Magnetic reconnection is a fundamental process in plasmas that enables magnetic energy to be rapidly converted into particle energy. It governs dynamics across scales, from Earth’s magnetosphere to astrophysical systems and laboratory plasmas, yet many aspects of its structure and energetics remain poorly constrained. This thesis addresses five outstanding problems: how to identify key spatial regions of reconnection, how energy is partitioned across these regions, what particle velocity distribution functions (VDFs) reveal about local kinetic physics, whether consistent patterns emerge across events, and what these findings imply for reconnection in other environments.
Using statistical surveys of Magnetospheric Multiscale (MMS) data, established trends in energy conversion are confirmed, with ion enthalpy flux dominating in outflows and Poynting flux concentrated in separatrices, while highlighting the limitations of proxy-based classification. To overcome these, scalable machine learning methods are used: k-means clustering to identify inflows, outflows, and separatrices in particle-in-cell simulations, and recurrent neural networks to map these regions onto MMS time series. This provides reproducible classifications and enables direct simulation to observation comparison.
Particle energisation is further investigated by applying density-based clustering to ion VDFs, revealing multiple populations whose contributions substantially increase bulk kinetic energy density, particularly in outflows. This demonstrates the importance of population-resolved analysis for understanding energisation pathways. Simulation studies show that guide fields alter energy transport by enhancing Poynting flux while reducing cross-tail particle fluxes, with implications for solar and astrophysical plasmas. Finally, the extension of these results to planetary magnetospheres, the solar corona, relativistic astrophysical environments, and laboratory plasmas is discussed, where reconnection plays a key role in energy release and transport.
Overall, this thesis demonstrates that combining spacecraft observations, simulations, and machine learning provides new insight into the spatial structure, energy partition, and kinetic physics of magnetic reconnection, and establishes scalable methods applicable across plasma environments.