Metadynamics sampling in atomic environment space for collecting training data for machine learning potentials

Abstract

The universal mathematical form of machine-learning potentials (MLPs) shifts the core of development of interatomic potentials to collecting proper training data. Ideally, the training set should encompass diverse local atomic environments but conventional approaches are prone to sampling similar configurations repeatedly, mainly due to the Boltzmann statistics. As such, practitioners handpick a large pool of distinct configurations manually, stretching the development period significantly. To overcome this hurdle, methods are being proposed that automatically generate training data. Herein, we suggest a sampling method optimized for gathering diverse yet relevant configurations semi-automatically. This is achieved by applying the metadynamics with the descriptor for the local atomic environment as a collective variable. As a result, the simulation is automatically steered toward unvisited local environment space such that each atom experiences diverse chemical environments without redundancy. We apply the proposed metadynamics sampling to H:Pt(111), GeTe, and Si systems. Throughout these examples, a small number of metadynamics trajectories can provide reference structures necessary for training high-fidelity MLPs. By proposing a semi-automatic sampling method tuned for MLPs, the present work paves the way to wider applications of MLPs to many challenging applications.

Modeling Chemical Reactions in Alkali Carbonate–Hydroxide Electrolytes with Deep Learning Potentials

“…A solution to overcome this barrier would be introducing a sampling method optimized for semiautomatically collecting manifold yet appropriate configurations. One possible approach would be to employ enhanced sampling techniques such as metadynamics with the descriptor for the local atomic environment as a collective variable . Consequently, the simulation is expected to be steered toward an unvisited local environment space so that each atom explores diverse chemical environments.…”

“…One possible approach would be to employ enhanced sampling techniques such as metadynamics with the descriptor for the local atomic environment as a collective variable. 75 Consequently, the simulation is expected to be steered toward an unvisited local environment space so that each atom explores diverse chemical environments. In the case of OH − , the overall profile almost reproduces the AIMD curve within the limits of the error bar.…”

Efficient generation of stable linear machine-learning force fields with uncertainty-aware active learning

“…The latter approach is perhaps more appealing and could naively be realized by sampling MD at a higher temperature than the one targeted in production runs, or by introducing advanced sampling methods developed in the context of rare events. Some promising attempts in the latter direction have been recently been taken [40,59,60].…”

Exploring the Relationship between Composition and Li-Ion Conductivity in the Amorphous Li–La–Zr–O System

“…However, it is important to note that AIMD simulations have limitations in sampling configurations near high energy barriers. In this work, to overcome this limitation, we adopted the metadynamics method based on the abstract atomic environment space (G-metaD) proposed by Yoo et al 30 This method introduces a bias potential to enhance sampling and explore configurations that may deviate from the Boltzmann distribution.…”