C59

Presentation Title: A New Disambiguation Method Based on Physics-Informed Machine Learning
Author(s): Kai E. Yang, Lucas A. Tarr, Matthias Rempel, Curt Dodds, Sarah A. Jaeggli, Peter Sadowski, Thomas A. Schad, Xudong Sun, Ian Cunnyngham, Jiayi Liu, Yannik Glaser

Abstract:

The spectropolarimetric inversion of vector magnetic fields faces a well-known problem of 180-degree ambiguity in the azimuth angle. Traditionally, this challenge is tackled by minimizing a target function, such as the divergence of the magnetic field within a single layer. We introduce a novel approach using Physics-Informed Machine Learning to solve the divergence-free equation of the magnetic field across multiple optical depth layers, while simultaneously estimating the geometric heights of these layers. Our approach has been tested on magnetohydrodynamic simulations of both quiet-Sun and plage regions, effectively recovering the horizontal field orientation on pixels with strong magnetic fields. By integrating the vector magnetic field with the estimated geometric heights, our method can reconstruct the three-dimensional vector electric currents in the solar photosphere.