| Content |
Critical metallicity of red supergiant formation: effects on mass loss and the physical origin
I will present our findings on the critical metallicity of red supergiant (RSG) formation. In our studies of mass loss of massive stars, we perform one-dimensional stellar evolution simulations and build a grid of ~2000 models with different initial masses and metallicities. The results reveal the existence of a critical metallicity Zc at Z~0.001, where the mass loss exhibits a dramatic jump. If Z>Zc, massive stars tend to evolve into RSGs, and a robust cool wind is operational. In contrast, if Z<Zc, massive stars usually remain as blue supergiants (BSGs), wherein the cool wind is not activated and the mass loss is generally weak. The effects of critical metallicity provide implications for the fates of metal-poor stars in the early universe. We further study the physical origin of the critical metallicity through several experiments with stellar models. The results show that changes in nuclear reaction rates and opacities can make a difference in the fate of stars (i.e., evolving toward RSGs or BSGs). I will discuss the factors that affect stellar structures and provide a detailed physical picture of the expansion of post-main-sequence stars toward RSGs.
|
Application of Machine learning on multi-messenger astrophysics
One of the most important missions in the gravitational waves data community is to provide a fast and accurate parameter estimation of an astrophysics event, especially sky-localization, which can help us send alerts to other optical telescopes. The main challenge is achieving it in low latency, usually around a few seconds. This incorporates denoising, detection, and parameter estimation. A systematic machine learning approach is the ideal method to achieve these tasks. In this talk, I will introduce the denoising machine learning model DeepClean, which takes the environment witness channel as an input and predicts the coupled strain noise for removal. I will also introduce the machine learning framework GW-Inference as a service (GW-IaaS) which provides tools from dynamical training to efficient inference. With the ML model and framework together, we demonstrate how the ML denoising pipeline works from end to end.
|
|
