Integrated machine learning assisted high-throughput discovery of novel magnetic double perovskite oxides

Magnetic properties of double perovskites (DPs) are rich but underlain by complex microscopic exchange mechanisms. Machine learning (ML) assisted high-throughput (HT) methods have witnessed great success but employ either simplified atomic inputs or exhaustive band-structure calculations which fall short where strong electron correlations play a significant role. Here we take advantage of hopping parameters generated by Wannierization and existing experimental data to train the integrated machine learning (ML) model capable of predicting DPs without experimental observations. We use classification learning to distinguish between anti-ferromagnetic (AFM) and ferromagnetic (FM) exchange dominated DPs with the high accuracy of 82%, regression learning to fit magnetic transition temperatures (T_c) with the mean square error of 19 and 67 K, respectively, and propose two AFM and four FM candidates with high T_c. Our methodology is able to replace resource-demanding HT calculations, open up the possibility of ML assisted HT computations and provide new insights into better understanding the underlying physical mechanisms of complex magnetic properties.