Deep Learning of Charged Particles Motion in Media with Loss-Induced Topological Transitions of Dispersion Surfaces
摘要
We study the dynamics of charged particles in uniaxial lossless and lossy magnetic media. For lossless case, the dispersion surfaces have spherical topology and the trajectories of the particles show no bifurcations. For the case where loss is present, the dispersion surfaces are hyperboloids and for some values of the particles momenta, bifurcations occur. This is shown by deploying the AI Poincaré algorithm to obtain the corresponding explained variance ratios. The behavior of these explained ratios change at bifurcations. We show this for some values of the permeability tensor components and some different choices of the initial conditions.