Inorganic electrides, characterized by localized electrons acting as anions within the cationic lattice framework, represent a unique and emerging class of materials with significant potential for applications in electrode materials, electronics, catalysis, and beyond. Despite their promise, the discovery of new electrides, especially those with complex structures, remains a formidable challenge. Current computational approaches, which predominantly employ "direct" strategies to screen for ideal electrides, often fail to identify promising candidates that could be derived from stable precursor compounds through anion removal. Recognizing that an electride is fundamentally a reduced derivative of its electron-saturated precursor, we introduce an innovative "indirect" screening strategy. This novel approach centers on identifying potential precursor materials that can be transformed into electrides, rather than directly searching for the final electride structures themselves. Consequently, this work endeavors to unlock the vast, yet underexplored, potential of mineral databases as a treasure trove for such precursor systems and to establish a robust, generalizable methodology for their efficient identification and discovery.

We proposed a comprehensive four-stage screening workflow, leveraging the American Mineralogist Crystal Structure Database (AMCSD) to systematically identify potential electride precursors. The initial phase involved the pre-processing of 20 488 mineral structures to standardize the dataset, ensuring uniformity and consistency. Subsequently, a preliminary screening stage was implemented, imposing constraints on compositional complexity, unit cell size, and explicitly excluding specific systems such as carbonates and metal oxides to narrow down the candidate pool. The cornerstone of our approach was the pattern screening stage, which utilized a novel topological descriptor. This descriptor was specifically engineered to identify local coordination environments where a removable anion (X = O, F, Cl, Br, I) is coordinated by 4 to 6 identical metal cations (the XM_n motif). This stage effectively filtered out structures that did not meet the predefined criteria for potential electride formation. In the final stage, the candidate structures identified through pattern screening were subjected to a rigorous fine screening process based on density functional theory (DFT). This involved structural optimization of both the precursor compounds and their corresponding anion-removed derivatives. Following optimization, electronic structure calculations were performed to assess the band gap transition and the spatial localization of electrons using the electron localization function (ELF). Specific thresholds were applied to identify the most promising electride precursors from this refined pool of candidates.

The proposed four-stage screening strategy demonstrated remarkable efficiency, successfully narrowing down from an initial dataset of 20 488 mineral structures to a final selection of 18 high-potential electride precursors. DFT calculations provided conclusive electronic evidence for their electride character: all identified precursors were wide-gap insulators (band gap > 2.0 eV), and upon removal of the central anion (X = O, F, Cl, Br, I), all underwent a characteristic electronic transition. This transition is characterized by the formation of new in-gap states at the Fermi level following anion removal. Specifically, 13 systems transitioned to metallic states, while 5 became narrow-gap semiconductors (< 0.5 eV). ELF analysis unambiguously confirmed the spatial localization of the resulting excess electrons within the anion vacancy sites. The effectiveness of our strategy was further validated by the identification of experimentally known sodalite- and apatite-type electrides among the candidates. Beyond merely validating the method, the discovery of 7 previously unreported structural prototypes underscores the strategy's power for novel discovery. This outcome not only highlights the rich diversity of electride precursors within mineral databases but also demonstrates the generalizability of our topological descriptor for identifying the essential chemical motif required for electride formation.

In conclusion, we have successfully established a robust and innovative screening strategy that is based on an indirect approach. This strategy has proven to be highly effective in identifying 18 high-potential inorganic electride precursors from an initial dataset of 20 488 mineral structures. The indirect strategy's success is underscored by its remarkable ability to not only reproduce known sodalite- and apatite-type electrides but also, more significantly, to uncover 7 previously unreported structural prototypes. The comprehensive spectrum of electron localization dimensionalities (0D, 1D, 2D) observed across the entire candidate pool further highlights the broad applicability and generalizability of our methodology. This work establishes mineral databases as a rich and valuable resource for the discovery of novel electrides. It also illustrates the power and potential of precursor-targeted screening in uncovering new materials. The generalizable framework presented here offers a versatile and powerful tool that can be extended to the systematic discovery of other functional materials based on specific local coordination patterns. The identified candidates provide a solid foundation for future experimental synthesis and in-depth investigation, paving the way for further advancements in the field of materials science.