Automatic Structure Analysis in High-Throughput Characterization of Porous Materials

Inspection of the structure and the void space of a porous material is a critical step in most computational studies involving guest molecules. Some sections of the void space, like inaccessible pockets, have to be identified and blocked in molecular simulations. These pockets are typically detected by visual analysis of the geometry, potential or free energy landscapes, or a histogram of an initial molecular simulation. Such visual analysis is time-consuming and inhibits characterization of large sets of materials required in studies focused on identification of the best materials for a given application. We present an automatic approach that bypasses manual visual analysis of this kind, thereby enabling execution of molecular simulations in an unsupervised, high-throughput manner. In our approach, we used a partial differential equations-based front propagation technique to segment out channels and inaccessible pockets of a periodic unit cell of a material. We cast the problem as a path planning problem in 3D space representing a periodic fragment of porous material, and solve the resulting Eikonal equation by using Fast Marching Methods. One attractive feature of this approach is that the to-be-analyzed data can be of varying types, including, for example, a 3D grid representing the distance to the material's surface, the potential or free energy of a molecule inside the material, or even a histogram (a set of snapshots) from a molecular simulation showing areas which were visited by the molecule during the simulation.