Jennifer Rosina Andersson - Half-time seminar: "Bayesian uncertainty quantification in high-dimensional stellar magnetic field models"

Discussion leader: Christian Glaser

Abstract

Spectropolarimetric inversion techniques, known as Zeeman Doppler Imaging (ZDI), have become the standard tools for reconstructing the surface magnetic field topology of distant stars. However, accurate and efficient uncertainty quantification of such magnetic field maps is an open problem in current research. The computational intensity of the forward models and the high dimensionality of the magnetic field parameterization makes inference inherently difficult. In this half-time seminar, we present a probabilistic ZDI framework for stellar magnetic field reconstruction, and demonstrate the proposed framework on the bright, massive star Tau Sco using several statistical models. We show that fast and stable convergence is possible using gradient-based MCMC methods, and demonstrate that for a large group of stars, including solar-like stars, closed-form solutions for the posterior magnetic field distribution exist under natural model assumptions. We conclude the talk with a discussion on future work in the direction of differentiable stochastic programs.