Neural network to analyze rising magnetic flux ropes in M-dwarf stellar convective dynamo simulations

Despite the central role that rising magnetic flux ropes are thought to play in the formation of active regions on the surfaces of the Sun and other stars, global MHD simulations of stellar interiors have struggled to self-consistently capture their dynamics. By their nature, these structures tend to be short-lived or wholly absent in all but the most turbulent high-resolution convective dynamo simulations — environments which make their hand-identification and study prohibitively time-consuming. We present here an analysis of hundreds of self-consistent, rising flux ropes identified autonomously in global simulations of M-dwarf convective dynamos with the help of a novel machine-learning pipeline developed for the task. We report on the details of the neural network models employed, as well as the variation of flux rope formation rates, locations, and trajectories across differing stellar parameters.