NAC 2018 – extra material
Above is a link to the poster I present at NAC “Speeding Up Simulations of Black Hole- Neutron Star mergers (and other rare events)”
For my master thesis I developed a variance reduction method, Adaptive Importance Sampling, that can reduce the computational costs of simulations by two orders of magnitude.
Below is a XKCD styled plot that shows the computational costs as a function of the performance (uncertainty) of the Adaptive Importance Sampling method
Uncertainty quantification methods can significantly enhance stellar evolution simulations. Binary population synthesis models are a versatile tool in astrophysics to compare theory and observations. They include a large variety of underlying binary interaction processes that are challenging to model and induce uncertainties in the outcome of the model. Hence, to fully understand the simulations outcome and subsequently the underlying physics, it is important to incorporate uncertainty from the beginning of the model instead of as an afterthought, to reduce its computational cost. This can be achieved by using methods from uncertainty quantification, which is a field in mathematics that tries to investigate, understand, and quantify uncertainties in models.