Machine Learning for Atomistic Simulation I: Applications in Materials Science

Artificial intelligence (AI) is increasingly changing the paradigm of scientific discovery to accelerate research and solve real-world scientific challenges. One of the major contributions came from machine learning interatomic potentials (MLIPs) that approximate the potential energy surface (PES) of an atomic system by the position and chemical identities of the atoms in their local environments. MLIPs have enabled the chance to scale atomic-level quantum chemical accuracy to large-scale simulations and demonstrated their success in multiple materials applications.

In this talk, we will discuss the status of current MLIPs, explaining their applicability and limitations in materials modeling. By building essential physics into the model, we show the capability of MLIPs to simulate energy storage materials. By building careful benchmarks, we demonstrate the key limitations for current MLIP models and point out the important next steps for building reliable MLIPs.