Research Digest — 2026-09-25¶
ML Potentials & AI-Assisted Simulation¶
1. SoLiD26: A First Principles Solid-Liquid Interface Dataset for Machine-learned Interatomic Potentials¶
Source: arXiv · 📅 2026-09-23 · ↗ Open paper
The DTU group releases SoLiD26, a curated open dataset of 15.4 million DFT structures (up to 576 atoms, 15 elements) sampled from liquid, solid, and crucially solid-liquid interface regions, mostly from AIMD trajectories. It provides train/test splits designed for benchmarking MLIPs in electrochemistry, catalysis, and corrosion settings.
Relevance to DENG.Group
Yanhao: directly relevant to how we build and validate ML potentials for electrode/electrolyte interfaces — a reference for interface-sampling coverage and benchmark protocol. Also a candidate evaluation set when choosing universal potentials for group projects.
2. An open benchmark for machine learning-based polymer property prediction¶
Source: arXiv · 📅 2026-09-22 · ↗ Open paper
Introduces PolyBench26, an open benchmark of ~250,000 polymer-property datapoints across eight physical properties from experiment, DFT, and MD, spanning homopolymers and alternating/random/block copolymers. Defines four tasks including dataset-size scaling and transfer to held-out polymer families.
Relevance to DENG.Group
Naibing: a standardized ML baseline for polymer property prediction that could anchor the ML component of our polymer/composite electrolyte simulation work — useful for judging whether ML surrogates are ready to complement MD on polymer electrolytes.
3. Agent-E2MD: Autonomous Translation of Interatomic Potential Equations into Physically Validated Pair Styles for Molecular Dynamics in LAMMPS¶
Source: arXiv · 📅 2026-09-22 · ↗ Open paper
An agentic LLM workflow that converts user-specified interatomic potential equations into executable, physically validated LAMMPS pair styles, handling neighbor-list architecture, many-body derivatives, MPI communication, and energy/force/virial conventions automatically.
Relevance to DENG.Group
Tim + Yanhao: automation of the potential-implementation bottleneck we hit when porting custom potentials to HPC MD codes; worth a read for our simulation infrastructure and for cutting group time from paper equations to running production simulations.
4. Machine learning assisted molecular dynamics of charge-transfer mechanisms at Li/Ga-doped Li7La3Zr2O12 (LLZO) interfaces¶
Source: arXiv · 📅 2026-09-20 · ↗ Open paper
Trains moment tensor potentials for cubic/tetragonal/Ga-doped LLZO and Li metal to run machine-learning MD of Li+ transport in bulk and across Li/solid-electrolyte interfaces at finite temperature. Introduces a residence-time window method to filter interfacial charge-transfer events, showing concerted multi-ion motion beyond 0 K NEB pictures.
Relevance to DENG.Group
Yanhao: methodologically the closest match to our ML-potential interface program (MTP + finite-T charge transfer at Li/SE). Umang: the interface charge-transfer framing and event-filtering analysis transfer directly to electrolyte/electrode interface studies.
Solid Electrolytes & Ion Transport¶
5. Electrolyte Dependent Structure Transport Relationships in Electrodeposited Prussian Blue Analogue Thin Films¶
Source: arXiv · 📅 2026-09-22 · ↗ Open paper
Electrodeposits Fe/Co/Ni hexacyanoferrate films from KCl, NaCl, NH4Cl, and LiCl baths and shows the cation identity controls lattice parameter, vacancy concentration, microstrain, and local coordination — with Na+ promoting expansion, vacancies, and structural heterogeneity — and links this structural disorder to ion-transport and charge-storage behavior.
Relevance to DENG.Group
Naibing: the electrolyte-cation-dependent disorder/transport relationship is a clean experimental case of how synthesis environment sets transport-relevant defects, complementing our simulation of defect-mediated ion transport in composite systems.
6. Ultralow-Barrier Ion Transport in the Subnitride Electride Ba3N for High-Rate Sodium Storage¶
Source: arXiv · 📅 2026-09-20 · ↗ Open paper
First-principles calculations show quasi-1D subnitride electride Ba3N supports spontaneous Na intercalation into open interchain channels with an exceptionally flat low-potential plateau and ultralow diffusion barriers, avoiding common Na-anode trade-offs between rate and voltage.
Relevance to DENG.Group
Group-wide methodology benchmark: DFT screening of low-barrier alkali-ion channels (with interstitial electron delocalization tuning transport) parallels our solid electrolyte descriptor work; relevant context for Naibing's sodium-polymer electrolyte systems on the anode side.