Accelerating the exploration of novel perovskite-structured metal borohydrides with enhanced dehydrogenation performance through machine learning

Abstract

Metal borohydrides are recognized as ideal solid-state materials due to their exceptional hydrogen storage capacity. However, their high dehydrogenation temperatures limit their practical use for hydrogen storage. Prior research has demonstrated that increasing the electronegativity of metal cations can lower the dehydrogenation temperature. In this study, we developed perovskite-structured metal borohydrides, leveraging the properties of perovskite structures with bimetallic or polymetallic cations and soft lattice characteristics. This approach allows for precise regulation of metal ion electronegativity within a single structure, thereby conveniently decreasing the dehydrogenation temperature. To accelerate the discovery of novel perovskite-structured metal borohydrides, we employed machine learning. To address the challenge of limited data, we proposed a strategy of utilizing metal halides as analogs to metal borohydrides for training. After comparing four distinct training algorithms, the high-performing gradient boosting decision tree (GBDT) was selected for prediction. Through the computation of single perovskite (ABX₃) and double perovskite (A₂B₁B₂X₆)-structured borohydrides, KMn(BH₄)₃ emerged as the most promising candidate, exhibiting a dehydrogenation temperature of just 207.5 °C and a hydrogen capacity of 8.63 wt%. We also found that the B-site cation significantly influences the dehydrogenation temperature. Density functional theory (DFT) was further used to calculate the electronic structure, and transition state calculations were conducted to demonstrate its superior hydrogen desorption performance. In conclusion, this work introduces novel perovskite-structured metal borohydrides and paves the way for the exploration of promising hydrogen storage materials with high capacity and moderate temperature, thereby promoting the large-scale development of hydrogen energy.