Active learning of linearly parametrized interatomic potentials

By Evgeny V. Podryabinkin, Alexander V. Shapeev. 2016

626 citations recorded by the source corpus.

Abstract This paper introduces an active learning approach to the fitting of machine learning interatomic potentials. Our approach is based on the D-optimality criterion for selecting atomic configurations on which the potential is fitted. It is shown that the proposed active learning approach is highly efficient in training potentials on the fly, ensuring that no extrapolation is attempted and leading to a completely reliable atomistic simulation without any significant decrease in accuracy. We apply our approach to molecular dynamics and structure relaxation, and we argue that it can be applied, in principle, to any other type of atomistic simulation. The software, test cases, and examples of usage are published at http://gitlab.skoltech.ru/shapeev/mlip/ .

Subfields: Biomedical Engineering, Computational Theory and Mathematics, Materials Chemistry

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

In Computational Theory and Mathematics

In Materials Chemistry

Citations in the corpus

Machine-Learning Interatomic Potentials for Materials Science (2021, background): “s e m a MLIP (Machine Learning Interatomic Potentials) [100,143] packge constructs MTP potentials [80,96,97,100] (including multicomonent potentials [80] ) by the active learning approach.”
Machine Learning for High-entropy Alloys: Progress, Challenges and Opportunities (2022, methodology): “proposed an active learning strategy based on the so-called D-optimality criterion [122, 130], which selects the training set by maximizing the determinant of the information matrix.”
Machine-learned potentials for next-generation matter simulations (2021, background): “…(based on Thompson sampling 97 ), to identify suitable collective variables (based on reinforcement learning 98 ), as well as for the prediction of molecular properties 99 (for example, atomization energies, polarizabilities) and interatomic potentials 100 (based on the D-optimality criterion).”
Nanoporous C3N4, C3N5 and C3N6 nanosheets; novel strong semiconductors with low thermal conductivities and appealing optical/electronic properties (2020, methodology): “After the initial training of MTPs, the accuracy of the trained potentials were evaluated over the full AIMD trajectories and the configurations with high extrapolations grades [45] were selected.”
The MLIP package: moment tensor potentials with MPI and active learning (2020, methodology): “An active learning algorithm was proposed in [57] based on which point defect diffusion [53] and lattice dynamics [37] of a number of mono-component crystals were studied.”