Presentation Information
[10p-S101-2]Development of r2SCAN level machine learning interatomic potential and its applications
〇So Takamoto1 (1.PFN)
Keywords:
molecular dynamics,machine learning interatomic potential,universal interatomic potential
We have started constructing a dataset based on the meta-GGA level r2SCAN to further improve the accuracy of the universal machine learning potential PFP. By simultaneously utilizing the existing large-scale PBE dataset and the high-accuracy r2SCAN dataset for training, we achieved a balance between reproducing a wide range of energy landscapes and maintaining high precision. The resulting model successfully reproduced properties such as crystal formation energy, melting point, and liquid density.