CHGNet: Pretrained universal neural network potential for charge-informed atomistic modeling

Abstract

The simulation of large-scale systems with complex electron interactions remains one of the greatest challenges for the atomistic modeling of materials. Although classical force-fields often fail to describe the coupling between electronic states and ionic rearrangements, the more accurate ab-initio molecular dynamics suffers from computational complexity that prevents long-time and large-scale simulations, which are essential to study many technologically relevant phenomena, such as reactions, ion migrations, phase transformations, and degradation.In this work, we present the Crystal Hamiltonian Graph neural Network (CHGNet) as a novel machine-learning interatomic potential (MLIP), using a graph-neural-network-based force-field to model a universal potential energy surface. CHGNet is pretrained on the energies, forces, stresses, and magnetic moments from the Materials Project Trajectory Dataset, which consists of over 10 years of density functional theory static and relaxation trajectories of ∼ 1.5 million inorganic structures. The explicit inclusion of magnetic moments enables CHGNet to learn and accurately represent the orbital occupancy of electrons, enhancing its capability to describe both atomic and electronic degrees of freedom. We demonstrate several applications of CHGNet in solid-state materials, including charge-informed molecular dynamics in LixMnO2, the finite temperature phase diagram for LixFePO4 and Li diffusion in garnet conductors. We critically analyze the significance of including charge information for capturing appropriate chemistry, and we provide new insights into ionic systems with additional electronic degrees of freedom that can not be observed by previous MLIPs.

Constructing and Evaluating Machine-Learned Interatomic Potentials for Li-Based Disordered Rocksalts

“…Most graph network potentials can be combined with deep learning to provide high accuracy for multicomponent systems, as illustrated by MEGNet 98 and CHGNet. 96 Long range interactions, such as electrostatics and dispersions, have also been included with such graph-based potentials. 99 Note that incorporating deep learning often results in better performance with "large" data sets, such as the Materials Project, 85 and, in turn, requires large computational training time.…”

“…For example, AENET and MTP are parametrizations of radial and angular distributions of atoms around a central atom of interest, while GAP and SNAP employ descriptors to quantify the local density around an atom. There are several graph-network-based (neural net) potentials that have been developed recently, such as SchNet, NequIP, MEGNet, and CHGNet with the work by Reiser et al providing a well-compiled summary of available graph-based MLIPs. Most graph network potentials can be combined with deep learning to provide high accuracy for multicomponent systems, as illustrated by MEGNet and CHGNet .…”

“…There are several graph-network-based (neural net) potentials that have been developed recently, such as SchNet, NequIP, MEGNet, and CHGNet with the work by Reiser et al providing a well-compiled summary of available graph-based MLIPs. Most graph network potentials can be combined with deep learning to provide high accuracy for multicomponent systems, as illustrated by MEGNet and CHGNet . Long range interactions, such as electrostatics and dispersions, have also been included with such graph-based potentials .…”

Ameliorating electrochemical performance of Li-rich Mn-based cathodes for Li-ion batteries by Fe substitution

“…For the Fe-substitution and Gibb's free energies of O 2 formation, M3GNet and CHGNet, which are generalized machine learning interatomic potentials (MLIPs), were used to select appropriate structures. 37,38 The structures were substituted or removed by comparing the energy preferences of MLIPs.…”

Patterns Lead the Way to Far-from-Equilibrium Materials

“…Beyond data collection, another challenge is effectively processing diverse data qualities . Recent literature demonstrates the utility of transfer learning as an effective tool to extract useful materials design patterns from diverse data sets, such as data sets from different levels of simulation and experiments. This capability facilitates the pretraining of machine learning models on more generic data sets for specific research domains including high entropy materials.…”