Equivariant Networks for Porous Crystalline Materials

Porous crystalline materials have the potential to play a key role in developing solutions for molecular storage, gas separation and carbon adsorption. For these solutions, we need to develop new materials with specific properties. Estimating the properties of such porous materials involves first principle simulation using classical molecular simulations. The computational complexity of these methods can be a barrier to high throughput screening of the potential materials as the space of possible materials is vast. Data-driven methods, specifically machine learning methods based on deep neural networks offer a significant opportunity to significantly scale the simulation of the behavior of these materials. However, to effectively achieve this the Deep Learning models need to utilize the symmetries present in the crystals. Crystals pose specific symmetries that are present in their space group. Existing methods for crystal property prediction either have symmetry constraints that are too restrictive or only incorporate symmetries between unit cells. In addition, these models do not explicitly model the porous structure of the crystal. In this paper, we develop a model which incorporates the symmetries of the unit cell of a crystal in its architecture and explicitly models the porous structure. We evaluate our model by predicting the heat of adsorption of CO$_2$ for different configurations of the Mordenite and ZSM-5 zeolites. Our results confirm that our method performs better than existing methods for crystal property prediction and that the inclusion of pores results in a more efficient model.