MIST
Magnetosphere, Ionosphere and Solar-Terrestrial
Classifying Magnetic Reconnection Regions Using k-Means Clustering: Applications to Energy Partition
By Cara Waters (Imperial College London)
Magnetic reconnection is a universal process which facilitates the repartition of magnetic energy to particle energies. It is important to understand the structure of reconnection regions and how they convert and partition energy. As visual identification of such structures can be subjective and time consuming, we take an unsupervised machine learning approach involving k-means clustering.
Carrying out this clustering on a 2.5-D particle-in-cell simulation of symmetric reconnection comparable to that in Earth’s magnetotail, we identify that the optimal number of clusters is six. We input only field and plasma variables to the clustering, giving a result which is independent of position. We identify two inflow regions, two outflow regions, and two pairs of separatrices. By looking at the distributions of the energy flux densities in these regions, we confirm that outgoing particle energy flux densities from reconnection decrease as guide field increases. The ion enthalpy flux density is the most dominant form of energy flux density in the outflows, agreeing with previous studies, and Poynting flux density may be dominant at some points in the outflows and is only half that of the Poynting flux density in the separatrices. This demonstrates an approach which may be applied to large volumes of data to determine statistically different regions within phenomena in simulations and could be extended to in situ observations, applicable to future multi-point missions.
Results of carrying out k-means clustering with six clusters on comparable simulation runs with (a) BG = 0, (b) BG = 0.1, and (c) BG = 0.2. Magnetic field lines are shown in black, with the colour showing the regions identified by the k-means clustering. These regions are labelled in relation to the equivalent directions in GSM coordinates in the case of magnetotail reconnection. Each simulation run has k-means carried out independently with variables scaled in the same manner and subsequent clusters re-numbered for comparison between each case.
See publication for details:
Cara L. Waters, Jonathan P. Eastwood, Naïs Fargette, David L. Newman, Martin V. Goldman. Classifying Magnetic Reconnection Regions Using k-Means Clustering: Applications to Energy Partition, JGR: Space Physics, 2024, 129, 10. https://doi.org/10.1029/2024JA033010