In operando active learning of interatomic interaction during large-scale simulations

Max Hodapp,Alexander Shapeev

DOI: https://doi.org/10.1088/2632-2153/aba373

2020-09-21

Abstract:A well-known drawback of state-of-the-art machine-learning interatomic potentials is their poor ability to extrapolate beyond the training domain. For small-scale problems with tens to hundreds of atoms this can be solved by using active learning which is able to select atomic configurations on which a potential attempts extrapolation and add them to the ab initio-computed training set. In this sense an active learning algorithm can be viewed as an on-the-fly interpolation of an ab initio model. For large-scale problems, possibly involving tens of thousands of atoms, this is not feasible because one cannot afford even a single density functional theory computation with such a large number of atoms. This work marks a new milestone toward fully automatic ab initio-accurate large-scale atomistic simulations. We develop an active learning algorithm that identifies local subregions of the simulation region where the potential extrapolates. Then the algorithm constructs periodic configurations out of these local, non-periodic subregions, sufficiently small to be computable with plane-wave density functional theory codes, in order to obtain accurate ab initio energies. We benchmark our algorithm on the problem of the screw dislocation motion in bcc tungsten and show that our algorithm reaches ab initio accuracy, down to typical magnitudes of numerical noise in DFT codes. We show that our algorithm reproduces material properties such as core structure, Peierls barrier, and Peierls stress. This unleashes new capabilities for computational materials science toward applications which have currently been out of scope if approached solely by ab initio methods.

Computational Physics,Materials Science

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in large - scale atomic simulations, the existing machine - learning interatomic potentials (MLIPs) have poor extrapolation ability outside the training data. For large - scale simulations involving thousands of atoms, due to the high cost of a single density - functional - theory (DFT) calculation, traditional active - learning methods are no longer applicable. Therefore, this paper proposes a new active - learning algorithm, which aims to improve the accuracy of large - scale simulations by identifying local sub - regions in the simulation area where MLIPs attempt to extrapolate, and converting these sub - regions into periodic configurations for accurate ab - initio calculations using plane - wave DFT codes. Specifically, the main contributions of the paper include: 1. **Developed an active - learning algorithm for large - scale problems**: This algorithm can identify the local sub - regions in the simulation area that need to be extrapolated and convert these sub - regions into periodic configurations, making DFT calculations feasible. 2. **Improved the accuracy of material property prediction**: Through this method, the algorithm can achieve ab - initio calculation accuracy and can even reproduce key properties of materials, such as dislocation core structures, Peierls barriers, and Peierls stresses. 3. **Expanded the application range of computational materials science**: This method makes large - scale atomic simulations possible, especially for problems that are currently difficult to handle only by ab - initio calculation methods, such as dislocation motion. Through these improvements, the paper provides a new tool for computational materials science, which can significantly reduce computational costs while maintaining high accuracy, thereby promoting multi - scale simulation research of complex material systems.

In operando active learning of interatomic interaction during large-scale simulations

Accelerating high-throughput searches for new alloys with active learning of interatomic potentials

Efficient Machine Learning Force Field for Large-Scale Molecular Simulations of Organic Systems

Accurate interatomic force field for molecular dynamics simulation by hybridizing classical and machine learning potentials

Automated atomistic simulations of dissociated dislocations with ab initio accuracy

Active learning of linearly parametrized interatomic potentials

Overcoming the Size Limit of First Principles Molecular Dynamics Simulations with an In-Distribution Substructure Embedding Active Learner

Accelerating Atomistic Simulations with Piecewise Machine Learned Ab Initio Potentials at Classical Force Field-like Cost

Accelerating crystal structure prediction by machine-learning interatomic potentials with active learning

Machine Learning Inter-Atomic Potentials Generation Driven by Active Learning: A Case Study for Amorphous and Liquid Hafnium dioxide

Active learning of effective Hamiltonian for super-large-scale atomic structures

Accelerating the Training and Improving the Reliability of Machine-Learned Interatomic Potentials for Strongly Anharmonic Materials through Active Learning

Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields

A transferable active-learning strategy for reactive molecular force fields

Adaptive-precision potentials for large-scale atomistic simulations

Lattice dynamics simulation using machine learning interatomic potentials

Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning

Ab-Initio Molecular Dynamics Acceleration Scheme with an Adaptive Machine Learning Framework

FINETUNA: Fine-tuning Accelerated Molecular Simulations

Enabling Large-Scale Condensed-Phase Hybrid Density Functional Theory Based $Ab$ $Initio$ Molecular Dynamics I: Theory, Algorithm, and Performance

Uncertainty-aware molecular dynamics from Bayesian active learning for phase transformations and thermal transport in SiC