Accelerated Inorganic Electrides Discovery by Generative Models and Hierarchical Screening

TL;DR

The paper tackles the challenge of discovering inorganic electrides by introducing an accelerated discovery framework that combines physics-guided chemical-space selection, diffusion-based crystal structure generation, and hierarchical screening with ML potentials and high-throughput DFT validation. Using this pipeline, the authors explore 1,510 binary and 6,654 ternary compositions and identify 264 electride candidates within $0.05$ eV/atom of the convex hull, including 13 thermodynamically stable electrides. They validate the prescreening workflow, quantify its accuracy against DFT and show substantial pruning of candidates while preserving true positives, and demonstrate concrete binary and ternary electride discoveries with interstitial electron localization. The findings illustrate a generalizable strategy for targeted, AI-assisted materials discovery in vast chemical spaces, with potential extensions to other functional materials and opportunities for experimental synthesis guided by the identified candidates.

Abstract

Electrides are exotic compounds in which excess electrons occupy interstitial regions of the crystal lattice and serve as anions, exhibiting exceptional properties such as low work function, high electron mobility, and strong catalytic activity. Although they show promise for diverse applications, identifying new electrides remains challenging due to the difficulty of achieving energetically favorable electron localization in crystal cavities. Here, we present an accelerated materials discovery framework that combines physical principles, diffusion-based materials generation with hierarchical thermodynamic and electronic structure screening. Using this workflow, we systematically explored 1,510 binary and 6,654 ternary chemical compositions containing excess valence electrons from electropositive alkaline, alkaline-earth, and early transition metals, and then filtered them with a high throughput validation on both thermodynamical stability and electronic structure analysis. As a result, we have identified 264 new electron rich compounds within 0.05 eV/atom above the convex hull at the density functional theory (DFT) level, including 13 thermodynamically stable electrides. Our approach demonstrates a generalizable strategy for targeted materials discovery in a vast chemical space.

Figures (5)

• Figure 1: The workflow for accelerated electride discovery combining physical principles, generative modeling (MatterGen), machine learning potential prescreening (MatterSim), and high-throughput DFT validation for electride and stability analysis. The top panel outlines key operations at each stage; the middle panel illustrates the definition of metrics used for candidate selection with the targeted compositions being highlighted by the light purple color; and the bottom panel specifies filtering criteria and the number of candidates retained after each stage.
• Figure 2: Validation of MLP Energy used in the screening workflow for (a) binary and (b) ternary systems, respectively. In both plots, the left panel shows the comparison in terms of absolute energy per atom ($E_\text{abs}$) between MLP and DFT results, while the right panel shows the comparison between energy with respect to the reference convex hull values ($E_\text{ref-hull}$) from the available materials project data.
• Figure 3: Convex hulls and new phases for (a) Ca-P and (b) Y-N systems. Top panels show the Materials Project (black circles) and newly discovered phases (red triangles for electrides and red squares for nonelectrides, on-hull structures are filled with blue, above hull structures are filled with orange), with the gray dashed line indicating the original convex hull and the solid black line representing the refined convex hull. The electron-rich compositions with $0 < N_\text{excess} \leq 4$ are highlighted in yellow rectangles. The middle panels display a zoomed-in sub-convex hull plot between the elemental metal and the stoichiometric binary with the most negative formation energy. The bottom panels illustrate representative structures from each system with atomic colors: Ca (steel blue), P (pale purple), Y (dark green), N (light blue); yellow isosurfaces represent excess electrons' distributions.
• Figure 4: Ternary convex hulls and new phases for (a) Cs-Al-P and (b) K-B-O systems. Top panels: dashed black lines indicate original convex hull boundaries; solid purple lines show updated convex hulls after incorporating newly identified electride compounds. Black circles represent Materials Project phases; red triangles denote electride candidates (stable ones are filled with purple). The electron-rich composition region ($0 < N_{\text{excess}} \leq 2$) is highlighted by yellow shading. Bottom panels: representative electride structures from each system with atomic colors: K (purple), Cs (cyan), Al (gray), B (green), O (red), P (pale purple); yellow isosurfaces represent excess electrons' distributions.
• Figure 5: Electronic structures of two representative stable ternary electrides: (a) hP11-K$_6$BO$_4$ and (b) mS26-Cs$_6$Al$_2$S$_5$. Left panels: band structures and projected density of states (PDOS). Right panels: partial charge density distributions for the top valence bands in hP11-K$_6$BO$_4$ and mS26-Cs$_6$Al$_2$S$_5$ (shown in top and side views). Atomic colors: K (purple), Cs (cyan), Al (gray), B (green), O (red), S (orange); yellow isosurfaces represent the electron charge density.