Fig. 8: Schematic of the steps in SGMA.
From: Competing nucleation pathways in nanocrystal formation
A database is created from (a) atomistic trajectories obtained using Molecular Dynamics. b The neighbors of each atom are found using Voronoi tessellation, c as well as the neighbors of only the same element. d The Steinhardt parameters are then computed for all atoms in each snapshot of the trajectory. A Gaussian Mixture Model is then trained on the database. e The parameters of the Gaussian clusters are initialized using the K-means algorithm and are then optimized using the Expectation-Maximization algorithm (f). Steps (e, f) are performed 100 times and the parameters with the best results are kept. g Classification is then performed on the structures in the database or on new test structures.
