Skip to main content

Reconstructing the physics of transients using machine learning

Supernova explosions are some of the most energetic and luminous events in the universe, andunderstanding them is crucial for many areas of astrophysics. One way to gain insight into the physicalprocesses involved in these explosions is through supernova tomography, which involves reconstructinga spatially resolved explosion model using a spectral time series. However, this requires a radiativetransfer model and is computationally intractable with traditional means, requiring millions of MCMCsamples.A new solution to this problem is the use of surrogate models or emulators, which employ machinelearning techniques to accelerate simulations. In this talk, we present a new emulator for the radiativetransfer code TARDIS that outperforms existing emulators and provides uncertainties in its prediction.This offers the foundation for a future active‐learning‐based machinery that will be able to emulate veryhigh dimensional spaces of hundreds of parameters crucial for unraveling urgent questions insupernovae and related fields. Our work provides a promising avenue for understanding the physicalprocesses involved in supernova explosions and their progenitors, with implications for a wide range ofastrophysical phenomena.

Cody Hall

Wolfgang Kerzendorf, Michigan State University

October 25, 2023
2:00pm - 3:00pm