considering applying the same model to the beginning stages, which would enhance the modeling of stellar birth as well.
The team’s research was published online last month in the journal Monthly Notices of the Royal Astronomical Society.
