Machine learner interatomic potentials for 2D systems

Studying nanostructures at the experimental scale often demands extremely large computational cells, encompassing thousands of atoms. This renders quantum-mechanical methods, such as density functional theory (DFT), impractical. Machine learning provides a solution to this challenge. By training interatomic potentials against DFT data, we can conduct simulations of potential energy surfaces and corresponding structures much more swiftly than with DFT alone, while maintaining the accuracy inherent in the quantum-mechanical approach. Our focus lies particularly on employing deep learning to train interatomic potentials for tasks, such as optimization, molecular dynamics, and phonon studies.

Publications:

R. Kempt, A. Kuc, T. Brumme, T. Heine, Edge Conductivity in PtSe2 Nanostructures, Small Structures 5, 2300222 (2024), DOI: https://doi.org/10.1002/sstr.202300222.