Research Digest — 2026-09-17¶
Solid Electrolytes¶
1. Non-reactive sintering enhances density and ionic conductivity of NASICON solid electrolytes¶
Source: arXiv · 📅 2026-09-11 · ↗ Open paper
Systematically varies calcination temperature (900-1200 °C) for Na3Zr2Si2PO12-based NASICON electrolytes before identical sintering conditions, showing that higher degrees of pre-reaction (non-reactive sintering) yield denser pellets with higher ionic conductivity. Demonstrates that the calcination step, often treated as a processing detail, critically controls final electrolyte quality.
Relevance to DENG.Group
Relevant to Naibing Wu's composite electrolyte simulations and Jerry's processing-microstructure interests: processing-induced microstructure (density, connectivity) is a first-order variable for transport, which our microstructure-resolved models should account for when comparing predicted vs measured conductivities.
ML Interatomic Potentials & Simulation Methods¶
2. Symbolic Ensemble Learning Enables Discovery of Fast Accurate Physics-Based Interatomic Potentials¶
Source: arXiv · 📅 2026-09-15 · ↗ Open paper
Introduces a hybrid symbolic-neural framework (Equation Learner Networks trained on DFT data) that unifies the interpretability of analytical potentials like EAM with the adaptability of machine learning. The resulting potentials are physically transparent, fast, and transferable to far-from-equilibrium regimes where conventional analytical forms fail.
Relevance to DENG.Group
Yanhao Deng: an interpretable-potential direction worth watching — symbolic regression components could improve extrapolation behavior of the group's electrolyte potentials while keeping the physical insights that plain black-box MLIPs hide.
3. Extreme-Scale Linear-Scaling Kohn-Sham DFT at 100 Million Atoms: Bridging Quantum Simulations and Experiments¶
Source: arXiv · 📅 2026-09-11 · ↗ Open paper
Presents XLSDFT, a linear-scaling Kohn-Sham DFT framework based on divide-and-conquer density-matrix decomposition with Chebyshev-filtered subspace iteration, run on an exascale supercomputer for systems of up to 100 million atoms with DFT-level accuracy. Demonstrates quantum-mechanical simulation at length scales approaching real microstructures and devices.
Relevance to DENG.Group
Tim Pook and Yanhao Deng: a benchmark for what extreme-scale first-principles computation looks like; relevant both for cluster capacity planning and as a reminder that large-scale training-data generation (e.g., for MLIPs of interfaces and grain boundaries) can ultimately be anchored in DFT at experimentally relevant scales.
4. El Agente Potente: High-Throughput Agentic Atomistic Simulations¶
Source: arXiv · 📅 2026-09-13 · ↗ Open paper
Introduces an agentic system combining typed execution graphs with a coding mode to translate high-level scientific intent into rigorous, adaptive MLIP-driven simulation campaigns. The framework maintains workflow rigour and reproducibility while letting LLM agents orchestrate high-throughput atomistic property calculations.
Relevance to DENG.Group
Tim Pook and Yanhao Deng: directly applicable to automating the group's high-throughput screening pipelines (potential fitting, property sweeps) on the cluster — typed execution graphs are one concrete pattern for keeping AI-orchestrated simulations auditable rather than ad hoc.
5. Benchmarking Universal Machine Learning Force Fields for Crystal Structure Prediction of High-Energy Molecular Systems¶
Source: arXiv · 📅 2026-09-07 · ↗ Open paper
Tests universal MLIPs for crystal structure prediction of energetic molecular crystals, finding they can be over-sensitive and break molecular connectivity during relaxation. Classical force-field pre-relaxation before UMLIP relaxation improves robustness on a large database of high-energy systems.
Relevance to DENG.Group
Yanhao Deng: a concrete caution for using universal potentials in structure prediction/relaxation workflows — the failure mode (bond breaking under relaxation) is a useful stress test to keep in mind when applying foundation potentials to novel electrolyte chemistries.
