Physically-informed artificial neural networks for atomistic modeling of materials

Abstract

Large-scale atomistic computer simulations of materials heavily rely on interatomic potentials predicting the potential energy and Newtonian forces on atoms. Traditional interatomic potentials are based on physical intuition but contain few adjustable parameters and are usually not accurate. The emerging machine-learning (ML) potentials achieve highly accurate interpolation between the energies in a large DFT database but, being purely mathematical constructions, suffer from poor transferability to unknown structures. We propose a new approach that can drastically improve the transferability of ML potentials by informing them of the physical nature of interatomic bonding. This is achieved by combining a rather general physics-based model (analytical bond-order potential) with a neural-network regression. The network adjusts the parameters of the physics-based model on the fly during the simulations according to the local environments of individual atoms. This approach, called the physically-informed neural network (PINN) potential, is demonstrated by developing a general-purpose PINN potential for Al. The potential provides a DFT-level accuracy of energy predictions and excellent agreement with experimental and DFT data for a wide range of physical properties. We suggest that the development of physics-based ML potentials is the most effective way forward in the field of atomistic simulations.

References in corpus (14)

• Gaussian Approximation Potentials: the accuracy of quantum mechanics, without the electrons
• SchNet - a deep learning architecture for molecules and materials
• Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields
• Automatic Selection of Atomic Fingerprints and Reference Configurations for Machine-Learning Potentials
• A neural network interatomic potential for the phase change material GeTe
• Accurate Interatomic Force Fields via Machine Learning with Covariant Kernels
• Extending the Accuracy of the SNAP Interatomic Potential Form
• A learning scheme to predict atomic forces and accelerate materials simulations
• Non-covalent interactions across organic and biological subsets of chemical space: Physics-based potentials parametrized from machine learning
• Efficient nonparametric $n$-body force fields from machine learning
• Ab initio quality neural-network potential for sodium
• Transferable atomic multipole machine learning models for small organic molecules
• First-principles interatomic potentials for ten elemental metals via compressed sensing
• Energetics of point defects in aluminum via orbital-free density functional theory

Cited by in corpus (39)

• Recent Advances and Applications of Deep Learning Methods in Materials Science
• Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems
• Machine-learning interatomic potentials for materials science
• Deep Potentials for Materials Science
• A Hybrid Science-Guided Machine Learning Approach for Modeling and Optimizing Chemical Processes
• Machine Learning on Neutron and X-Ray Scattering
• A Deep Dive into Machine Learning Density Functional Theory for Materials Science and Chemistry
• Automated discovery of a robust interatomic potential for aluminum
• Uncertainty Prediction for Machine Learning Models of Material Properties
• Quantum deep field: data-driven wave function, electron density generation, and atomization energy prediction and extrapolation with machine learning
• Optimization and Validation of a Deep Learning CuZr Atomistic Potential: Robust Applications for Crystalline and Amorphous Phases with near-DFT Accuracy
• Low-temperature plasma simulation based on physics-informed neural networks: frameworks and preliminary applications
• Development of a general-purpose machine-learning interatomic potential for aluminum by the physically-informed neural network method
• Elinvar effect in $\beta -$Ti simulated by on-the-fly trained moment tensor potential
• Universal correlation between electronic factors and solute-defect interactions in bcc refractory metals
• Development of a physically-informed neural network interatomic potential for tantalum
• Multilayer atomic cluster expansion for semi-local interactions
• A Physics-Enforced Neural Network to Predict Polymer Melt Viscosity
• Accelerating physics-informed neural network based 1D arc simulation by meta learning
• Machine-learning potentials enable predictive $\text{and}$ tractable high-throughput screening of random alloys
• Innate Dynamics and Identity Crisis of a Metal Surface Unveiled by Machine Learning of Atomic Environments
• Multi-task graph neural networks for simultaneous prediction of global and atomic properties in ferromagnetic systems
• A coarse-grained deep neural network model for liquid water
• Combining stochastic density functional theory with deep potential molecular dynamics to study warm dense matter
• Physics-consistent machine learning: output projection onto physical manifolds
• Algorithmic Differentiation for Automated Modeling of Machine Learned Force Fields
• Fast and stable deep-learning predictions of material properties for solid solution alloys
• Accelerated calculation of configurational free energy using a combination of reverse Monte Carlo and neural network models: Adsorption isotherm for 2D square and triangular lattices
• Fitting to magnetic forces improves the reliability of magnetic Moment Tensor Potentials
• Physical Neural Networks with Self-Learning Capabilities
• Automated atomistic simulations of dissociated dislocations with ab initio accuracy
• AI-accelerated Materials Informatics Method for the Discovery of Ductile Alloys
• A physics-defined recurrent neural network to compute coherent light wave scattering on the millimetre scale
• Equivariant Tensor Network Potentials
• Machine learning interatomic potential for the low-modulus Ti-Nb-Zr alloys in the vicinity of dynamical instability
• Low-rank matrix and tensor approximations for compression of machine-learning interatomic potentials
• Machine-Learned Bond-Order Potential for Exploring the Configuration Space of Carbon
• Exact average many-body interatomic interaction model for random alloys
• Validation of moment tensor potentials for fcc and bcc metals using EXAFS spectra