Colloquium by Jens Mahlmann and Viraj Pandya, Columbia
Extreme plasma processes around compact objects
Jens Mahlmann
In this talk, I will discuss how stars with extremely strong magnetic fields generate some of the most powerful events in space—including X-ray events and fast radio bursts (FRBs). Through advanced computer simulations, we model magnetic energy dissipation and radiative reactions in the extreme environments of magnetars and binary neutron stars with unprecedented realism. This approach helps us identify key system characteristics for the first time, including magnetic field structures and first principles photon spectra that directly connect to X-ray observations.
Decoding the Complexity of Galaxy Formation with Precision Astrophysics
Viraj Pandya
Galaxies are complex dynamical systems evolving against the backdrop of cosmology. One of the grand challenges of modern astrophysics is to build a fully predictive theory for galaxy formation so that we can reliably use them to understand the fundamental physics of dark matter, dark energy and inflation. However, the relevant physical processes are poorly understood and non-linearly coupled over an enormous range of spatiotemporal scales. This, combined with our inability to observe the time evolution of individual systems, necessitates phenomenological approaches. I will describe my interdisciplinary efforts to enable transformative cosmology-style precision astrophysics for galaxy formation. First, I will introduce a new model in which supernovae and black holes over-pressurize galactic atmospheres, which naturally explains why galaxies are so inefficient at forming stars. Then, for the first time in the ~50 year history of dynamical modeling of galaxy populations, I will present an exact, rapid sensitivity analysis using automatic differentiation to identify the key astrophysical parameters responsible for galaxy self-regulation. With gradients and massive GPU parallelization in hand, I will highlight some promising early applications of gradient descent optimization, Hamiltonian Monte Carlo and population-level implicit likelihood inference. These reveal that observations of the circumgalactic medium may hold the key to breaking numerous modeling degeneracies. I will conclude with my vision for how next-generation physics-informed, data-driven, AI-accelerated, Bayesian galaxy formation models can help us optimally interpret expensive upcoming galaxy surveys with, e.g., Rubin, Roman, Euclid, SDSS-V and NASA's future Great Observatories.
Cookies will be available, starting at 3:45.
Host: Mary Putman