This paper proposes a graph neural network model to predict the synthesizability of perovskites, a crucial material in various applications. The model achieves a high out-of-sample true positive rate (0.957), outperforming empirical rule-based methods. Validation shows the model accurately identifies previously synthesized perovskites and excludes those not yet reported. Unlike previous methods limited to metal oxides, this model is generalizable across various perovskite classes. The authors apply the model to identify synthesizable candidates for Li-rich ion conductors and metal halide optical materials.
