Research Digest — 2026-05-27¶
ML Interatomic Potentials & Workflows¶
1. DPA3: A Graph Neural Network for the Era of Large Atomistic Models¶
Source: npj Computational Materials (s41524-026-02146-2) · 📅 2026-05-25 · ↗ Open paper
Presents DPA3, a multi-layer graph neural network built on line graph series (LiGS) designed for large atomistic models (LAMs). The authors demonstrate that DPA3's generalization error follows a scaling law with model size and training data, enabled by stacking additional layers and a dataset encoding mechanism that decouples training data scaling from model size via multi-task learning. When trained as problem-oriented potential energy models, DPA3 achieves competitive accuracy across diverse chemical systems.
Relevance to DENG.Group
Directly relevant to Yanhao Deng's MLIP development work. DPA3 represents the next evolution of the DPA (Deep Potential) family that is widely used for materials simulations. The scaling law analysis provides practical guidance on how much data and compute the group should invest when building MLIPs for halide and oxide solid electrolytes. The multi-task framework could allow the group to leverage their existing DFT datasets across multiple electrolyte chemistries.
2. Harnessing AtomisticSkills for Agentic Atomistic Research¶
Source: arXiv:2605.24002 · 📅 2026-05-18 · ↗ Open paper
Introduces AtomisticSkills, an open-source harness framework that empowers general-purpose AI coding agents to conduct atomistic research across materials science, chemistry, and drug discovery. The framework decomposes scientific workflows into modular agent skills and tools, integrating over 100 human-curated capabilities including database access, thermodynamics and kinetics modeling, and diverse simulation engines using MLIPs and DFT. The authors validate functional coverage against scientific literature and demonstrate campaigns including generative design of Li-ion solid-state electrolytes, autonomous MLIP benchmarking and fine-tuning, and multi-stage virtual screening.
Relevance to DENG.Group
Highly relevant to Yanhao Deng and the group's computational workflow. AtomisticSkills provides a pre-built agentic infrastructure for automating many tasks the group performs manually — from DFT calculations to MLIP training to high-throughput screening. The demonstrated campaign on generative design of Li-ion solid-state electrolytes directly mirrors the group's research. The framework could accelerate the group's discovery pipeline for new halide and sulfide electrolyte compositions by automating the cycle of structure generation, calculation, and analysis.
Halide Solid Electrolytes¶
3. A Cost-competitive Amorphous Oxychlorophosphate Polyanion Cluster Solid Electrolyte for All-Solid-State Lithium Batteries¶
Source: Nano Energy (S2211285526003630) · 📅 2026-05-22 · ↗ Open paper
Reports a cost-effective polyanion-incorporated amorphous oxyhalide solid electrolyte, xLi₃PO₄-TaCl₅ (xLPTC), synthesized via low-energy ball-milling. The optimized 1/3-LPTC composition achieves 1.3 mS cm⁻¹ room-temperature ionic conductivity with 0.310 eV activation energy. The incorporation of PO₄ groups stabilizes an amorphous TaCl₆-based framework and induces local distortion of the chlorine coordination environment. DFT calculations and MLFF modeling confirm that PO₄ units distort the Cl sublattice to enable superionic conduction. Full cells (Li₀.₇In | LPSCl | 1/3-LPTC | NMC811) showed 126.7 mAh g⁻¹ initial capacity and 70% retention after 738 cycles at 1C.
Relevance to DENG.Group
Directly relevant to the Deng group's core research on halide solid electrolytes. The polyanion-doping strategy represents a new design principle for engineering amorphous halide SSEs that the group could explore computationally using their MLIP and DFT tools. The use of MLFF (machine-learned force field) modeling to understand the conduction mechanism in the amorphous structure is exactly the type of computation Yanhao Deng performs. The cost-competitive angle (TaCl₅-based) is also notable for practical applications.
4. NASICON-type LATP Solid Electrolytes for Lithium Metal Batteries: Fundamentals to AI-driven Materials Design¶
Source: Energy Storage Materials (S2405829726002710) · 📅 2026-05-20 · ↗ Open paper
A comprehensive review covering the crystal structure, Li⁺ transport mechanisms, and key limitations of LATP solid electrolytes. The review systematically covers synthesis methods, LATP/Li-metal interfacial challenges, composite and hybrid electrolyte designs, and emerging AI/ML strategies for LATP optimization. It provides a roadmap from lab studies to industrial implementation, connecting structure–processing–property relationships with bulk and grain-boundary transport alongside interface and composite strategies.
Relevance to DENG.Group
Relevant as a reference resource for the group's broader solid electrolyte work. While LATP is an oxide SSE, the review's treatment of grain-boundary resistance, interface stability, and AI/ML-driven materials design provides transferable insights for the group's halide electrolyte research. The discussion of composite electrolyte strategies and interfacial engineering parallels challenges the group faces with halide SSE/cathode interfaces. The AI/ML section provides a useful benchmark for the group's computational approaches.
Polymer Electrolytes¶
5. Artificial Crystalline-Amorphous Architecture Enables Continuous Ion Transport in Poly(Vinylidene Fluoride)-Based Solid-State Electrolytes¶
Source: Small (10.1002/smll.73788) · 📅 2026-05-20 · ↗ Open paper
Designs an 'artificial crystalline-amorphous' architecture for solid-state polymer electrolytes by infiltrating fully amorphous PVT into an oriented electrospun fibrous framework of semi-crystalline PVT. The fibrous framework serves as an 'artificial crystalline phase' providing mechanical support and facilitating lithium-salt dissociation, while the amorphous phase enables long-range ion conduction. The resulting SPEs achieve 1.23 mS cm⁻¹ at 25°C, outperforming most all-polymeric SPEs, with stable Li/Li cycling over 1300 h at 0.2 mA cm⁻² and stable NCM811/Li full-cell cycling.
Relevance to DENG.Group
Relevant to the group's interest in solid polymer electrolytes. The artificial crystalline-amorphous architecture provides a new design principle for resolving the classic conductivity-mechanical strength trade-off in polymer electrolytes. The approach of using an electrospun fibrous framework to create continuous amorphous conduction pathways could inspire similar composite designs incorporating inorganic fillers (e.g., halide SSE particles) for the group's composite electrolyte work. The 1.23 mS cm⁻¹ conductivity is competitive with many ceramic SSEs.