Interfaces & Transport¶
6. Direct observation of reversible ionic polarization at a buried solid-state battery interface¶
Source: arXiv · 📅 2026-09-11 · ↗ Open paper
Combines time-of-flight Elastic Recoil Detection Analysis with Neutron Depth Profiling to directly observe reversible lithium redistribution at the buried amorphous LiCoO2/LAGP solid electrolyte interface under ±1 V bias. First direct experimental evidence that buried solid-solid interfaces possess intrinsic ionic reversibility under applied fields.
Relevance to DENG.Group
Umang Agarwal: a quantitative experimental anchor for the group's electrolyte/electrode interface simulations — predicted space-charge layers and bias-driven Li redistribution at cathode/halide-type interfaces can now be benchmarked against direct compositional profiling data.
7. Multi4D: an end-to-end neural network for structural determination at complex material interfaces¶
Source: arXiv · 📅 2026-09-13 · ↗ Open paper
A physics-informed neural network framework for automated multi-component crystallographic identification from 4D-STEM diffraction data, designed to decode heterogeneous interfaces with phase overlap and local disorder at atomic resolution across large fields of view. Enables linking local interfacial structural variation to macroscopic degradation mechanisms.
Relevance to DENG.Group
Umang Agarwal and Yanhao Deng: experimental 4D-STEM interfacial analysis is becoming automated and quantitative — simulated interface structures (disorder, interphase thickness, phase distribution) from the group can be validated against, and jointly interpreted with, these learned diffraction analysis pipelines.
8. Microstructure-Resolved Impedance Modeling of Solid-State Batteries¶
Source: arXiv · 📅 2026-09-14 · ↗ Open paper
Develops a modeling framework connecting the microstructure of solid electrolytes (particle networks, porosity, interfaces) to electrochemical impedance spectra, enabling quantitative interpretation of EIS in terms of underlying microstructural features. Bridges the gap between measured impedance signatures and morphology of composite cathodes and electrolytes.
Relevance to DENG.Group
Shoutong Jin and Umang Agarwal: microstructure-to-impedance forward modeling complements the group's phase-field and interface work — the same microstructural inputs that drive dendrite/interface simulations also determine EIS fingerprints, offering a route to connect our simulations with routine experimental characterization.
Defects & Grain Boundaries¶
9. Ab initio Study of Substitutional Defects in Li3OCl Solid Electrolyte for Li-ion Batteries¶
Source: arXiv · 📅 2026-09-11 · ↗ Open paper
Explores extending neutron-transmutation defect engineering (previously shown for LiBO2) to the antiperovskite solid electrolyte Li3OCl, using ab initio calculations of substitutional defects and their effect on ionic conductivity. Assesses whether transmutation-generated vacancies can raise room-temperature conductivity toward practical levels.
Relevance to DENG.Group
Cheng Peng: directly on point-defect engineering in an antiferroelectric-structured solid electrolyte — the defect-conductivity design logic pairs well with the group's grain boundary transport studies, since GB and point-defect populations jointly control polycrystalline conductivity.
10. A cylindrical sintering method for more realistic grain boundaries in nanocrystalline thin films¶
Source: arXiv · 📅 2026-09-14 · ↗ Open paper
Introduces a cylindrical sintering sample-construction method that produces nanocrystalline aluminum films with wider, more disordered grain boundaries than standard Voronoi tessellation, while keeping deterministic control over grain size, shape, and orientation. Benchmarks show the resulting GB structures and mechanical response agree better with experiment.
Relevance to DENG.Group
Cheng Peng: construction methodology matters as much as the potential in GB simulations — the sintering approach could generate more realistic GB networks and structures for the group's solid electrolyte grain boundary models, where GB width/disorder directly affects calculated ion transport.
11. Prescreening Point Defects in Semiconductors With Machine Learning¶
Source: arXiv · 📅 2026-09-13 · ↗ Open paper
Develops physics-guided ML models for predicting defect formation energies and optical transitions, designed as a fast prescreening stage ahead of DFT-level high-throughput defect workflows. Demonstrates the accuracy/throughput trade-offs needed to make defect screening tractable across large chemical spaces.
Relevance to DENG.Group
Cheng Peng: the prescreening workflow pattern (cheap ML filter, DFT verification) transfers well to screening charge-compensating defect schemes and dopants in solid electrolytes, complementing the group's potential-based sampling of defect-containing GB structures.