Accelerating Solid-State Hydrogen Storage Discovery: Machine Learning, DFT, and Phonon Calculations of Equimolar NaCa(BH4)3

Solid-state hydrogen storage is pivotal for the transition to zero-emission energy infrastructure, yet high-capacity complex metal hydrides are chronically hindered by unfavorable thermodynamics. While heterovalent mixed-metal borohydrides, such as equimolar NaCa(BH₄)₃, theoretically offer an exceptional capacity of 11.24 wt.% and the potential to bridge the thermodynamic extremes of their parent phases, their immense configurational complexity has historically obscured their true most stable structure. Standard crystallographic algorithms routinely fail to navigate these low-symmetry domains, often falling into local-minima traps due to artificial supercell bloating. Here, we overcome this fundamental bottleneck by engineering a custom deterministic combinatorial engine driven by nearest-neighbor environmental fingerprinting. Starting from an exhaustive, highly degenerate configurational space of 2844 theoretical cationic permutations, our tailored topological sieve deterministically distilled the landscape down to 143 unique, symmetry-inequivalent configurations. This composition-resolved screen was complemented by an independent fixed-composition cation-ordering enumeration of 427 equimolar Na/Ca arrangements (ATAT); the two screens converge on the same most stable configuration. By coupling this highly refined structural funnel with machine-learning interatomic potentials (MACE-MP) and first-principles lattice dynamics, we definitively isolate the most stable structure of the system — which overcomes the soft-mode vibrational instabilities of previously assumed high-symmetry lattice structures. These findings resolve long-standing structural ambiguities and establish our localized symmetry-breaking workflow as a superior blueprint for tailoring reversible, high-density energy storage materials.