Graphene is a 2D carbon material that is impermeable to all gases. By engineering pores into graphene, we can adjust its transport properties. But current methods for synthesizing graphene pores can create unintentional vacancies that have the potential to reconstruct themselves. By predicting how these porous structures will rearrange to seek the lowest energy state, we can better predict the material’'s transport properties. Density functional theory (DFT) presents an accurate means of searching for these minimum energy structures, but it is too computationally expensive to implement for large structures. We have developed a neural network potential (NNP) to calculate these energies with accuracy similar to DFT, but greatly increased computation times.

Additional information available at http://devonwa.com/2016/10/21/poster-chegsa-symposium/