Research Digest — 2026-08-29¶
ML Potentials & Ion Transport¶
1. Vibrational, structural, and chemical fingerprints of ion diffusion in crystalline solids¶
Source: arXiv · 📅 2026-08-21 · ↗ Open paper
Predicting mobile-ion self-diffusivity from MD is essential for screening solid-state electrolytes, but accurate MLIP-MD requires long trajectories because diffusion is slow and emergent. This work shows that vibrational, structural, and chemical fingerprints computed from short-time simulations and static structure can serve as proxies for Li-ion diffusivity, drastically cutting the simulation cost of electrolyte screening.
Relevance to DENG.Group
Yanhao Deng: a direct workflow upgrade for his MLIP-based halide/sulfide electrolyte screening — estimating transport from short runs instead of nanosecond trajectories. Mengke Li and Yan Li: faster transport proxies for halide electrolyte comparison studies.
Degradation & Data-Driven Battery Analytics¶
2. Data-efficient crack quantification in lithium-ion cathodes using foundation model transfer¶
Source: arXiv · 📅 2026-08-27 · ↗ Open paper
Quantitative microscopy of cathode particle cracking is bottlenecked by pixel-level expert annotation of gigapixel cross-sections. Transferring a vision foundation model enables data-efficient crack segmentation and quantification of degradation, dramatically reducing labeling effort for cathode aging studies.
Relevance to DENG.Group
Shoutong Jin: image-to-quantitative-degradation pipelines complement his phase-field dendrite/fracture simulations with experimental validation data. Umang Agarwal: chemo-mechanical cracking statistics feed directly into interface degradation models.
AI for Solid Electrolyte Discovery (Catch-up Pick)¶
3. Breaking Bottlenecks in Solid Electrolyte Discovery with Large Artificial Intelligence Models¶
Source: arXiv · 📅 2026-06-23 · ↗ Open paper
A comprehensive perspective that surfaced during the digest gap and has not yet appeared in this archive: it lays out a framework for autonomous solid electrolyte discovery using large AI models — foundation-model MLIPs, LLM-driven data extraction, generative structure prediction, and closed-loop robotic validation — tailored to the coupled requirements of bulk ion transport, defect chemistry, and mechanical integrity that make SEs harder than catalysis.
Relevance to DENG.Group
Group-wide strategy: benchmarks the group's ML-potential program (Yanhao Deng) against the emerging autonomous-discovery stack and offers a roadmap Jerry can use when positioning the group's computational identity in grant proposals. Timothy Pook: workflow/infrastructure implications for cluster automation.