Series
IAU G5 talk: Accelerating Computational Modeling via Neural Networks: Application to Exoplanet Atmospheric Retrieval/The first magnetic Helium-sdOs: which mergers are magnetic?

Michael Himes, Dr. Matti Dorsch

Abstract

Accelerating Computational Modeling via Neural Networks: Application to Exoplanet Atmospheric Retrieval

IIn physics and astronomy, computationally expensive forward models are often an integral part of preparing experiments/observations, analyzing data, and/or planning future instrumentation/telescopes.  In many of these cases, machine learning (ML) models, such as neural networks (NNs), can offer a significant reduction in compute time with minimal loss in accuracy.  We demonstrate this approach on the problem of exoplanet atmospheric retrieval, which involves on the order of 10^5 -- 10^6 radiative transfer (RT) model evaluations.  We find that the ML RT approach yields the same scientific conclusions as the traditional method, while requiring ~1000x less compute cost for typical setups.  We present our open-source software packages that implement this technique, and we discuss broader applications of this NN surrogate modeling approach.

 

The first magnetic Helium-sdOs: which mergers are magnetic?

Magnetic fields play an important role throughout stellar evolution, and among white dwarfs, the end stage of 95% of all stars, the fraction of strongly magnetic systems is larger than 20%. The origins of magnetic white dwarfs are still under discussion, but it is likely that a significant fraction of them are formed by stellar mergers.

Several types of merger remnants are thought to ignite helium fusion, such as the merger of a helium-WD (He-WD) with a second He-WD, a He/C/O hybrid WD, or a low-mass main sequence star, thus forming a hot subdwarf star. The majority of hot subdwarf stars are helium burning stars with very thin or no hydrogen envelopes. In particular, most of the hot and helium-poor He-sdO stars are thought to be formed by mergers. However, out of hundreds of hot subdwarfs studied over several decades, none showed detectable magnetic fields.
This changed recently, when four almost identical magnetic He-sdO stars were discovered, with mean field strengths between 300 and 500kG.

Why are these stars magnetic while vast majority of other He-sdOs are non-magnetic? This question is still open. In this talk I will give a short introduction to He-sdO stars and their formation and then try to highlight the differences between the four magnetic stars and their non-magnetic cousins.

About the talk

IAU G5 talk: Accelerating Computational Modeling via Neural Networks: Application to Exoplanet Atmospheric Retrieval/The first magnetic Helium-sdOs: which mergers are magnetic?
Michael Himes
Univ. of Central Florida, USA
Dr. Matti Dorsch
Univ. of Central Florida, USA
Tuesday May 24, 2022 - 17:00 GMT+1  (Online)
en     en
iCalendar Google Calendar