2024 AIChE Annual Meeting
(169bv) Systematic Development of Machine Learning Interatomic Potentials for MAPbI3 Perovskites
Authors
Joshi, N. - Presenter, University of Washington
Intan, N., Pacific Northwest National Laboratory
Zorman, M., University of Washington
Pfaendtner, J., University of Washington
Obtaining potential energy surface (PES) of a molecular system has been shown to provide valuable insights of molecules, understanding their conformational changes and predicting their properties. Balancing the accuracy vs computational cost trade-off between first principle calculations and empirical force fields has been a major limitation to studying accurately large systems and complex phenomena with significant activation barriers. Machine learning potentials (MLP) have proved to bridge this gap and provide high accuracy from its first principle data at a fraction of the computational cost1. However, MLP simulations tend to encounter unexplored trajectories where its reliability and robustness is low. To overcome this challenge, the workaround surrounds gathering more training data with methods such as active learning2. In this study we propose a protocol for systematic development of MLP and apply them for MAPbI3 perovskite systems. We demonstrate training protocols of MLP for aqueous systems and show their generalizability for interfacial and MAPbI3 perovskites systems. We show using the development protocols how MLP can be used to understand the degradation mechanisms of perovskites in liquid water. Furthermore, we envision its applications for different surfaces to explore the extent to which they can be used for systems of varied size and different thermodynamic ranges.
References:
- Unke OT, Chmiela S, Sauceda HE, et al. Machine Learning Force Fields. Chem Rev. Published online March 11, 2021. doi:10.1021/acs.chemrev.0c01111
- Zhang, L., Lin, D.-Y., Wang, H., Car, R. & E, W. Active learning of uniformly accurate interatomic potentials for materials simulation. Phys Rev Mater 3, 023804 (2019).