High-Efficient ab initio Bayesian Active Learning Method and Applications in Prediction of Two-dimensional Functional Materials

TL;DR

This paper introduces an ab initio Bayesian active learning method that combines active learning with high-throughput calculations to efficiently and accurately discover functional two-dimensional materials, significantly reducing computational costs.

Contribution

The paper presents a novel ab initio Bayesian active learning approach that effectively handles unbalanced property distributions in materials discovery, improving efficiency and accuracy.

Findings

Successfully screened 3,119 binary compounds for electric polarization and photovoltaic gaps.

Achieved high accuracy with only a fraction of the candidate calculations compared to random search.

Demonstrated significant reduction in computational costs while maintaining prediction quality.

Abstract

Beyond the conventional trial-and-error method, machine learning offers a great opportunity to accelerate the discovery of functional materials, but still often suffers from difficulties such as limited materials data and unbalanced distribution of target property. Here, we propose the ab initio Bayesian active learning method that combines active learning and high-throughput ab initio calculations to accelerate prediction of desired functional materials with the ultrahigh efficiency and accuracy. We apply it as an instance to a large family (3,119) of two-dimensional hexagonal binary compounds with unbalanced materials property, and accurately screen out the materials with maximal electric polarization and proper photovoltaic band gaps, respectively, whereas the computational costs are significantly reduced by only calculating a few tenths of possible candidates in comparison to the random search. This approach shows enormous advantages for the cases with unbalanced distributions of target property. It can be readily applied to seek for a broad range of advanced materials.