Double perovskite structure has brought a lot of attention due to its great potentials. Despite of its importance, it is difficult to find the thermodynamically stable structure satisfying target properties due to its enormous chemical space. The chemical formula of the oxide-type of double perovskite is ABB’O3 and a number of atomic elements, which can locate at A and B site, reaches 70. This leads a number of formable structures to around 350,000 and high-throughput materials screening cannot be performed with the usual method such as experiments or ab initio calculations.<br/>In this study, to overcome this issue, machine learning algorithm is employed to search for new stable double perovskite materials. First, the inorganic materials properties are adopted from well-established Materials Project database as a trainset. Then, the chemical and structural descriptors are applied to describe the given materials. The predict algorithms are implemented to train the database for construction of the regression and classification model for predicting the formation energy and the convex hull energy, respectively. For prediction of the formation energy, R2 and RMSE value for the test set reaches 0.97 and 0.2020 eV/atom. Meanwhile, the classification accuracy for prediction of the convex hull energy exhibits F1-score of 0.763. Both models demonstrate that proposed method can accurately capture the thermodynamical nature of the inorganic materials. Finally, trained machine learning model is applied to the whole chemical space of the double perovskite structures (13,234 structures; they are chosen based on the tolerance factor) and it exhibits that 2,487 structures are potentially stable to be synthesized. Finally, for validation of the proposed machine learning model, DFT calculations are performed to compute the formation energy.<br/>Current results clearly demonstrate that the machine learning method can be used to perform the high-throughput screening for the double perovskite structures with much reduced computational resources and reasonable accuracy.

Predicting synthesizability of double perovskite via machine learning

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.