Geometry-based Discovery of Calcium Battery Cathodes Accelerated by Foundational Machine-Learned Models

Calcium batteries (CBs) are an attractive post-Li-ion technology, offering the appeal of Ca's natural abundance and high volumetric energy density. However, practical realization of CBs remains limited by the scarcity of positive electrode (cathode) materials that support reversible Ca$^{2+}$ (de)intercalation under electrochemical conditions. To address this challenge, we screen the materials project (MP) database for novel host structures that can intercalate Ca using geometry- and chemistry-based design principles. Specifically, we employ the Voronoi polyhedral volume as a descriptor of site compatibility for hosting Ca in potential frameworks. Further, we down-select candidate structures progressively through criteria including charge neutrality, absence of non-Ca mobile cations, thermodynamic (meta)stability, average voltage, and Ca migration barriers ($E_m$) using foundational machine-learning (ML) models. Subsequently, we validate the ensemble-ML-predicted $E_m$ in a subset of the final candidates using density functional theory based nudged elastic band calculations. Overall, from an initial pool of 52,945 MP structures, our workflow identifies 37 promising Ca cathode candidates, several of which exhibit favorable combinations of thermodynamic (meta)stability, voltage, and Ca-mobility, marking them as strong candidates for synthesis and electrochemical characterization. Particularly, we identify two Ca cathode candidates with markedly low Ca$^{2+}~E_m$ (CaSc$_2$V$_2$O$_8$ and CaVSO$_4$F$_3$), and four cathode candidates with thermodynamically stable charged states (Ca$_3$(CoO$_2$)$_4$, Ca$_3$Mn$_4$(TeO$_6$)$_2$, CaVF$_5$, and CaVSO$_4$F$_3$). Beyond identifying Ca-cathodes, our work establishes geometry-based descriptors and ML-based workflows as transferable methods for high-throughput screening, enabling the rapid discovery of novel materials for battery and other applications.