Data Mining Molecular Dynamics’ Trajectories

Cationic diffusion in solid state ionics can be simulated by molecular dynamics (MD). The dump output from a MD simulation easily leads to several GBs of data, whereas the useful information extracted is only the diffusion coefficient, obtained from either the mean-squared displacement or the Green-Kubo formula. A large portion of the data is therefore discarded.

In oxygen conductors, ofter time oxygen trajectories are characterized by hopping from a site to another. In our work, we study such transport of charged defects by machine learning. Specifically, we cluster the cationic trajectories to sites computed using data-mining algorithm. This approach allows us to reduce the dimensionality of the MD data and to determine important quantities such as site-specific residence times and occupancies. Our data-mining approach coupled to statistical analysis clarifies the role of transport and the link to the local cationic environment and atomic arrangement.


[1] C. Chen, D. Chen, and F. Ciucci. A Molecular Dynamics Study of Oxygen Ion Diffusion in A-site Ordered Perovskite PrBaCo2O5.5: Data Mining the Oxygen Trajectories. Physical Chemistry Chemical Physics, 17, 12, 7831-7837 (2015) link[2] C. Chen, Z.M. Baiyee, and F. Ciucci. Unraveling the Role of La A-site Substitution on Oxygen Ion Diffusion and Oxygen Catalysis in Perovskite BaFeO3by Data-driven Molecular Dynamics and Density Functional Theory. Physical Chemistry Chemical Physics, 17, 24011-24019 (2015) link