Machine-learned potentials (MLPs) have become widely adopted alternatives to traditional electronic structure and molecular mechanics methods; however, most MLPs remain prone to instability when deployed in molecular dynamics simulations, particularly at elevated temperatures. Here, the authors present physics-informed Gaussian process (GP)-based atomic energy models that achieve enhanced stability in NVT simulations at temperatures as high as 1000 K, demonstrating their robustness in simulations of flexible organic molecules (peptide-capped glycine and serine, malondialdehyde, and aspirin) with a cumulative simulation time of 0.5 microseconds completed within two CPU days.
- Bienfait Kabuyaya Isamura
- Olivia Aten
- Paul Lode Albert Popelier