Machine Learning Model for High-Throughput Screening of Perovskite Manganites with the Highest Néel Temperature

Abstract

Neel temperature (TN), the antiferromagnetic to paramagnetic phase transition temperature, is also one of the important magnetic properties of antiferromagnets. Antiferromagnets could be used to write and read information below TN in electronic memory devices. Perovskite manganites are an important antiferromagnetic material with superior magnetic properties. However, it is challengeable to seek novel perovskite manganites with TN higher than room temperature by trial-and-error method. In this work, the maximum correlation minimum redundancy (mRMR) integrated machine learning approaches were used to screen the optimal subset of features, which included chemical compositions and atomic parameters of perovskite manganites. The machine learning model called support vector regression (SVR) was constructed to predict the TN of perovskite manganites. The correlation coefficient (R) between experimental TN and predicted TN reached as high as 0.87 for the training set in leave-one-out cross-validation (LOOCV) and 0.86 for the independent testing set, respectively. The high-throughput screening of new perovskites with higher TN temperature was then carried out by using our online computation platform for materials data mining (OCPMDM). The TN of designed perovskite manganites (Sr0.7Pm0.3MnO3) was predicted to be 307.5K, increasing by 6% compared to the maximum TN of Sr0.9Ce0.1MnO3 (290K) reported. Our machine learning model can be accessible for public on web server: http://materials-data-mining.com/lkl/lkl_model .