CHGNet: A Pre-trained Universal Interatomic Potential for Electron-Coupled Ionic Systems

Explore a comprehensive lecture on CHGNet, a groundbreaking pre-trained universal interatomic potential for studying electron coupled ionic systems. Delve into the challenges of large-scale simulations involving complex electron interactions and discover how CHGNet addresses these issues. Learn about the Crystal Hamiltonian Graph Neural Network (CHGNet), a graph-neural-network-based machine-learning interatomic potential that models the universal potential energy surface. Understand how CHGNet is pretrained on data from the Materials Project Trajectory Dataset, incorporating energies, forces, stresses, and magnetic moments from approximately 1.5 million inorganic structures. Examine the potential of CHGNet to accurately represent electron orbital occupancy and its enhanced capability to describe both atomic and electronic degrees of freedom. Explore various applications of CHGNet in solid-state materials and energy storage, demonstrating its practical significance in the field of atomistic modeling.

B. Deng: CHGNet, pre-trained universal interatomic potential to study electron coupled ionic systems