Magnetic Field topology reconstruction in a 3-D simulation box using Gaussian Process Regression

Abstract:

Unlike the vast majority of astrophysical plasmas, the solar wind is accessible to spacecraft, which for decades have carried in-situ instruments for directly measuring its particles and fields. Though such single-spacecraft measurements provide precise and detailed information, one such spacecraft on its own can neither disentangle spatial and temporal fluctuations nor fully reveal the plasma's 3-D structure. To address this, a few missions have flown with 4 or 5 spacecraft (e.g., Cluster, THEMIS-ARTEMIS, and MMS), and missions with even more spacecraft have been proposed. However, none of the missions have succeeded in generating a full three dimensional image of the magnetic vector field in the solar wind, mostly because of insufficient number of spacecraft. A full 3-D image would provide the information related to structure and topology which are extremely important for understanding turbulence and its evolution in space plasma specially how energy is stored in and transported through the plasma. Though an active field of research, not much has been done in this from the vantage point of machine learning. In this study we present a proof of concept of magnetic field's topology reconstruction using multi-point observation in a 3-D simulation box. For multi-point observation, we fly a constellation of virtual spacecraft through the simulation box, and carry out the interpolation on observed vector data in the 3-D space along its trajectory using machine learning algorithms with emphasis on Gaussian Process Regression. The study also explores number of spacecraft, relative separation between them and their configuration required for resolving structures of various scale.