(599e) Combined Deep Learning and Classical Potential Approach for Modeling Diffusion in Uio-66

Modeling the diffusion of adsorbates through porous materials using atomistic molecular dynamics (MD) can be challenging when the flexibility of the adsorbent needs to be included. The development of potentials that accurately account for the motion of the adsorbent in response to the presence of adsorbate molecules is necessary. This study proposes a hybrid potential approach that utilizes accurate machine learning atomistic potentials for metal-organic frameworks in combination with classical force fields, like Lennard-Jones (LJ) potentials, for adsorbates to accurately compute diffusivities. We demonstrate this approach by developing an accurate deep learning potential (DP) for UiO-66, a metal-organic framework, and performing hybrid potential simulations that model the diffusion of neon and xenon through the crystal. The hybrid potential approach allows for adsorbent-adsorbate interactions with classical potentials while modeling the response of the adsorbent to the presence of the adsorbate through near-DFT accuracy DPs. The approach does not require refitting the DP for new adsorbates. The study shows that the DP/LJ results are in excellent agreement with DFT-MD and can be applied to other MOFs and adsorbates, making it possible to model adsorption and diffusion within the porous material, including adsorbate-induced changes to the framework.