Machine-learned Force Fields
Main content start
Shaswat Mohanty, Sang Hyuk Yoo, Hyoung Ryul Park
Machine-learned force fields or interatomic potentials (MLFF or MLIP) are candidate solutions to executing ab-initio accuracy molecular dynamics simulations while being computationally inexpensive in comparison to traditional ab-initio methods.
We develop a testbed for benchmarking MLFFs which is available as TB-MLFF. We present a series of benchmarking tests for the MLFFs in thesolid phase and liquid phase to test the generalizability of the trained model which may vary in performance depending on the chemistry of the system and the interaction potential of the system that comprises the training data.
Publications
- Shaswat Mohanty, SangHyuk Yoo, Keonwook Kang and Wei Cai, "Evaluating the Transferability of Machine-Learned Force Fields for Material Property Modeling", Computer Physics Communications, 288, 108723 (2023). [arXiv] [Scholar]