A Combined First‐Principles and Machine Learning Framework for XMnH 3 (X = Li, Na, K) Perovskite Hydrides: Hydrogen Storage Capacity and Stability Under Extreme Conditions

ABSTRACT Perovskite‐type hydrides have emerged as promising candidates for next‐generation solid‐state hydrogen storage owing to their tunable structural, mechanical, and thermal properties. Here, we combine density functional theory (DFT) and machine learning (ML) to investigate cubic XMnH 3 perovskites (X = Li, Na, K) and extend predictions to unexplored compositions and pressures. DFT calculations confirm that all compounds stabilize in a ferromagnetic cubic phase with negative formation energies, competitive hydrogen storage capacities (4.66 wt.% for LiMnH 3 , 3.73 wt.% for NaMnH 3 , and 3.12 wt.% for KMnH 3 ), and hydrogen desorption temperatures between 517–886 K. Elastic constants establish mechanical stability up to 30 GPa, though with inherent brittleness, while thermal analysis via the quasi‐harmonic Debye model highlights pressure‐driven stiffening of the lattice. To circumvent the limitations of small DFT datasets, we trained Random Forest, XGBoost, and neural network models on direct and derived DFT descriptors, and we demonstrated that ensemble tree models yield the most accurate predictions. LiMnH 3 continuously outperforms heavier analogues, and expected monotonic trends are reproduced when the Debye and melting temperatures are extended to 60 GPa. Moreover, screening of about 46 ABH 3 hydrides was made possible by composition‐based descriptors created with Matminer, which showed systematic ionic‐radius trends and identified BeMnH 3 and MgMnH 3 as promising candidates with superior vibrational stability and hardness. This integrated DFT–ML framework establishes a predictive strategy for accelerating the discovery of hydrides with optimized hydrogen storage performance.