Computational generation of porous material structures for neural net based discovery

Abstract

We present a novel computational approach using the artificial neural networks (ANNs) that can generate the hypothetical adsorption properties. For the learning of ANNs, the molecular simulation screened more than 330,000 zeolite structures. In addition, we developed the structure prediction algorithm that is working on the energy grid space. We reproduced the eleven zeolite structures when the experimentally synthesized structures are used as the input.