Insights of cation ordering in double perovskites oxides from first-principles calculations and machine learning

In this work, we have employed first-principles density functional calculations and traditional machine learning (ML) techniques to explore insights of cation ordering in AA′BB′O₆ double perovskites. We have studied various possible oxidation states of A, A’, B and B’ maintaining the charge neutrality of the system and various B(3d)- B′(3d), B(3d)- B′(4d) and B(3d)- B′(5d) systems to construct a dataset consisting of a wide compositional space. A non-exclusive list of these features includes optimized structure parameters, bond length and angles, magnetic moments, orbital field-matrix and structural modes. Random Forest Classifier is then employed to build a decision-tree type ML model. Our model is successful in classifying various A-site cation orderings. It also identifies important features including structural modes relevant to predict A[Layer]B[Rock-salt] ordering which leads to polar space group (P2₁) if (a-a-c+) distortion is imposed. Based on the ML model combined with causal analyses, we aim to predict new compositions for stable A[Layer]B[Rock-salt] with functional properties.