(2gh) Exploring the Frontiers of Molecular Diffusion through Machine Learning-Based Forcefields and Electron Density Predictors

Computational chemistry is an effective tool for predicting the diffusion of ions and atoms, which can facilitate the design of advanced materials with targeted transport properties. However, methods such as density functional theory (DFT) are limited to small system sizes and short timescales. To overcome these limitations, machine learning (ML) potentials can be developed to study diffusion in larger systems for longer times. ML potentials are generally trained using data generated from DFT by exploring a large potential energy surface for a given system.

We used a workflow that combines deep learning potentials (DP) and graph lattice models (GLM) to design proton-conducting membranes for fuel cells. Our workflow showed that graphanol, a hydroxylated graphane, can conduct protons anhydrously with very low diffusion barriers. We used this information to develop specific design rules for developing next-generation proton-conducting membranes with even lower diffusion barriers. To account for electron densities and polarization effects, we developed the DeepCDP method for predicting electron densities for infinitely large system sizes at a fraction of the computational cost. We show that electron density predictors build using the DeepCDP method attain high accuracy for charged systems.

We also used ML potentials to model the diffusion of Ne and Xe through UiO-66, a metal-organic framework (MOF), using a hybrid potential approach. This approach combines accurate DPs for MOFs with classical potentials for adsorbates to accurately compute diffusivities. ML potentials were also used to model interface diffusion of chalcogenide alloys and electrodes that are used in next generation non-volatile memory cells. These potentials were constructed using the moment tensor approach (MTP). MTP-based active learning helped in exploring amorphous phases of these materials. This methodology will be used to find advanced chalcogenide alloys – electrode interfaces that are non-diffusive for long MD time.

Overall, our workflow and methods demonstrate the potential for using ML potentials to design advanced materials with targeted transport properties, and to improve our understanding of diffusion in complex systems.