Affiliations: [a] School of Electrical Engineering and Computer Science, Louisiana State University, Baton Rouge, LA, USA | [b] Department of Geology and Geophysics, Louisiana State University, Baton Rouge, LA, USA
Correspondence:
[*]
Corresponding author: Shobhit Shakya, School of Electrical Engineering and Computer Science, Louisiana State University, Baton Rouge, LA 70803, USA. Tel.: +1 225 578 1495; E-mail:[email protected]
Abstract: In materials science, micro-level (atomic-scale) activities are considered to be
the key to understanding various macro-level (bulk) properties of a material.
Molecular dynamics simulations are widely used to obtain atom dynamics.
However, researchers are often too overwhelmed by the amount of simulation data
to discover relevant atomic activities.
Furthermore, mining the structure graph of a material (the graph where
the constituent atoms form nodes and the bonds between the atoms form edges)
offers little help in this scenario.
It is the patterns among the atomic dynamics that may reveal the
mechanisms underlying a particular material property.
Discovery of such patterns can lead to a better model and better predictions
of the properties and behaviors of the materials.
We propose an event graph to model the atomic dynamics and
propose a graph mining algorithm to discover popular subgraphs
in the event graph. Because the event graph is a directed acyclic graph,
our mining algorithm uses a new graph encoding scheme that is
based on topological-sorting. This encoding scheme ensures that
our algorithm enumerates candidate subgraphs without any duplication.
Experiments with simulation data of silica liquid demonstrate the effectiveness
of our mining system which we call ``ToPoMine''.
