Computational Discovery of Thermochemical Heat Storage Materials Based on Chalcogenide and Complex Anion Salt Hydrates

Technologies for thermal energy storage (TES) are limited by the performance of the heat storage material. Therefore, it is desirable to develop materials with superior heat storage properties. The present study employs first-principles calculations to predict the properties of 7012 hypothetical hydrates based on chalcogenide and complex anion salts. Accounting for thermodynamic stability and energy densities, promising hydrates were identified for temperatures below 200 °C, including Li2S·9H2O, Ca(OH)2·8H2O, and Li2CO3·10H2O. System-level projections indicate that several of the proposed materials surpass the energy densities of known materials when incorporated into a solar-thermal storage system. Interpretable machine learning models were trained on the hydrate data set and used to identify features that control the enthalpy of dehydration. This analysis reveals similarities and differences in the thermodynamic behavior of hydrates based on chalcogenides, complex anions, and the previously studied halides. Hydrates based on chalcogenide anions exhibit a wide distribution of dehydration enthalpies; the low average enthalpies of these hydrates reflect the fact that relatively few are stable. In contrast, hydrates based on complex anions and halides exhibit enthalpies that are, on average, larger and more narrowly distributed. The enthalpies of the chalcogenide hydrates can be predicted by using only two machine-learned features, both of which implicate the electronegativity of the cation as a controlling property. This correlation agrees with a trend reported previously for halide-salt hydrates. In contrast, the behavior of the complex-anion hydrates requires twice as many features for machine-learning predictions, and some of these features are complex. Nevertheless, a combination of the molar volume and boiling point data is identified as a useful descriptor. In total, the hydrate compositions and design insights identified in this study are anticipated to catalyze the development of more efficient TES systems.