Towards universal neural network interatomic potential
So Takamoto, Daisuke Okanohara, Qing‐Jie Li, Ju Li
Abstract
• Universal interatomic potential for modeling chemically and structurally complex systems. • Tensorial message passing in equivariant graph convolutional network. • Iterative construction of a diverse and broad training dataset. • User-friendly atomistic simulation platform Matlantis™ powered by universal interatomic potential for 72 elements.
Topics & Concepts
Interatomic potentialGraphComputer scienceConvolutional neural networkTheoretical computer scienceComputational scienceStatistical physicsMolecular dynamicsArtificial intelligencePhysicsQuantum mechanicsMachine Learning in Materials ScienceX-ray Diffraction in CrystallographyElectron and X-Ray Spectroscopy Techniques